Loss of effector T_reg signature in APOB-reactive CD4⁺ T cells in patients with coronary artery disease

Abstract

Atherosclerosis underlies most coronary artery disease (CAD). It involves a significant autoimmune component against apolipoprotein B (APOB). In this study, we used short activation-induced marker (AIM) assays to characterize APOB-reactive CD4⁺ T cells in patients with angiographically verified CAD. APOB-reactive CD4⁺ T cells expressing CD25 and 4–1BB markers were the most abundant. Their frequency correlated positively with CAD severity. Transcriptomic analysis revealed that these cells were clonally expanded and significantly enriched in genes expressed in tissue-homing effector regulatory T (eT_reg) cells. They shared signatures with CD4⁺ T cells in mouse and human plaques, including expression of the plaque-homing chemokine receptor CXCR6. With increasing disease severity, the T_reg signature was progressively and significantly lost. Conversely, APOB-specific T_reg cells from patients with severe CAD gained glycolytic and interferon response signatures. We conclude that mild CAD is associated with a regulatory program in APOB-reactive CD4⁺ T cells, which is replaced by a pro-inflammatory program in patients with severe CAD.

Cardiovascular diseases (CVDs), including coronary artery disease (CAD), are the leading cause of death worldwide. The underlying pathology, atherosclerosis, involves the formation of fibrofatty plaque lesions that obstruct the arterial lumen¹. Plaque rupture leads to catastrophic thrombosis and organ damage. In the late 1980s, two groups reported the presence of immune cells, including activated T cells, in human atherosclerotic plaques^2,3. Since then, many studies have consolidated the pivotal role of immune-mediated inflammation in modulating atherosclerosis⁴.

Recent single-cell RNA sequencing (scRNA-seq) studies identified T cells as the most abundant immune cell subset in human atherosclerotic plaques^5,6. As the central regulators of cellular and humoral immunity, CD4⁺ T cells play a pivotal role in determining disease progression⁴. Although inflammatory T helper 1 (T_H1) cells and effector molecules such as interferon gamma (IFNγ)^7–10 promote atherogenesis, regulatory T (T_reg) cells and anti-inflammatory cytokines such as interleukin (IL)-10 and TGFβ are protective^11–15. In humans, blood T_reg numbers and their functional efficacy were significantly lower in patients with acute coronary syndrome (ACS) than those with stable angina or with normal coronary arteries^16–18. On the other hand, T_H1 and T_H17 subtypes and related cytokines were elevated in patients with ACS^19–21. Another study found no correlation between circulating T_reg cell numbers and extent or severity of atherosclerosis in patients with ACS²².

High oligoclonality and enrichment of activated phenotypes in human plaque-associated CD4⁺ T cells suggest an ongoing local response to specific atherosclerosis-related antigens in the vessel wall^5,6,23. Robust T cell activation and proliferation in response to native and modified low-density lipoprotein (LDL) were detected in human atherosclerotic plaques^24,25 and characterized in humanized mouse models of atherosclerosis^26,27. These studies highlighted that LDL and its core protein apolipoprotein B (APOB), which accumulate in atherosclerotic lesions, are relevant ‘CVD risk-associated’ autoantigens²⁸.

Maladaptive inflammatory conditions in atherosclerosis are associated with a failure of multiple immune regulatory checkpoints, thereby triggering loss of protective T_reg functions²⁹. Although the triggers for this breakdown remain to be identified, the consequences are clear: a vigorous inflammatory T cell response against epitopes in self-antigens, including APOB^30–32, can exacerbate the disease process. Previous work from our laboratory demonstrated that APOB-specific CD4⁺ T cells in the blood of individuals with CVD exhibit increased expression of pro-inflammatory T_H1/T_H17 signatures as compared to non-CVD controls^31,32. Using human leukocyte antigen (HLA) DRB1*07:01-APOB_3036–3050 tetramers, we recently reported a single-cell transcriptomic map of approximately 100 APOB epitope-specific CD4⁺ T cells from CVD cases and controls³³. The transcriptomes of these cells exhibited a mixed phenotype, without clear separation between atheroprotective regulatory and atherogenic inflammatory T cell signatures.

Because tetramers cannot cover the range of disease-relevant autoreactive epitopes in human atherosclerosis, we optimized an activation-induced marker (AIM) assay that enabled simultaneous evaluation of CD4⁺ T cell responses to multiple APOB epitopes in human subjects expressing diverse HLA-II alleles. Short (6–24-h) re-stimulation-based AIM assays with transcriptomic analyses have helped to identify diverse T helper signatures and T cell receptor (TCR) specificities of CD4⁺ T cells responding to allergens³⁴, viral infections³⁵ and autoantigens³⁶.

In the present study, we used the AIM assay to define the transcriptomic heterogeneity of APOB-reactive CD4⁺ T cells against six immunodominant APOB epitopes that are presented by more than 75% of common HLA-II alleles³⁰. We sorted 3 subsets of AIM-expressing APOB-reactive and control non-reactive CD4⁺ T cells from the blood of 40 patients with angiographically graded CAD. Transcriptomic analysis revealed distinct enrichment of conventional (T_conv) and regulatory (T_reg) signatures in clonally expanded APOB-reactive CD4⁺ T cells. The predominant APOB-reactive subset expressed a clear effector T_reg signature. In patients with severe CAD, we discovered a loss of regulatory signatures and a concomitant gain in inflammatory and cytotoxic markers in the APOB-reactive CD4⁺ T_reg cells.

Results

Differential abundance of APOB-reactive CD4⁺ T cells in CAD

In a 24-h AIM assay, we detected significant human CD4⁺ T cell responses to 6 immunodominant HLA class II-restricted epitopes (APOB₆)³⁰. Three kinds of negative controls (peptides from human actin, a pool of weakly antigenic APOB peptides and scrambled versions of immunodominant APOB epitopes) demonstrated specificity³⁰. As expected³³, the frequency of autoreactive APOB-specific CD4⁺ T cells in human peripheral blood mononuclear cells (PBMCs) was low (<<1%). To maximize our chances of capturing responder CD4⁺ T cells, we screened nine combinations of five commonly used T cell activation markers: CD40L, CD69, CD25, 4–1BB and OX-40 (ref. 37). We focused on the top three marker combinations—AIM1 (CD40L⁺CD69⁺), AIM2 (CD25⁺4–1BB⁺) and AIM3 (CD25⁺OX-40⁺)—based on their superior yield and specificity (Extended Data Fig. 1a,b). The robustness and versatility of using a combinatorial approach was validated in a recent study³⁸.

Current cell sorting technology requires a sequential sorting scheme to identify pure populations of AIM1⁺, AIM2⁺ and AIM3⁺ APOB-reactive responder CD4⁺ T cells (Fig. 1a). To understand the differences across the three AIM⁺ groups, we analyzed fluorescence-activated cell sorting (FACS) data (gating strategies in Fig. 1a and Extended Data Fig. 1c) from APOB₆-reactive AIM⁺ cells in 18 patients with angiographically verified CAD (males, age 51–72 years) enrolled in the clinical Coronary Assessment in Virginia (CAVA) cohort³⁰. We grouped the donors by their disease burden, as measured by Gensini scores. The group with low CAD severity (Gensini <20, n = 9) had an average Gensini score of 10.6, whereas the high-severity group (Gensini >20, n = 9) had an average Gensini score of 70.9 (P = 0.003, Mann–Whitney test). No significant difference was observed in other parameters, including demographics, CAD-related medications and lab values between the groups (detailed clinical table provided in ref. 30).

Fig. 1 |. Human APOB-reactive CD4⁺ T cell subsets expressing specific AIM combinations exhibit differential abundance in CAD.

a, Representative FACS plots depicting a serial gating scheme to assess surface expression of CD40L⁺CD69⁺ (AIM1⁺, gated on all CD4⁺ T cells), CD25⁺4–1BB⁺ (AIM2⁺, gated on AIM1⁻CD4⁺ T cells) and CD25⁺OX40⁺ (AIM3⁺, gated on AIM2⁻CD4⁺ T cells) markers on human CD4⁺ T cells in APOB₆-stimulated human PBMCs. CD4⁺ T cells that did not express these combinations are labeled as AIM⁻. Unstimulated cells served as negative controls to set the gates. b, Paired analysis of frequencies of APOB-reactive %AIM2⁺ versus %AIM1⁺ (left) or %AIM3⁺ (right) cells in each donor (n = 18). c, Frequencies (mean ± s.e.m.) of AIM2⁺ cells in CAD^lo (n = 9) versus CAD^hi (n = 9) groups. log₁₀-transformed y axes (b,c); data points with 0 or negative values were collapsed onto the minimum value on the scale. d, Scatter plot with Gensini score (x axis) and frequencies of APOB-reactive AIM2⁺ cells (y axis) in each donor. Simple regression line (blue, solid line) and 95% confidence intervals (black, dashed lines) are shown. e, Representative FACS plots (left) and quantification (right) of frequencies (mean ± s.e.m.) of central (T_CM) and effector (T_EM) memory CD4⁺ T cells among AIM1⁺ (red), AIM2⁺ (green) and AIM3⁺ (blue) subsets (n = 18). Analyses in b–e are based on FACS data related to Roy et al.³⁰. Two-tailed statistical comparisons were performed using Wilcoxon matched-pairs signed-rank test (b), Mann–Whitney U-test (c), Pearson’s correlation test (d) and Kruskal–Wallis test with Dunn’s multiple comparison adjustment (e). T_EMRA, CD45RA-expressing memory T cells.

AIM2 was the predominant APOB-reactive CD4⁺ T group (Extended Data Fig. 1d), with significantly higher frequencies than AIM1⁺ (Fig. 1b, left) and AIM3⁺ (Fig. 1b, right) cells in all but one donor. We observed significantly elevated frequencies of AIM2⁺CD4⁺ T cells (Fig. 1c), and not %AIM1⁺ or %AIM3⁺ cells (Extended Data Fig. 1e), in the high CAD severity group. The percentage of AIM2⁺ cells exhibited a strong positive correlation with Gensini scores (Fig. 1d; Pearsonʼs r = 0.7807; P = 0.0001). Analysis of central (T_CM, CD45RA⁻CCR7⁺) and effector (T_EM, CD45RA⁻CCR7⁻) memory markers on APOB-reactive AIM⁺ subsets revealed that the AIM2 cells were clearly skewed toward a T_EM phenotype (Fig. 1e).

These observations were consistent with previous studies that reported positive correlations between circulating effector memory T cells and CVD in multiple human cohorts^39,40.

Transcriptomes of APOB-reactive CD4⁺ T cells in patients with CAD

To acquire deeper insights into the molecular programs of the APOB-reactive CD4⁺ T cells, we obtained PBMCs from 40 HLA-typed (Supplementary Table 1) patients with CAD from the CAVA cohort (clinical table in Extended Data Fig. 2a). The donors were uniformly distributed into quartiles based on the level of angiographically measured disease severity (Gensini <10, 10–20, 20.5–30 and >30). We stimulated PBMCs from each donor with a pool of APOB₆ peptides for 24 h and generated deep transcriptomes from sorted AIM1⁺CD4⁺, AIM2⁺CD4⁺, AIM3⁺CD4⁺ and control AIM⁻CD4⁺ T cells (input cell numbers in Extended Data Fig. 2b). Input numbers of control AIM⁻CD4⁺ T cells were capped to 1,000 cells to make them comparable to the average input cell numbers of AIM⁺ subsets (Extended Data Fig. 2b). Low-input Smart-seq2 libraries were prepared from 1 ng of cDNA per sample. Most samples yielded high-quality transcriptomes (Extended Data Fig. 2b). After filtering out low-expressed genes, 10,805 protein-coding genes were retained.

These genes were used to run gene set enrichment analysis (GSEA) with mouse single-cell transcriptomes from aortic CD4⁺ T cells and tetramer⁺ APOB-reactive CD4⁺ T cells³¹. Mouse gene sets were converted to human orthologs (Supplementary Table 2). Two CD4⁺ T clusters in atherosclerotic aortas containing APOB-reactive cells (clusters 1 and 2)³¹ showed significant enrichment in human AIM⁺CD4⁺ T cell transcriptomes as compared to the AIM⁻ group (Fig. 2a). Independent comparison of each AIM⁺ subset with the AIM⁻ transcriptomes revealed most significant enrichment of both clusters 1 and 2 in the AIM2 subset (Fig. 2b).

Fig. 2 |. Identification of genes enriched in APOB-reactive AIM⁺ subsets as compared to AIM⁻ control CD4⁺ T cells.

a, Enrichment of signatures (GSE149068 and GSE149069) of APOB-reactive (peptide p6, mouse APOB_978–993) CD4⁺ T cell clusters in atherosclerotic aortas from Apoe^−/− mice within human AIM⁺CD4⁺ versus AIM⁻CD4⁺ T cell comparisons. b, Table showing normalized enrichment scores (NESs) and P values for APOB-enriched mouse aortic clusters 1 and 2 within the AIM1, AIM2 or AIM3 transcriptome versus the AIM⁻ group. c, Volcano plots showing the total number of genes differentially expressed between AIM⁻ and AIM1 (left) or AIM2 (middle) or AIM3 (right) subsets. y axis capped at P = 10⁻¹⁴. Horizontal line at −log₁₀(P value) = 1.3 (representing adjusted P = 0.05). Vertical lines at log₂fold ± 1. Statistical analyses were performed using a two-tailed Wald test with Benjamini–Hochberg P value adjustment. Criteria for significant differentially expressed: log₂fold ± 1, adjusted P < 0.05. d, Venn diagram representing overlap across gene sets that were significantly upregulated in AIM1, AIM2 and AIM3 subsets as compared to the AIM⁻ group. e, GSEA plots of AIM2 (left) or AIM3 (right) versus AIM⁻ transcriptomes showing enrichment of genes expressed at higher levels in tetramer-enriched mouse APOB-reactive (peptide p6, mouse APOB_978–993) versus MCMV m25-reactive CD4⁺ T cells from spleen and lymph nodes under basal conditions (unimmunized and uninfected wild-type mice, GSE263393). Enrichment scores (Fig. 2a,b,e) were calculated using a weighted Kolmogorov–Smirnov-like statistic and phenotype-based permutation test in-built in GSEA.

Pairwise analysis of differentially expressed genes between each AIM⁺ subset versus the control AIM⁻ group identified 448, 716 and 544 upregulated genes and 75, 356 and 177 downregulated genes in AIM1⁺, AIM2⁺ and AIM3⁺ cells compared to AIM⁻ cells, respectively (Fig. 2c and Supplementary Table 3). Venn diagram analysis identified a non-redundant set of 975 upregulated genes, of which 245 (25%) were shared across AIM1⁺CD4⁺, AIM2⁺CD4⁺ and AIM3⁺CD4⁺ T cells (Fig. 2d and Supplementary Table 3). Only nine genes were commonly downregulated in the three subsets (Extended Data Fig. 2c and Supplementary Table 3).

We recently reported⁴¹ gene signatures that were enriched in mouse self-epitope-reactive APOB-tetramer-positive CD4⁺ T cells as compared to murine cytomegalovirus (MCMV)-reactive m25-tetramer-reactive CD4⁺ T cells. The genes (Supplementary Table 2) that were expressed in the naive APOB-tetramer-positive CD4⁺ T cell repertoire⁴¹ showed significant positive enrichment in AIM2 and AIM3 transcriptomes as compared to the control AIM⁻ group (Fig. 2e). AIM1 transcriptomes exhibited no significant enrichment.

Collectively, these results revealed an enrichment of atherosclerosis-related APOB-specific gene signature in human AIM⁺ subsets from patients with CAD. AIM⁺ cells, particularly AIM2⁺ cells, strongly resembled APOB-tetramer-reactive CD4⁺ T cells in mouse aortas.

T_reg cell signatures in AIM2⁺CD4⁺ and AIM3⁺CD4⁺ T cells

Previous studies with mouse³¹ and human³³ tetramer-reactive CD4⁺ T cells showed that the transcriptomes of APOB-reactive CD4⁺ T cells are heterogeneous and express mixed pro-inflammatory and anti-inflammatory gene signatures. Based on this, we reasoned that the responding AIM1⁺CD4⁺, AIM2⁺CD4⁺ and AIM3⁺CD4⁺ T cells may have different phenotypes. Indeed, uniform manifold approximation and projection (UMAP) visualization of the 77 principal components that explained more than 90% of the variance in our dataset (Extended Data Fig. 3a) showed that the AIM1 group (red) formed a distinct cluster that was well separated from AIM2 (green) and AIM3 (blue) subsets, which were intermingled (Fig. 3a). All three AIM⁺ transcriptomes were clearly separated from the control AIM⁻ (purple) group (Fig. 3a). Pairwise differential expression analysis across the three AIM⁺ transcriptomes identified 472 differentially expressed genes between AIM1 and AIM2, 264 genes between AIM1 and AIM3 and 101 genes between AIM3 and AIM2 (Extended Data Fig. 3b and Supplementary Table 4). Taken together, these data show that AIM2 and AIM3 shared overlapping gene programs.

Fig. 3 |. Signature analysis reveals enrichment of conventional T cell genes in APOB-reactive AIM1 transcriptomes, whereas AIM2 and AIM3 are T_reg cell subsets.

a, UMAP visualization of 77 principal components that explain more than 90% of the variance in the bulk RNA-seq data from 37 AIM1 (red), 35 AIM2 (green), 34 AIM3 (blue) and 38 AIM⁻ (purple) transcriptomes. b, Table (left) and radar plots (right) showing enrichment (NES) of top genes differentially expressed in memory regulatory (T_reg) and conventional (T_H1, T_H2, T_FH and T_H17) CD4⁺ T cells within human APOB-reactive AIM1 (red), AIM2 (green) and AIM3 (blue) cells. Signatures were obtained from bulk transcriptomic analysis of sorted CD4⁺ T cells from 91 healthy participants reported in the DICE project. c, GSEA plots showing enrichment of human T_reg gene signatures (GSE25087 in ImmuneSigDB) within AIM2 (left) or AIM3 (right) versus AIM1 transcriptomes. d, GSEA of human T_reg genes (GSE149090) in AIM2 versus AIM3 subsets. e, Median levels of normalized gene expression (TPM) values in AIM1 (red, n = 37), AIM2 (green, n = 35) and AIM3 (blue, n = 34) subsets. Each dot represents an independent sample. y axis: log₁₀ transformed. f–h, Quantification (mean ± s.e.m., right) of FOXP3 (f), HELIOS (g) and CTLA4 (h) protein expression in AIM1 (red), AIM2 (green) and AIM3 (blue) subsets (n = 6). Statistical tests (e–h) were done with the Kruskal–Wallis test with Dunn’s multiple comparison adjustment. Enrichment scores (Fig. 3c,d) were calculated using a weighted Kolmogorov–Smirnov-like statistic and phenotype-based permutation test in-built in GSEA.

We compared enrichment of memory CD4⁺ T helper signatures (Supplementary Table 2) reported in the Database of Immune Cell Expression, eQTLs and Epigenomics (DICE) project⁴². AIM2 and AIM3 transcriptomes specifically resembled memory T_reg cells, whereas AIM1 cells showed enrichment for multiple T_conv (T_H1, T_H2, T follicular helper (T_FH) and T_H17) signatures (Fig. 3b). An unbiased analysis with human immunologic gene sets available in ImmuneSigDB confirmed that both AIM2 and AIM3 transcriptomes are highly enriched in adult T_reg signatures as compared to AIM1 (Fig. 3c). The AIM2⁺ cells exhibited higher enrichment of T_reg genes⁴³ than AIM3⁺ cells (Fig. 3d).

Ranking of the top 100 genes expressed in AIM⁺ cells revealed a clear dichotomy in their enrichment in AIM1 versus AIM2 and AIM3 transcriptomes (Extended Data Fig. 3c). Top-ranked genes in the AIM1 transcriptomes included CD40L, NR4A1, IFNG and GZMB (Fig. 3e, top). Genes that were expressed at significantly higher levels in AIM2 and AIM3 subsets included core T_reg lineage-defining transcription factors and regulatory effector proteins such as FOXP3, IKZF2 (encoding HELIOS), IKZF4 (encoding EOS) and CTLA4 (Fig. 3e, bottom). Flow cytometry (gating strategy and representative FACS plots in Extended Data Fig. 4a–d) confirmed higher protein expression of the T_reg-related markers FOXP3 (Fig. 3f), HELIOS (Fig. 3g) and CTLA4 (Fig. 3h) in AIM2 and AIM3 subsets as compared to AIM1⁺ APOB-reactive CD4⁺ T cells. Expression of these markers in AIM⁻ cells was negligible (Extended Data Fig. 4b–d).

Based on transcriptomic and flow cytometry analysis, we conclude that AIM2 and AIM3 subsets consistently expressed T_reg-related signatures, but AIM1 transcriptomes exhibited a mixed T_conv phenotype.

TCRs from APOB-reactive CD4⁺ T cells show oligoclonality in patients with CAD

The clonal profile of the hypervariable complementarity-determining region 3 (CDR3) regions in TCR sequences reflect the ongoing in vivo status of an antigen-driven T cell response⁴⁴. Our previous analysis of mouse³¹ and human⁴⁵ TCR repertoire revealed oligoclonality within the APOB-specific compartment. Thus, we expected clonal expansion in AIM⁺CD4⁺ compared to AIM⁻CD4⁺ T cells. We extracted TCRβ chain clonotypes from each full-length transcriptomic dataset and analyzed the repertoire of APOB-reactive subsets using MiXCR and immunarch (Supplementary Table 5). To ensure matched comparisons across subsets from the same donor, we considered CDR3β sequences from a subset of 31 donors that yielded transcriptomes from all 3 AIM⁺ and the AIM⁻ groups (Extended Data Fig. 2b). Indeed, AIM⁻ cells predominantly (67%) contained low copy number (2–10) clones (Fig. 4a). By contrast, most (>60%) clones in the APOB-reactive AIM⁺ subsets had more than 10 copies (Fig. 4a). Compared to the other subsets, the AIM2 repertoire had the highest frequency of large clones with copy numbers more than 100 or more than 1,000 (Fig. 4a). Paired comparison of Simpson’s clonality index between each AIM⁺ subset versus the control AIM⁻ group revealed that both AIM1 and AIM2, but not AIM3, exhibited significantly higher oligoclonality than the AIM⁻ repertoire (Fig. 4b).

Fig. 4 |. Clonal expansion of APOB-specific CD4⁺ T cells in CAD.

a, Frequency of CDR3β clones with copy numbers within a specific range (rows) in each specified subset (columns). Combined analysis of sequences from 31 donors. White: lowest; red: highest. b, Simpson’s clonality in AIM⁻ (gray) with matched AIM1 (left, orange), AIM2 (middle, green) and AIM3 (right, blue) repertoire (n = 31 for each subset). c, Pairwise comparisons of CDR3 overlap between AIM2 and AIM3 repertoire versus those between AIM⁻ and AIM3 (right). d, Paired comparison of CDR3 repertoire sharing between AIM2 and AIM3 versus sharing between AIM1 and AIM2 repertoire. In c and d, data are expressed as Jaccard index (n = 31 for each subset) within matched donors. The names of the compared groups are labeled on the x axis. e, Sharing of expanded (>1 copy) APOB-enriched (not detected in AIM⁻) CDR3 clones across the 3 AIM groups. Statistical tests (b–d) were performed using two-tailed Wilcoxon matched-pairs signed-rank test.

To examine overall sharing of TCR sequences among the three AIM⁺ groups, we measured the extent of repertoire overlap (expressed as Jaccard index, range 0–1). As AIM2 and AIM3 transcriptomes strongly resemble each other, we first analyzed whether there was any overlap in their CDR3 repertoire. Sharing of CDR3 clones between AIM2 with AIM3 was significantly higher than the overlap of either subset with non-APOB-reactive AIM⁻ control TCRs (Fig. 4c). The TCR repertoire of AIM2⁺ cells (CD25⁺4–1BB⁺ T_reg cells) showed significantly higher overlap with AIM3⁺ cells (CD25⁺OX-40⁺ T_reg cells) than with AIM1⁺ cells (CD40L⁺CD69⁺ T_conv cells) (Fig. 4d). This is in line with previous reports of antigen-specific TCRs that showed largely non-overlapping repertoire from human CD40L⁺ T_conv and 4–1BB⁺ T_reg cells³⁴.

As expected, most of the expanded (present at >1 copy number) APOB-reactive (AIM⁺) clones were not detected in the control (AIM⁻) repertoire (Extended Data Fig. 5). Those that overlapped were filtered out to focus on APOB-enriched clones (expanded exclusively in AIM⁺). In total, 110 clones were shared between AIM2 and AIM3 subsets from 24 donors (Fig. 4e). In contrast, only 10 clones were shared between AIM1 and AIM2 repertoires from 9 donors (Fig. 4e). Only two APOB-enriched clones were detected in all three subsets from two donors (Fig. 4e).

Thus, APOB-specific autoreactive CD4⁺ T cells are clonally expanded in patients with angiographically verified CAD. We detected clonal expansion in both T_conv and T_reg subsets, with higher frequency of large clones in AIM2⁺ T_reg cells. Significantly higher repertoire overlap was observed between AIM2⁺CD4⁺ and AIM3⁺CD4⁺ T cells than between other groups.

Tissue-homing chemokine receptors in APOB-reactive T_reg cells

Evaluation of AIM⁺ transcriptomes revealed that AIM2 and AIM3 shared a common T_reg-related gene program. We focused on AIM2 transcriptomes because this was the predominant APOB-reactive CD4⁺ T subset in patients with CAD (Fig. 1b); AIM2 frequencies correlated strongly and significantly with Gensini scores (Fig. 1d); and T_reg signatures were more pronounced in AIM2 compared to AIM3 (Fig. 3d).

For further analysis of APOB-reactive cells, we first contrasted APOB-reactive T_reg (AIM2) and T_conv (AIM1) transcriptomes using signatures from magnetically enriched 4–1BB⁺ (mostly T_reg) and CD40L⁺ (mostly T_conv) CD4⁺ T cells reactive to an airborne fungal allergen³⁴. Expression of this signature from human antigen-specific T cells was uniformly low in our control non-APOB-reactive cells (Fig. 5a and Supplementary Table 6). Expectedly, APOB-reactive AIM1 cells expressed T_conv genes such as pro-inflammatory cytokines, whereas AIM2 cells expressed T_reg-related genes and transcription factors. However, the AIM2 transcriptomes were also enriched in multiple tissue-homing chemokine receptors (Fig. 5a). It is known that the trafficking pattern of T_reg cells between lymphoid and non-lymphoid organs changes during an ongoing immune response⁴⁶. Recent thymic emigrants express CCR7 and CXCR4 receptors that allow them to patrol lymphoid tissues⁴⁷. Upon antigen priming, these receptors are downregulated, and a new array of tissue-homing receptors, including CCR4 and CCR8, is expressed⁴⁷. Expression of these chemokine receptors was reported on mouse and human T_reg cells^48–50. They allow T_reg cells to migrate to inflamed organs for immune suppression and tissue repair⁵¹. CCR4 expression was recently reported in T_reg cells from human atherosclerotic plaques⁶. Previous work showed that CCR5 (ref. 52) and CXCR6 (ref. 53) are involved in T cell homing to atherosclerotic plaques in the Apoe^−/− mouse model of atherosclerosis. Expression of CCR8 and CCR4 was significantly highest in AIM2⁺ APOB-reactive T_reg cells (Fig. 5b, left). The atherosclerosis-related chemokine receptors CCR5 and CXCR6 were also more highly expressed in APOB-reactive T_reg cells (Fig. 5b, middle). By contrast, expression of CXCR4, a homeostatic chemokine receptor⁵⁴, was downregulated in APOB-reactive T_reg cells (Fig. 5b, right) compared to non-APOB-specific CD4⁺ T cells.

Fig. 5 |. APOB-reactive T_reg cells from patients with CAD express tissue-homing chemokine receptors.

a, Heatmap showing expression patterns of genes enriched in antigen-specific CD4⁺ T cells (from GSE77081) in APOB-reactive T_conv (AIM1, n = 37), T_reg (AIM2, n = 35) and non-APOB-reactive control (AIM⁻, n = 38) samples. b, Median expression (TPM) of chemokine receptor genes related to tissue T_reg cells (CCR8 and CCR4), atherosclerotic plaque-homing T cells (CCR5 and CXCR6) and the constitutively expressed homeostatic control receptor (CXCR4). APOB-reactive T_conv (AIM1, n = 37) and T_reg (AIM2, n = 35), non-APOB-reactive control (AIM⁻, n = 38). y axis: log₁₀ transformed. c, Normalized expression (TPM, mean ± s.e.m.) of Ccr4, Ccr8, Cxcr6 and Cxcr4 genes in sorted T_conv and T_reg CD4⁺ T cells from the spleen (n = 5 for T_conv and T_reg) and lymph nodes (n = 5 for T_conv; n = 3 for T_reg) of atherosclerotic T_reg lineage tracker mice (GSE217010). d, Bar graphs showing log₂ fold change (x axis) and adjusted P value (right y axis) for genes (left y axis) differentially expressed between mouse APOB(p6)-tetramer-reactive and MCMV(m25)-tetramer-reactive CD4⁺ T cells from spleen and lymph nodes under basal conditions (GSE263393). e, Table showing Cxcr6 gene expression (fold change and percent positive cells) in mouse CD4⁺ T cells within APOB-tetramer-positive cluster 1 in atherosclerotic aortas (GSE149068 and GSE149069). f, Representative FACS plots (left) and quantification (mean ± s.e.m., right) comparing expression of CXCR6 protein marker on APOB-reactive T_reg subset (green) and on all CD25⁺CD127^lo T_reg cells (black) in human PBMCs (n = 6). Non-APOB-reactive (AIM⁻) cells and fluorescence minus one (FMO) were used as negative controls. Statistical tests were done using the Kruskal–Wallis test with Dunn’s multiple comparison adjustment (b), two-way ANOVA with Sidakʼs multiple comparisons test (c), two-tailed Wald test with Benjamini–Hochberg correction (d), non-parametric two-tailed Wilcoxon rank-sum test (e) and two-tailed Mann–Whitney U-test (f). FSC-H, forward scatter height; LN, lymph node; M, million.

Next, we analyzed chemokine receptor expression in a published dataset (GSE217010) of sorted T_conv and T_reg cells from the spleen and lymph nodes of T_reg lineage tracker Apoe^−/− mice⁵⁵. In both spleen and lymph nodes, expression of Ccr4 and Ccr8 was significantly higher in mouse T_reg cells than in T_conv cells (Fig. 5c). Significantly enhanced expression of Cxcr6 on T_reg cells was observed specifically in the lymph node compartment (Fig. 5c). No significant difference was observed in Cxcr4 expression between T_reg and T_conv cells (Fig. 5c).

In a published dataset (GSE263393) of mouse tetramer⁺ APOB-positive versus virus-specific CD4⁺ T cells from spleen and lymph nodes⁴¹, we observed significant upregulation of Cxcr6, Ccr4 and Ccr8 genes and downregulation of the Cxcr4 gene in Foxp3 and Il2ra expressing T_reg-like APOB-reactive CD4⁺ T cells (Fig. 5d). Cxcr6, which was most highly upregulated in these transcriptomes from mouse APOB-tetramer-reactive cells (Fig. 5d), was also specifically enriched (Fig. 5e) in an APOB-tetramer-reactive CD4⁺ T cell cluster that was abundant in the atherosclerotic aortas from Western diet-fed Apoe^−/− mice³¹. FACS analysis (gating strategy in Extended Data Fig. 6a) confirmed significant upregulation of CXCR6 protein expression on APOB-reactive T_reg cells with negligible expression on total T_reg cells in human PBMCs (Fig. 5f). Expression of the other plaque-homing chemokine receptor CCR5 was higher in APOB-reactive than in total T_reg cells but detectable in both groups (Extended Data Fig. 6b).

Taken together, these data suggest that APOB-reactive T_reg cells detected in the peripheral blood of patients with CAD express chemokine receptors CCR4 and CCR8, which are characteristic of tissue-homing memory T_reg cells, and CCR5 and CXCR6, which can mediate their migration to atherosclerotic plaques.

Plaque effector T_reg markers in blood APOB-specific T_reg cells

As expression of the plaque-homing T cell chemokine receptor CXCR6 was enriched in human APOB-reactive AIM2⁺ T_reg cells (Fig. 5b,f) and in mouse APOB-tetramer-positive aortic CD4⁺ T cells (Fig. 5e), we analyzed CXCR6 expression in the single-cell transcriptomes from 3 blood and 12 coronary artery plaque (GSE196943) samples²³. Comparison of CD4⁺ T cells from blood and plaque tissue (Fig. 6a, left) revealed significantly increased expression of CXCR6 in the plaque samples (Fig. 6a, right). A recent scRNA-seq study⁵⁶ of T cells from different tissue sites in atherosclerotic Apoe^−/− mice showed that expression of the Cxcr6 gene is particularly enhanced in plaque T cells compared to their clonotypic counterparts in other tissues, such as secondary and tertiary lymph nodes. Other plaque T-cell-associated genes such as S100A4, S100A6 and REEP5, reported in that study⁵⁶, also exhibited enhanced expression in our CXCR6-expressing human APOB-reactive AIM2⁺ T_reg subset (Fig. 6b).

Fig. 6 |. APOB-reactive T_reg cells are enriched in eT_reg signatures.

a, Left: UMAP representation of blood (red, n = 3) and plaque (blue, n = 12) CD4⁺ T cells filtered from human single-cell transcriptomic dataset (GSE196943). Right: feature plot and bar graph (mean ± s.e.m.) showing CXCR6 expression in CD4⁺ T cells. b,c, Median expression levels (TPM) of mouse atherosclerotic plaque-associated genes (b) and human plaque-associated eT_reg genes (c) in APOB-reactive T_conv (AIM1, n = 37), T_reg (AIM2, n = 35) and non-APOB-reactive control (AIM⁻, n = 38) samples. d, Circulating human CD4⁺ T cell clusters accessed at the Broad Institute’s SingleCellPortal (SCP1963). FOXP3-expressing T_reg clusters are marked with a red boundary. Effector (turquoise), activated (gray) and naive (purple) T_reg subsets identified in the study are shown. e, UMAP visualization of HLA-DRA, HLA-DRB1 and HLA-DRB5 in T_reg subsets. Color scale: yellow (lowest) to blue (highest) expression. f, Quantification (mean ± s.e.m.) of FACS-based assessment of surface-expressed TIGIT (top) and HLA-DR (bottom) proteins on APOB-reactive T_reg subset (green) and all CD25⁺CD127^lo T_reg cells (black) in human PBMCs (n = 5). g, Heatmap showing expression of a shared signature of cancer and autoimmune disease-related human eT_reg genes (GSE161426) in APOB-reactive T_reg (AIM2) and non-APOB-reactive (AIM⁻) transcriptomes. h, FACS plots depicting PD-1 expression on human AIM2⁺ eT_reg (left) and total T_reg cells (right) in two independent donors. Statistical comparisons were done using two-tailed Mann–Whitney U-test (a,f) and Kruskal–Wallis test with Dunn’s adjustment for multiple comparisons (b,c).

Recent studies have highlighted that memory T_reg cells are not homogeneous and exhibit distinct adaptations depending on the tissue involved and the nature of the inflammatory response, including a phenotypically distinct effector T_reg (eT_reg) lineage^57,58. A recent scRNA-seq study reported the presence of eT_reg cells in human coronary plaques²³. We observed significantly higher expression of these human plaque-related eT_reg signature genes such as RTKN2, CORO1B and LTB in our AIM2⁺ APOB-reactive T_reg transcriptomes (Fig. 6c).

The human plaque scRNA-seq study²³ also identified HLA-DRA gene expression in activated plaque T cells²³. Interestingly, an HLA-DR⁺ effector memory subpopulation in the blood was reported to exhibit the strongest association with atherosclerotic disease in patients with chronic stable angina or acute myocardial infarction compared to controls³⁹. Recent scRNA-seq and assay for transposase-accessible chromatin using sequencing (ATAC–seq) studies of human T_reg cells observed upregulation of HLA-DR genes in eT_reg lineages⁵⁹ expressing chemokine receptors such as CCR8 (ref. 49). In our dataset, APOB-reactive AIM2⁺ T_reg cells expressed general T_reg markers such as TIGIT as well as HLA-DRA, HLA-DRB1 and HLA-DRB5 genes that encode HLA-DR α and β chains (Extended Data Fig. 7a). To confirm that HLA-DR expression is upregulated on tissue-homing eT_reg cells and not on in vitro activated T_reg cells, we compared expression of TIGIT and HLA-DR genes in a publicly available scRNA-seq dataset wherein distinct FOXP3⁺ subclusters of naive, activated and eT_reg cells (Fig. 6d) were clearly annotated among circulating human CD4⁺ T cells⁵⁰. The eT_reg subpopulation was marked by high expression of chemokine receptors such as CCR4, CCR8 and CXCR6 (Extended Data Fig. 7b). We found that TIGIT was uniformly expressed in all T_reg cells (Extended Data Fig. 7c), but high expression of HLA-DRA, HLA-DRB1 and HLA-DRB5 genes was restricted to the eT_reg cluster (Fig. 6e and Extended Data Fig. 7d). FACS analysis (gating strategy in Extended Data Fig. 6a) validated that surface expression of HLA-DR protein was significantly higher on APOB-reactive T_reg cells than on total T_reg cells, but TIGIT expression was similar on both (Fig. 6f and Extended Data Fig. 7e).

Next, we examined the expression of a common human eT_reg gene signature (Supplementary Table 6) reported to be upregulated in synovial fluid T_reg cells from patients with juvenile idiopathic arthritis or rheumatoid arthritis and in tumor-infiltrating T_reg cells⁶⁰. This signature, which was highly expressed in APOB-reactive AIM2⁺ T_reg cells, as compared to the AIM⁻ control (Fig. 6g), was enriched in pathways related to T_reg function, such as IL-2 signaling and TGFβ regulation of extracellular matrix, as well as those implicated in promoting T_reg instability²⁹, such as mTORC1, IL-6/JAK/STAT3 and IL-12 signaling (Extended Data Fig. 7f).

Human eT_reg cells at inflamed tissues, such as the tumor microenvironment, have been shown to express the inhibitory molecule PD-1, which limits their function. Anti-PD-1 therapy unleashes them, resulting in poor outcome in patients with malignant disease with high frequencies of PD-1^hi eT_reg cells in the tumor^61,62. The effect is opposite in autoimmune diseases. T_reg-specific deletion of PD-1 in mouse models of multiple sclerosis and type 1 diabetes resulted in superior suppressive activity and attenuated disease⁶³. We recently reported that high PD-1 expression on mouse APOB-reactive CD4⁺ T cells renders them anergic, which was rescued by pharmacological blockade with PD-1 inhibitors⁴¹. In this study, we detected PD-1 protein expression on 10–14% of the APOB-reactive AIM2⁺ eT_reg cells, with negligible expression on bulk T_reg cells (Fig. 6h and Extended Data Fig. 7g; gating strategy in Extended Data Fig. 6a). This highlights the opportunity to target the autoimmune aspect of atherosclerosis via specific inhibition of inhibitory PD-1 signals on autoreactive eT_reg cells using precise tools such as bispecific antibodies.

We conclude that APOB-reactive T_reg cells represent an activated eT_reg subset that exhibit signs of adaptation to an inflamed microenvironment at atherosclerotic plaque tissues.

Loss of APOB-specific regulatory program in severe CAD

Inflammation and antigen priming upregulates FOXP3 and reinforces the suppressive capacity of eT_reg cells to induce tolerance and repair^57,64. However, tissue-homing eT_reg cells have been shown to be more susceptible to inflammation-driven instability than central T_reg cells that recirculate between blood and lymphoid tissues⁶⁵. We reasoned that gene expression patterns in the eT_reg-like AIM2⁺ APOB-reactive T_reg cells may change with increasing Gensini score, a measure of CAD severity.

First, we analyzed a gene signature reported in human APOB-tetramer-positive CD4⁺ T cells that were clonally related to T_reg cells, as evidenced by TCRβ sharing³³. In our AIM2⁺ APOB-reactive T_reg transcriptomes, median expression of most of these genes was uniformly low in patients with Gensini >30 (Fig. 7a). Core T_reg functional genes such as FOXP3, IKZF2, IL2RA, IL10RA and TIGIT (highlighted in red) were more highly expressed in patients with Gensini scores between 10 and 30 but decreased in patients with Gensini scores above 30. GSEA with resting and activated T_reg signatures reported in the DICE database^42,66 showed that resting T_reg signatures were higher in mild disease (Gensini <20) and were lost in patients with Gensini >20 (Fig. 7b). Activated T_reg signatures peaked in the group with intermediate CAD severity (Gensini 20.5–30) but declined sharply and significantly in the high-severity group (Fig. 7b and Extended Data Fig. 8a). This suggests that mild inflammatory conditions associated with less severe CAD likely promotes activation of APOB-reactive eT_reg cells, which is compromised in severe CAD. Analysis of 168 human memory T_reg signatures from DICE⁴², which were also significantly expressed in our APOB-reactive AIM2⁺ T_reg transcriptomes (Supplementary Tables 3 and 7), led to the identification of 87 genes that were downregulated in the group with severe CAD (Extended Data Fig. 8b).

Fig. 7 |. APOB-reactive eT_reg cells lose their regulatory program under conditions of severe atherosclerosis.

a, Heatmap showing expression (median TPM in APOB-reactive T_reg transcriptomes grouped by Gensini scores) of genes significantly downregulated in human APOB-reactive cells that shared the same TCRβ clonotypes with T_reg cells. b, Radar plot (top) and table (bottom) showing enrichment of resting (black line) and activated (red line) T_reg signatures from DICE in APOB-reactive T_reg transcriptomes grouped by Gensini scores. c, Top 3 hallmark pathways significantly enriched among 50 highest-ranked genes that were upregulated in APOB-reactive T_reg transcriptomes from patients with CAD with severe (Gensini >30) versus mild (Gensini 10–20) disease. Dotted line at −log₁₀ adjusted P = 1.3. d, Heatmap showing median expression of inflammatory and cytotoxic genes in APOB-reactive T_reg transcriptomes grouped by Gensini scores. e, Median normalized expression of IFNG in APOB-reactive AIM2⁺ T_reg transcriptomes from patients grouped by CAD severity (n = 9, <10; n = 9, 10–20; n = 7, 20.5–30; n = 10, >30). y axis: log₁₀ transformed. Data with zero values were not plotted. f, Enrichment of hallmark pathways oxidative phosphorylation and IL2/STAT5 signaling in APOB-reactive T_reg transcriptomes from patients with CAD with low severity (Gensini <10) compared to groups with higher severity (Gensini ≥10). Statistical comparisons were done using the two-tailed Fisher’s exact test and Benjamini–Hochberg adjustment (c). Enrichment scores (Fig. 7b,f) were calculated using a weighted Kolmogorov–Smirnov-like statistic and phenotype-based permutation test in-built in GSEA. MSigDB, Molecular Signatures Database.

We observed positive enrichment of inflammatory T_H1 and T_H17 signatures in APOB-reactive AIM2⁺ T_reg transcriptomes, specifically in the high-severity group (Gensini >30; Extended Data Fig. 8c). Top-ranked genes expressed in APOB-reactive T_reg transcriptomes from patients with severe CAD were enriched in pathways related to glycolysis and IFNα and IFNγ responses (Fig. 7c), hallmarks of unstable and plastic T_reg cells^64,67. In addition to inflammatory genes such as IFNG, expression of cytotoxic genes such as GZMB and CCL4 was also upregulated in APOB-reactive T_reg transcriptomes from patients with severe CAD (Fig. 7d,e).

Therapeutic strategies to boost T_reg cells using low-dose IL-2 (ref. 68) or tolerogenic vaccines⁶⁹ are currently the focus of intense research to treat or prevent CVD in humans. In a recent clinical trial, low-dose IL-2 successfully boosted T_reg numbers in patients with athero sclerotic cardiovascular disease⁷⁰. Single-cell transcriptomic analyses showed that IL-2 treatment⁷⁰ induced pathways related to T_reg function and stability⁷⁰, such as oxidative phosphorylation and IL2/ STAT5 signaling^67,71. Here, we found that these pathways were significantly enriched in APOB-reactive T_reg transcriptomes from patients with low CAD severity (Gensini <10) but lost in more severe CAD (Gensini >10) (Fig. 7f).

Taken together, these data provide evidence that the perturbation in the homeostasis of autoreactive T_reg cells in CAD is driven by a compromised functional program of the plaque-homing AIM2⁺ effector T_reg subset. In patients with severe CAD, the T_reg gene program found in AIM2⁺ APOB-reactive CD4⁺ T cells from those with mild disease was replaced with an inflammatory and cytotoxic signature, reminiscent of unstable T_reg cells and exT_reg cells⁵⁵.

Discussion

It is now established that atherosclerosis involves breach of self-tolerance and increased autoimmune activity^6,29,56, with a clinically relevant T cell response to APOB-derived epitopes³⁰. However, information available regarding the molecular phenotypes of APOB-specific CD4⁺ T cells in human atherosclerosis is limited³³. Here we provide a detailed report of deep transcriptomic signatures of atherosclerosis-related APOB-reactive CD4⁺ T cells in blood samples from 40 patients with CAD with angiographically graded low to high disease severity (Gensini score 0–128).

Using a short (24-h) restimulation-based AIM assay, we tested 9 combinations of APOB peptide-induced T cell activation markers and chose 3—CD40L⁺CD69⁺ (AIM1), CD25⁺4–1BB⁺ (AIM2) and CD25⁺OX-40⁺ (AIM3)—based on cell yield and specificity. FACS-based analysis of the AIM⁺ subsets in the blood of patients with CAD identified AIM2 as the predominant subset whose frequencies correlated strongly and significantly with Gensini scores. We observed preferential expression of effector memory markers on AIM2⁺ cells, which is consistent with previous studies that reported strong correlation between T_EM subsets and CVD status in humans^39,40.

The AIM2 transcriptomes showed enrichment for T_reg, and specifically effector T_reg, signatures. Based on analysis of TCR sequences, the AIM2 repertoire was significantly oligoclonal and enriched in large clones (>100 copies). Deeper analysis of AIM2⁺ transcriptomes using antigen-specific T_reg signatures³⁴ highlighted an enrichment of tissue-homing chemokine receptors CCR4, CCR8, CCR5 and CXCR6 in AIM2⁺ APOB-reactive T_reg cells. CXCR6 gene expression was also high in datasets of CD4⁺ T cells from mouse atherosclerotic aortas³¹ and human coronary plaques²³. In mice, CCR5 (ref. 52) and CXCR6 (ref. 53) mediate T cell homing to atherosclerotic lesions. Flow cytometry analysis confirmed that APOB-reactive AIM2⁺ T_reg cells express higher levels of these receptor proteins as compared to total T_reg cells in the blood. This is in line with previous studies that suggested that a majority of the circulating T_reg cells patrol lymphoid organs⁴⁶. A minor fraction of antigen-experienced effector T_reg cells express a variety of molecules that allow them to migrate to tissue sites⁵⁷. This is critical for their ability to suppress ongoing inflammation⁵¹. Indeed, AIM2 transcriptomes highly expressed mouse⁵⁶ and human²³ plaque-associated T cell genes. They were enriched in a common human eT_reg signature⁶⁰, known to affect T_reg function and stability in an inflamed tissue microenvironment. They expressed the inhibitory checkpoint molecule PD-1, which has been shown to limit suppressive activity in mouse⁶³ and human^61,62 T_reg cells. Taken together, these data suggest that AIM2⁺ cells represent an APOB-reactive recirculating eT_reg subset that can likely visit atherosclerotic plaques. Their activity can be boosted with anti-PD-1 therapies.

Compared to lymphoid-organ-homing T_reg cells, tissue-homing eT_reg cells are more susceptible to dynamic changes in the tissue microenvironment⁶⁴. Although inflammation initially reinforces the suppressive program in eT_reg cells, prolonged and unresolved inflammatory conditions can compromise it⁶⁵. In the present study, we observed that the expression of several T_reg signature genes initially increased in patients with mild disease but were significantly downregulated in the severe CAD group. This included downregulation of the T_reg lineage-defining transcription factor FOXP3, activation markers such as TNFRSF18 (encoding GITR), functional genes such as the ATP-degrading enzyme ENTPD1 (encoding CD39), T_reg signaling-related cytokine receptors such as IL2RB, the atherosclerosis-related plaque-homing chemokine receptor CXCR6 and more recently reported human T_reg markers⁷² such as LAYN (encoding Layilin). This reflected a progressive loss in the regulatory effector program in APOB-reactive T_reg cells in CAD. In contrast, they gained the expression of pathways related to T_reg instability, such as glycolysis and type I and II interferon responses. They expressed inflammatory and cytotoxic effector genes such as TNF, IFNG, GZMB and CCL4 in patients with severe CAD, resembling a newly discovered atherosclerosis-related cytotoxic human exT_reg subset⁵⁵. In addition, metabolic and signaling pathways reported to be boosted by targeted T_reg therapies such as low-dose IL-2 treatment⁷⁰ were enriched in AIM2⁺ transcriptomes from patients with mild CAD but were lost in more severe CAD. It remains to be seen whether these changes can serve as surrogate biomarkers to assess the efficacy of therapies that aim to restore T_reg homeostasis in human CVD.

A limitation of this study is that the instability observed in APOB-reactive T_reg cells from patients with CAD with high disease severity could not be examined using functional assays due to the low numbers of autoreactive CD4⁺ T cells in human blood (~0.1–0.3% of CD4⁺ T cells). The volume of available patient PBMC samples is limited to 3–5 million total cells per vial, yielding only a few hundred APOB-reactive CD4⁺ T_reg cells. Another limitation is that we focused on one atherosclerosis-related autoantigen, APOB. Recent studies discovered new autoantigens in human atherosclerosis by focusing on autoreactive antibodies⁷³. The assays reported in this study can be used to identify immunodominant epitopes in other atherosclerosis-relevant self-proteins. This will facilitate comprehensive evaluation of the molecular programs in multiple classes of autoreactive CD4⁺ T cells that modulate the inflammatory response in human atherosclerosis.

In conclusion, we found that the AIM2⁺ APOB-reactive CD4⁺ T cells, by far the largest group of APOB-reactive cells, show gene and protein expression signatures consistent with eT_reg cells. Their frequencies in blood positively correlate with the severity of human CAD. Their oligoclonal TCR repertoire suggests antigen-driven expansion in patients with CAD. Their homing receptor expression pattern suggests that they may be able to traffic between blood and plaque tissue. In patients with severe CAD, APOB-reactive T_reg cells showed downregulation of regulatory genes and pathways and, instead, upregulated inflammatory and cytotoxic genes.

Methods

Human subjects

We obtained de-identified cryopreserved PBMCs from patients with CAD undergoing standard cardiac catheterization at the cardiac catheterization laboratory at the University of Virginia Health System in Charlottesville, Virginia (CAVA cohort). Written informed consent was obtained from all participants before enrollment. Blood samples were collected prior to cardiac catheterization. Quantitative coronary angiography was performed using automatic edge detection from an end-diastolic frame, which was selected for each lesion, based on demonstration of the most severe stenosis with minimal foreshortening and branch overlap. The minimum lumen diameter, reference diameter, percent diameter stenosis and stenosis length were calculated by blinded, experienced investigators who assessed disease severity based on the Gensini score. In brief, each artery segment was assigned a score of 0–32 based on the percent stenosis. For each segment, this score was multiplied by 0.5–5, depending on the location of the stenosis. Scores for all segments were then added together to give a final score of angiographic disease burden. Score adjustment for collateral was not performed for this study. De-identified records about patient demographics, medications and lab values were made available to us. Patients with a range of CAD severity (Gensini scores 0–128) were selected for the study.

PBMCs from healthy volunteers, used in some flow cytometry experiments, were recruited by the Clinical Core at the La Jolla Institute for Immunology (LJI). Both male and female donors older than 18 years of age were eligible to participate in the study. Donors self-reported ethnicity and race details and tested negative for hepatitis B, hepatitis C and HIV. None of the donors had any ongoing infection. They had no known conditions of cancer, diabetes, heart or kidney or liver disease. Donors were neither pregnant nor nursing. All participants received financial compensation according to guidelines approved by LJI’s institutional review board. Written informed consent was obtained from all participants before enrollment. The study was approved by the Human Institutional Review Board (IRB number 15328) at the University of Virginia and by the LJI institutional review board (numbers VD-057 and IB-248–0821).

PBMC isolation by density gradient centrifugation

Venous blood samples from donors recruited through LJI’s Clinical Core were collected in K2-EDTA-coated tubes. The tubes were centrifuged at 800g for 15 min at room temperature with brakes off. The top layer of plasma was removed, and an equivalent amount of 1× PBS (without Ca/Mg; Gibco) was added and resuspended thoroughly. Seven parts of diluted blood were gently layered on 3 parts of Ficoll-Paque PLUS (Millipore Sigma) and centrifuged at 800g for 30 min at room temperature with brakes off. The layer of cells at the interface was carefully harvested and washed twice with 1× PBS. Cell counts and viability were evaluated using the trypan blue dye exclusion method. Cells were cultured in serum-free cell culture medium (Miltenyi Biotec, TexMACS) supplemented with 1% penicillin–streptomycin (Thermo Fisher Scientific).

Peptides

Six human HLA-II-restricted APOB-derived peptides (APOB₆₇₆ TLTAFGFASADLIEI; APOB₈₈₁ VEFVTNMGIIIPDFA; APOB₁₂₂₆ VGSKLIVAMSSWLQK; APOB₂₄₉₁ LIINWLQEALSSASL; APOB₂₈₀₁ LEVLNFDFQANAQLS; APOB₄₂₄₁ ILFSYFQDLVITLPF) were synthesized at more than 95% purity (TC Peptide Lab). Peptides were resuspended in dimethyl sulfoxide (DMSO; Sigma Aldrich) and combined in equal proportions to prepare APOB₆ pool.

AIM assay and cell sorting

Cryopreserved PBMCs from patients with CAD (CAVA cohort) were thawed at 37 °C and washed once in 1× PBS (without Ca/Mg) by centrifuging at 400g for 10 min at 24 °C. Cell numbers were determined using a hemocytometer, and viability was examined using the trypan blue dye exclusion method. An aliquot of the PBMCs was used to isolate genomic DNA using a REPLI-g DNA Midi Kit (Qiagen) and HLA typed using services provided by an American Society for Histocompatibility and Immunogenetics (ASHI)-accredited laboratory at the Institute for Immunology & Infectious Diseases, Murdoch University, Western Australia. Remaining PBMCs were resuspended in TexMACS medium and plated at a density of 1.5 × 10⁶ cells per well in 96-well plates. A CD40 blocking antibody (Novus Biologicals, Centennial) was added at a final concentration of 1 μg ml⁻¹. After 15 min of incubation at 37 °C with 5% CO₂, APOB₆ peptide pool was added at a dose of 20 μg ml⁻¹ per peptide. Cells were incubated for 24 h at 37 °C with 5% CO₂. After the incubation period, stimulated PBMCs were washed with FACS buffer (1× PBS without Ca/Mg, 2% FBS) and resuspended in a staining master mix containing anti-human Fc-Block (BioLegend), fixable viability dye (1:1,000) and antibodies against T cell and non-T cell (Dump) surface markers (1:200) and activation markers (1:100). Dump channel markers included anti-hCD8a-APC-Cy7 (clone RPA-T8), anti-hCD14-APC-Cy7 (clone M5E2), anti-hCD16-APC-Cy7 (clone 3G8), anti-hCD19-APC-Cy7 (clone HIB19) and anti-hCD56-APC-Cy7 (clone HCD56). T cell markers included anti-hCD3-PerCp-Cy5.5 (clone UCHT1) and anti-hCD4-Pacific Blue (clone RPA-T4). T cell activation markers included anti-hCD40L-PE (clone 24–31), anti-hCD69 BV650 (clone FN50), anti-hCD25-PE-Cy7 (clone BC96), anti-hOX-40-APC (clone Ber-ACT35) and anti-h4–1BB-BV605 (clone 4B4–1). Cells were stained for 45 min at 4 °C and then washed with cold FACS buffer (PBS without Ca/Mg, 2% FBS). Viability dye was used at 1:1,000 dilution; antibodies against non-T cell (Dump) and T cell lineage markers were used at a final dilution of 1:200; and antibodies against T cell activation markers were used at 1:100 dilution. Cells were analyzed and sorted using a FACSAria II (BD Biosciences). Voltages were set up using single color-stained cells and compensation beads (UltraComp eBeads; Invitrogen). Gates for activation markers were set using unstimulated controls. AIM1⁺CD4⁺, AIM2⁺CD4⁺, AIM3⁺CD4⁺ and AIM⁻CD4⁺ T cells from APOB₆-stimulated PBMCs were collected in 0.2-ml RNase-free tubes (Axygen) containing low-input lysis buffer (LI-LB) with 0.1% Triton X-100 (v/v), 1 U μl⁻¹ RNase inhibitor (Takara) and 2.5 mM dNTPs (Thermo Fisher Scientific). The number of AIM⁻ cells was capped to make them comparable to AIM⁺ subsets (input cell numbers for cDNA synthesis in Extended Data Fig. 2b). After collection, tubes were promptly closed, vortexed for 30 s and centrifuged (3,000g, 2 min, 4 °C) to ensure that all cells were collected in the lysis buffer.

Low-input bulk RNA-seq

For cDNA synthesis and amplification, the Smart-seq2 protocol was used. In brief, poly-dT (5′-AAG CAG TGG TAT CAA CGC AGA GTA CT(30) VN-3′; Integrated DNA Technologies) and template-switch oligos (5′-AAG CAG TGG TAT CAA CGC AGA GTA CAT rGrG+G-3′; Qiagen) were used for mRNA capture and generation of full-length double-stranded cDNA using a SuperScript II Reverse Transcriptase Kit (Thermo Fisher Scientific). The cDNA was amplified using KAPA HiFi HotStart ReadyMix buffer and ISPCR oligo (5′-AAG CAG TGG TAT CAA CGC AGA GT-3′; Integrated DNA Technologies) and purified with AMPure XP magnetic bead (Beckman Coulter). Quality and quantity of cDNA were assessed with TapeStation (Agilent) and Qubit (Invitrogen). For each sample, 1 ng of cDNA was used to prepare dual-index barcoded libraries using a Nextera XT DNA library prep kit and index kits (Illumina). Libraries were purified with AMPure XP magnetic beads. Good-quality libraries, as examined with TapeStation and Qubit, were retained for sequencing. Samples from low-range, mid-range and high-range Gensini scores were processed together to minimize batch effects. Equimolar concentrations of uniquely indexed libraries were pooled and sequenced using an S4 flow cell on a NovaSeq 6000 (Illumina) to obtain 70 million 100-bp paired-end reads per cluster.

Transcriptomic and TCR analyses

FASTQ files were generated with bcl2fastq (Illumina). Sequencing quality control was performed with FastQC version 0.11.9 and MultiQC version 1.12. Adapter trimming and filtering of low-quality reads were done with DRAGEN FASTQ Toolkit version 1.0.0. (Illumina). The STAR aligner (version 2.7.10 with default parameters) was used to map the reads onto the human GENCODE reference GRCh38.p13. Raw read counts from STAR aligner were used for differential gene expression analysis using DESeq2 (version 1.34). Quality of read counts was assessed before normalization with DESeq2. Variance stabilizing transformed (VST) counts were used to visualize results from principal component analysis. Genes with ≤10 counts in more than 75% of samples were filtered out. Volcano plots to visualize differentially expressed genes were generated using R (version 4.0.1).

Scaled heatmaps of normalized gene expression values (transcripts per million (TPM)) and Venn diagrams were made using tools available at https://www.bioinformatics.com.cn/en, a free online platform for data analysis and visualization. GSEA was done with GSEA version 4.3.2 using gene sets in Human ImmuneSigDB, HALLMARK genes and from curated T-cell-specific published gene sets. Mouse gene sets were converted to human orthologs in R (version 12.0). For TCR analysis, CDR3 sequences were extracted from RNA-seq data using MiXCR version 4.0. Clonality and repertoire overlap analyses were performed using immunarch 0.7.0. Pathway analysis was done with the online tool Enrichr (https://maayanlab.cloud/Enrichr/, webserver 2016 update). Human circulating T_reg clusters were analyzed using the Broad Institute’s interactive SingleCellPortal (https://singlecell.broadinstitute.org/single_cell/study/SCP1963). Pie charts, scatter plots, columns and bar graphs for data visualization were generated using GraphPad Prism (version 10.0.0). Radar plots were generated using ggplot2 version 3.5.1.

Flow cytometry

For FACS analysis of T_reg markers, human PBMCs were first stimulated using the AIM assay. After the 24-h incubation period, cells were stained with fixable viability dye (1:1,000) and surface marker antibodies (Dump, T cell lineage and activation markers) according to the protocol mentioned previously in the AIM assay section. Antibodies against non-T cell (Dump) and T cell lineage markers were used at a final dilution of 1:200, and antibodies against T cell activation markers were used at 1:100 dilution. Additionally, fluorochrome-conjugated anti-hCD127 (clone eBioDR5; Invitrogen), anti-hTIGIT (clone MBSA43; Invitrogen), anti-hHLA-DR (clone L243; BioLegend) antibodies (1:100) and anti-hPD-1 (clone EH12.2H7; BioLegend) antibodies (1:100) were added. Cells were incubated for 45 min at 4 °C. For intracellular staining, cells were fixed and permeabilized with Foxp3/Transcription Factor Staining Buffer Set (eBioscience) for 20 min at 24 °C. Cells were washed with 1× Perm Buffer (eBioscience) and incubated with fluorochrome-labeled antibodies (1:50) against hFOXP3 (clone 206D; BioLegend), hHELIOS (clone 22F6; BioLegend) and hCTLA4 (clone BNI3; BioLegend) for 45 min at 24 °C. Cells were washed with 1× Perm Buffer and analyzed on an LSR II (BD Biosciences) or a Cytek Aurora (Cytek Biosciences).

For FACS analysis of chemokine receptors, human PBMCs were stimulated using the AIM assay. Fluorochrome-conjugated anti-hCCR5 (clone J418F1; BioLegend) and anti-hCXCR6 (clone K041E5; BioLegend) were added to each well (1:100 dilution, v/v) at the start of the incubation period. After 24 h, cells were washed with cold FACS buffer (PBS without Ca/Mg, 2% FBS) and stained according to the protocol mentioned previously in the AIM assay section. Antibodies against chemokine receptors were not added again. Cells were washed with cold FACS buffer. Data were collected and analyzed on the LSR II (FACSDiva; BD Biosciences) or the Cytek Aurora (Cytek Biosciences). Single color-stained beads (UltraComp eBeads; Invitrogen) were used for compensation. Data were analyzed with FlowJo (version 10.8.1).

Processing and analysis of human scRNA-seq data from blood and plaque

The publicly available dataset (GSE196943) was processed using Seurat (version 5) in R. Cells with fewer than 300 detected genes, RNA counts exceeding 20,000 or mitochondrial transcript percentages above 10% were excluded to remove low-quality cells. Doublet detection was performed using scDblFinder. Each Seurat object was converted into a SingleCellExperiment (SCE) object, and doublets were identified and removed based on classification scores. Filtered data underwent normalization using SCTransform, regressing out mitochondrial effects. Principal component analysis and ElbowPlot determined the optimal number of dimensions for downstream clustering. A shared nearest neighbor graph was constructed using the FindNeighbors() function, followed by clustering with FindClusters() at a resolution of 0.5. To correct batch effects across samples, an integration workflow was implemented. SelectIntegrationFeatures() identified highly variable genes across samples, and FindIntegrationAnchors() computed batch correction anchors. Samples were integrated using IntegrateData(). Cell types were assigned with SingleR using the HumanPrimaryCellAtlas reference. CD4⁺ T cells were identified based on marker expression (CD3E⁺, CD4⁺, CD8A⁻, CD19⁻, CD14⁻) and embedded on a UMAP. Expression of CXCR6 was visualized using FeaturePlot(). Sample-based expression of CXCR6 was analyzed using AverageExpression().

Statistical analyses

Data analysis and statistical comparisons were done using GraphPad Prism version 10.0.0 and R version 4.0.1. We used a two-tailed Wald test with Benjamini–Hochberg P value adjustment for comparing differential expression of genes. Two-sample comparisons were done with non-parametric two-tailed Wilcoxon matched-pairs signed-rank test (for paired comparisons) or Mann–Whitney U-test (for unpaired comparisons). We used the Kruskal–Wallis test with Dunn’s multiple comparison testing for analyses involving more than two samples. All statistical tests, sample sizes, P values and error bar descriptions are detailed in the legends or respective figures.

Extended Data

Extended Data Fig. 1 |. Flow cytometry-based evaluation of activation marker expression in APOB-reactive human CD4⁺T cells.

a-b) Human PBMCs (n = 18 independent donors) were stimulated with APOB₆ peptide pool and expression of different activation-induced marker (AIM) combinations in stimulated vs unstimulated PBMCs were compared. a) Mean ± SEM yield per million CD4⁺T cells (numbers in stimulated minus those in unstimulated sets) for nine AIM combinations are shown. b) Average (mean ± SEM) fold change (frequency in stimulated divided by that in unstimulated) for CD40L + CD69 +, CD25 + 4–1BB +, CD25 + OX-40+ and CD69 + OX-40 + AIM combinations are shown. c-e) Analysis of APOB-reactive CD4⁺T cells in CAD patients (n = 18) related to Fig. 1. c) Gating strategy in flow cytometry to identify AIM1,2,3 and AIM− subsets shown in Fig. 1a. d) Pie-chart showing relative abundance of each APOB-reactive CD4⁺T subset (AIM1, 2 and 3) within total AIM+ cells. Average frequencies were plotted to calculate percent abundance. e) Frequencies (mean ± SEM) of AIM1+ (left) and AIM3+ (right) cells in CAD^lo (Gensini<20, n = 9, mean Gensini score 10.6; SD ± 8.3) vs CAD^hi groups (Gensini>20, n = 9, mean Gensini 70.9; SD ± 28.7). Log₁₀ transformed Y-axes; data points with 0 or negative values were collapsed onto the minimum value on the scale. Statistical tests done with two-tailed Mann-Whitney U test (e).

Extended Data Fig. 2 |. Gene expression in APOB-reactive and control transcriptomes from CAD patients.

a) Clinical table summarizing details related to demographics, medications, lab values and disease severity of patients in the CAD cohort used for transcriptomic analyses. Categorical variables are plotted as counts and percentages, while continuous variables are shown as mean ± SEM within the cohort (sample size = 40 patients). b) Table showing donor IDs (as used in the CAVA cohort), CAD severity scores (Gensini), and input cell numbers for cDNA preparation for each APOB-reactive (AIM1,2,3) and control (AIM−) libraries that were sequenced. Samples labelled as “Dropout” were excluded due to poor quality and yield of cDNA or library. All samples from donor #472 were excluded due to low ( < 50%) viability of PBMCs from this donor. Median viability of PBMCs from other donors was 93.1% (Min-max 80.34 – 98.1, interquartile range 4.7). c) Venn diagram depicting overlap across genes downregulated in AIM1, 2 and 3 subsets as compared to the AIM− group.

Extended Data Fig. 3 |. Analysis of gene expression across three APOB-reactive subsets.

a) Elbow plot analysis to identify the number of PCs that contribute to >90% of the variance in the transcriptomes. b) Volcano plots showing number of DE genes between AIM2 and AIM1 (left), AIM3 and AIM1 (middle), AIM3 and AIM2 (right). Y axis capped at p = 10⁻¹⁴. Horizontal line at −log₁₀ (p-value) = 1.3 (representing adjusted p-value 0.05). Vertical lines at log₂fold ± 1. Statistical analyses were performed using a two-tailed Wald test with Benjamini–Hochberg p-value adjustment. Criteria for significant DE: log₂fold ± 1, adjusted p-value < 0.05. Red: upregulated; Blue: downregulated; Grey dots: not significantly different. c) Heatmap of APOB-enriched genes ranked based on their enrichment in AIM1 (top 50) or AIM2&3 (bottom 50) transcriptomes.

Extended Data Fig. 4 |. Flow cytometry analysis of Treg-related markers in three APOB-reactive subsets.

a) Gating strategy in flow cytometry to identify AIM1,2,3 and AIM− subsets. b-d) Representative FACS plots showing FOXP3 (b), HELIOS (c) and CTLA4 (d) protein expression in AIM1, AIM2, AIM3 and AIM− subsets.

Extended Data Fig. 5 |. Clonality analysis of APOB-reactive TCRβ CDR3 repertoire.

Table showing the number of expanded CDR3 clones from APOB-reactive subsets (AIM1–3 and total AIM + ) within each donor. The final column shows the number of AIM+ clones that were also detected in the control AIM− CDR3 repertoire from the same donor.

Extended Data Fig. 6 |. Chemokine receptor expression in APOB-reactive Tregs and all Tregs.

a) Gating strategy in flow cytometry to identify AIM2+ Tregs, total Tregs and Non-APOB (AIM−) subsets. b) Representative FACS plots (left) and quantification (mean ± SEM, right) showing expression of CCR5 protein marker on APOB-reactive Treg subset (green) and on all CD25⁺CD127^lo Treg cells (black) in human PBMCs (n = 6). Non-APOB reactive (AIM−) cells and Fluorescence minus one (FMO) were used as negative controls. Statistical comparisons (b) were done using two-tailed Mann–Whitney U test.

Extended Data Fig. 7 |. APOB-reactive Tregs are enriched in effector Treg markers.

a) Median expression levels (TPMs) of general and effector Treg-related genes in APOB-reactive Tconv (AIM1, n = 37), Treg (AIM2, n = 35), and non-APOB reactive control (AIM−, n = 38) samples. b-d) Gene expression analysis in circulating human Treg subsets accessed at Broad Institute’s SingleCellPortal (SCP1963). UMAP visualization showing CCR4, CCR8, CXCR6 (b) and TIGIT (c) in Treg subsets. Color scale: yellow (lowest) to blue (highest) expression. Red outline marks FOXP3 expressing Treg cells. d) Violin plots of HLA-DRA, HLA-DRB1, and HLA-DRB5 genes (n = 13 subjects). e) FACS plots showing surface expression of TIGIT (left) and HLA-DR (right) proteins on APOB-reactive Treg subset and on all CD25⁺CD127^lo Treg cells in human PBMCs. f) Significantly enriched pathways related to effector Treg genes described in Fig. 6g. Dotted line at −log₁₀ adjusted p = 1.3. g) FACS plots showing FMO control for PD-1 expression on human AIM2 (left) and total Treg (right) cells. Gating strategy to identify AIM2+ Tregs and total Tregs (e,g) shown in ED Fig. 6a. Statistical comparisons were done using the Kruskal-Wallis test with Dunn’s adjustment (a) and two-tailed Fisher’s exact test and Benjamini-Hochberg adjustment (f).

Extended Data Fig. 8 |. Dynamics of Treg gene expression in APOB-reactive Treg transcriptomes from patients with varying CAD severity.

a) GSEA plot showing negative enrichment of activated Treg signature in APOB-reactive Treg transcriptomes from patients with highest CAD severity (Gensini >30) compared to those with less severe CAD (Gensini <30). Enrichment score was calculated using a weighted Kolmogorov–Smirnov-like statistic and phenotype-based permutation test in-built in GSEA. b) Heatmaps showing expression (median TPMs) of Treg genes from DICE in AIM2 transcriptomes from donors grouped by CAD severity. c) Table showing NES values for Th1/Th17 signatures in APOB-reactive Treg transcriptomes from CAD patients with Gensini 10–20, 20.5–30 and >30.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s44161-025-00671-9.

Data availability

The raw sequencing data are publicly available at the National Center for Biotechnology Informationʼs Gene Expression Omnibus under accession code GSE279783. All other processed data are available in the main text or Supplementary Information. No new material resources were generated in this study. GSEA was done using gene sets in Human ImmuneSigDB (https://www.gsea-msigdb.org/gsea/msigdb/human/genesets.jsp?collection=C7), HALLMARK genes (https://www.gsea-msigdb.org/gsea/msigdb/human/genesets.jsp?collection=H) and from curated T-cell-specific published gene sets (GSE149068, GSE149069, GSE263393 and GSE149090). Human circulating T_reg clusters were analyzed using the Broad Institute’s interactive SingleCellPortal (https://singlecell.broadinstitute.org/single_cell/study/SCP1963). Heatmaps were generated based on data in GSE77081 and GSE161426. For analyzing chemokine receptor expression in human coronary plaque T cells, the publicly available dataset (GSE196943) was used.

Code availability

No new algorithms were generated for this study.