CD8⁺HLA-DR⁺CD27⁺ T cells define a population of naturally occurring regulatory precursors in humans

Abstract

Regulatory T_reg cells are essential for immune homeostasis. While CD4 T_reg cells are well characterized, CD8 T_reg cells remain less understood and are primarily observed in pathological or experimental contexts. Here, we identify a naturally occurring CD8 regulatory precursor T_rp cell at the steady state, defined by a CD8⁺HLA-DR⁺CD27⁺ phenotype and a transcriptome resembling CD4 T_reg cells. Multiomics analyses reveal activation of TCF7 and costimulatory and co-inhibitory molecules in CD8 T_rp cells. CD8 T_rp cells suppress T cell expansion in vitro and in vivo. In a humanized xenogeneic graft-versus-host disease (GVHD) model, they dampen T cell activation, alleviate GVHD pathology, and prolong survival without impairing antileukemia activity. Mechanistically, CD8 T_rp cells promote immune regulation by inducing FOXP3 expression in both CD4 T_reg cells and themselves. Their expansion also correlates with immune homeostasis restoration post–allogeneic stem cell transplantation. These findings establish CD8 T_rp cells as a naturally occurring regulatory precursor population that promotes transplantation tolerance.

Naturally occurring human CD8⁺ regulatory precursor T cells contribute to maintaining peripheral immune homeostasis.

INTRODUCTION

Regulatory T cells (T_reg cells) are essential for maintaining immune homeostasis by establishing tolerance to self-antigens and controlling the magnitude of the adaptive immune response to non–self-antigens, thereby preventing damage to the host. CD4 T_reg cells have been extensively characterized, defined by their specific immune phenotype (CD4⁺CD25⁺FOXP3⁺) and well-documented developmental trajectory. On the basis of their origin, T_reg cells are classified as thymic T_reg cells, also known as natural T_reg cells, which develop in the thymus and recognize self-antigens, or peripheral T_reg cells, also called induced T_reg cells, which differentiate from the conventional T cell (T_con cell) pool in peripheral tissues (1).

CD8 T_reg cells exhibit substantial phenotypic heterogeneity and have been described in various pathophysiological contexts, including cancer, infectious diseases, autoimmunity, organ transplantation, and graft-versus-host disease (GVHD) (2). The major subsets of CD8 T_reg cells include the following: (i) Qa-1–restricted CD8⁺ T cells and KIR⁺ human CD8⁺ T cells (with Ly49 being the murine counterpart), which suppress autoreactive CD4⁺ T cells implicated in autoimmune disorders; (ii) forkhead box P3 (FOXP3)–expressing CD8 T_reg cells, which can be induced by factors such as transforming growth factor–β (TGF-β); and (iii) tissue-resident CD8 T_reg cells, such as intestinal unconventional T cells expressing the CD8 α homodimer (CD8αα⁺ T cells) (2–4). Although these studies have confirmed the existence of CD8⁺ T cells with immunoregulatory properties, the specific surface markers, along with their transcriptomic and epigenomic profiles, have yet to be fully elucidated for CD8⁺ T_reg cells. Furthermore, given that FOXP3 expression is minimal in CD8⁺ T cells within the thymus, secondary lymphoid organs, and nonlymphoid tissues under the steady state (5), numerous studies have focused on generating CD8 T_reg cells by inducing FOXP3 expression in CD8⁺ T cells, either manually or under pathological conditions such as GVHD (6–9). However, it is still unclear whether natural CD8 T_reg cells, similar to CD4 T_reg cells, exist at the steady state.

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) serves not only as a curative therapy for hematological malignancies but also as a strategy to remodel the immune system into a homeostatic state. This approach has been used to induce allograft tolerance in solid organ transplantation and to treat autoimmune diseases (10–12). In our previous study, we delineated the transcriptomic atlas of the immune system in recipients who successfully restored immune homeostasis after allo-HSCT (13). Using single-cell RNA sequencing (RNA-seq) analysis, we characterized a population of CD8⁺ regulatory precursors, termed T_rp cells, which contribute to remodeling immune homeostasis in patients following allo-HSCT. It is suggested that this cluster might represent a potential subset of T_reg cells within the CD8⁺ T cell population.

In the current study, we comprehensively characterized the immune phenotype of CD8 T_rp cells and validated their immunosuppressive function both in vitro and in vivo. We identified the immune phenotype marker of CD8 T_rp cells as CD8⁺HLA-DR⁺CD27⁺. Further analysis revealed that CD8⁺HLA-DR⁺CD27⁺ T cells exhibit distinct transcriptomic signatures compared to classical CD8⁺ T cell subsets while displaying expression patterns similar to those of CD4 T_reg cells. Functionally, CD8⁺HLA-DR⁺CD27⁺ T cells showed robust immunosuppressive capabilities both in vitro and in vivo, effectively suppressing the proliferation and activation of both CD4⁺ and CD8⁺ T cells. This study introduces a promising approach for treating autoimmune disorders and achieving peripheral immune homeostasis in transplantation.

RESULTS

Identification of the CD8 T_rp cell immune phenotype as CD8⁺HLA-DR⁺CD27⁺ in the homeostatic state

In our previous study, we identified a distinct subtype of CD8⁺ T cells that was expanded in recipients who achieved immune homeostasis following allo-HSCT compared to their paired donors (13). Single-cell analysis showed that the gene expression profile of this CD8 T cell cluster closely resembles that of CD4 T_reg cells and is positioned upstream of CD8 effector memory T cells (T_em cells) in trajectory analysis. Therefore, we named this cluster as CD8 regulatory precursors (CD8 T_rp cells). To further identify the immune phenotype of CD8 T_rp cells, we collected peripheral blood mononuclear cells (PBMCs) from healthy donors (n = 10), as well as from homeostatic recipients following haploidentical stem cell transplantation (Haplo-SCT; n = 12) or matched sibling donor transplantation (MSDT; n = 9). These samples were subjected to analysis using a 16-color spectrum flow cytometry panel (Fig. 1A). First, we analyzed the expression levels of phenotype markers using a manual gating strategy (fig. S1A). Our previous single-cell analysis indicated that CD8 T_rp cells are characterized by the high expression of major histocompatibility complex (MHC) class II genes, such as HLA-DRB1. Flow cytometry analysis further demonstrated that the CD8⁺HLA-DR⁻ subset was enriched with naïve T cells (T_naive cells), while the CD8⁺HLA-DR⁺ subset contained more T_em cells and terminal effector memory T cells (T_emra cells) (Fig. 1B). As single-cell trajectory analysis suggested that CD8 T_rp cells exhibit a precursor-like signature, additional surface markers were required to distinguish T_em cells from T_emra cells. We found that CD27 effectively distinguished these subsets, with CD8⁺CD27⁻ T cells being enriched for T_emra cells and CD8⁺CD27⁺ T cells being enriched for T_em cells (Fig. 1C). By combining human leucocyte antigen DR (HLA-DR) and CD27 markers, we identified the CD8⁺HLA-DR⁺CD27⁺ subset as predominantly enriched with T_em cells in both healthy individuals and homeostatic recipients of Haplo-SCT and MSDT (Fig. 1D and fig. S1B). Simultaneously, we performed unsupervised clustering to reduce the dimensionality of the 16-parameter dataset (Fig. 1, A and E). Focusing specifically on CD8⁺ T cells, we extracted flow cytometry data for CD8⁺ T cells from all samples for further clustering (Fig. 1F). Our analysis identified Cluster18 as corresponding to the CD8 T_rp cell phenotype, characterized by the high expression of MHC class II markers such as HLA-DR and HLA-DQ, as well as inhibitory markers such as programmed cell death 1 (PD-1) and T cell immunoreceptor with Ig (immunoglobulin) and ITIM ( immunoreceptor tyrosine-based inhibitory motif) domains (TIGIT) (Fig. 1, G and H). Cluster18 also exhibited high levels of CD27 and a lack of CD45RA expression (Fig. 1, G and H). Together, these findings confirm the natural presence of CD8 T_rp cells in the homeostatic state at the protein level and define their immune phenotype as CD8⁺HLA-DR⁺CD27⁺.

Fig. 1. Immunophenotype characterization of CD8⁺ regulatory precursor cells in the homeostatic state.

(A) Schematic illustration of the experimental workflow. Comprehensive immunological analysis is performed on PBMCs obtained from healthy individuals (n = 10) and homeostatic recipients following Haplo-SCT (n = 12) or MSDT (n = 9). (B) Flow cytometry analysis of CD8⁺ T cell subsets including T_naive (T_n), T_cm, T_em, and T_emra cell distribution within the CD8⁺HLA-DR⁺ and CD8⁺HLA-DR⁻ populations from the healthy individual group, Haplo-SCT recipient group, and MSDT recipient group. (C) Flow cytometry analysis of CD8⁺ T cell subsets including T_naive, T_cm, T_em, and T_emra cell distribution within the CD8⁺CD27⁺ and CD8⁺CD27⁻ populations from the healthy individual group, Haplo-SCT recipient group, and MSDT recipient group. (D) Distribution of CD8⁺ T cell subsets including T_naive, T_cm, T_em, and T_emra cells within the four quadrants extinguished by HLA-DR and CD27 in CD8⁺ T cells. Representative graph from the Haplo-SCT recipient group. (E) UMAP visualization of spectrum flow cytometry–detected CD3⁺ T cells from the healthy individual group, Haplo-SCT recipient group, and MSDT recipient group. (F) UMAP visualization of spectrum flow cytometry–detected CD8⁺ T cells from the healthy individual group, Haplo-SCT recipient group, and MSDT recipient group. (G and H) Heatmaps of the expression level of indicated markers in CD8⁺ T cells from the healthy individual group, Haplo-SCT recipient group, and MSDT recipient group. Unpaired t test for comparison between two groups. Error bars represent the means ± SEM. **P < 0.01, ***P < 0.001, ****P < 0.0001.

CD8⁺HLA-DR⁺CD27⁺ T cells display a distinct expression profile

To further characterize the expression signature of CD8⁺HLA-DR⁺CD27⁺ T cells, we analyzed the protein level expression profile of the four quadrants distinguished by HLA-DR and CD27 (fig. S1A). CD8⁺HLA-DR⁺CD27⁺ T cells exhibited high levels of MHC class II markers such as HLA-DQ, inhibitory markers including PD-1 and TIGIT, and activation markers such as CD28 and CD69 (Fig. 2, A to C). This expression profile is consistently observed in both healthy individuals and homeostatic recipients following allo-HSCT (Fig. 2, B and C), suggesting the stable presence of CD8 T_rp cells in the homeostatic state. To further investigate the transcriptomic signature of CD8⁺HLA-DR⁺CD27⁺ T cells, we sorted CD8⁺HLA-DR⁺CD27⁺ and CD8⁺HLA-DR⁻CD27⁻ T cells from healthy individuals for RNA-seq. Classical T cell subsets, including T_naive cells, central memory T cells (T_cm cells), T_em cells, and T_emra cells, were also sorted as controls (fig. S2, A and B). Principal components analysis (PCA) showed that CD8⁺HLA-DR⁺CD27⁺ T cells are closely related to CD8 T_em cells, consistent with the flow cytometry results and single-cell analysis (Fig. 2D). Compared to CD8 T_naive cells, CD8⁺HLA-DR⁺CD27⁺ T cells highly express inhibitory genes such as PDCD1, TIGIT, LAG3, and CD160; cytotoxic genes including IFNG, KLRG1, and GZMB; and T cell tolerance–related transcription factors like PRDM1 and TOX (Fig. 2E). To further evaluate the transcriptomic similarity between CD8⁺HLA-DR⁺CD27⁺ cells and the CD8 T_rp cell cluster identified in our previous single-cell results, we performed an integrated analysis of single-cell RNA-seq data and bulk RNA-seq from sorted T cell subsets. The results showed that CD8⁺ T cells coexpressing HLA-DRB1/HLA-DRA and CD27 are predominantly localized within the CD8 T_rp cell cluster in single-cell data (fig. S3), and the CD8⁺HLA-DR⁺CD27⁺ subset exhibited the highest enrichment of the CD8 T_rp cell–specific gene signature (Fig. 2F and table S1). Furthermore, CD8⁺HLA-DR⁺CD27⁺ T cells exhibited a transcriptional profile similar to that of CD4 T_reg cells, characterized by consistent up-regulation of costimulatory molecules (CD28 and ICOS), inhibitory markers (TIGIT and TOX), and key CD4 T_reg cell–associated genes (TGFB1 and IKZF2) (Fig. 2, G and H). These shared features suggest that CD8⁺HLA-DR⁺CD27⁺ T cells represent a regulatory subset within the CD8⁺ T cell population. In addition, CD8⁺HLA-DR⁺CD27⁺ T cells display high levels of TCF7, reflecting their precursor-like characteristics (Fig. 2I). The expression level of CD28 on CD8⁺HLA-DR⁺CD27⁺ T cells is relatively high (Fig. 2I), suggesting that these cells differ from CD8⁺CD28⁻ T_reg cells, a heterogeneous population that encompasses exhausted and senescent T cells (14, 15). Compared to classical CD8 T cell subsets, CD8⁺HLA-DR⁺CD27⁺ T cells exhibit a distinct transcriptional profile characterized by the high expression of TCF7, costimulatory molecules such as CD28 and ICOS, co-inhibitory molecules including PD-1 and TIGIT, and transcription factors NR4A2 and TOX2 (Fig. 2I). These results suggest that CD8⁺HLA-DR⁺CD27⁺ T cells have both precursor and regulatory signatures at the transcriptomic level.

Fig. 2. Protein and transcriptomic expression patterns of CD8⁺HLA-DR⁺CD27⁺ T cells.

(A and B) Representative histogram (A) and summary of median fluorescence intensity (MFI) (B) showing protein expression levels for the indicated markers of HLA-DR⁺CD27⁺, HLA-DR⁻CD27⁺, HLA-DR⁻CD27⁻, and HLA-DR⁺CD27⁻ within CD8⁺ T cells from the healthy individual group (n = 10), Haplo-SCT recipient group (n = 12), and MSDT recipient group (n = 9). (C) Heatmap of protein levels of CD8⁺HLA-DR⁺CD27⁺ and CD8⁺HLA-DR⁻CD27⁻ T cells from the indicated groups. (D) Principal components analysis (PCA) of classic CD8⁺ T cell subsets, including T_naive, T_cm, T_em, and T_emra cells, along with the CD8⁺HLA-DR⁺CD27⁺ and CD8⁺HLA-DR⁻CD27⁻ subsets. (E) Volcano plot showing the differentially expressed genes (DEGs) between CD8⁺HLA-DR⁺CD27⁺ T cells and CD8 T_naive cells. (F) The top 50 CD8 T_rp cell signature genes, identified from single-cell data in our previous study, were extracted, and their enrichment scores across different groups were calculated using the GSVA package. (G) Heatmap showing the expression levels of selected genes across different groups. (H) Gene enrichment in CD8⁺HLA-DR⁺CD27⁺ T cells and CD4 T_reg cells relative to CD8 and CD4 T_naive cells, respectively. (I) Transcripts per million (TPM) of the indicated genes across different groups. CD8 T_rp, CD8⁺HLA-DR⁺CD27⁺; CD8 T_naive, CD8⁺CCR7⁺CD45RA⁺; CD8 T_cm, CD8⁺CCR7⁺CD45RA⁻; CD8 T_em, CD8⁺CCR7⁻CD45RA⁻; CD8 T_emra, CD8⁺CCR7⁻CD45RA⁺; CD4 T_reg, CD4⁺CD25⁺CD127^low; CD4 T_naive, CD4⁺CD25^lowCCR7⁺CD45RA⁺. A one-way ANOVA was applied for comparisons among four groups. For the RNA-seq data of sorted T cell subsets from healthy individuals, classical subsets were sorted from four donors, and CD8⁺HLA-DR⁺CD27⁺ and CD8⁺HLA-DR⁻CD27⁻ subsets were sorted from five donors. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

CD8⁺HLA-DR⁺CD27⁺ T cells exhibit immunosuppressive functions both in vitro and in vivo

To validate whether CD8⁺HLA-DR⁺CD27⁺ T cells have immunosuppressive functions, we isolated these cells from healthy individuals and cocultured them with PBMCs activated by anti-CD3 and anti-CD28 antibodies in vitro. The presence of CD8⁺HLA-DR⁺CD27⁺ T cells significantly inhibited the proliferation of PBMCs and CD8⁺ T cells (Fig. 3A). These findings suggest that CD8⁺HLA-DR⁺CD27⁺ T cells have immunosuppressive functions.

Fig. 3. CD8 T_rp cells suppress T cell proliferation and activation in vitro and in vivo.

(A) Proliferation analysis of PBMCs with or without coculture with CD8⁺HLA-DR⁺CD27⁺ T cells (n = 8). Combined data from two independent experiments. (B) Schematic illustration of the experimental workflow for in vivo MLR. T_regs, T_reg cells; T_rps, T_rp cells. (C) Representative flow cytometry graph and statistical analysis of hCD45⁺ cell engraftment in humanized mice from the peripheral blood (PB) and spleen (SP). (D) Absolute number of hCD45⁺ cells in the spleen of humanized mice. (E) Representative flow cytometry graph showing FOXP3 expression in CD4⁺ T cells and statistical analysis of the proportion and absolute number of CD4 T_reg cells in the spleen of humanized mice from the indicated groups. (F) Representative flow cytometry graph and statistical analysis of cytokine secretion levels in CD4⁺ or CD8⁺ T cells of humanized mice from the indicated groups. A paired t test was applied for comparison between two groups, and a one-way ANOVA was applied for comparisons among four groups. Each group included five mice. The data shown represent one experiment of three independent experiments. Symbols indicate individual mice, and error bars represent the means ± SEM. *P < 0.05, **P < 0.01.

We further assessed the immunosuppressive function using an in vivo mixed lymphocyte reaction (MLR) model. PBMCs from HLA-A02:01–negative healthy individuals were transfused into MHC-knockout NSG mice (Prkdc^scidIL2rg^−/−H2-K1^−/−H2-Ab1^−/−H2-D1^−/−), and engraftment and GVHD clinical scores were dynamically monitored. On day 14, mice that showed successful engraftment [human CD45-positive (hCD45⁺) cell ratio over 20%] without signs of GVHD were randomly assigned to one of four groups and challenged with the following: (i) autologous PBMCs (same HLA-A02:01–negative donor); (ii) allogeneic PBMCs (HLA-A02:01–positive donor); (iii) allogeneic PBMCs and CD4 T_reg cells from the same HLA-A02:01–negative donor; and (iv) allogeneic PBMCs and CD8⁺HLA-DR⁺CD27⁺ T cells (CD8 T_rp cells) from the same HLA-A02:01–negative donor (Fig. 3B). The results indicated that cotransfusion with CD8 T_rp cells significantly inhibited the expansion of human CD45⁺ cells (Fig. 3, C and D). CD4 T_reg cells were expanded not only in the group receiving allogeneic PBMCs combined with CD4 T_reg cells but also in the group receiving allogeneic PBMCs combined with CD8 T_rp cells (Fig. 3E). In addition, FOXP3 expression levels were elevated in both CD4⁺ and CD8⁺ T cells in the group receiving allogeneic PBMCs combined with CD8 T_rp cells (Fig. 3E and fig. S4, A and B). These findings suggest that CD8 T_rp cells promote the expansion of CD4 T_reg cells and induce the differentiation of CD8⁺FOXP3⁺ cells. Moreover, CD8 T_rp cells slightly inhibited the secretion of inflammatory cytokines including interleukin-2 (IL-2) and tumor necrosis factor–α (TNF-α) by both CD4⁺ and CD8⁺ T cells (Fig. 3F). Collectively, these results indicate that CD8 T_rp cells exhibit immunosuppressive functions both in vitro and in vivo.

CD8⁺HLA-DR⁺CD27⁺ CD8 T cells mitigate GVHD without compromising GVL activity in a humanized mouse model

For further evaluation of the in vivo inhibitory effects of CD8⁺HLA-DR⁺CD27⁺ T cells, we established a humanized xenogeneic GVHD (x-GVHD) model by transplanting 5 × 10⁶ PBMCs alone (group 1), 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD4 T_reg cells [group 2; effector:target (E:T) = 1:10], or 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD8 T_rp cells (group 3; E:T = 1:10) into NSG mice (Fig. 4A). Cotransfusion with either CD4 T_reg cells or CD8 T_rp cells mitigated GVHD damage and prolonged the survival of NSG mice (Fig. 4B). Mice receiving PBMCs combined with CD8 T_rp cells exhibited reduced lymphocyte infiltration and decreased GVHD-related tissue damage (fig. S5). Moreover, CD8 T_rp cells inhibited the expansion of human CD45⁺ cells, CD4⁺ T cells, and CD8⁺ T cells (Fig. 4C). Consistent with the in vivo MLR results, CD8 T_rp cells promoted the expansion of CD4 T_reg cells, with a more pronounced effect observed in the spleen compared to the CD4 T_reg cell cotransfusion group (Fig. 4D). CD8 T_rp cells also suppressed the activation of CD4⁺ and CD8⁺ T cells, as indicated by the decreased expression levels of activation markers CD25 and CD38 (Fig. 4E). Furthermore, CD8 T_rp cells reduced the secretion of inflammatory cytokines, including IL-2 and TNF-α (Fig. 4F). Collectively, these findings demonstrate that CD8 T_rp cells effectively mitigate GVHD damage and immune activation in vivo.

Fig. 4. CD8 T_rp cells inhibit GVHD damage in humanized xenogeneic GVHD models.

(A) Schematic illustration of the experimental workflow for the humanized xenogeneic GVHD (x-GVHD) model. (B) Survival curve, weight changes, and GVHD scores of humanized x-GVHD models. Twenty-two mice for group 1, 22 mice for group 2, and 19 mice for group 3. Median survival: group 1, 21 days; group 2, 30 days; group 3, 31 days. Combined data from three independent experiments. (C) Representative flow cytometry graph and statistical analysis of hCD45⁺ cell engraftment in x-GVHD mice from the peripheral blood (PB) and spleen (SP). (D) Representative flow cytometry graph and statistical analysis of CD4 T_reg cells in x-GVHD mice from the indicated groups. (E) Representative flow cytometry graph and statistical analysis of activation markers expression levels in CD4⁺ and CD8⁺ T cells in x-GVHD mice from the indicated groups. (F) Representative flow cytometry graph and statistical analysis of cytokine secretion levels in CD4⁺ and CD8⁺ T cells in x-GVHD mice from the indicated groups. (G) and (H) Tumor growth was monitored using bioluminescence imaging (BLI) on the dates indicated. (I) Survival curve of humanized x-GVHD and GVL models. Group 1: THP-1 alone, median survival of 41 days; group 2: PBMCs + THP-1, median survival of 33 days; group 3: PBMCs + THP-1 + CD4 T_reg cells, median survival of 39 days; group 4: PBMCs + THP-1 + CD8 T_rp cells, median survival of 42 days. Each group included five mice. Unless otherwise indicated, the data shown are representative of one of three independent experiments. For the comparison of recipient survival among groups, the log-rank test was used to determine statistical significance. A one-way ANOVA was applied for comparisons among three groups. Symbols indicate individual mice, and error bars represent the means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

To further evaluate whether CD8 T_rp cells can suppress GVHD while preserving graft-versus-leukemia (GVL) activity, we cotransferred acute myeloid leukemia cells (THP-1), PBMCs, and CD8 T_rp cells into NSG mice. Four experimental groups were established: G1, THP-1 alone; G2, THP-1 + PBMCs; G3, THP-1 + PBMCs + CD4 T_reg cells; G4, THP-1 + PBMCs + CD8 T_rp cells. Compared to the G1 group, all PBMC-transplanted groups (G2 to G4) showed reduced leukemia burden (Fig. 4, G and H). Mice in the G2 group, which received both PBMCs and THP-1 cells, exhibited the shortest survival, likely due to concurrent GVHD and leukemia progression. In contrast, cotransfer with CD4 T_reg cells (G3) or CD8 T_rp cells (G4) alleviated GVHD while maintaining antileukemia activity, resulting in prolonged survival (Fig. 4I). These findings indicate that CD8 T_rp cells can mitigate GVHD without compromising GVL effects.

CD8⁺HLA-DR⁺CD27⁺ T cells represent the precursor stage of CD8⁺FOXP3⁺ T cells

As we observed that CD8 T_rp cells promote CD4 T_reg cell expansion and induce CD8⁺FOXP3⁺ differentiation in vivo through the MLR model, we also detected FOXP3 expression levels in CD4⁺ and CD8⁺ T cells within xenogeneic GVHD models. Our results showed that FOXP3 expression was elevated in CD4⁺ T cells in both the cotransfusion groups with CD4 T_reg cells and CD8 T_rp cells (Fig. 5A). Notably, FOXP3 was specifically up-regulated in CD8⁺ T cells in the CD8 T_rp cell cotransfusion group (Fig. 5A). These findings suggest that CD8 T_rp cells routinely promote CD4 T_reg cell expansion as well as induce CD8⁺FOXP3⁺ differentiation.

Fig. 5. CD8 T_rp cells enrich precursors of CD8⁺FOXP3⁺ T cells.

(A) Representative flow cytometry graph and statistical analysis of FOXP3 expression levels in CD4⁺ and CD8⁺ T cells from x-GVHD mice across the indicated groups. Thirteen mice for group 1, 14 mice for group 2, and 15 mice for group 3. Combined data from three independent experiments. (B) FOXP3 expression levels in CD4 T_reg cells with or without coculture with CD8 T_rp cells (n = 4). (C) FOXP3 expression levels in CD4 T_con cells stimulated with either IgG or HLA-DR agonist antibody (n = 5, P = 0.0099, paired t test). (D) Representative flow cytometry graph and statistical analysis of FOXP3 expression levels in the specific groups under the steady state and TGF-β stimulation. (E) Proliferation analysis of HLA-A02:01–positive PBMCs with or without coculture with HLA-A02:01–negative CD8⁺HLA-DR⁺CD27⁺ T cells (n = 8). Combined data from two independent experiments. (F) Representative flow cytometry graph of FOXP3 expression levels in the indicated groups. (G) Statistical analysis of FOXP3 expression levels in the indicated groups (n = 4). (H) Statistical analysis of FOXP3 expression levels in CD8 T_rp cells before and after coculture with PBMCs (n = 4). (I) Expression levels of CD4 T_reg cell signature proteins, including FOXP3, CD25, CTLA-4, and HELIOS, in long-term in vitro cultures of CD4 T_reg cells and CD8 T_rp cells (n = 5). d, days. (J) TGF-β levels in supernatants from the indicated groups during long-term in vitro cultures (n = 5). (K) Relative levels of 5-methylcytosine (5mC) DNA methylation measured by qMSP (n = 3). A paired/unpaired t test for comparison between two groups, and a one-way ANOVA was applied for comparisons among three groups. Symbols indicate individual samples, and error bars represent the means ± SEM. Unless otherwise noted, the data shown represent one of two or three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001.

CD4 T_reg cells can arise either through differentiation from CD4 T_con cells or by the expansion of existing T_reg cells. To investigate how CD8 T_rp cells promote the increase in CD4 T_reg cells, we cocultured CD8 T_rp cells with CD4 T_con cells (CD4⁺CD25⁻, CD4 T_con cells) to assess whether CD8 T_rp cells induce their differentiation into CD4 T_reg cells. Under standard culture conditions [Iscove’s modified Dulbecco’s medium (IMDM) + 10% BIT 9500 + anti–CD3/28 microbeads + IL-2 at 100 U/ml], CD8 T_rp cells modestly promoted the differentiation of CD4 T_con cells into CD4 T_reg cells (fig. S6A). In contrast, under optimized induction conditions (ImmunoCult XF + anti–CD3/28 microbeads + IL-2 at 200 U/ml + IL-15 at 10 ng/ml + rapamycin at 100 ng/ml + TGF-β at 1 ng/ml; see Materials and Methods for details), FOXP3 expression in CD4 T_con cells was significantly increased (fig. S6B). To determine whether CD8 T_rp cells promote CD4 T_reg cell expansion, we cocultured them with CD4 T_reg cells in vitro. Although no significant difference in proliferation was observed (fig. S6, C and D), FOXP3 expression was significantly up-regulated in CD4 T_reg cells cocultured with CD8 T_rp cells (Fig. 5B). Stimulation with HLA-DR agonist antibodies in the culture system promoted increased FOXP3 expression in CD4 T_con cells (Fig. 5C). In addition, blocking HLA-DR with antibodies impaired the ability of CD8 T_rp cells to suppress PBMC proliferation during coculture (fig. S6E) but did not affect FOXP3 up-regulation in CD4 T_reg cells when CD4 T_reg cells were cocultured with CD8 T_rp cells (fig. S6F). To determine whether direct cell contact is required for CD8 T_rp cell–mediated suppression, PBMCs were cocultured with CD8 T_rp cells, CD4 T_reg cells, or CD4 T_con cells either in direct contact or separated by a transwell system. The results showed that the suppressive effect of CD8 T_rp cells was abolished when physical separation was introduced (fig. S6G). Together, these findings suggest that CD8 T_rp cells promote FOXP3 expression in CD4 T cells and suppress PBMC proliferation through a contact- and HLA-DR–dependent mechanism. However, HLA-DR signaling is not essential for FOXP3 induction in CD4 T_reg cells.

As we observed that CD8⁺HLA-DR⁺CD27⁺ T cells (CD8 T_rp cells) also promote CD8⁺FOXP3⁺ differentiation in vivo, we speculated that CD8⁺HLA-DR⁺CD27⁺ T cells might be enriched with more precursors of CD8⁺FOXP3⁺ T cells. To investigate this, we sorted CD8⁺HLA-DR⁻CD27⁻ [CD8 double negative (DN)], CD8⁺HLA-DR⁺CD27⁺ (CD8 T_rp), CD4⁺CD25⁻ (CD4 T_con), and CD4⁺CD25⁺CD127^−/low (CD4 T_reg) cells and cultured them with TGF-β in vitro. The results showed that FOXP3 expression was significantly up-regulated in CD8 T_rp cells after TGF-β induction (Fig. 5D). In addition, CD8 T_rp cells exhibited higher FOXP3 expression levels compared to CD8 DN T cells after TGF-β induction (Fig. 5D). Given that PBMCs also contain CD8⁺HLA-DR⁺CD27⁺ subsets, it was challenging to determine whether the up-regulated CD8⁺FOXP3⁺ T cells originated from CD8 T_rp cells or PBMCs. To address this, we sorted CD8 T_rp cells from HLA-A02:01–negative healthy individuals and cocultured them with HLA-A02:01–positive PBMCs. The results showed that the immunosuppressive effects of CD8 T_rp cells are not MHC I restricted, as these cells still inhibited the proliferation of CD4⁺ and CD8⁺ T cells from HLA-A02:01–positive healthy individuals (Fig. 5E). Furthermore, FOXP3 expression levels were significantly higher in CD8 T_rp cells compared to CD8⁺ T cells from PBMCs (Fig. 5, F and G). In addition, FOXP3 expression in CD8 T_rp cells was significantly up-regulated after coculture with PBMCs compared to preculture levels (Fig. 5H). Together, these results indicate that CD8 T_rp cells are enriched with precursors of CD8⁺FOXP3⁺ T cells and can be robustly induced to generate CD8⁺FOXP3⁺ T cells.

Furthermore, we found that the expression level of CD38 was higher in CD8 T_rp cells compared to other classical T cell subsets in the steady state, but it was down-regulated after coculture with PBMCs (fig. S6H), suggesting a context-dependent role of CD38 beyond simply serving as a marker of T cell activation.

We observed that CD8 T_rp cells can serve as precursors to CD8⁺FOXP3⁺ T cells, with FOXP3 expression increasing in the in vivo xenogeneic GVHD model (Fig. 5A) and following in vitro TGF-β stimulation (Fig. 5D). Further investigating when FOXP3 expression becomes stabilized, we evaluated the expression dynamics of key CD4 T_reg cell signature proteins in CD8 T_rp cells during long-term in vitro culture. The results showed that FOXP3, CD25, CTLA-4, and HELIOS were progressively up-regulated in CD8 T_rp cells over time, reaching stable levels by day 12 (Fig. 5I). In contrast, CD4 T_reg cells exhibited a continuous increase in CTLA-4 expression, while FOXP3, CD25, and HELIOS peaked at day 12 and then declined to levels comparable to those in CD8 T_rp cells (Fig. 5I). In addition, TGF-β levels steadily increased in CD8 T_rp cells, while a decline was observed in CD4 T_reg cells by day 16 (Fig. 5J). Methylation of the FOXP3 locus, including the promoter region and the T_reg cell–specific demethylated region (TSDR), also decreased over time in CD8 T_rp cells, as measured by both qMSP [quantitative methylation-specific polymerase chain reaction (PCR)] and bisulfite sequencing (Fig. 5K and fig. S6I). Collectively, these findings suggest that long-term in vitro culture induces progressive up-regulation and eventual stabilization of CD4 T_reg cell signature proteins in CD8 T_rp cells.

Epigenomic profiling of CD8⁺HLA-DR⁺CD27⁺ T cells unveils their precursor and regulatory characteristics

To identify the transcription factors associated with the immunosuppressive signature of CD8 T_rp cells, we performed assay for transposase-accessible chromatin using sequencing (ATAC-seq) analysis on sorted cell populations, including CD8⁺HLA-DR⁻CD27⁻ (CD8 DN), CD8⁺HLA-DR⁺CD27⁺ (CD8 T_rp), CD4⁺CD25⁻ (CD4 T_con), and CD4⁺CD25⁺CD127^−/low (CD4 T_reg) cells, from healthy individuals. Consistent with the chromatin accessibility observed in CD4 T_reg cells, we found that the costimulatory molecule ICOS and chemokine receptor CCR4 showed increased chromatin accessibility in CD8 T_rp cells (Fig. 6, A and B). As expected from prior studies (16, 17), the transcription factors IKZF2 and CTLA-4 were activated in CD4 T_reg cells (Fig. 6B). In CD8 T_rp cells, transcription factors TCF7, TOX2, and PRDM1 were activated (Fig. 6A). Combined analysis with RNA-seq data revealed 176 genes that were up-regulated at both the transcriptomic and epigenomic levels in CD8 T_rp cells compared to CD8 DN T cells (Fig. 6C). Among these, costimulatory molecules CD28 and ICOS, along with the transcription factor TCF7, were activated in CD8 T_rp cells, while the cytotoxic gene GZMB and transcription factor STAT3 were suppressed (Fig. 6D). Analysis of regions with increased accessible chromatin accessibility showed an enrichment of motifs for the transcription factors TCF1 and nuclear factor κB (NF-κB)-p65, whereas regions with decreased accessibility were enriched for motifs of FOS and JUN-AP1 (Fig. 6E). These findings suggest that the activation of TCF7, along with costimulatory signals CD28 and ICOS, is a key feature of CD8 T_rp cells, accompanied by the suppression of cytotoxic programs. Consistent with the RNA-seq results, the chromatin accessibility profile of CD8 T_rp cells demonstrated distinct patterns, including increased activation of TCF7, TOX2, and NR4A2 and co-inhibitory molecules PD-1 and TIGIT and decreased activation of GZMB (Figs. 2G and 6F). These findings suggest that TCF7 activation corresponds to the precursor signature of CD8 T_rp cells, while the activation of co-stimulatory molecules CD28 and ICOS, as well as the transcription factor TOX2, may predispose CD8 T_rp cells to the efficient activation of FOXP3.

Fig. 6. Epigenomic signatures of CD8 T_rp cells.

(A) Volcano plots showing genes proximal to differential accessible regions in CD8 T_rp cells compared to CD8 DN T cells. (B) Volcano plots showing genes proximal to differential accessible regions in CD4 T_reg cells compared to CD4 T_con cells. (C) Overlap between DEGs and different accessibility-proximal genes in CD8 T_rp cells compared to CD8 DN T cells. (D) Integrated analysis of RNA-seq and ATAC-seq data. DEGs and DA peak-proximal genes in CD8 T_rp cells compared to CD8 DN T cells. (E) Enriched transcription factor motifs in the differential accessible peaks of CD8 T_rp cells compared to CD8 DN T cells are shown. (F) WashU browser views show the gene expression level (RNA-seq) and chromatin accessibility (ATAC-seq) of specific genes in the indicated groups. n = 5. CD8 T_rp, CD8⁺HLA-DR⁺CD27⁺; CD8 DN, CD8⁺HLA-DR⁻CD27⁻; CD8 T_naive, CD8⁺CCR7⁺CD45RA⁺; CD8 T_cm, CD8⁺CCR7⁺CD45RA⁻; CD8 T_em, CD8⁺CCR7⁻CD45RA⁻; CD8 T_emra, CD8⁺CCR7⁻CD45RA⁺.

Elevated CD8 T_rp cell frequencies correlate with immune homeostasis restoration following allo-HSCT

To investigate the correlation between CD8 T_rp cells and the maintenance of peripheral homeostasis, we detected the distribution of CD8 T_rp cells under both steady state and pathological conditions. Peripheral blood samples were collected from healthy individuals (n = 10), homeostatic recipients following allo-HSCT (n = 21), and chronic GVHD (cGVHD) recipients following allo-HSCT (n = 12) (table S2). We analyzed the proportion of HLA-DR⁺CD27⁺ subsets within the CD8⁺ T cells. The results showed that the ratio of HLA-DR⁺CD27⁺ is significantly increased in recipients who had successfully restored immune homeostasis following allo-HSCT compared to healthy individuals or cGVHD recipients (Fig. 7, A and B). These findings suggest that CD8 T_rp cells may contribute to the remodeling of immune homeostasis following allo-HSCT.

Fig. 7. Proportion and signature of CD8 T_rp cells in healthy individuals, homeostatic recipients, and cGVHD recipients following allo-HSCT.

(A) Representative flow cytometry graph of CD8 T_rp cells in healthy individuals, homeostatic recipients, and cGVHD recipients following allo-HSCT. (B) Statistical analysis of the proportion of HLA-DR⁺CD27⁺ in CD8⁺ T cells from healthy individuals (n = 10), homeostatic recipients (n = 21), and cGVHD recipients (n = 12) following allo-HSCT. (C) Distribution of classical T cell subsets in CD8⁺HLA-DR⁺CD27⁺ T cells across different groups. (D) Expression level of signature genes in CD8⁺HLA-DR⁺CD27⁺ T cells across different groups. CD8 T_naive, CD8⁺CCR7⁺CD45RA⁺; CD8 T_cm, CD8⁺CCR7⁺CD45RA⁻; CD8 T_em, CD8⁺CCR7⁻CD45RA⁻; CD8 T_emra, CD8⁺CCR7⁻CD45RA⁺. A one-way ANOVA was applied for comparisons among three groups. Symbols indicate individual participants, and error bars represent the means ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001.

To further assess the signature of CD8⁺HLA-DR⁺CD27⁺ T cells across different clinical contexts, we analyzed the distribution of classical T cell subsets within this population. In recipients with cGVHD, CD8⁺HLA-DR⁺CD27⁺ T cells exhibited an increased proportion of terminal effector T cells (T_emra cells) compared to healthy individuals and recipients who successfully restored immune homeostasis following allo-HSCT, indicating a reduced memory-like signature in the cGVHD setting (Fig. 7C). While MHC class II expression remained consistent across all groups, the expression of costimulatory molecule CD28 and the inhibitory markers TIGIT and PD-1 varied across clinical conditions (Fig. 7D). These results suggest that the signature of CD8⁺HLA-DR⁺CD27⁺ T cells is dynamic and influenced by the immune context. Further studies are needed to evaluate their potential immunoregulatory function across diverse clinical contexts.

DISCUSSION

T_reg cell subsets within CD8⁺ T cells are highly heterogeneous, with phenotypes that often lack direct correspondence between humans and mice. Consequently, despite numerous studies devoted to identifying CD8⁺ T_reg cells, these cells comprise subsets with diverse phenotypes and properties, typically arising under specific induction conditions. Our previous study identified a cluster of CD8 regulatory precursors (CD8 T_rp cells) that contribute to immune homeostasis remodeling in recipients after allo-HSCT at the single-cell level (13). In this study, we further characterized the immune phenotype of CD8 T_rp cells and confirmed their natural existence in humans under the steady state. Our in vitro and in vivo studies demonstrated that CD8⁺HLA-DR⁺CD27⁺ T cells exert immunosuppressive functions by promoting the expansion of CD4 T_reg cells and inducing the differentiation of CD8⁺FOXP3⁺ T cells. Notably, in humanized mouse models, CD8⁺HLA-DR⁺CD27⁺ T cells effectively minimized GVHD damage. Furthermore, a high proportion of CD8⁺HLA-DR⁺CD27⁺ T cells was associated with the remodeling of immune homeostasis following allo-HSCT. These findings suggest that CD8⁺HLA-DR⁺CD27⁺ T cells represent a regulatory precursor subset crucial for maintaining peripheral immune homeostasis in the steady state.

Previous studies have reported the presence of CD8⁺HLA-DR⁺ T cells in adult and umbilical venous blood, with the ability to suppress PBMC proliferation (18, 19). Another independent study observed an increase in CD8⁺HLA-DR⁺ T cells in the peripheral blood of older adults, while these cells exhibited reduced immunosuppressive function (20). Consistent with our findings, these CD8⁺HLA-DR⁺ T cells were shown to highly express inhibitory markers such as PD-1 and TIGIT. However, CD8⁺HLA-DR⁺ T cells remain a heterogeneous population, and previous studies lacked in vivo validation of their immunosuppressive function. Our study identified CD27 as a marker that can exclude terminal effector T cells from the CD8⁺HLA-DR⁺ population, enriching for a subset with potent regulatory and precursor-like functions. CD27, a member of the TNF receptor family, is necessary for the development of T cell memory (21, 22). The loss of CD27 primarily impairs memory T cell formation, leading to delayed kinetics and reduced numbers of virus-specific CD8⁺ T cells (21). Engagement of CD27 on CD8⁺ T cells enhances their expansion, promotes resistance to apoptosis, and improves both persistence and antitumor activity in chimeric antigen receptor T cells (23–25). In addition, within the CD4⁺CD25⁺ T_reg cell compartment, CD27⁺ subsets have been shown to have a highly suppressive function (26, 27). These findings align with the features of the CD8⁺HLA-DR⁺CD27⁺ population identified in our study, which exhibits a precursor-like regulatory phenotype with the capacity to differentiate into CD8⁺FOXP3⁺ T cells.

We validated the immunosuppressive capabilities of CD8⁺HLA-DR⁺CD27⁺ T cells both in vitro and in vivo. Notably, this subset exhibited relatively high CD28 expression, distinguishing them from conventional CD8⁺CD28⁻ T_reg cells (28–30). HLA-DR⁺ T cells have been shown to have enhanced inhibitory capacity within CD4 T_reg cells, suppressing T cell proliferation and cytokine production through an early contact-dependent mechanism associated with FOXP3 induction (31). In other studies, HLA-DR expression on CD8⁺ T cells has been associated with activation rather than regulation (32–37). These activated CD8⁺HLA-DR⁺ T cells share a common signature of high CD38 expression. CD38 is a transmembrane protein involved in NAD⁺ [nicotinamide adenine dinucleotide (oxidized form)] consumption, cyclic adenosine diphosphate ribose generation, and calcium mobilization through its enzymatic activity (38). While CD38 is widely recognized as a T cell activation marker and is highly expressed during certain viral infection (32, 33, 39), its expression patterns differ in the steady state. T_naive cells from healthy donors exhibit remarkably higher levels of CD38 expression compared to effector T cells (40). Another study demonstrated that CD38 expression consistently identifies CD8⁺ and CD4⁺ recent thymic emigrant T_naive cells in circulation in healthy donors (41). In our study, CD38 expression was higher in CD8⁺HLA-DR⁺CD27⁺ cells under the steady state and decreased following coculture with PBMCs. These findings suggest that CD38 represents a progenitor-associated marker in the homeostatic state. Consistent with previous research (31), we also observed that HLA-DR⁺ T cells within specific subsets (CD4 T_reg cells and CD8 T_rp cells) exhibited enhanced inhibitory capacity. Moreover, consistent with our findings, Ma et al. (42) used xenoantigen stimulation to expand CD4 T_reg cells in vitro and demonstrated that HLA-DR⁺CD27⁺CD4⁺ T_reg cells are enriched for a stable, antigen-specific regulatory population with enhanced suppression function, effectively protecting islet xenografts from human T cell–mediated rejection. These findings suggest that HLA-DR⁺CD27⁺ T cells may represent a regulatory subset within both CD4⁺ and CD8⁺ T cell populations.

Another important feature of the identified CD8 T_rp cells is their role as precursors to CD8⁺FOXP3⁺ T cells. In the steady state, CD8⁺HLA-DR⁺CD27⁺ T cells exhibit a precursor signature and readily differentiate into CD8⁺FOXP3⁺ T cells upon stimulation or in humanized GVHD models. These findings suggest that CD8⁺HLA-DR⁺CD27⁺ T_rp cells represent a progenitor stage of CD8⁺FOXP3⁺ T cells. Transcriptomic and epigenomic profiling revealed the activation of transcription factor TCF7 in CD8 T_rp cells, which likely corresponds to their precursor signature. Another characteristic of CD8 T_rp cells is the activation of costimulatory molecules CD28 and ICOS. Previous studies have shown that the activation of ICOS and CD28 promotes FOXP3 activation in CD4 T_reg cells through an NF-κB–dependent pathway (43–45). Consistently, we observed increased chromatin accessibility of CD28 and ICOS in CD8 T_rp cells, with enriched motifs for NF-κB-p65. These findings suggest that the activation of CD28 and ICOS in CD8 T_rp cells may similarly induce FOXP3 activation through the NF-κB signaling pathway. Another notable feature of CD8 T_rp cells is the activation of transcription factor TOX2. Unlike TOX, which serve as a central regulator of exhaustion in CD8⁺ T cells, TOX2 is essential for the differentiation of T_cm cells (46, 47). Therefore, the activation of TOX2 further supports the precursor signature of CD8 T_rp cells. However, given that CD8 T_rp cells serve as precursors to CD8⁺FOXP3⁺ T cells, an important question remains: Does the CD8⁺HLA-DR⁺CD27⁺ subset maintain the same characteristics after stimulation as it exhibits in its steady-state phenotype? Further investigations are needed to evaluate the stability of the phenotype, transcriptome, and epigenome of CD8 T_rp cells under stimulation.

In summary, our study characterized a naturally existing subset of CD8 regulatory precursors with the immunophenotype CD8⁺HLA-DR⁺CD27⁺. Further research is needed to optimize conditions for inducing FOXP3 activation in CD8 T_rp cells, which could provide a promising therapeutic strategy for promoting immune homeostasis in transplantation or autoimmune disease.

MATERIALS AND METHODS

Clinical samples

Peripheral blood samples were collected from healthy individuals and recipients following allo-HSCT, purified to isolate PBMCs, and subsequently subjected to spectrum flow cytometry analysis, RNA-seq, ATAC-seq, or functional experiments. The clinical characteristics of recipients and their paired donors are listed in table S2. This study was approved by the Ethics Committee of Peking University People’s Hospital with approval number 2020PHB067, and written informed consent was obtained from all subjects in accordance with the Declaration of Helsinki.

Mice

MHC-knockout NSG mice (Prkdc^scidIL2rg^−/−H2-K1^−/−H2-Ab1^−/−H2-D1^−/−, 6 to 8 weeks old) were purchased from Beijing Vitalstar Biotechnology Co., Ltd. (Beijing China). NSG mice (Prkdc^scidIL2rg^−/−) were purchased from Beijing Vitalstar Biotechnology Co., Ltd. (Beijing China), or Biocytogen Pharmaceuticals (Beijing China) Co., Ltd. All mice were maintained in the specific pathogen–free animal facility of Peking University People’s Hospital. This study was approved by the Ethics Committee of Peking University People’s Hospital with approval number 2020PHE006. All experiments were performed according to the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals.

High-dimensional flow cytometry data analysis

FCS (Flow Cytometry Standard) 3.0 files were imported into FlowJo software (version 10) for analysis. Standard gating strategies were used to remove aggregated and dead cells and then gated CD3⁺ T cells. All CD3⁺ T cells per sample were subsequently imported in FlowJo, biexponentially transformed, and exported for further analysis in R (version 4.3.0). Cells from healthy individuals (n = 10), haplo-SCT recipients (n = 12), and MSDT recipients (n = 9) were labeled with a unique computational barcode for further identification and concatenated in a single matrix by using the R package cytofkit2 (version 0.99.8) with the parameter transformMethod = “autoLgcl”. Cells exhibiting abnormal fluorescence intensity values were removed by boxplot.stats with the parameter coef = 2.5. Then, 2000 cells were randomly selected from each sample for further analysis, and the limma package (version 3.56.2) was used to remove batch effects. The dataset was analyzed using the “FlowSOM” algorithm from the cytofkit2 package with the parameter k = 30 and visualized by uniform manifold approximation and projection (UMAP). For the analysis of CD8 T cell subsets, we selected cells with CD8 expression levels above 4 and CD4 expression levels below 2, ultimately isolating a total of 6810 purified CD8 T cells.

Sorting targeted T cells for bulk sequencing

T cells were purified from PBMCs by Pan T MicroBeads (Miltenyi Biotec, cat. no. 130-096-535) according to the manufacturer’s instructions. Cells were stained with surface antibodies against CD4, CD25, CD127, CD8, HLA-DR, CD27, CD45RA, and CCR7. Cells were incubated on ice for 20 min, washed with 3 ml of phosphate-buffered saline (PBS), and then resuspended in 500 μl of PBS for sorting on a FACS Aria III (BD Biosciences). Antibody information is supplied in table S3.

Bulk RNA-seq and data processing

RNA was extracted using the RNeasy Micro Kit (QIAGEN, 74004) and purified by NEBNext Oligo d(T)₂₅ beads (NEB, E7490). An NEBNext Ultra II RNA Library Prep Kit (NEB, E7770S) was used to generate cDNA libraries. AMPure XP beads (Beckman Coulter, A63881) were used to purify cDNA libraries between 300 and 500 base pairs (bp). Paired-end sequencing was performed on a NovaSeq6000 (Illumina) and produced between 28 and 35 million 150-bp paired-end reads per sample.

Raw RNA-seq FASTQ files were processed according to a previously described method (48). Low-quality reads and adapter sequences were cleaned up, the remaining reads were quantified against the GRCh38 reference at the transcript level using Hisat2 software, and gene expression was quantified by StringTie. Then, differential expression analysis was run using the DESeq2 package.

Bulk ATAC-seq and data processing

A total of 5 × 10⁴ sorted cells was resuspended and lysed in 50 μl of lysis buffer (10 mM NaCl, 10 mM tris-HCl, pH 7.4, 3 mM MgCl₂, and 0.1% ICEPALEA-630) after being washed twice in PBS. DNA was fragmented using a TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme, TD501-01) and purified by using a MinElute PCR Purification Kit (QIAGEN, 28006). The purified DNA was barcoded with TruePrep Index Kit V2 for Illumina (Vazyme, TD202) and amplified by PCR using TruePrep DNA Library Prep Kit V2 for Illumina. VAHTS DNA Clean Beads (Vazyme, N411) were used to purify cDNA libraries between 100 and 1000 bp in size. Paired-end sequencing was performed on a NovaSeq6000 (Illumina).

Raw ATAC-seq FASTQ files from paired-end sequencing were processed according to a previously described method (49). Clean FASTQ files were aligned to the GRCh38 reference genome using Bowtie2. Samtools was used to remove unmapped, unpaired, and mitochondrial reads. PCR duplicates were removed using Picard. Reads were shifted +4 bp and −5 bp for positive and negative strands, respectively. Peak calling was performed using MACS2 with a false discovery rate q value of 0.01. We combined the peaks of all samples to create a union peak list and merged overlapping peaks with the BedTools merge command. The number of reads in each peak was determined using BedTools coverage. Differentially accessible (DA) regions were identified following DESeq2 normalization using a false discovery rate cutoff of <0.05. Motif enrichment was calculated using HOMER (default parameters) on peaks that were DA across the compared group. Transcription binding site prediction analysis was performed using a known motif discovery strategy.

CD8 T_rp cell signature score

We extracted 11 cell types from our previously generated single-cell dataset, including the following: CD8 T_naive, CD8 T_cm, CD8 T_em, CD8 T_eff1, CD8 T_eff2, CD8 T_rp, CD4 T_naive, CD4 T_cm, CD4 T_em, CD4 T_eff, and CD4 T_reg. Using the “rank_genes_groups” function, we identified genes that were highly expressed in CD8 T_rp cells. From these, we selected the top 50 genes to define a CD8 T_rp cell–specific gene signature (table S1). This gene set was then used to score bulk RNA-seq data from freshly sorted samples, including CD8 T_n (T_naive), CD8 T_em, CD8 T_emra, CD8 T_cm, CD8⁺HLA-DR⁺CD27⁺, and CD8⁺HLA-DR⁻CD27⁻, using the GSVA package (version 1.50.0).

In vitro suppression assays

Responder PBMCs (1 × 10⁵) were labeled with 1 μM CFSE (carboxyfluorescein diacetate succinimidyl ester; BD Pharmingen, 565082) or 5 μM CellTrace Violet Fluorescent (Thermo Fisher Scientific, C34557) and cocultured with CD4 T_reg cells or CD8 T_rp cells at a 1:1 ratio in a 96-well round-bottom plate. The cells were stimulated with soluble anti-CD3 antibodies (Invitrogen, 16-0037-85) at 5 μg/ml and anti-CD28 antibodies (Invitrogen, 16-0289-85) at 5 μg/ml, along with recombinant human IL-2 (rhIL-2; 100 U/ml), and cultured in IMDM (Gibco, Invitrogen) supplemented with 10% BIT 9500 (Stemcell Technologies, 09500). After 96 hours, proliferation was detected using a BD FACSCanto II flow cytometer (BD Biosciences). PBMCs cultured without anti-CD3 and anti-CD28 antibodies served as a negative control.

Flow cytometry

Surface staining was performed at room temperature with antibodies for 20 min. For cytokine detection, T cells were stimulated with phorbol 12-myristate 13-acetate (50 ng/ml, Sigma-Aldrich, 880134P) and ionomycin (1000 ng/ml, Sigma-Aldrich, I3909) for 4 hours at 37°C in the presence of GolgiPlug (BD Pharmingen, 555029). Intracellular staining was carried out by using the Transcription Factor Buffer Set kit (BD Pharmingen, 562574) after being resuspended according to the manufacturer’s instructions and then incubated for 20 min with antibodies at room temperature. Detailed antibody information is listed in table S3.

T cell proliferation analysis

Cells were stained with CFSE (BD Pharmingen, 565082) at a final concentration of 1 μM for 10 min at 37°C and washed twice with complete medium containing 10% fetal bovine serum. Labeled T cells (2 × 10⁵ cells per well) were stimulated with CD3/28 beads in flat-bottom 96-well plates in IMDM containing 10% BIT 9500 with rhIL-2 (100 U/ml). After 96 hours of culture, the cells were harvested and then analyzed by flow cytometry.

FOXP3 expression induction by TGF-β

Freshly isolated cells were seeded into flat-bottom 96-well plates at a density of 2 × 10⁵ cells per well in 100 μl of IMDM (Gibco, Invitrogen) supplemented with 10% BIT 9500 (Stemcell Technologies, 09500). The cells were stimulated at 37°C with 5% CO₂ for 5 days in the presence of anti-CD3 antibodies (5 μg/ml), anti-CD28 antibodies (5 μg/ml), and rhIL-2 (100 U/ml). To induce FOXP3 expression, TGF-β (1 ng/ml; Abcam, AB50036) was added to the culture. FOXP3 expression levels were assessed after 120 hours of culture.

Long-term in vitro culture of CD8 T_rp cells

CD8 T_rp cells or CD4 T_reg cells were seeded in 24-well plates at a density of 5 × 10⁵ cells per well in 500 μl of ImmunoCult XF medium (Stemcell Technologies, 100-0956), supplemented with IL-2 (200 U/ml), IL-15 (10 ng/ml), rapamycin (100 ng/ml), and TGF-β (1 ng/ml). Cells were stimulated with anti–CD3/CD28 microbeads at a 1:1 cell-to-bead ratio. On day 7, the microbeads were removed, and the cell concentration was adjusted to 5 × 10⁵ cells per well, followed by restimulation. During the culture period, 150 μl of fresh complete medium was added on days 3, 6, 10, and 13. Cells were harvested on days 0, 8, 12, and 16 for the analysis of CD4 T_reg cell signature protein expression.

Detection of TGF-β in culture supernatants during long-term culture

TGF-β levels in cell culture supernatants were measured using a human TGF-β ELISA Kit (Proteintech, KE00002) according to the manufacturer’s instructions. Supernatants were collected on days 8, 12, and 16 during the long-term culture period. As the culture medium was supplemented with exogenous TGF-β, blank medium wells (without cells) were set up and collected at the corresponding time points to serve as controls.

Assessment of FOXP3 methylation by qMSP

Genomic DNA was extracted using the QIAamp DNA Blood Mini Kit (QIAGEN, 51104). Between 60 and 400 ng of genomic DNA was subjected to bisulfite conversion using the Ultra-fast DNA Methylation Conversion Kit (UMC kit, EM009, Beijing Junfeix Technology Co., Ltd., Beijing, China). After elution, the DNA concentration was adjusted to 8 ng/μl. Real-time PCR amplification of methylated and nonmethylated FOXP3 sequences (from −2000 to +1000 relative to the transcription start site) was performed using 8 ng of bisulfite-converted DNA, with the following primers: nonmethylated FOXP3 sequences, 5′-TAAAAGAATATATAGGTGGGTGTGG-3′ (forward) and 5′-CAACTTAAAAACCTAAAAACTACAAA-3′ (reverse); methylated FOXP3 sequences, 5′-TTAAAAGAATATATAGGTGGGTGCG-3′ (forward) and 5′-CCCAACTTAAAAACCTAAAAACTACG-3′ (reverse); nonmethylated β-actin sequences (control), 5′-GGGTTGAATTGGGTATTGTTTAGT-3′ (forward) and 5′-AAAAAAAATTTTAACATTAACCACC-3′ (reverse). The relative level of methylated FOXP3 was calculated using the following formula: Methylated FOXP3 (%) = 1/[1 + 2^{(ΔCtM − ΔCtU)}] × 100%, where ΔCtM and ΔCtU represent the threshold cycle (Ct) values for methylated and unmethylated reactions, respectively.

Bisulfite sequencing of the FOXP3 TSDR

Genomic DNA was extracted using the QIAamp DNA Blood Mini Kit (QIAGEN, 51104) and subjected to bisulfite conversion using the EpiArt Ultrafast DNA Methylation Bisulfite Kit (Vazyme, Nanjing, China). The FOXP3 TSDR containing a unique CpG site was amplified by PCR using the following primers: AAGTATAGATATATTTTTAGATAGGGAT (forward) and AACTAAAAACTTATCATAATAACCAA (reverse). One microliter of the PCR product was then used as a template for a nested PCR with the following primers: TGTTTGGGGGTAGAGGATTTAGAGGG (forward) and ACATCACCTACCACATCCACCAACACCCAT (reverse).

The amplification products were purified using the Trelief DNA Gel Extraction Kit (Tsingke) and cloned into the pClone007 Versatile Simple Vector (Tsingke). Plasmid DNA was isolated from 1 ml of overnight bacterial cultures using the Trelief Plasmid Mini Kit Plus (Tsingke), and 10 individual clones were sequenced using the M13 reverse primer.

HLA-DR antibody stimulation

CD4⁺CD25⁻ T_con cells were cultured in flat-bottom 96-well plates at a density of 2 × 10⁵ cells per well in 100 μl of IMDM (Gibco, Invitrogen) supplemented with 10% BIT 9500 (Stemcell Technologies, 09500). The plates were coated with anti–HLA-DR antibodies (1 ng/ml; BioLegend, cat. no. 327002), while anti-CD3 antibodies (5 μg/ml), anti-CD28 antibodies (5 μg/ml), and rhIL-2 (100 U/ml) were added to the medium. IgG (BioLegend, cat. no. 402201) was used as an isotype control. After 120 hours of culture, the cells were harvested and analyzed using a BD FACSCanto II flow cytometer (BD Biosciences).

HLA-DR antibody neutralization assay

CD8 T_rp cells were cocultured with PBMCs or CD4 T_reg cells at a 1:1 ratio in round-bottom 96-well plates containing 100 μl of IMDM (Gibco, Invitrogen) supplemented with 10% BIT 9500 (Stemcell Technologies, 09500). The total cell density was maintained at 2 × 10⁵ cells per well across all groups. An anti–HLA-DR antibody (BioLegend, 1 ng/ml, cat. no. 307602) or isotype-matched IgG control antibody (BioLegend, 1 ng/ml, cat. no. 400201) was added to the cultures. In all neutralization assays, cells were stimulated with soluble anti-CD3 (Invitrogen, 16-0037-85) and anti-CD28 (Invitrogen, 16-0289-85) antibodies at a final concentration of 5 μg/ml each.

Cell contact–dependent assay

To determine whether direct cell contact is required for CD8 T_rp cell–mediated suppression, 4 × 10⁵ PBMCs were cocultured with 2 × 10⁵ CD8 T_rp cells, CD4 T_reg cells, or CD4 T_con cells either in direct contact or separated using a transwell system. For direct coculture, a total of 6 × 10⁵ cells was seeded per well in a 24-well plate. In the transwell setup, 4 × 10⁵ PBMCs were seeded in the bottom chamber, while 2 × 10⁵ CD8 T_rp cells, CD4 T_reg cells, or CD4 T_con cells were seeded in the top chamber, separated by a 0.4-μm–pore membrane (Corning Life Sciences, cat. no. 3413). After 120 hours of coculture in the presence of soluble anti-CD3 (Invitrogen, 5 ng/ml) and CD28 (Invitrogen, 5 ng/ml), PBMCs from the bottom chamber were harvested and analyzed for proliferation using flow cytometry.

In vivo MLR

MHC-knockout NSG mice (Prkdc^scidIL2rg^−/−H2-K1^−/−H2-Ab1^−/−H2-D1^−/−, 6 to 8 weeks old, female) were humanized by intravenous tail vein injection of 5 × 10⁶ PBMCs from HLA-A02:01–negative healthy individuals. Following transplantation, the body weight and GVHD scores of all mice were monitored. Mice that exhibited no weight loss or other signs of GVHD were included in the subsequent study. On day 14, the humanized NSG mice were challenged with one of the following groups: (i) 5 × 10⁶ mitomycin C (MMC; MedChemExpress, HY-13316)–pretreated PBMCs from the same HLA-A02:01–negative donor (autologous group); (ii) 5 × 10⁶ MMC-pretreated PBMCs from HLA-A02:01–positive healthy individuals (allogeneic group); (iii) 5 × 10⁶ MMC-pretreated PBMCs from HLA-A02:01–positive PBMCs combined with 5 × 10⁵ CD4 T_reg cells; (iv) 5 × 10⁶ MMC-pretreated PBMCs from HLA-A02:01–positive PBMCs combined with 5 × 10⁵ CD8⁺HLA-DR⁺CD27⁺ T cells (CD8 T_rp cells). Flow cytometry was used to assess the percentage of hCD45⁺ cells. MMC was used at a concentration of 20 μg/ml, and rhIL-2 was administered at 50 U/ml. On day 19, the mice were euthanized, and flow cytometry was performed to analyze the expansion of hCD45⁺, T cell activation, and T cell function.

Xenogeneic humanized GVHD model

NSG mice (Prkdc^scidIL2rg^−/−, 6 to 8 weeks old, female) were irradiated at 1 gray (Gy) and subsequently injected with one of the following: (i) 5 × 10⁶ PBMCs from healthy individuals alone; (ii) 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD4 T_reg cells; (iii) 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD8 T_rp cells. Mice were monitored three times a week for changes in body weight and clinical signs to assess GVHD severity. GVHD scoring included evaluations of weight loss, posture, mobility, skin integrity, and fur conditions. Each parameter was assigned a score of 0 (absent), 1 (moderate), or 2 (severe). To analyze the T cell function, humanized GVHD mice were euthanized 16 days after transplantation, and peripheral blood and spleens were harvested for flow cytometric analysis.

Xenogeneic humanized GVL model

NSG mice (Prkdc^scidIL2rg^−/−, 6 to 8 weeks old, female) were irradiated at 1 Gy and subsequently injected with 1 × 10⁶ luciferase-expressing THP-1 cells. Four hours later, the mice were randomized into four groups: (i) THP-1 only (no treatment); (ii) 5 × 10⁶ PBMCs; (iii) 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD4 T_reg cells; (iv) 5 × 10⁶ PBMCs combined with 5 × 10⁵ CD8 T_rp cells. Leukemia burden was monitored weekly using bioluminescence imaging (BLI).

Bioluminescence imaging (BLI)

Mice were injected intraperitoneally with 3 mg of d-luciferin (PerkinElmer, US) dissolved in PBS and imaged 6 min later. Tumor growth was monitored by measuring the bioluminescence of THP-1-luc cells on days 8, 15, 22, 29, and 33 after transplantation using the IVIS Imaging System (PerkinElmer, IVIS Lumina III). Mice were anesthetized with isoflurane (2.5% in oxygen) during imaging. For analysis, the total photon flux (photons per second) was quantified from a fixed region of interest encompassing the entire abdominal area using Living Image software (version 4.3.1).

Statistical analyses

All the results are shown as the means ± SEM. Student’s t test was used for the analysis of two groups. A one-way analysis of variance (ANOVA) was used to compare the means of more than two groups. P values <0.05 were considered to be significant. The absence of P values in the graphs indicates that there is no statistic difference between groups. Statistical analyses were performed on GraphPad 8.0 software.

Supplementary Materials

The PDF file includes:

Figs. S1 to S6

Legend for table S1

Tables S2 and S3

Other Supplementary Material for this manuscript includes the following:

Table S1