Extraislet expression of islet antigen boosts T cell exhaustion to partially prevent autoimmune diabetes

Significance

T cell exhaustion is one of the ways that the body is protected from excessive activity of the immune system in chronic infections and cancer. T cell exhaustion also stops clearance of infections and cancers. Chronic infections and cancer have similarities with autoimmune diseases such as type 1 diabetes (T1D). T1D is also chronic but the target tissue is destroyed. We have compared T cell markers of exhaustion in T1D with markers in chronic infections and cancer. We find that T cell exhaustion is incomplete in T1D. Exhaustion could be enhanced by exposure to islet antigen outside the islets. This led to protection from diabetes. Enhancing T cell exhaustion with antigen therapy could be useful as a therapy for T1D.

Abstract

Persistent antigen exposure results in the differentiation of functionally impaired, also termed exhausted, T cells which are maintained by a distinct population of precursors of exhausted T (T_PEX) cells. T cell exhaustion is well studied in the context of chronic viral infections and cancer, but it is unclear whether and how antigen-driven T cell exhaustion controls progression of autoimmune diabetes and whether this process can be harnessed to prevent diabetes. Using nonobese diabetic (NOD) mice, we show that some CD8+ T cells specific for the islet antigen, islet-specific glucose-6-phosphatase catalytic subunit–related protein (IGRP) displayed terminal exhaustion characteristics within pancreatic islets but were maintained in the T_PEX cell state in peripheral lymphoid organs (PLO). More IGRP-specific T cells resided in the PLO than in islets. To examine the impact of extraislet antigen exposure on T cell exhaustion in diabetes, we generated transgenic NOD mice with inducible IGRP expression in peripheral antigen-presenting cells. Antigen exposure in the extraislet environment induced severely exhausted IGRP-specific T cells with reduced ability to produce interferon (IFN)γ, which protected these mice from diabetes. Our data demonstrate that T cell exhaustion induced by delivery of antigen can be harnessed to prevent autoimmune diabetes.

Chronic antigen exposure in cancer and chronic viral infection causes CD8⁺ T cells to lose cytokine production and cytotoxicity, a state termed T cell exhaustion (1–3). T cell exhaustion is a physiological adaptation to continuous antigen stimulation that protects against excessive immune-mediated tissue damage. However, it also results in failure to clear the antigen leading to viral or tumor persistence. T cell exhaustion has also been described in autoimmunity (4). In type 1 diabetes (T1D), CD8⁺ T cell–mediated beta cell destruction takes years in humans and months in the nonobese diabetic (NOD) mouse model (5–7). This chronicity is likely to be due to a balance between the autoimmune attack and processes such as T cell exhaustion, that reduce the effectiveness of the autoimmune attack. This balance is complex and at present insufficiently understood.

Exhausted T cells are functionally heterogenous. They are maintained by precursors of exhausted T (T_PEX) cells (8–10). These cells, which retain high proliferative potential and undergo long-term self-renewal, express the transcription factor TCF1 (encoded by TCF7). They give rise to short-lived, terminally exhausted effector T (T_EX) cells with restrained functionality that express TIM-3 but lack TCF1 (11). T_PEX cells mediate the response to therapies that block the immune checkpoint programmed cell death 1 (PD-1) and thereby reinvigorate exhausted T cells in cancer (8–10). TCF1⁺ exhausted T cells with precursor properties have also been described in autoimmune conditions (12–14). Transcriptomic profiling of T cells from patients with conditions such as vasculitis, Crohn’s disease, and systemic lupus erythematosus, showed that a T cell exhaustion signature correlated with a more benign form of autoimmune disease. This benign outcome suggested that mechanisms associated with T cell exhaustion may control autoimmunity (15). PD-1 blocking therapies can precipitate or exacerbate autoimmunity (16–18).

The T cell response to islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP), a major autoantigen recognized by CD8⁺ T cells in NOD mice (19, 20) allows the study of CD8⁺ T cell exhaustion in spontaneously developing, polyclonal, antigen-specific T cells in a clinically relevant model of autoimmunity. Recent studies have identified exhausted CD8⁺ T cells in islets of NOD mice that increase with age (12, 21, 22); however, one study did not (23). Progression of T1D is slower in individuals with islet-specific CD8⁺ T cells displaying an exhausted phenotype with expression of multiple inhibitory receptors (24). While reinvigoration of exhausted CD8⁺ T cells with checkpoint inhibitors is the aim of cancer treatment, we reasoned that driving terminal differentiation of exhausted CD8⁺ T cells may be a strategy to prevent diabetes in NOD mice. T_PEX cells reside in lymphoid tissues. When these cells encounter an antigen, they proliferate, leave the lymph nodes, and enter the peripheral tissues via the blood. During activation and migration, they lose TCF1 expression and differentiate to T_EX CD8⁺ T cells. We have previously shown that CD8⁺ T cells specific for IGRP_206–214 (IGRP-specific T cells) recirculate between peripheral lymphoid organs (PLO) and islets (25), and we noted more IGRP-specific T cells in the PLO than in the islet. Hence, we hypothesized that CD8⁺ T cell exhaustion could be boosted by continuously exposing T cells to antigen when they are resident in lymph nodes. To test this idea, we generated tetracycline-inducible IGRP transgenic NOD mice to enable temporal expression of IGRP in antigen-presenting cells (APCs) in a doxycycline-dependent manner (26). Antigen expressed in the PLO induced terminal exhaustion in CD8⁺ T cells that subsequently migrated to the islets. Surprisingly, while deletion of IGRP-specific T cells did not prevent diabetes, boosting exhaustion in IGRP-specific T cells prevented diabetes in NOD mice. Overall, we show that boosting T cell exhaustion in autoimmune diabetes is a promising strategy to limit the progression of disease.

Results

Islet Antigen–Specific T Cells Undergo an Exhaustion Program in NOD Mice.

Several studies described features of exhaustion in islet infiltrating CD8⁺ T cells (12, 21, 22). Furthermore, a recent study showed that IGRP-specific T cells in the draining pancreatic lymph node expressed high levels of TCF-1. These cells gave rise to functional effector T cells, which infiltrated the pancreas and destroyed insulin-producing beta cells (23), indicating at most partial exhaustion. To better understand T cell exhaustion and heterogeneity, we performed single-cell RNA-seq analysis of immune cells in the islets (SI Appendix, Fig. S1). Consistent with previous analyses, the infiltrate was dominated by T cells, including CD4 and CD8 T cells, and B cells with smaller populations of natural killer cells and myeloid cells (SI Appendix, Fig. S1 A–D). Subsequent analyses, restricted to antigen-responsive (CD44high) T cells (Fig. 1A), identified exhausted CD8⁺ T cells expressing Tox and Pdcd1 (encoding PD-1) (Fig. 1B) that could be further segregated to T_PEX and T_EX cell subsets (Fig. 1 A–D and SI Appendix, Fig. S1E and Table S1). T_PEX cells were characterized by expression of Tcf7, Slamf6, and Il7r genes associated with self-renewal and memory function (Fig. 1 A–D and SI Appendix, Table S1). T_EX cells lacked Tcf7 and Slamf6 but expressed a range of transcripts encoding coinhibitory receptors, transcription factors, and molecules associated with terminal exhaustion, including Havcr2 (TIM-3), Lag3, Id2, Prdm1 (Blimp1), Cxcr6, Cd38, Gzmb, and Entpd1 (CD39) (Fig. 1 A–D and SI Appendix, Table S1). Both the subsets expressed Tox, Pdcd1, Maf, Eomes, and Tigit (Fig. 1 A–D and SI Appendix, Table S1). We next investigated whether the islet infiltrating exhausted CD8⁺ T cells expressed the classical exhaustion signatures. Gene set enrichment analysis using a core set of genes differentially expressed between T_PEX and T_EX cells in both chronic lymphocytic choriomeningitis (LCMV) infection and the B16-OVA model of tumor immunity (11) showed that both our T_PEX and T_EX cell clusters were enriched for expression of genes upregulated in classical T_PEX and T_EX cell populations (Fig. 1E and SI Appendix, Tables S1 and S2). To validate the subsets of exhausted CD8+ T cells identified by sc-RNA-seq, we analyzed islet-infiltrating CD8+ T cells from 12- to 15-wk-old NOD mice by flow cytometry. In agreement with our transcriptomic data (Fig. 1 A–D) and previous studies (8, 9) TCF-1 (Tcf7) and TOX staining identified two subsets of exhausted T cells namely, TOX+ TCF-1+ T_PEX cells and TOX+ TCF-1- T_EX cells (Fig. 1F). Consistent with our sc-RNA-seq results both T_PEX and T_EX cell subsets showed similar expression of PD-1 and TIGIT, whereas TIM-3 and CD39 were highly expressed in T_EX cells (Fig. 1F). As previously reported, Slamf6 and TCF-1 were coexpressed and thus Slamf6 could be used as a surrogate of TCF-1 (11) (SI Appendix, Fig. S1F). Thus, T_PEX (PD1⁺ Slamf6⁺ TIM-3⁻ and CD39⁻) and T_EX (PD1⁺ Slamf6⁻ TIM-3⁺ and CD39⁺) subsets of exhausted T cells could be reliably identified by their expression of unique cell-surface molecules. Notably, among the exhausted T cells, T_PEX cells were the predominant population (Fig. 1 A and B), which was surprising because T_PEX cells are predominantly found in lymphoid tissues in chronic viral infection and in tumors. T_EX cells preferentially reside in nonlymphoid tissues (8).

Fig. 1.

Islet antigen–specific T cells undergo an exhaustion program in NOD mice. (A) UMAP plot of islet infiltrating CD44+ CD8+ T cells in 16-wk-old NOD mice showing short-lived effector (SLEC), central memory (Tcm), effector memory (Tem), precursor of exhausted (T_PEX), and terminal exhausted (T_EX) subsets of CD8+ T cells. (B) Expression of indicated genes in individual cells from (A). (C) Dot plot visualization of CD8 + T cell subsets in (A) as indicated on the y-axis and genes (features) defining the clusters are listed on the x-axis. Dot size indicates percentage of cells in a cluster expressing each gene; dot color reflects expression level. (D) Volcano plot showing differential gene expression between T_PEX and T_EX cell subsets with Log₂ fold change (FC) on the x-axis and adjusted P values on the y axis (P_adj < 0.05, Wilcoxson rank sum test). Key genes defining cell populations are indicated. NS = not significant. (E) Gene set enrichment analysis of a signature of islet infiltrating T_PEX and T_EX CD8+ T cells from NOD mice using a core set of genes differentially expressed between T_PEX and T_EX cells in both chronic LCMV infection and the B16-OVA model of tumor immunity. (F) Flow cytometric analyses of exhausted CD8+ T cells from the islets of unmanipulated NOD mice. Contour plot (Left) shows gating of CD8+TOX+ TCF1+ T_PEX cells (blue box), CD8+TOX+ TCF1− T_EX cells (red box) and CD8+ TOX− TCF1 (intermediate) nonexhausted cells (black box), numbers in the contour plot indicate the frequency of each subset. Histograms (Right) showing the expression of indicated genes associated with the exhausted T cell subsets. Numbers next to each histogram profile indicate mean fluorescence intensity (MFI). Data representative of n = 4 mice. (G) Expression of Klrg1 and Tox in individual cells from (A).

In addition to exhausted T cells, we also found T cells with memory characteristics that lacked expression of Tox and transcripts of coinhibitory receptors. These T cells included central memory cells expressing Tcf7, Il7r, and Sell (encoding CD62L), and an effector memory subset lacking expression of Sell (Fig. 1 A–C and SI Appendix, Table S3). Furthermore, we identified a cluster that was enriched for Klrg1-expressing cells (short-lived effector cells, SLEC), which did not overlap with Tox-expressing exhausted cells (Fig. 1 A–C and G and SI Appendix, Table S3). Genes associated with cytotoxicity, including Gzmb and Prf1 (encoding Perforin), were expressed in both the T_EX cell and SLEC clusters (Fig. 1C and SI Appendix, Table S3). Thus, T cells that have entered an exhaustion program comprised a large fraction of islet-infiltrating CD8⁺ T cells.

IGRP-Specific T Cells Undergo Exhaustion but Are Maintained as T_PEX Cells in NOD Mice.

We next analyzed IGRP-specific T cells from islets, pancreatic lymph nodes (PLN), and PLO (inguinal and mesenteric lymph nodes, and spleen) of NOD mice by magnetic bead enrichment and tetramer staining (SI Appendix, Fig. S2). We found that all the IGRP-specific T cells, whether isolated from islets, PLN or PLO, expressed TOX and could thus be classified as exhausted T cells. The expression of TOX was highest in IGRP-specific T cells isolated from the islets followed by PLN and PLO (Fig. 2A). We detected high levels of the inhibitory receptors PD-1, TIM-3, and TIGIT on islet-infiltrating IGRP-specific T cells (Fig. 2 B and C). Based on our sc-RNA-seq data (Fig. 1 C, D, and G), differential Slamf6 and TIM-3 staining was used to identify T_PEX and T_EX subsets of IGRP-specific T cells. About 60 to 80% of the IGRP-specific T cells in the islets displayed a Slamf6⁺TIM-3⁻ T_PEX cell phenotype (Fig. 2D), while the remaining 20 to 40% of the cells were TIM-3^hi T_EX cells.

Fig. 2.

Precursor of exhausted and terminal exhausted antigen-specific T cells are found in islets of NOD mice. IGRP-specific T cells and naive CD8+ T cells from islets and PLO (nondraining lymph nodes, and spleen) of 16- to 18-wk-old NOD mice were analyzed by flow cytometry after tetramer staining and magnetic bead enrichment. (A) Representative histogram and MFI of TOX expression in IGRP-specific T cells from islets, PLN and PLO as compared to naive CD8+ T cells in the PLO. (B) Representative histograms showing the expression of indicated coinhibitory receptors and (C) quantification of MFI in the islets and PLO of NOD mice. (D) Proportion of the Slamf6^hiTIM3^lo precursor of exhausted and Slamf6^loTIM3^hi terminal exhausted IGRP-specific CD8+ T cells from PLO and islets of 16- to 18-wk-old NOD mice. (E) Absolute number of IGRP-specific T cells in islets (open circle) and PLO (closed circle) of individual NOD mice. Pooled data (mean ± SEM) in (B–D) from three independent experiments, each symbol in the scatter plots represents data from an individual mouse. Values in the FACS plots (D) show percentages. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 calculated using one-way ANOVA with Tukey’s multiple comparisons test (C), two-tailed unpaired t test (D), and paired t-test (E).

In the PLN, IGRP-specific T cells were TOX^hiPD-1^hi. Most expressed Slamf-6 and TCF-1, only about 5 to 10% expressed TIM-3 (SI Appendix, Fig. S3A). Therefore, IGRP-specific T cells in the PLN were mainly T_PEX cells. Consistent with our previous observation that antigen-experienced IGRP-specific T cells in the PLO of NOD mice are derived from islet-infiltrating cells (25), IGRP-specific T cells in other lymphoid organs resembled the phenotype of T cells in the PLN. While the expression of PD-1 was lower on IGRP-specific T cells in the PLO compared to the islets (Fig. 2 B and C), it was greater than on naive or CD44^hi, KLRG-1^hi effector CD8⁺ T cells not specific for IGRP (SI Appendix, Fig. S3B). We examined the absolute number of IGRP-specific T cells in the islets vs. PLO. We found that there were at least 10-fold higher IGRP-specific T cells in the PLO (Fig. 2E). Thus, IGRP-specific T cells in NOD mice undergo an exhaustion program within the islets where they are predominantly maintained as self-renewing T_PEX cells and they recirculate in the PLO.

Temporal Expression of IGRP in the APCs of NOD Mice.

The majority of IGRP-specific T cells reside in PLO of NOD mice. We therefore exposed IGRP-specific T cells to IGRP in PLO and observed the effect on IGRP-specific T cells and diabetes. We used transgenic NOD mice in which IGRP is expressed under the control of the MHCII (I-Eακ) promoter and the expression of IGRP in APCs is prevented by exposure to doxycycline (26) (Fig. 3A), termed Tetracycline Inhibited IGRP (NOD-TII) mice. NOD-TII mice were induced to express IGRP at 10 wk of age. At 12, 14, and 16 wk, spleens were harvested for longitudinal analysis of transgenic IGRP expression via qPCR (Fig. 3 B and C). These were compared to NOD-TII mice that expressed IGRP from birth (never given doxycycline, these mice were called NOD IGRP mice) and those that never expressed IGRP (doxycycline from gestation). There was no difference in IGRP expression in the APCs from spleens of 10-wk-old transgenic NOD-TII mice that were never expressed IGRP (doxycycline from gestation) and wild-type NOD mice, indicating that the IGRP expression was not “leaky” in the transgenic NOD-TII mice (SI Appendix, Fig. S3C). From here on, this cohort is called “NOD control.” IGRP expression was significantly higher in 12-wk-old NOD-TII mice after doxycycline cessation at 10 wk compared with NOD control mice, and this transcriptional level was maintained (Fig. 3C). However, this expression was lower than when IGRP was induced from birth (NOD-IGRP mice, Fig. 3C). To confirm IGRP antigen expression, we CFSE (carboxyfluorescein succinimidyl ester)-labeled T cells from NOD8.3 TCR transgenic mice that have ∼90% of CD8⁺ T cells specific for IGRP_206–214 and transferred them into NOD-TII and NOD control mice. As expected, in NOD control mice, transferred NOD8.3 T cells proliferated only in the islets where native IGRP is found and in the draining PLN (Fig. 3D). In contrast, extensive proliferation of the transferred NOD8.3 T cells in NOD-TII mice was detected in both the PLN and nondraining lymph nodes confirming that IGRP is expressed by the APCs of NOD-TII mice.

Fig. 3.

Ongoing antigen exposure induces markers of T cell exhaustion on IGRP-specific T cells. (A) Scheme of generation of tetracycline regulated NOD-IEα-tTA (TA-NOD) and TetO-IGRP dual transgenic mice (NOD-TII mice). (B) Study design depicting treatment groups. Continuous IGRP expression in the absence of doxycycline (Dox) shown in red (NOD.IGRP mice), no IGRP expression while always on Dox in white (control NOD mice) and IGRP expression induced after removal of Dox at 10 wk of age shown in orange (NOD-TII mice). (C) Quantitative RT-PCR for G6pc2 in splenic lysates of NOD.IGRP, NOD, and NOD-TII mice at indicated ages. (D) CFSE-labeled CD8+ T cells from NOD8.3 mice were transferred into NOD recipients at 10 to 12 wk of age and into NOD-TII recipients 6 wk after stopping Dox treatment to induce IGRP expression. CFSE-labeled cells were analyzed in the islets, PLN and ILN 5 d post-transfer. Data are representative of two independent experiments. Numbers in each histogram indicate percentage of CFSE low cells. (E and F) IGRP-specific T cells were stained with Kd-IGRP206-214 tetramer and enriched from pooled PLO of 16- to 18-wk-old NOD and NOD-TII mice using magnetic beads and enumerated by flow cytometry. IGRP-specific T cells were analyzed in dispersed islets by tetramer staining without enrichment. (E) Representative FACS plots and (F) quantification of the absolute number of IGRP-specific CD8+ CD44hi T cells after enrichment from PLO. Values in the FACS plots indicate the absolute number of tetramer-binding cells. (G) Representative histograms and (H) MFI quantification of PD-1, TIM-3, TIGIT, and TOX on IGRP-specific T cells in islets of indicated mice. (I) Representative histograms and (J) MFI quantification of PD-1, TIM-3, TIGIT, and TOX on IGRP-specific T cells in PLO of indicated mice. (K) Representative FACS plots and (L) frequency of Slamf6 and TIM-3 expressing IGRP-specific T cells in PLO of NOD and NOD-TII mice. (M) Representative FACS plots and (N) frequency of Slamf6 and TIM-3 expressing IGRP-specific T cells in islets of NOD and NOD-TII mice. Numbers in (K and M) are the frequency of cells in each circled population. (O) Representative FACS plots and (P) frequency of KLRG-1 expressing IGRP-specific T cells in PLO of NOD and NOD-TII mice. (Q) Representative FACS plots showing frequency of KLRG-1 expressing cells in TIM-3+ and TIM-3− IGRP-specific T cells in PLO of NOD and NOD-TII mice. Data in the scatter plots show mean ± SEM from individual mice, pooled from two to four independent experiments. Statistical analysis performed using one-way ANOVA with Tukey’s multiple comparisons test (D) or unpaired t-test (H, J, L, N, and P). ns = not significant, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

There was no difference in proliferation of transferred cells in the islets of NOD control or NOD-TII mice, indicating that there is already abundant endogenous IGRP in the islets. The extent of proliferation of transferred NOD8.3 T cells in inguinal lymph nodes (ILN) of NOD-TII mice was less than that seen in islets of either NOD control or NOD-TII mice, indicating that IGRP-specific T cells were stimulated less in the lymph nodes than in islets of NOD-TII mice (Fig. 3D). This was likely to be due to the lower expression of costimulatory markers CD86, CD40, and MHC class II on APC from PLO compared to APCs from islets (SI Appendix, Fig. S3D). Together, these observations confirm the fidelity and robustness of NOD-TII transgenic mice.

Autoantigen Expression in the Periphery Induces T Cell Exhaustion.

As we have previously described (20), IGRP-specific T cells in transgenic NOD-TII mice expressing IGRP from birth were almost completely deleted (Fig. 3E, NOD-IGRP). We next induced IGRP expression from 10 wk of age. Notably, antigen-experienced CD44^hi IGRP-specific T cells in the PLO were detectable at similar frequencies in 20-wk-old NOD-TII mice (induced to express IGRP from 10 wk of age) and NOD control mice (never induced to express IGRP) (Fig. 3 E and F). Therefore, unlike naive T cells, antigen-experienced IGRP-specific T cells are not deleted by exposure to antigen.

Islet infiltrating IGRP-specific T cells in both NOD-TII and NOD control mice uniformly expressed high levels of TOX, PD-1, TIM-3, and TIGIT (Fig. 3 G and H). IGRP-specific T cells in the PLO of NOD-TII mice expressed higher levels of TOX, PD-1, TIM-3, and TIGIT compared to NOD control mice (Fig. 3 I and J). We next compared the subsets of exhausted IGRP-specific T cells from islets of NOD control and NOD-TII mice (Fig. 4 I and J). Substantially larger proportions of IGRP-specific T cells were T_EX cells in the PLO (Fig. 3 K and L) and islets (Fig. 3 M and N) of NOD-TII mice compared to NOD control mice (Fig. 3 K, L, M, and N). Indeed, only about 30% of IGRP-specific T cells were T_PEX cells in the PLO of NOD-TII mice (Fig. 3 K and L). Also, while a subset of IGRP-specific T cells from PLO of NOD mice expressed KLRG-1, this was significantly decreased in NOD TII mice (Fig. 3 O and P). KLRG-1 was predominantly expressed by the T_PEX subset of IGRP-specific T cells in both NOD and NOD TII mice (Fig. 3Q). These results indicate that continuous exposure to autoantigen in the periphery drives T_EX cell differentiation.

Fig. 4.

Exhausted IGRP-specific T cells induce dominant tolerance and reduce diabetes incidence. (A) Representative FACS plots and (B) frequency of CX3CR1 and TIM-3 expressing IGRP-specific T cells in islets (Top) and enriched from PLO (Bottom) of NOD and NOD-TII mice. Panels A and B show cells gated on TIM-3+, IGRP-specific T cells. Numbers in (A) are the frequency of CX3CR1 expressing cells. (C) Representative FACS plots and (D) frequency of IFNγ expressing IGRP-specific T cells analyzed after restimulation with PMA and ionomycin for 4 h. Numbers in (C) are the frequency of IFNγ expressing cells. (E) Study design depicting treatment groups. NOD never expressed IGRP (white), NOD-TII expressed IGRP from 12 wk of age (orange) and NOD-TII transient expressed IGRP from 16 to 20 wk of age (green). (F) MFI quantification of PD-1, TIM-3, and TIGIT on IGRP-specific T cells enriched from PLO of NOD, NOD-TII, and NOD-TII mice with transient IGRP expression at 20 wk of age, analyzed by flow cytometry. (G) Expression of exhaustion-associated genes on IGRP-specific T cells sorted from PLO of NOD, NOD-TII, and NOD-TII mice with transient IGRP expression, assessed by quantitative RT-PCR. Heatmaps showing relative expression normalized to 18srRNA. (H) Incidence of spontaneous diabetes until 300 d of age in cohorts of female NOD mice (transgenic NOD-TII mice given doxycycline from birth to suppress transgenic IGRP expression, open circles), NOD-TII mice expressing IGRP from birth (red circles) and NOD-TII mice expressing IGRP from 10 wk of age (orange circles). Numbers in parentheses indicate the number of mice analyzed. (I) Representative FACS plots and frequency of IL-10-expressing IGRP-specific T cells enriched from PLO of 18- to 20-wk-old NOD and NOD-TII mice and analyzed after restimulation with PMA and ionomycin for 4 h. Numbers in FACS plots are frequency of IL-10-expressing cells. (J) Representative histograms showing expression of CD39 in the indicated CD8+ T cell subsets in islets (Top) and PLO (Bottom) of NOD and NOD-TII mice. Nontetramer-binding CD8+ T cell subsets shown are from NOD mice. (K) MFI quantification of CD39 in IGRP -specific T cells from islets (Top) and PLO (Bottom) of NOD and NOD-TII mice. (L) Representative FACS plots and frequency of CD4⁺FoxP3⁺ Treg cells in PLO of 18- to 20-wk-old NOD and NOD-TII mice. Numbers in FACS plots are frequency of Treg cells. (M) Model of ex vivo suppression assay. (N) Representative histograms showing cell division profiles of CTV labeled responder cells incubated for 72 h with sorted suppressors as indicated and (O) Quantification of suppression by the indicated suppressor populations. Data pooled from two to four independent experiments. Data in the scatter plots show mean ± SEM from individual mice. Statistical analysis performed using the unpaired t-test (B, D, F, I, K, L, and O). Survival curves in (H) compared using the log-rank (Mantel–Cox) test. ns = not significant, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

IGRP-Specific T Cells Are Functionally Disabled When Exposed to Antigen and Recover after Termination of Antigen Exposure.

We next assessed the function of IGRP-specific T cells in NOD TII mice. The pool of TIM-3⁺ T_EX cells can be further dissected based on the expression of the chemokine receptor CX3CR1. While CX3CR1+ cells maintain the highest effector function, loss of CX3CR1 demarcates acquisition of a terminally exhausted state with low effector function and survival (27–29). About 40% of TIM-3⁺ IGRP-specific T cells in the islets expressed CX3CR1 and there was no difference between NOD control and NOD-TII mice (Fig. 4A). However, in the PLO of NOD control mice, the majority of TIM3⁺ cells expressed CX3CR1 while only about 20% of the cells expressed CX3CR1 in NOD-TII mice (Fig. 4 A and B). IGRP-specific T cells from the PLO of NOD-TII mice expressed less IFNγ following stimulation with PMA and Ionomycin in vitro compared with NOD mice (Fig. 4 C and D).

Although most IGRP-specific T cells in NOD-TII mice had phenotypic and functional markers of terminal exhaustion, the number of IGRP-specific T cells was not decreased suggesting that the self-renewal and precursor function of T_PEX cell remained intact.

We induced IGRP expression in NOD-TII mice only for a defined period, between 10 and 15 wk of age (Fig. 4E) to test if exhaustion would remain after antigen removal. Following 5 wk without antigen expression (at 20 wk of age), IGRP-specific cells in the PLO had downregulated expression of inhibitory receptors PD-1, TIGIT, and TIM-3 (Fig. 4F) to a level that was not different to IGRP-specific cells from NOD control mice never exposed to transgenic expression of IGRP. There was also no difference in the number of IGRP-specific T cells in the PLO (SI Appendix, Fig. S3E) or their ability to secrete IFNγ following stimulation with PMA and ionomycin in vitro (SI Appendix, Fig. S3F) at 20 wk. We next sorted IGRP-specific T cells from the PLO of NOD-TII mice and performed RT-PCR for the expression of exhaustion marker genes. The expression of exhaustion-related genes [Pdcd1, Tox, Havcr, Prdm1, Maf, Cd38 (30), Lag3, and Tigit] was similar in IGRP-specific T cells from NOD-TII mice after transient transgene expression and NOD control mice, while it was increased in NOD-TII mice with continuing transgene expression (Fig. 4G). Tcf7, Eomes, and Tbx21 were decreased in NOD-TII mice with current antigen expression and recovered after removal of antigen to the level seen in NOD-TII mice that never expressed antigen (Fig. 4G). Overall, these results demonstrate that T_PEX cells remain functional during periods of high antigen expression and mediate recovery of autoaggressive IGRP-specific T cells following antigen withdrawal.

Exhausted IGRP-Specific T Cells Can Induce Dominant Tolerance and Reduce Diabetes Incidence.

There was no difference in diabetes between transgenic NOD mice that never expressed IGRP or expressed IGRP throughout life (Fig. 4H, red vs. white symbols) consistent with our previous studies indicating that deletion of IGRP-specific T cells had no impact on diabetes (20, 26). Strikingly, however, induction of IGRP expression from week 10 onward led to significant protection from diabetes (Fig. 4H, red vs. orange symbols). This observation suggested that exhausted IGRP-specific T cells can inhibit T cells specific for other beta cell antigens. We tested the inhibitory potential of these T cells of NOD-TII mice in an adoptive transfer model. Splenocytes from prediabetic NOD control mice transferred diabetes efficiently into irradiated NOD recipients (median survival 73 d), however spleen cells from NOD control donors combined at a 1:1 ratio with splenocytes from 18-wk-old NOD-TII mice (expressing IGRP from 10 wk of age), transferred diabetes with delayed kinetics (median survival 126 d) (SI Appendix, Fig. S3G).

T_EX cells upregulate CD39 and IL-10 which are both utilized by regulatory T cells (Tregs) to suppress other T cells (31, 32). IGRP-specific T cells in NOD-TII mice expressed high levels of IL-10 (Fig. 4I) and CD39 (Fig. 4 J and K), while there was no difference in the frequency of FoxP3⁺ Tregs between NOD and NOD-TII mice (Fig. 4L). We tested the suppressive capacity of CD39⁺TIM-3⁺ IGRP-specific T_EX cells in an ex vivo assay. For this, we used NOD8.3 TCR transgenic mice that were crossed with NOD-TII mice (NOD-TII/8.3 mice) and induced to express IGRP for more than 10 wk. We sorted PD1⁺ CD39⁺ TIM-3⁺ (T_EX) and PD1⁺ CD39⁻ TIM-3⁻ (T_PEX) IGRP-specific CD8⁺ T cells from these mice (suppressors) and cocultured them with cell trace violet (CTV) dye-labeled and congenically marked naive IGRP-specific CD8⁺ T cells (responders) from NOD 8.3/CD45.2 mice at a 2:1 ratio. We used T cell depleted spleen cells from wild-type NOD mice loaded with the IGRP _206-214 peptide as stimulators (Fig. 4M). CD39+TIM-3⁺ IGRP-specific CD8⁺ T_EX cells significantly suppressed responder T cells as compared to their T_PEX counterparts (Fig. 4 N and O). Thus, persistent extraislet exposure to cognate antigen drives pathogenic IGRP-specific T cells to terminal exhaustion. IGRP-specific T_EX cells acquire the capacity to suppress islet-specific T cell populations. This attenuates islet autoimmunity.

Discussion

T cells express exhaustion-associated inhibitory receptors due to chronic antigen exposure in the islets. However, the islet microenvironment prevents differentiation to terminal exhaustion and the T cells are maintained in a less differentiated precursor of an exhausted state with retained self-renewal features. These exhausted T cells undergo partial phenotypic and functional reversal upon antigen withdrawal when they leave the islets. The antigen-experienced, islet-reactive T cells undergo differentiation to the terminally exhausted state when exposed to antigen in the extraislet microenvironment. Importantly, inducing differentiation to the terminally exhausted state in T cells specific for a single antigen led to protection from diabetes.

Our results are different from those of a recent study (23) that showed that pancreas-infiltrating IGRP-specific T cells in NOD mice are short-lived effector T cells and do not display a transcriptional or phenotypic TOX-driven exhaustion program. We used tetramer staining of T cells isolated from purified islets and found IGRP-specific T cells uniformly expressed high levels of TOX (Fig. 1C), a critical transcription factor for the induction of CD8⁺ T cell exhaustion (33). The difference in the findings may in part reflect the use of different methods to isolate and analyze islet infiltrating T cells. We first isolate islets from the pancreas and then disperse them to single cells rather than the method used by Gearty to disperse the whole pancreas. This may result in nonislet and blood T cells being included in the analysis.

The predominance of T_PEX cells in the islets contrasts with what is seen in tumors and chronic viral infection (8, 34). In chronic viral infection, the T_PEX cells are mainly present in lymphoid tissues but are rare in the nonlymphoid tissues whereas the terminally exhausted T cells localized to both lymphoid and nonlymphoid tissues. Islet inflammation in NOD mice and human subjects with T1D leads to development of tertiary lymphoid structures (35, 36) and it is possible these structures support the development of T_PEX cells. Cancers have also been shown to develop intratumoral tertiary lymphoid structures through chronic inflammatory signals. The presence of tertiary lymphoid structures in tumors has been linked to higher CD8⁺ T cell infiltration and improved responsiveness to immunotherapy. This suggests that tertiary lymphoid structures in the tumor support development of T_PEX cells (37). Interestingly, our sc-RNA-seq data show high expression of Ccr7 in T_PEX cells in the islets of NOD mice (Fig. 3D). CCR7 is normally expressed in T cells in lymph nodes (e.g., central memory T cells) as compared to T cells in blood (e.g., effector memory T cells) or tissue (e.g., resident memory T cells).

Our data indicate that NOD mice were protected from diabetes by inducing exhaustion in T cells specific for a single antigen, IGRP, at a time when there is an immune response against multiple antigens. This suggests a dominant mechanism, i.e., that the exhausted IGRP-specific T cells can inhibit T cells specific for other islet antigens. CD39 (Entpd1) is expressed in terminally exhausted T cells (38) and acts via catabolizing ATP and ADP to AMP and further degradation to adenosine via CD73. CD39 expressing T_EX but not T_PEX cells suppressed T cell proliferation. Recent studies have shown that tumor-derived T_EX cells are able to suppress proliferation of T cells in a CD39-dependent manner (31) and administration of soluble CD39 delayed the onset of diabetes in NOD mice (39). Several reports have identified that IL-10-secreting CD8⁺ T cells play a regulatory role in protecting against autoimmune disease (32, 40, 41). Conversely, blocking of IL-10 signaling improved the function of exhausted T cells in chronic viral infections (42) and simultaneous blockade of IL-10 and PD-1 pathways resulted in elimination of persistent viral infection (43). It is possible that these IL-10-secreting IGRP-specific T_EX cells also contributed to protection of NOD-TII mice from diabetes. Previous studies have shown that antigen delivery via nanoparticles coated with IGRP peptide bound to either MHC class 1 or class II molecules protected NOD mice from diabetes by inducing IL-10-secreting T cells (44, 45).

We observed that IGRP-specific T cells partially lose their exhaustion-associated inhibitory markers as they traffic away from the site of the antigen. This is similar to exhausted hepatitis C virus (HCV)-specific CD8⁺ T cells from infected patients recovering memory-like features and losing features of exhaustion after elimination of virus with direct-acting antiviral therapy. Recovery from exhaustion has also been seen in chronic LCMV and CAR T cell models of exhaustion (46). Analysis of TCR diversity of HCV-specific T cells following HCV cure suggests that transcriptional and epigenetic reprogramming may play a role in reversal of the exhaustion-associated phenotype (46). Future studies will need to determine whether the recovery of phenotypic exhaustion and memory-like features that we observed following antigen removal are due to dedifferentiation of terminally exhausted T cells or exclusively due to outgrowth of new progeny from the T_PEX cell population.

The exhaustion program of T cells can be beneficial in restraining T cells in autoimmunity. Examples include patients with thyroid or islet antibodies who do not progress to clinical disease. These individuals rapidly develop autoimmune disease following checkpoint inhibitor treatment (17, 47). Maintenance of the ability to destroy target tissue despite chronic antigen exposure occurs in autoimmunity and by definition T cell exhaustion as a tolerance mechanism is ineffective in people with autoimmune diseases. In tumors, there is a progressive increase in exposure to antigen as the tumor grows, and the T cells progressively become more exhausted with attrition of T_PEX cells (48). Studies in the LCMV model of T cell exhaustion showed that an increased amount of antigen can drive more severe T cell exhaustion. Moreover, the antigen exposure in LCMV infection is systemic rather than organ specific as seen in T1D and other autoimmune diseases. In the immune response against the beta cells, the amount of target tissue is fixed or declining (49), unlike in diseases such as cancer and viral infections in which the antigens have the capacity to expand. In tumors and chronic viral infections, a small population of T_PEX cells maintains the long-term immune response and generates short-lived T_EX cells with effector function. In autoimmunity, T_PEX cells are abundant and hence it is easy to maintain the immune response. The T_PEX cells differentiate into CX3CR1⁺ transitory T_EX cells with effector function or into CD39⁺TIM3⁺ T_EX cells with regulatory functions (27). The transitory effector-like CX3CR1-expressing CD8⁺ T cell subset is required for viral control and the differentiation of this population depends on CD4⁺ T cell provision of IL-21 (28). In addition, IL-21 is also important for maintaining T_PEX cells (50)_. We have shown that widespread antigen exposure induces T_EX cells with regulatory function. The T cells are exposed to a fixed and limited amount of transgenic antigen in our model. Modulating the amount of antigen in the PLO could have an impact on the proportion of precursor vs. terminally exhausted T cells. Inducing T cell exhaustion by T cell–directed treatment has been associated with slower progression of T1D (24, 51–53). T cell–directed intervention either with antigens or anti-CD3 monoclonal antibody could be combined with IL-21 blockade to control T_PEX cells and T_EX cells with effector functions. This approach could be applied to other autoimmune diseases in which T cell exhaustion may be important (15).

In summary, we found that T cells undergo an exhaustion program in NOD mice. However, this process is not complete as these cells persist mainly as precursors of exhausted T cells. Antigen exposure in the extraislet environment induced terminal differentiation of T cells. Terminally exhausted T cells specific for a single antigen led to partial protection from diabetes. The predominance of T_PEX cells in the islets implies that factors within the islet support them and their elimination from that site is the highest priority.

Materials and Methods

Mice.

NOD/Lt mice were bred and housed at the Bioresources Centre, St. Vincent’s Hospital, Fitzroy. NOD-IEα-tTA mice that drive the expression of tetracycline transactivator (tTA) under control of the MHC class II IEα promoter have been previously described (54) and were obtained from C. Benoist and D. Mathis (Dept of Pathology, Harvard Medical School, Boston, Massachusetts, USA). TetO-IGRP transgenic mice were crossed with NOD-IEα-tTA mice to generate dual transgenic NOD-TII mice as previously described (26). NOD8.3 mice, expressing the TCRαβ rearrangements of the H-2Kd-restricted, β cell-reactive, CD8+ T cell clone NY8.3, have been previously described (55). All mice were bred, maintained, and used under specific pathogen-free conditions at St. Vincent’s Institute (Melbourne, Australia). All experimental procedures followed the guidelines approved by the institutional animal ethics committee.

Doxycycline Treatment.

Untreated NOD-TII mice constitutively express IGRP in APCs. To turn off IGRP expression, doxycycline hyclate (Dox) (Sigma-Aldrich) was administered via drinking water at a concentration of 10 mg/L. Water bottles were changed thrice weekly. NOD-TII mice receiving water with Dox were then given unmedicated water to re-express IGRP.

Real-time PCR (qPCR).

For total RNA extraction, spleens were harvested in cold Phosphate Buffered Saline. Tissue homogenates were prepared in RNA lysis buffer RLY1 (Bioline) from a 15-mg slice of tissue using a tissue homogenizer. RNA was isolated using the ISOLATE II RNA Mini Kit (Bioline) according to the manufacturer’s instructions. Eluted RNA was further purified with the TURBO DNA-free™ Kit (Invitrogen). A NanoDrop 2000 UV-Vis Spectrophotometer was used to verify RNA content. Complementary DNA synthesis was performed using the High-Capacity cDNA Reverse Transcription Kit with RNase Inhibitor (Applied Biosystems). AmpliTaq Gold Taqman Assay (Applied Biosystems) was performed in a Roche LightCycler 480 Instrument II. Taqman gene expression probes used in the study are listed in SI Appendix, Table S4. Relative expression of genes of interest was determined by normalization to indicated reference genes using differential threshold cycle (ΔCT, $2^{{-\mathrm{\Delta }\mathrm{\Delta }\mathrm{C}}_T}$ ) analysis.

CFSE Labeling and Adoptive Transfer.

Single-cell suspensions of NOD 8.3 splenocytes were prepared and labeled with CFSE (Thermo Fisher) as previously described (20). First, 5 × 10⁶ CFSE-labeled cells were injected intravenously into tail veins of NOD-TII mice with or without induced IGRP expression. Then, hosts were killed after 5 d, and their ILN, PLN, and pancreatic islets were examined for CFSE+ cells by flow cytometry.

Islet Isolation.

Islets were isolated using collagenase P (Roche, Basel, Switzerland) and Histopaque-1077 density gradients (Sigma-Aldrich) as previously described (56). Islets were dissociated via bovine trypsin (Calbiochem) hydrolysis and single-cell suspensions were processed for flow cytometry.

Flow Cytometry.

Single-cell suspensions were prepared from pooled PLO (spleen and nondraining lymph nodes, PLO) and islets. Cell suspensions from PLO were treated with ammonium chloride buffer to lyse red blood cells. For surface staining, cells were stained with fluorescently labeled antibodies on ice for 30 min. All antibodies used are listed in SI Appendix, Table S5. Propidium Iodide (PI, Calbiochem) was added prior to flow cytometry analysis to exclude dead cells. IFNγ and IL-10 were detected intracellularly using the Cytofix/Cytoperm Kit (BD Biosciences) following incubation with PMA and Ionomycin for 4 h at 37 °C. Intracellular staining for TOX, TCF-1, and FoxP3 was performed using the FoxP3/Transcription Fixation/Permeabilization kit (eBiosciences). Data were collected with a LSR Fortessa (Becton Dickinson) or Cytek Aurora (Millennium science) flow cytometer and analyzed with FlowJo (v10.8.1) (Tree Star, Ashland, USA) software.

Tetramer Staining and Magnetic Bead-Based Enrichment.

The tetramer and magnetic bead-based enrichment assay has been previously described (25). Briefly, single-cell suspensions from PLO (pooled spleen and nondraining lymph nodes comprising of two inguinal and four mesenteric lymph nodes per sample) were stained with phycoerythrin (PE)-conjugated IGRP_204-216 (VYLKTNVFL) H2-Kd tetramer (ImmunoID, Parkville, Victoria, Australia), for 1 h on ice, washed and incubated with anti-PE magnetic beads (Miltenyi Biotec, Cologne, Germany) followed by magnetic separation using an AutoMACSpro (Miltenyi Biotec) according to manufacturer’s instructions. The tetramer-enriched fractions were stained with cell-surface markers and analyzed by flow cytometry. Gating strategy for tetramer enrichment was as follows: Single cells were gated on forward and side scatter, and dead cells excluded using propidium iodide. From the live cell population, CD3⁺ dump⁻ (dump = CD11c, CD11b, B220 and F4/80) cells were gated as the T cell population for analysis of tetramer-binding CD4⁻ CD8⁺ T cells (SI Appendix, Fig. S2).

Cell Sorting, Library Preparation, and Sequencing for sc-RNA-Seq.

For sc-RNA-seq analysis, live CD45⁺ cells were FACS sorted from dispersed islets pooled from 2 to 3 NOD mice (15 to 16 wk of age) per sample. Sorted CD45⁺ cells were washed and resuspended in RPMI cell culture medium (Gibco) containing 10% FCS at a density of 1,200 cells/μL and ~20,000 cells were loaded on to Chromium Controller (10× Genomics). 3 samples from two independent experiments were processed further using the Chromium Single cell 3′ Gel bead kit (v 3.1) and library construction kit as per the manufacturer’s instructions. The libraries were quantified using the Agilent Bioanalyzer High sensitivity Chip and sequenced on the Illumina NovaSeq PE150 platform (Novogene AIT Genomics, Singapore).

Single-Cell RNA-Seq Dataset Alignment and Clustering.

Sequencing files were demultiplexed and aligned to Mus musculus transcriptome reference (mm10), and count matrices were extracted using Cell Ranger software v6.0.1 (10× Genomics). The expression matrices were then imported into R (v4.0.4) and processed with Seurat (v4.0.0). Following cell-cycle regression, doublet removal and filtering of cells based on counts, genes, and percentage of mitochondrial genes, the remaining cells in each sample were normalized and integrated by SCTransform to correct for batch effects. Dimensionality reduction was performed by unsupervised principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) embedding for each sample. 22,616 CD45⁺ cells from three samples were reclustered using a K-nearest neighbor graph followed by Louvain clustering using 12 PCs and a resolution of 1.2. The resulting cell clusters were visualized using t-distributed stochastic neighbor embedding and UMAP. Cell-type meta-clusters were annotated using a list of predefined marker genes and the FindMarkers function to identify cell-type specific gene signatures within each cluster (Supplementary figure S1). A subset of 16,585 cells identified as T cells were reclustered using 14 PCs and a resolution of 1.5. Cd8a and Cd4 expression was used to identify islet CD8⁺ and CD4⁺ T cells. Finally, 4,520 CD8⁺ T cells were reclustered using 14 PCs and a resolution of 1.2 and a predetermined set of genes was used to annotate CD8⁺ T cell subsets. PCA cut-offs were established by determining the number of PCs accounting for 90% of the variance.

Diabetes Incidence.

Female mice were monitored for spontaneous diabetes over a 300-d time course. Diabetes onset was monitored by weekly measurement of urine glucose levels using Diastix (Bayer Diagnostics). Blood glucose levels were measured in mice with glycosuria (>110 mmol/L) using Advantage II Glucose strips (Roche). Animals displaying two consecutive blood glucose measurements of ≥15 mmol/L were considered diabetic. For adoptive transfer of diabetes, 2 × 10⁷ splenocytes from 12 to 15 wk old prediabetic NOD mice were transferred (i.v) alone or cotransferred at a 1:1 ratio with splenocytes from 18-wk old NOD-TII mice into irradiated (9 Gy) 10 wk old NOD recipients and diabetes development was monitored as above.

Ex Vivo T Cell Suppression Assay.

Splenic CD8⁺ CD4⁻ from NOD 8.3/CD45.2 TCR transgenic mice were sorted as responder cells using magnetic beads and labeled with CTV dye (5uM, Invitrogen). CD4⁻ CD8⁻ spleen cells isolated from wild-type NOD mice were coated with 1 μM IGRP _206-214 peptide (VYLKTNVFL) and used as APCs. CD4⁻ CD8⁺ PD1⁺ TIM3⁺CD39⁺ (T_EX) and CD4⁻ CD8⁺ PD1⁺TIM3⁻CD39⁻ (T_PEX) subsets were flow sorted from spleens of NOD-TII/8.3 TCR transgenic mice expressing IGRP and used as suppressors. Responders (2-2.5 × 10⁴) and suppressors (4-5 × 10⁴) (1:2 ratio) were activated with peptide loaded APCs (1 × 10⁵) in a 96-well U-bottom tissue culture plate with 200 μL complete RPMI per well for 72 h. Proliferation of CTV-labeled responders was determined by flow cytometry after 72 h. Suppression was calculated using the formula, % Suppression = (% Proliferation unsuppressed responders − % Proliferation suppressed responders/% Proliferation unsuppressed responders) × 100.

Quantification and Statistical Analysis.

All statistical analyses were performed using GraphPad Prism 9 software (GraphPad, San Diego, USA). A two-tailed unpaired Student’s t-test was used for comparisons between two groups. Multiple comparisons were performed using one-way ANOVA with Tukey’s post hoc test. Diabetes incidence curves were compared using the log-rank (Mantel–Cox) test. In all graphs, each symbol represents an individual sample, and the error bars represent the mean ± SEM. ^∗P < 0.05, ^∗∗P < 0.01, ^∗∗∗P < 0.001.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

RNAseq data have been deposited in NCBI repository (GSE247956) (57). All other data are included in the manuscript and/or SI Appendix.