Immune checkpoints in a genetically engineered mouse model of spontaneous autoimmune uveitis

Abstract

Engineered mutant Aire^GW/+Lyn^−/− C57BL/6 mice are a model of spontaneous autoimmunity in which 50% of mice develop highly destructive uveitis due to compromised central and peripheral T cell tolerance. The key retinal autoantigen in these mice is interphotoreceptor retinoid-binding protein (IRBP). The CD4⁺ T cells recognizing the dominant epitope of IRBP, called P2, from eye-draining lymph nodes of mice with or without uveitis and from the retinas of mice with uveitis, were characterized by single-cell RNA sequencing (scRNA-seq) and flow cytometry. Mice with uveitis had autoantigen-specific T cells in the eye-draining lymph node and in the retinas, exhibiting a range of activation states and a Th1 polarization, along with a small fraction of Tregs. Mice without uveitis had low numbers of P2-specific T cells in the eye-draining lymph nodes, and in most mice a substantial proportion of them were FoxP3⁺ Tregs. Transient depletion of Tregs by treatment of heathy Aire^GW/+Lyn^−/− Foxp3DTR^+/Y mice with diphtheria toxin resulted in rapid expansion of P2-specific CD4⁺ T cells in the eye-draining LN, and some of the Treg depleted mice went on to develop uveitis. Thus, development of inflammation in the retina was limited by a checkpoint in the eye-draining lymph nodes involving Tregs, and also apparently by an additional peripheral tolerogenic mechanism.

Introduction

Although human autoimmune disease is seen in individuals with loss-of-function mutations in genes such as FoxP3 and Aire,^1–3 which control critical immune tolerance processes, much more commonly individuals with autoimmune disease have variant alleles in multiple genes, each of which contributes a small degree of genetic susceptibility.^4,5 This multigenic susceptibility to autoimmune disease typically is not fully penetrant in genetically identical twins. Thus, a central question related to autoimmune disease is what determines whether or not an individual with genetic susceptibility develops disease. In a few diseases, a particular infection has been shown to trigger an autoimmune attack in susceptible individuals,⁶ but causal triggers have not been established for the large majority of autoimmune diseases.

Animal models of spontaneous autoimmune disease with incomplete penetrance offer an opportunity to address this question. Of such animal models, the non-obese diabetic (NOD) mouse model has been especially well studied. In this model, in which 20 different genetic loci contribute to disease susceptibility,⁷ all young mice develop an autoimmune inflammation in the pancreas adjacent to the islets.⁸ Subsequently, some individuals progress to inflammatory invasion of the islets, destruction of insulin-producing pancreatic β cells, and eventual development of diabetes. Thus, the main checkpoint determining the development of diabetes or not in NOD mice is thought to involve the progression, or not, of a destructive response of autoimmune T cells in the target organ, but the reasons for variable progress through this checkpoint in different individuals are not well established.

Recently, we have created a new mouse model of spontaneous autoimmune disease with incomplete penetrance by combining engineered defects in 2 genes participating in different immune tolerance checkpoints,⁹ namely Aire, which promotes the expression of tissue-restricted self-proteins in the thymus to induce central tolerance of T cells,¹⁰ and Lyn, an intracellular protein tyrosine kinase that promotes the function of inhibitory receptors in dendritic cells and therefore restrains the T cell-activating ability of dendritic cells in lymph nodes (LN).^11,12 We used a genetically dominant, partial loss-of-function Aire mutation (G228W, here referred to as Aire^GW), identified in an Italian family with dominantly inherited autoimmunity, and introduced the corresponding mutation into the Aire gene of mice. This Aire mutation promoted only mild inflammatory disease in lacrimal and salivary glands in C57BL/6 mice but induced more severe autoimmunity when introduced into NOD mice.¹³ Deletion of Lyn by itself does not cause any organ-specific autoimmunity, but does lead to a lupus-like disease starting at about 3 mo of age due to Lyn’s inhibitory role in dendritic cells and in B cells.^14,15

Combination of the Aire^GW mutation with a Lyn loss-of-function mutation in C57BL/6 mice (Aire^GW/+ Lyn^−/− mice) resulted in highly destructive bilaterally symmetric autoimmune eye disease in 50% of mice, demonstrating that these 2 genetic defects compromising distinct immune tolerance mechanisms, synergistically conferred a strong autoimmune disease susceptibility.⁹ The resulting retinitis, also referred to as uveitis, because inflammation is also evident in the surrounding uveal tissue, was caused by CD4 T cells recognizing a protein expressed only in retinal rod and cone cells, interphotoreceptor retinoid binding protein (IRBP).⁹ Genetic removal of IRBP prevented disease onset.¹⁶ In addition to being expressed in retinal light sensing cells, IRBP is expressed by medullary epithelial cells in the thymus under the control of Aire, and this expression promotes negative selection and Treg development of IRBP-reactive CD4 T cells.^13,16 Aire-induced expression of IRBP resulted in negative selection of IRBP-specific CD4 T cells, especially those recognizing amino acids 271–290 of IRBP; referred to as P2.¹³ Whereas CD4 T cells in the retinas of mice with uveitis also recognized at least 2 other epitopes of IRBP presented by I-A^b class II MHC molecules, CD4 T cells recognizing P2 were more frequent than T cells recognizing the other two epitopes.⁹ Fifty percent of the Aire^GW/+ Lyn^−/− mice first developed uveitis at 6–9 wk of age, whereas the remaining 50% of mice showed no signs of inflammation in the retina then or later in life, but did have a small number of IRBP P2-specific CD4 T cells present in the eye-draining LN.⁹

Thus, Aire^GW/+ Lyn^−/− mice exhibit a multigenic autoimmune susceptibility resulting from dysfunction in 2 distinct elements involved in self-tolerance of T cells, but it is unclear what is the nature of the CD4 T cell autoimmune response in the mice that develop uveitis, and conversely what immune tolerance checkpoint(s) continue to operate in the mice that fail to develop uveitis. In this study, we have addressed these questions by comparing the numbers and characteristics of IRBP P2-specific CD4 T cells from mice with uveitis to those from animals without uveitis. The autoimmune CD4 T cells in the eye-draining LN of mice with or without uveitis similarly consisted of cells apparently representing different stages in the activation program directed toward Th1 effector cells expressing CXCR3. Immediately prior to the onset of uveitis, variations between individual mice were seen in the numbers of autoimmune CD4 T cells and in the fraction of them that were FoxP3⁺ Tregs. A minority of the mice had very few CD4 T cells with this specificity in their eye-draining LNs, perhaps explaining why a subset of the mice failed to develop uveitis. Among the remaining mice, there were considerable variations in the frequency of the P2-specific CD4 T cells that were FoxP3⁺ Tregs. Moreover, depletion of Tregs in 6 wk-old mice without uveitis induced a substantial expansion of IRBP P2-specific CD4 T cells in the eye-draining LN of most of these mice, demonstrating an ongoing role of Tregs in suppressing the proliferation of autoreactive T cells in the draining LN of mice without uveitis. Remarkably, transient Tregs depletion triggered uveitis in about 25% of the mice, demonstrating a critical role of Tregs in preventing autoimmune uveitis in these mice and also implying that at least 1 additional immune tolerance checkpoint was operating to prevent the expanding autoreactive CD4 T cells from initiating damage in the retina.

Materials and methods

Mice

Aire^GW/+ and Lyn^−/− mice of this study were previously described.^9,16 B6.129P2-Rbp3 tm1Gil/J(IRBP^−/−) (stock no. 023080)¹⁷ and C.Cg-Foxp3^tm2Tch/J(Foxp3^GFP) (stock no 006769) mice were obtained from the Jackson Laboratory. B6.129(Cg)-Foxp3 tm3(DTR/GFP)Ayr/J (Foxp3^DTR) mice were from Dr Michael D. Rosenblum’s lab at UCSF. All animal experiments were approved by the UCSF Animal Care and Use Committee.

Fundoscopy

Fundoscopy was performed by using a Micron III camera (Phoenix Research Labs Inc.), as previously described.⁹ Mice were anesthetized with isoflurane during imaging. Tropicamide ophthalmic 1% (Sigma Pharmaceuticals, LLC, 3035-1) and Phenylephrine hydrochloride 2.5% (Sigma Pharmaceuticals, LLC, 3020-1) eye drops were used to relax eye muscles and dilate the pupil. To avoid dryness of the cornea, GenTeal^® Gel (Sigma Pharmaceuticals, LLC, 5009-1) and TheraTears (Sigma Pharmaceuticals, LLC, 5035) were used during imaging. The presence and severity of uveitis were determined in Aire^GW/+Lyn^−/−, Aire^GW/+Lyn^−/−Foxp3^GFP, Lyn^−/−Foxp3^GFP, Lyn^−/−Foxp3^DTR and Aire^GW/+Lyn^−/−Foxp3^DTR mice according to previously described grading system.¹⁸

Isolation of lymphocytes from mouse retina

Briefly, both retinas of individual mice were dissected and transferred to digestion buffer containing 90% RPMI 1640 medium and 10% FBS. Retinas were minced with scissors and incubated in digestion buffer containing 100 U/ml DNase I (Roche Diagnostics Deutschland GmbH, 10104159001) and 0.5 mg/ml collagenase D (Roche Diagnostics Deutschland GmbH, 11088858001) for 1 h at 37 °C. For further analysis, lymphocytes of the retina were purified by centrifugation through a 30%/37%/70% Percoll (Cytiva, 17-5445-02) step gradient.

Mass cytometry antibodies

Antibodies for mass cytometry were prepared using a MaxPAR antibody conjugation kit (Fluidigm, South San Francisco, California, USA) according to the manufacturer’s recommended protocol sourcing metals from Fluidigm (Fluidigm, South San Francisco, California, USA) or Trace Sciences International (Richmond Hill, Canada). Conjugated antibodies were diluted in Candor PBS Antibody Stabilization solution (Candor Bioscience GmbH, Wangen, Germany) supplemented with 0.02% NaN₃ to between 0.1 and 0.3 mg/ml and stored long-term at 4 °C. Each antibody clone and lot was titrated to optimal staining concentrations using primary mouse splenocytes.

Mass-tag cellular barcoding

Mass cytometry samples were barcoded and pooled together for staining and running on mass cytometer. Mass-tag cellular barcoding was performed as previously described.¹⁹ Briefly, 1 × 10⁶ cells of each sample were incubated with distinct combinations of stable Pd isotopes in 0.02% saponin in PBS. Cells were washed once with cell staining media (PBS with 0.5% BSA and 0.02% NaN₃), and once with 1× PBS, and pooled into a single FACS tube (BD Biosciences, San Jose, California, USA). After data collection, each condition was deconvoluted using a single-cell debarcoding algorithm.¹⁹

Mass cytometry staining and measurement

Previously barcoded cells from multiple samples were resuspended in cell staining media (PBS with 0.5% BSA and 0.02% NaN₃), and antibodies against CD16/32 (BioLegend, San Diego, California, USA) were added at 20 μg/ml for 5 min at RT on a shaker to block Fc receptors. Surface marker antibodies (Table S1) were then added to a 500 µl final reaction volume and stained for 30 min at RT on a shaker. After staining, cells were washed 2 times with cell staining media and then permeabilized with methanol for 10 min at 4 °C. Cells were then washed twice in cell staining media to remove the remaining methanol and stained with intracellular antibodies in 500 µl for 1 h at RT on a shaker. Cells were washed twice in cell staining media and then stained with 1 ml of 1:4000 191/193Ir DNA intercalator (Fluidigm, South San Francisco, California, USA) diluted in PBS with 4% PFA overnight. Cells were then washed once with cell staining media, once with 1× PBS and once with Cell Acquisition Solution (Fluidigm, South San Francisco, California, USA). Mass cytometry samples were diluted in Cell Acquisition Solution containing bead standards (Fluidigm, South San Francisco, California, USA) to approximately 10⁶ cells per ml and then analyzed on a Helios mass cytometer (Fluidigm, South San Francisco, California, USA) equilibrated with Cell Acquisition Solution.

Mass cytometry bead standard data normalization

Data normalization was performed as previously described.²⁰ Briefly, just before analysis, the stained and intercalated cell pellet was resuspended in freshly prepared Cell Acquisition Solution containing the bead standard at a concentration ranging between 1 and 2 × 10⁴ beads/ml. The mixture of beads and cells were filtered through filter cap FACS tubes (BD Biosciences) before analysis. All mass cytometry files were normalized together using the mass cytometry data normalization algorithm,²¹ which uses the intensity values of a sliding window of these bead standards to correct for instrument fluctuations over time and between samples.

CyTOF data analysis

UMAP dimensionality reduction and Louvain community detection (Fig. 1) were performed using CellEngine (CellCarta, Montreal, Canada) software. UMAP parameters: 90 neighbors, 200 Total iterations, 0.4 Minimum Distance. Louvain clustering was performed using all markers except Ki67 with the following parameters: 90 neighbors, 0.85 Resolution. Principal component analysis was performed using the prcomp function in R and frequencies of manually gated immune cell populations. Contributions were visualized using the fviz_pca_var function from the factoextra package in R.

Figure 1.

Characteristics of CD4 T cells in retinas of Aire^GW/+Lyn^−/− mice with uveitis. CD4 T cells present among inflammatory cells in the retinas of mice with uveitis were analyzed by CyTOF. UMAP dimensionality reduction and unsupervised clustering of 4,681 CD4⁺ T cells from the retinas of mice with late stage uveitis (n = 5) were characterized for expression of informative cell surface and intracellular molecules by mass cytometry (A). CD4⁺ T cells were gated as Cisplatin⁻CD45⁺Ter119^-SiglecF⁻ Ly6G⁻FcεR1a⁻PDCA1⁻B220⁻CD3⁺ NK1.1⁻CD4⁺CD8^- cells. Clusters were assigned and cluster identities were interpreted based on expression of molecules of interest as explained in the text. Expression of key molecules (FoxP3, CD27, CD103, PD1, TCF1, and Ly6C) across CD4⁺ T cells in panel A (B).

Flow cytometry

Single-cell suspensions of LN and lymphocytes of retina were washed, blocked with purified Anti-CD16/32 Antibody (BioLegend, 101302, clone 93), and stained with antibodies of indicated specificities in HBSS buffer with 2% FBS. Staining reagents include APC-eFluor 780 Anti-TCR beta (47-5961-82, clone H57-597), APC-eFluor 780 Anti-Ly-6C (47-5932-82, Clone HK1.4), PE-Cyanine7 Anti-CD223 (LAG-3) (25-2231-82, Clone C9B7W) from Thermo Fisher Scientific, purified Anti-CD3 (70-0032, clone 17A2) and PE-Cyanine7 Anti-CD4 (60-0042, clone RM4-5) from Tonbo Biosciences, purified Anti-CD16/32 antibody (101302, clone 93), Pacific Blue™ Anti-CD19 (115523, clone 6D5), Pacific Blue™ Anti-F4/80 (123124, clone BM8), Pacific Blue™ Anti-CD11b (101224, clone M1/70), Pacific Blue™ Anti-CD11c (117322, clone N418), Brilliant Violet 421™ Anti-mouse CD62L (104436, Clone MEL-14), Brilliant Violet 510™ Anti-PD-1 (135241, Clone 29F.1A12), PE/Dazzle™ 594 Anti-mouse CD186 (CXCR6) (151117, Clone SA051D1), BV605 Anti-CD8 (563152, clone 53-6.7) from BD bioscience. Dead cell exclusion was based on DAPI positivity (Thermo Fisher Scientific, D1306). Allophycocyanin (APC) or phycoerythrin (PE)-conjugated I-A^b P2 tetramer (QTWEGSGVLPCVG) corresponding to mouse IRBP amino acids 277–290 were obtained from the National Institutes of Health (NIH) tetramer facility, Atlanta, Georgia, USA. For flow cytometry to characterize P2-specific T cells, single cells from LN or retina were incubated with tetramer in staining buffer (2% FBS, 5% normal rat serum, 5% normal mouse serum [Jackson ImmunoResearch Inc, 015-000-120] and 10 μg/ml purified Anti-CD16/32 antibody) for 1 h at room temperature. The cells were stained, and P2 tetramer-reactive cells were gated on TCRβ⁺CD4⁺CD8⁻DUMP⁻ (F4/80⁻ CD11c⁻CD19⁻CD11b⁻) lymphocytes for further analysis. All data of flow cytometry were collected on an LSR II cytometer (BD), LSRFortessa (BD) or Full spectrum at the Flow Cytometry Core at UCSF. Analyses of flow cytometry data were performed using FlowJo software (TreeStar).

scRNA-seq and scRNA-seq data processing

For scRNA-seq, P2 tetramer-specific T were sorted as described previously.^9,16 Briefly, single-cell suspensions of cervical LNs of Aire^GW/+Lyn^−/− mice with and without uveitis or of retinas of mice with uveitis were incubated with P2 tetramer for 1 h at room temperature in staining buffer (2% FBS, 5% normal rat serum, 5% normal mouse serum and 10 µg/ml purified anti-CD16/32 antibody) and APC and PE-conjugated I-A^b P2 tetramer, followed by magnetic bead enrichment for tetramer-positive cells with Anti-APC (Miltenyi Biotec, 130-090-855) and Anti-PE (Miltenyi Biotec, 130-048-801) magnetic beads. The positively selected cells were stained, and P2 tetramer-reactive cells were gated on TCRβ⁺CD4⁺CD8⁻DUMP⁻ (F4/80⁻CD11c⁻CD19⁻CD11b⁻) lymphocytes for sorting and scRNA-seq. According to the manufacturer’s instructions, sorted P2⁺ cells were loaded onto the 10× Genomics Chromium platform for droplet-based massively parallel scRNA-seq by the genomics core of UCSF. Libraries were prepared by using the 10x Genomics Chromium Single Cell 5′ Reagent (GEX + VDJ) Kit version 1.1 (10xGenomics, PN-1000165, PN-1000020, PN-1000005, PN-1000120, PN-1000213) according to the manufacturer’s protocol (Illumina), and sequencing were performed on an Illumina Novaseq 6000. FASTQ were aligned using the Cell Ranger pipeline (10× Genomics Inc., version 3) by using mouse genome reference data set (mm10). The Seurat pipeline was applied to identify and cluster the cell subsets with the data set. Data set was read into R (version 4). For further analysis, cells with low gene detection (<200 genes) and high mitochondrial gene content (>5%) were filtered out. “LogNormalize” (scale factor 1:10,000) was used to normalize gene expression in each cell. The influence of batch effects, percentage of mitochondrial genes, percentage of ribosome genes, and cell cycle-related genes in each group/condition were regressed out using Seurat. For the UMAP dimensional reduction shown in Fig. 2, the first 13 principal components were used for clustering with resolution set at 0.25. For the UMAP analysis shown in Fig. 3, the first 30 principal components were used with resolution set at 1.0. The FindAllMarkers function was applied to identify differentially expressed genes.

Figure 2.

Single cell transcriptome analysis of P2⁺CD4⁺T cells in retina of Aire^GW/+Lyn^−/− mice with uveitis. The P2⁺CD4⁺ T cells in retina of Aire^GW/+Lyn^−/− mice with uveitis at 6–7 wk were isolated by magnetic bead enrichment and cell sorting and subjected to single cell transcriptome sequencing. After quality filtering, 1,602 cells from 6 pooled mice were represented by UMAP plot and segregated into 4 phenotypic clusters (A), with their frequencies shown in (B). The identities of the clusters were interpreted as described in the text. Superimposed on the UMAP plot is the expression of Foxp3, Mki67, Tbx21, Rorc, Gata3, Maf, Bcl6, Prdm1 (C). Heatmap of top 10 differentially expressed genes for the clusters (D) in panel A. Dot plots showing average expression level across all cells within a cluster (color of circle) and percentage of cells within a cluster (size of circle) for functionally informative genes Sell, Cd44, Cd27, Mki67, Icos, Pdcd1, Lag3, Tigit, Ctla4, Il2ra, Il2rb, Foxp3, Il7a, Ifng, Il10, Il4, Ikzf2, Cxcr5, Cxcr6, Ccr8, Selplg, Cx3cr1 and Runx3 expression in different P2⁺CD4⁺ cell clusters in retina of Aire^GW/+Lyn^−/− mice with uveitis (E).

Figure 3.

Single cell transcriptome analysis and clustering of P2-binding CD4⁺ T cells in LN of Aire^GW/+Lyn^−/− mice with uveitis and without uveitis, and in LN of Aire^GW/+Lyn^−/−IRBP^−/− mice immunized with IRBP protein. After scRNAseq and filtering, a single UMAP representation was constructed from 1,516 total P2⁺CD4⁺T cells, which were combined from 540 filtered cells in two datasets from eye-draining LNs of 7 pooled and 8 pooled Aire^GW/+Lyn^−/− mice at 7–12 wk of age with uveitis, 518 filtered cells in one dataset from eye-draining LNs of 16 pooled Aire^GW/+Lyn^−/− mice without uveitis, and 458 filtered cells from draining LNs of 11 pooled Aire^GW/+Lyn^−/−IRBP^−/− mice immunized 5 d earlier with IRBP protein in CFA. Clustering with Seurat identified 7 phenotypic clusters, which are represented by different colors for each of the types of P2-specific CD4⁺ T cells separately shown on the combined UMAP (A). See the text for explanation of the identities assigned to the different clusters, the frequencies within each type of sample are shown (B). Expression of Ly6c1, Lag3, Mki67, Ifit1, Il7r, Foxp3, Il2ra, Il2rb, Tbx21, Ifng, Il4, Selplg within individual cells in the combined UMAP is shown by scaled color (log normalized expression level) (C). A heatmap shows expression levels in the P2-specific CD4⁺ T cells of the top 10 differentially expressed genes for each cluster for the combined samples (D). Relative expression within each cluster of the functionally informative genes Ly6c1, Sell, Cd44, Cd27, Mki67, Icos, Pdcd1, Lag3, Tigit, Ctla4, Il2ra, Il2rb, Foxp3, Ikzf2, Cxcr5, Cxcr6, Ifng, Il2, Mif, Ifit1, Il7r, Cxcr3, Runx3, Selplg with average expression level across all cells within a cluster by color intensity and the percentage of cells within a cluster of the gene represented by size of circle (E). Note, the data shown in C and E are broken out by type of sample in Fig. S3A and S3C.

IRBP protein immunization

IRBP was isolated from bovine retinas by affinity chromatography on concanavalin-A-Sepharose and fast performance liquid chromatography (FPLC), as previously described.^22,23 IRBP^−/− mice at 7-9 weeks of age were injected subcutaneously with 100 μg purified bovine IRBP protein emulsified 1:1 v/v in CFA. A total of 200 μl emulsion was injected subcutaneously (SC) on both sides of the chest. After 5 d, the draining LN were harvested and then stained with P2-tetramer reagents, sorted, and subjected to single cell sequencing, as described above.

Diphtheria toxin (DT) administration

The 6 wk old Aire^GW/+Lyn^−/−Foxp3^DTR or Lyn^−/−Foxp3^DTR mice without uveitis as determined by fundoscopy were intraperitoneally injected with 1 dose of 300 ng DT (Sigma, D0564) dissolved in PBS at 1.5 ng/ µl concentration, or injected with PBS as a control. After 2 wk, the treated mice were analyzed by fundoscopy for uveitis and then were euthanized and lymphocytes prepared for FACS analysis as described above.

Statistics

Statistical analysis and graphing were performed by using Prism 9.3.1 (GraphPad). Statistical tests for comparison of 2 groups involved 2-tailed t tests. The number of independent experiments, statistical tests, and P values are indicated in figure legends.

Results

Characteristics of CD4 T cells in the retinas of Aire^GW/+Lyn^−/− mice with autoimmune uveitis

Previously, it was found that approximately 50% of Aire^GW/+Lyn^−/− mice spontaneously developed autoimmune uveitis between 6 and 9 wk of age, and disease incidence correlated with an expansion of CD4 T cells recognizing epitopes from the Aire-regulated retinal protein IRBP.⁹ Moreover, deletion of the gene encoding IRBP completely protected these mice from developing uveitis,¹⁶ indicating that the immune response to this intracellular rod cell protein is required for disease induction. To analyze the nature of the CD4 T cells infiltrating the retinas in Aire^GW/+Lyn^−/− mice with early-stage disease at 5–7 wk of age or with severe disease at 9–12 wk of age (Fig. S1A and B), we used high-dimensional mass cytometry (cytometry by time-of-flight, CyTOF) analysis to assess expression in T cells of proteins of interest, including CD62L, CD44, Ly6C, PD1, LAG-3, CD103, FoxP3, TCF1, CD27, and other proteins. With these parameters, 6 distinct phenotypic clusters (clusters 0–5) of CD4⁺ T cells were observed in the retinas of Aire^GW/+Lyn^−/− mice with advanced uveitis (Fig. 1A). Almost all of these CD4 T cells were CD44^hi, indicating that they were activated T cells. Only cluster 4 had cells with a Ly-6C^hi phenotype, consistent with those cells being early activated T cells, although a subset of more differentiated Th1 cells re-induce expression of Ly6C²⁴ (Fig. 1B). Prominent expression of FoxP3 was seen in cluster 3, indicating that this cluster was composed of Tregs (Fig. 1B). Clusters 2 and 5 expressed TCF1, consistent with memory populations of CD4 T cells. The remaining clusters (0 and 1) were PD-1^hi and likely represented effector CD4 T cells (Fig. 1B).

We also used additional cell surface markers to characterize the other immune cells infiltrating the retinas at both stages of disease. These cells included many monocytes, NK cells, and CD8⁺ T cells, but few neutrophils, which is consistent with a Th1-based inflammation. Comparing the inflammatory cells seen in mice with disease soon after inflammation was first detected vs. later-stage disease characterized by extensive and widespread inflammation, the infiltrating cells at the earlier stage included monocytes, dendritic cells, and B cells, whereas in later-stage disease the inflammatory cells were increasingly composed of CD4⁺ and CD8⁺ T cells and of myeloid cells with a CD11b⁺ CD64⁺ phenotype (Fig. S1C and D).

We next wanted to better understand the characteristics of the autoantigen-specific CD4 T cells infiltrating the retina at an early stage of the disease, that is at 6–7 wk of age, when mild inflammation was detected by fundoscopy. For this purpose, we focused on the CD4 T cells recognizing the predominant peptide epitope from IRBP, amino acids 271–290 (“P2”).⁹ These cells were isolated using P2-loaded I-A^b tetramers and subjected to single cell RNA-seq (scRNA-seq) coupled with TCR sequencing using the 10× Genomics Chromium platform. The results of the TCR analysis of the cells have been reported elsewhere.¹⁶ After quality control, we retained the data from 1602 P2⁺CD4⁺T cells pooled from the retinas of 6 Aire^GW/+Lyn^−/− mice with uveitis for further analysis. Clustering with Seurat revealed 4 clusters of cells, which were visualized in two dimensions with UMAP and annotated as described below (Fig. 2A). The relative frequencies of the different clusters are shown in Fig. 2B. To gain insight into the nature of the cells within each cluster, lineage-defining transcription factors and other informative genes were mapped onto the UMAP projection (Fig. 2C). Especially relevant are the expression of FoxP3, Tbx21, RORc, and Bcl6, which define Treg, Th1, Th17, and Tfh, respectively; Prdm1, encoding BLIMP-1, which is expressed by effector T cells; and Mki67, which is highly expressed by proliferating cells. The clusters were also analyzed for the most differentially expressed genes (Fig. 2D and Table S2) and for the expression of functionally relevant genes such as key checkpoint regulators, cytokines, and cytokine receptors (Fig. 2E). The cluster that expressed high levels of FoxP3 was identified as Tregs. The cells of another cluster expressed Mki67 as well as many other proliferation-related genes and were thus described as proliferating T cells (Tprolif). The remaining ∼80% of cells were separated into 2 clusters, a smaller cluster that we interpreted to be Th1 effector cells (Th1eff) based on expression of Tbx21 (Tbet) and IFN-γ. These cells expressed Runx3, which is one of the repressors of the Tfh program²⁵ and cell trafficking molecules consistent with Th1 effectors,^24,26 namely P-selectin glycoprotein ligand 1 (PSGL-1; encoded by Selplg) and CXCR6 (Fig. 2E and Fig. S2A). Finally, the most abundant cluster expressed the transcription factor c-Maf, the anti-inflammatory cytokine IL-10, and also high levels of mRNA encoding inhibitory receptors LAG3, PD-1, Tigit, and CTLA4. The cells may be related to a FoxP3-negative anti-inflammatory CD4⁺ T cell population referred to as T regulatory 1 (Tr1).²⁷ These cells were also enriched for expression of transcripts of Nt5e and Izumo1r (Fig. S2B), which are expressed by anergic CD4 T cells.²⁸ Remarkably, this final cluster represented more than 50% of the P2-binding CD4⁺ T cells in the retinas exhibiting on-going inflammatory damage. Both the Th1 cluster and the putative Tr1 cluster exhibited strong expression of CXCR6 (Fig. 2E), which is expressed on Th1 effector cells and not on T_FH cells.²⁶ Pathway analysis, utilizing each cluster’s top 50 differentially expressed genes, was consistent with the above interpretations of cluster identity (Fig. S2C–F). In comparing these results to the characterization of total CD4⁺ T cells infiltrating the retina (Fig. 1), the P2-specific CD4 T cells had all downregulated L-selectin and Ly6C, indicating that the early activated cells in the retina had specificities against antigens other than IRBP. Taken together, the analysis of CD4 T cells in the autoimmune retinas in Aire^GW/+Lyn^−/− mice indicated that the pathogenesis was primarily a Th1 inflammatory process, which was accompanied by an anti-inflammatory process similar to what has been described as Tr1, which was, however, insufficient to halt the autoimmune tissue destruction.

Diverse states of P2-binding CD4⁺ T cells in the eye-draining LN of Aire^GW/+Lyn^−/− mice

Previously, we had observed that in addition to P2-binding autoimmune CD4 T cells being present in the retinas of Aire^GW/+Lyn^−/− mice with uveitis, these T cells were also elevated in the eye-draining LNs of the mice, and moreover were elevated but to a lesser extent in the draining LN of double mutant mice without uveitis.⁹ To gain insight into the characteristics of the autoimmune T cells in the draining LNs of mice with and without disease, P2⁺CD4⁺ T cells were isolated and subjected to scRNA-seq as above. After quality control, we retained 1516 P2⁺CD4⁺ T cells of 3 data sets for further analysis. Two data sets, containing 540 P2⁺CD4⁺T cells, were from LN of pooled Aire^GW/+Lyn^−/− mice with uveitis (pooled from 7 mice in 1 sample and 8 mice in the other). In addition, 518 P2⁺CD4⁺ T cells were obtained from the LN of 16 pooled Aire^GW/+Lyn^−/− mice without uveitis, comprising the third data set. To enable the additional comparison of the autoimmune response to the CD4 T cell response to immunization with a foreign antigen, we immunized IRBP^−/− mice with purified bovine IRBP protein in complete Freund’s adjuvant and, after 5 d, harvested the draining LN, isolated P2-binding CD4⁺ T cells, and subjected them to scRNA-seq (458 cells from 11 mice). Clustering analysis with Seurat was performed with the combined P2⁺CD4⁺ T cells of all 4 data sets and yielded 7 clusters, which were plotted on a 2-dimensional UMAP representation (Fig. 3A). Although each data set contributed to every cluster, there were substantial differences in the frequency of cells within the different clusters (Fig. 3B). For example, the uveitis sample had an elevated number of antigen-specific CD4 T cells expressing an interferon-response signature (Ifit1^hi). In contrast, the immunization sample had a very low fraction of the CD4 T cells that were FoxP3⁺ Tregs (Fig. 3B). Two adjacent clusters expressed Mki67 at higher and lower levels. The cells with higher Mki67 expression are referred to as proliferating T cells (Tprolif), whereas those with lower Mki67 were likely exiting proliferation and transitioning to other states and therefore are referred to as transitional T cells (Ttrans). Among the remaining three clusters, one was characterized by the expression of Ly6c1 and L-selectin (Sell) mRNAs, so they likely represented early activated T cells (T early act) (Fig. 3C–E). In agreement with this interpretation, pathway analysis of each cluster’s top 50 differentially expressed genes (Fig. S3D–J and Table S3) indicated that these cells had elevated expression of genes associated with early cellular activation such as increased ribosome biosynthesis. Another cluster expressed LAG3 and other inhibitory receptors but not Ly6c1 or L-selectin, which collectively are characteristics of a later stage of T cell activation, and therefore these cells are referred to as T effectors (Teff) (Fig. 3C–E). This cluster of cells taken from the eye-draining LNs of Aire^GW/+Lyn^−/− mice expressed Tbx21 (encoding Tbet), IFN-γ, Selplg, Runx3, and IL-2Rβ, all of which are consistent with assignment as Th1 effectors.^24,25,29,30 In contrast, the subset of cells of this cluster at the right edge of the UMAP, which were mostly from the immunization dataset, expressed IL-4 (Fig. S3A), and thus are more likely to be T_FH. Cells of the final cluster expressed IL-7R but not Ly6c1 and therefore are referred as memory T cells (Il7r^hi) (Fig. 3C–E). Consistent with this assignment, the IL-7R⁺ cluster also expressed CXCR6 or CXCR3 and Id2, in common with the Teff cluster and the transcription factor Klf2, in common with the early activated cells. The presence of this memory-like cluster with considerable overlap in gene expression with Th1 effectors is similar to what has been observed with LCMV-specific CD4 T cells during a chronic virus infection, although that response had a considerably higher proportion of T_FH cells.³⁰ Expression of characteristic genes suggesting the above identifications is shown on the combined UMAP plots in Fig. 3C and the UMAP plots separated by sample type in Fig. S3A. Expression of Izumo1r and Nt5e are shown in Fig. S3B. The top differentially expressed genes of each cluster are shown in Fig. 3D and relative expression of genes of interest among cells of each cluster are represented in Fig. 3E and Fig. S3C. Pathway analysis of the top differentially expressed genes (Table S3) of each cluster is shown in Fig. S3D–J. These results also support the interpretations of cluster identities listed above. In summary, RNA transcriptome characterization identified seven states of antigen activation in the eye-draining LN that were seen upon immunization with a foreign antigen or upon spontaneous autoimmune activation.

The most significant difference between IRBP P2-specific CD4 T cells from mice that had uveitis versus those that did not was the presence in the former of more cells with a robust interferon transcriptional response (the Ifit1^hi cluster; Fig. 3B). A corresponding cluster of CD4 T cells with a robust interferon response signature has been observed in several other studies using virus or bacterial infection models of mice.^26,30–32 In our samples, the prevalence of this cluster in mice with uveitis may reflect secretion of IFN-γ by IRBP-specific CD4⁺ T cells in these mice, as expression of Tbet and IFN-γ mRNAs was evident by some of the P2⁺CD4⁺cells of LN (Fig. 3C–E and Fig. S3A). For both the Teff and Il7r^hi populations, we also compared the gene expression patterns for the cells within the clusters from the mice with uveitis vs. those from the mice without uveitis and found a handful of genes (Fig. S3K and L) and signal pathways (Fig. S3M and N) with statistically significant greater expression in the cluster cells from mice with uveitis and other genes with greater expression in the cluster cells from mice without uveitis. Pathway analysis of these data indicated that IL-2 receptor signaling may have been greater in the Th1 effectors from the mice with uveitis compared to those from mice without uveitis (Fig. S3M).

Variations in the numbers and activation states of P2⁺CD4⁺ T cells in individual Aire^GW/+Lyn^−/−Foxp3^GFP mice

The results of scRNA-seq of P2⁺CD4⁺ T cells from the eye-draining LN demonstrated that mice with or without uveitis both contained autoantigen-specific CD4 T cells in multiple stages of activation and differentiation. A limitation of these data, however, is that the samples were obtained by pooling P2⁺CD4⁺ T cells from multiple mice to get enough cells to analyze by this technique. To assess mouse-to-mouse variation, we developed a flow cytometry approach to determine the numbers of P2⁺ CD4⁺ T cells in different differentiation states within individual mice. We then correlated these data with the presence or absence of uveitis. For this purpose, we introduced into the mice a FoxP3-GFP reporter that has been validated to accurately identify FoxP3⁺ Treg,³³ albeit with somewhat reduced FoxP3 function.³⁴ When introduced into Aire^GW Lyn^−/− mice, this allele did not increase the fraction of mice that developed uveitis to a detectable degree (16/38 of these mice developed uveitis). We combined use of the FoxP3-GFP reporter with staining for a variety of cell surface proteins that may be informative about the activation and differentiation state of the Th1 effectors, including the inhibitory receptors PD-1 and LAG3, and the Th1 chemokine receptor CXCR6 (Fig. 4A). Cells defined by scRNAseq as having a type 1 interferon signature, a memory phenotype, or proliferating were not explicitly identified by this flow cytometry analysis and thus were likely included within the 2 effector types distinguished by degree of LAG3 expression. In agreement with previous results,⁹ the mice with uveitis had a roughly 37-fold greater expansion of P2-binding CD4 T cells in the eye-draining LN compared to those that did not (Fig. 4B; Fig. S4A). In addition, the P2-tetramer binding CD4 T cells in mice with uveitis had greater fractional representation of cells at later states of activation, based on lack of expression of Ly6C and CD62L and increased expression of PD-1 and LAG3 (Fig. 4C and Fig. S4C–D). Applying the same methodology to P2-binding CD4 T cells in the retinas of mice with uveitis (Fig. S4B) showed that the P2-specific CD4 T cells in retina were predominately PD-1^hi LAG3^hi, with lower % representation of cells that were FoxP3⁺ Tregs or PD-1^hi LAG3^int (Fig. 4C). The differences in the activation status of P2-binding CD4 T cells in the LN of mice with vs. without uveitis were particularly evident when the results are shown for each mouse analyzed (Fig. 4D), which also illustrates significant mouse-to-mouse variation within the mice without uveitis. Of the 6 mice without uveitis, 4 mice had P2-specific FoxP3⁺ Tregs, ranging from 1 out of 7 cells to 66%, and the non-Treg cells were mostly earlier activated T cells (Ly6C⁺ or Ly6C⁻CD62L⁺). The remaining two mice, those without detected Tregs, still had relatively few P2-specific CD4 T cells. The distribution of activation states of the P2-specific CD4 T cells in these 2 mice resembled those of the mice with uveitis except that the latter had many more such cells in the LN, in addition to having such cells in the retinas causing inflammation (Fig. 4D). A fraction of the CD62L⁻ Ly6C⁻ PD-1^hi CD4 T cells also expressed CXCR6, which was robustly expressed on nearly 50% of P2-binding cells in the retinas of mice with uveitis (Fig. 4E and Fig. S4E-G).

Figure 4.

High frequency of Tregs in some Aire^GW/+Lyn^−/−Foxp3^GFP mice without uveitis. Aire^GW/+Lyn^−/−Foxp3^GFP mice at 7–9 wk of age with uveitis (n = 6) or without uveitis (n = 7) were analyzed by flow cytometry to determine the distribution among P2-specific CD4⁺ T cells of Treg (FoxP3⁺), early activated T cells (Ly-6C⁺ or Ly-6C^-CD62L⁺) and late activated T cells (PD-1^hi LAG-3^int or PD-1^hi LAG-3^hi). Shown are the gating strategy (A), numbers of P2⁺ CD4⁺ T cells in eye-draining LNs per mouse (B), frequencies of Treg (LN without uveitis (n = 8); LN with uveitis (n = 9)), early activated T cells, and late activated T cells in eye-draining LN (left) or retina (right) (C), the number (above bar) and fractions of P2⁺ CD4⁺ T cells in LN in each category of activation state in individual mice with or without uveitis (D), and fractions of various P2⁺ CD4⁺ T cell subsets that expressed CXCR6 in eye-draining LN or retina (E). Frequencies of Tregs, Ly-6C⁺, Ly-6C^-CD62L⁺, PD-1^hi LAG-3^int, and PD-1^hi LAG-3^hi cells were frequencies of cells within P2⁺CD4⁺cells, frequencies of CXCR6⁺ cells were frequencies of cells within P2⁺CD4⁺cells. P2⁺CD4⁺cells were gated on TCRβ⁺CD4⁺CD8^-DUMP^- cells. *P < 0.05. For B, left panel, 2-tailed t test; For C, left panel, unpaired t test, error bars are mean ± SD.

As the numbers of P2-binding CD4 T cells in the eye-draining LNs were strikingly different between mice with and without uveitis, we next examined by flow cytometry the numbers and phenotypes of these T cells in the eye-draining LNs of individual mice just before the onset of uveitis. The lack of detectable inflammation in the retina in these mice was verified by fundoscopy. As expected, very few if any P2⁺CD4⁺ T cells were observed in the eye-draining LNs of age-matched control Lyn^−/− FoxP3^GFP mice (Fig. 5A and B), presumably reflecting negative selection due to Aire-induced expression of IRBP in the thymus.^16,35 In contrast, in the eye-draining LNs of young Aire^GW/+Lyn^−/− FoxP3^GFP mice, small numbers of P2⁺ CD4⁺ T cells were present, with a considerable mouse-to-mouse variation (∼ 5-fold), centered around a mean of ∼30 P2⁺CD4⁺ T cells per mouse. The 2 mice with the fewest P2⁺ CD4⁺ T cells also had very few Tregs (Fig. 5C), whereas for the remaining mice, there was a statistically significant negative correlation between the number of these autoantigen-specific CD4⁺ T cells and the fraction of them that were FoxP3⁺ (Fig. 5C; dashed red line). This analysis required sacrificing the mice, and hence we cannot know which of these mice would have developed uveitis within the next several weeks. Based on historical data, roughly 50% of these 8 young mice would have gone on to develop uveitis. Strikingly, these pre-disease younger mice averaged about 30 P2⁺ CD4⁺ T cells in their eye-draining LNs, whereas the older mice without uveitis, with one exception, had fewer than 10 such cells (Fig. 4B), This observation suggests that over the subsequent several weeks in the mice that did not develop uveitis, many of the P2⁺ CD4⁺ T cells either died or migrated out of the eye-draining LNs.

Figure 5.

Variation in numbers of P2-binding CD4⁺ T cells in Aire^GW/+Lyn^−/−Foxp3^GFP mice prior to disease onset. The number of P2⁺CD4⁺ cells and the fraction of them that were FoxP3⁺ in the eye-draining LN of Aire^GW/+Lyn^−/−Foxp3^GFP mice (n = 8) or Lyn^−/− Foxp3^GFP (n = 2) mice without uveitis at 4–6 wk of age were determined by flow cytometry. Representative example of flow cytometry analysis of P2 tetramer staining in the TCRβ⁺CD4⁺CD8⁻DUMP⁻ cells with calculated number of P2⁺ cells shown (left), and GFP staining of the P2⁺ cells with fraction of GFP⁺ cells shown (right) (A) and summary of numbers of P2⁺CD4⁺ T cells observed per mouse (B). Correlation of the number of P2⁺ CD4⁺ T cells vs. the % of them that expressed Foxp3^GFP(C). A linear regression line (red dashed line) is shown for the 6 mice that had 21 or more P2⁺ CD4⁺ T cells (R = 0.69), that is, excluding the 2 mice with fewest number of P2⁺ CD4⁺ T cells. *P < 0.05. Two-tailed t test; error bars are mean ± SD.

Transient depletion of Tregs in Aire^GW/+Lyn^−/− mice triggers expansion of P2-specific CD4 T cells

The negative correlation observed between the number of P2⁺CD4⁺ T cells in the eye-draining LNs of young pre-disease mice and the fraction of these cells that were FoxP3⁺ (Fig. 5C), suggested that P2-specific Tregs may have limited the clonal expansion of P2-specific effector CD4 T cells. To test this hypothesis, we introduced into the mice the FoxP3^DTR allele, which expresses the primate diphtheria toxin receptor selectively in FoxP3⁺ Tregs.³⁶ Six-week old Aire^GW/+Lyn^−/−Foxp3^DTR mice were screened for uveitis by fundoscopy, and mice without any detectable uveitis and control Lyn^−/−Foxp3^DTR mice were injected a single time with 300 ng of diphtheria toxin, an approach that has been shown to deplete a large majority of Tregs.³⁶ This depletion has been observed to be transient, as the remaining Tregs can proliferate and/or conventional CD4 T cells can differentiate into induced Tregs. Two weeks following diphtheria toxin treatment, the mice were analyzed for uveitis by fundoscopy, and for the numbers of P2⁺ CD4⁺ T cells (Fig. 6A). Remarkably, about 25% of Aire^GW/+Lyn^−/−Foxp3^DTR mice developed uveitis 2 wk following transient Treg depletion (Fig. 6B and C). Despite the lack of disease in the remaining mice, almost all of them exhibited a major expansion of P2⁺ CD4⁺ T cells (∼15-fold; Fig. 6D and E). The three mice that developed uveitis (indicated by the solid circles in Fig. 6E) all had average or above-average numbers of P2⁺CD4⁺ T cells in the eye-draining LNs, but some mice with above-average numbers of these cells did not develop uveitis. As expected based on previous studies, by 2 wk after depletion of Treg, their frequencies had recovered (Fig. 6F–G). These results indicate that FoxP3⁺ Tregs restrained the expansion of P2⁺CD4⁺ T cells in Aire^GW/+Lyn^−/−Foxp3^DTR mice, providing additional evidence that the dendritic cells in the eye-draining LN were presenting IRBP even in the absence of detectable tissue damage in the eye. Moreover, the expansion of IRBP-reactive conventional CD4 T cells that occurred after Treg depletion was not sufficient to induce autoimmune uveitis in most mice. This result implies that a second peripheral tolerance regulatory mechanism may have prevented autoimmune attack on the retina in these mice.

Figure 6.

Transient depletion of Tregs in Aire^GW/+Lyn^−/−Foxp3^DTR mice without uveitis promotes substantial expansion of autoreactive CD4 T cells and also induces uveitis in some mice. Schematic showing diphtheria toxin (DT) administration schedule for Aire^GW/+Lyn^−/−Foxp3^DTR and Lyn^−/−Foxp3^DTR mice without uveitis at 6 wk (A). Representative funduscopic images (B) and frequencies (C) of Aire^GW/+Lyn^−/−Foxp3^DTR (PBS: n = 4; DT: n = 12) and Lyn^−/−Foxp3^DTR (PBS: n = 3; DT: n = 3) mice with uveitis or without uveitis after PBS or DT treatment for 2 wk starting at 6 wk age in Aire^GW/+Lyn^−/−Foxp3^DTR and Lyn^−/−Foxp3^DTR mice lacking detectable uveitis. Representative flow cytometric analysis (D) and cell number (E, left panel) of P2⁺ CD4⁺T cells in eye-draining LN of Aire^GW/+Lyn^−/−Foxp3^DTR (PBS: n = 4; DT: n = 12) and Lyn^−/−Foxp3^DTR (DT: n = 3) mice at 8 wk of age with PBS or 300 ng DT treatment for 2 wk. Cells shown were gated on the TCRβ⁺CD4⁺CD8⁻DUMP⁻ cells. Red solid circles indicate the mice which developed uveitis after DT treatment (E, left). Fold change in average number of P2⁺CD4⁺T cells between the DT-treated and PBS-treated groups is shown (E, right panel). Two weeks after DT of Aire^GW/+Lyn^−/−Foxp3^DTR mice (n = 12), the fraction of P2⁺CD4⁺T cells that were FoxP3⁺ was determined using the GFP reporter contained within FoxP3^DTR allele. Shown is a representative flow cytometric plot (F) and a summary plot comparing the number of P2⁺CD4⁺ T cells to the fraction of them that were GFP⁺ Treg (G). Shown in red is the linear regression line (R = 0.050). Red solid circles indicate the mice that developed uveitis after DT treatment. *P < 0.05. in panel E, 2-tailed t test; error bars are mean ± SD.

Discussion

The results presented here address central issues related to the spontaneous initiation of autoimmune uveitis in Aire^GW/+Lyn^−/− mice and conversely to the mechanisms by which approximately 50% of these genetically susceptible mice avoid disease. We characterized the nature of the autoantigen-specific CD4 T cell response in the retinas of mice with disease and in the eye-draining LN of mice with or without disease. CD4 T cells specific for the key retinal autoantigen, IRBP, were found to represent various stages of activation, with earlier activated phenotypes and proliferating CD4 T cells being present in the LN and later activated effector phenotype cells being seen in both LN and retina. The effector-type cells expressed Tbet, consistent with their being polarized to Th1. Moreover, some of the autoantigen-specific CD4 T cells in the LN of mice with uveitis had a strong interferon-inducible gene signal in their mRNA, which may have been induced by interferon-γ produced by effector T cells. Also seen were two types of anti-inflammatory IRBP-specific CD4 T cells: FoxP3^- cells expressing c-Maf and IL-10, and thus possibly related to T regulatory 1 cells (Tr1), which were present only in retinas of mice with tissue destruction, and FoxP3⁺ Tregs, which were present in retinas of mice with disease and in the eye-draining LN of mice with or without disease. In the mice without uveitis, FoxP3⁺ Treg were critical for restraining proliferation of IRBP P2-specific conventional CD4 T cells, as transient depletion of FoxP3⁺ cells led to a 15-fold increase in the number of P2-specific CD4 T cells. Analysis of mice at 6 wk of age, prior to the initiation of detectable tissue damage to the retina, also supported the conclusion that FoxP3⁺ Treg suppressed proliferation of P2-specific CD4 T cells prior to disease initiation. Remarkably, only a minority of the mice developed uveitis after Treg depletion despite a ∼15-fold expansion of the autoantigen-specific CD4 T cells in the eye-draining LN. These results suggest that, in addition to the inhibitory role of FoxP3⁺ Tregs, there is at least one additional peripheral tolerogenic mechanism preventing autoimmune uveitis in these mice.

The findings presented here also reinforce the conclusion from prior studies^9,16 in this model that the key checkpoint controlling spontaneous autoimmune disease in Aire^GW/+Lyn^−/− mice is located in the draining LN, not in the target tissue, the eye. In particular, longitudinal analysis of Aire^GW/+Lyn^−/− mice by fundoscopy found that approximately 50% of mice first show signs of eye inflammation between 6 and 9 wk of age, and that mice lacking eye inflammation at 9 wk of age do not subsequently develop uveitis.⁹ Moreover, the mice that showed signs of inflammation of the eye almost always progressed to severe disease in both eyes. Thus, although the IRBP P2-specific CD4 T cells in the eye included some FoxP3⁺ Treg and many c-Maf⁺IL-10⁺ cells with an anti-inflammatory phenotype similar to Tr1, these 2 anti-inflammatory types of autoreactive CD4 T cells were unable to stop disease progression once an early autoimmune checkpoint was breached. This is in contrast to observations in the NOD mouse model of autoimmune diabetes, where the critical checkpoint determining whether diabetes develops or not is thought to occur in the pancreas, since peri-islet insulitis starts well in advance of major pancreatic β-cell destruction and is seen in all mice.³⁶ Thus, what can be learned from these 2 mouse models of spontaneous organ-specific autoimmunity about how autoimmune disease is initiated may be complementary.

What determined whether an expansion of IRBP-specific CD4 T cells in the eye-draining LN progressed to autoimmune attack in the retina or alternatively was limited to the eye-draining LN? One hypothesis suggested by the data presented here is that this checkpoint is controlled by the numbers of IRBP-specific CD4 T cells reaching a critical threshold and/or the balance between effector-type and FoxP3⁺ Tregs of these cells in the eye-draining LN. Even in mice with the Aire^GW/+ genotype, which have decreased expression of IRBP in the thymus and decreased negative selection of P2-specific thymocytes,¹⁶ the numbers of IRBP-specific CD4 T cells leaving the thymus is likely to be small and subject to stochastic variation. Indeed, we observed considerable mouse-to-mouse variation in the numbers of the IRBP-specific CD4 T cells in the eye-draining LN and in the percentage of them that were FoxP3⁺ when assessed prior to disease onset (Fig. 5). Briefly, 6 of 8 such mice had >20 P2-specific CD4 T cells in their eye-draining LN, and for these mice, there appeared to be a negative correlation between the number of these cells and the fraction of them that were Treg. This negative correlation is consistent with FoxP3⁺ Treg restraining the proliferation of P2-specific conventional CD4 T cells, as was indicated by the observation that transient depletion of Treg caused a 15-fold expansion of P2-specific CD4 T cells (Fig. 6). Moreover, a role of Treg in restraining proliferation of autoreactive CD4 T cells was also indicated for CD4 T cells specific for a stomach autoantigen by an in situ imagining approach.³⁹ Interestingly, 2 of the 8 mice analyzed in the experiment shown in Fig. 4 had very few P2-specific CD4 T cells (20 or less) along with only ∼10% of them being Treg. In these mice, clonal expansion of the P2-specific conventional CD4 T cells may have been poor due to a peripheral tolerance mechanism of conventional CD4 T cells such as anergy²⁸ or simply a failure of the cells to expand due to low numbers. More detailed future studies will be needed to address this issue more fully.

Pointing to an important role of FoxP3⁺ Tregs in preventing disease in mice without uveitis, transient depletion of Treg in 6 wk old Aire^GW/+Lyn^−/− mice without uveitis induced a considerable expansion of the IRBP-specific CD4 T cells in the draining LN in almost all of the mice. Surprisingly, inflammation of the retina was only induced in roughly one fourth of these mice despite this proliferation. This observation suggests that avoidance of uveitis in 9 wk old Aire^GW/+Lyn^−/− mice is achieved by a combination of FoxP3⁺ Tregs and either a second immune tolerance mechanism or a checkpoint related to effector cell programming for pathogenic potential and/or infiltration into the retina. An important role for Treg in restricting autoimmunity to the retina was also observed in prior studies of experimental allergic uveitis (EAU) in B10.RIII mice immunized with IRBP in adjuvant. Studies in this experimentally induced model of eye autoimmunity also found a role of Treg in limiting the autoimmune response to IRBP in the draining LN and additionally revealed a significant role for Treg in limiting tissue destruction in the eye.^37–39 We did not specifically address the degree to which Treg slow the progression of autoimmune uveitis in Aire^GW/+Lyn^−/− mice, although for the 50% of these mice that exhibit inflammation in the eye, destruction proceeds quickly and extensively,⁹ despite the presence in the retinas of IRBP-specific Treg and CD4 T cells with a Tr1-like phenotype (Fig. 2).

In this study, we have focused on retinal autoantigen-specific CD4 T cells, and the results support the hypothesis that whether mice develop uveitis or not is determined by the absolute numbers of IRBP-specific CD4 T cells and/or the relative balance between effector and Treg subtypes of these T cells. While we favor the interpretation that the main immune checkpoint where dampening of the pathogenic effectors by Tregs takes place is the eye-draining LN, the role of the eye itself cannot be overlooked. The eye is known as the prototypic immune-privileged organ.^40,41 In addition to physical blood-tissue barriers, which prevent free trafficking of cells in and out of the eye, studies have documented multiple cell-bound and soluble molecules that together actively maintain a profoundly inhibitory intraocular environment. Immune privilege efficiently controls naïve retina-specific T cells that enter the eye passively (minor bleeding into the eye is common and does not lead to uveitis) by converting them to Foxp3⁺ Tregs or anergic cells,⁴² although primed T-effector cells that can actively cross blood-tissue barriers are refractory.⁴³ Nevertheless, some level of inhibition even of primed T cells is implied by the findings that Tregs bring about resolution and maintain remission of IRBP-induced EAU at least in part through local effects within the eye.³⁸ In the experiments reported here, we found autoantigen-specific Tr1-like cells in the retinas of mice with disease (Fig. 2), suggesting that conversion of effector CD4 T cells to Tr1 may occur in the retina directed by the immunosuppressive environment there.

Some investigators examining pathogen-specific Th1 and T_FH responses in mouse infection models have used expression of CXCR6 and CXCR5 to distinguish between Th1 effectors (CXCR6⁺) and T_FH (CXCR5⁺).^30,32 In contrast, we found that in the autoimmune response of Aire^GW/+Lyn^−/−mice, only a small fraction of the autoreactive Th1 cells in the eye-draining LN of mice with uveitis expressed CXCR6 (Fig. 4E). These cells did express numerous other genes that are prominent in Th1 effectors in multiple studies and absent in T_FH, including T-bet (encoded by Tbx21), Runx3, IFN-γ, PSGL-1 (encoded by Selplg), and IL-2Rβ,^25,26,29 and therefore their identity as Th1 seems quite likely. Moreover, they did express another chemokine receptor that has been associated with Th1 effectors, CXCR3.³¹ Interestingly, in the retinas of mice with uveitis, a higher percentage of the P2-specific Th1 expressed CXCR6 and also some expressed CX3CR1, which has been correlated with cytotoxic potential.³¹ It is possible that induced expression of CXCR6 in the eye-draining LN is related to the acquisition of pathogenic potential or alternatively that CXCR6 was induced on the Th1 cells in the retina, for example, by inflammatory cytokines. Additionally, it is striking that a substantial fraction of the effector Th1 in the LN of Aire^GW/+Lyn^−/−mice expressed high levels of inhibitory receptors, including PD-1 and LAG3 (Figs. 3E and 4). Interestingly, expression of inhibitory receptors by virus-specific Th1 cells was more prominent in mice with chronic LCMV infection than in mice with acute infections,³⁰ perhaps indicating that the autoimmune Th1 response in Aire^GW/+Lyn^−/−mice reflects a prolonged response to the autoantigen. While these studies have provided new insight into the nature of the Th1 autoreactive T cells in Aire^GW/+Lyn^−/−mice, more detailed studies will be required to characterize the phenotypic and functional properties of these cells and to determine how they relate to disease initiation.

A recent study analyzed the transcriptomes of autoreactive CD4 T cells recognizing 4 different islet cell-specific epitopes from the pancreatic islets of NOD mice with ongoing autoimmune destruction.⁴⁴ Autoreactive CD4 T cell were observed with a variety of phenotypes, including Th1 effector, Treg, anergic T cells, and a large cluster that the authors referred to as weak Th1 effectors, which had transcriptome similarity to our putative Tr1 population. Interestingly, CD4 cells recognizing 2 types of hybrid insulin peptides were highly skewed to Th1 effectors or possible Tr1 type in NOD pancreas tissue, similarly to what was seen in our study for IRBP P2-specific CD4 T cells in the retina, In contrast, CD4 T cells recognizing 2 variant forms of a peptide epitope of insulin B chain were highly skewed to Treg and anergic T cell phenotypes.⁴⁴ Peptides derived entirely from insulin are presented in the thymus, and therefore can induce clonal deletion and/or Treg differentiation of CD4 T cells recognizing them, whereas hybrid insulin peptides are created inside islet β cells, and thus are likely not presented in significant amounts in the thymus. As Aire is responsible for driving expression of IRBP in the thymus,¹³ central tolerance to IRBP is defective in mice with absent or diminished Aire function,^13,16 and thus the transcriptional similarity of IRBP P2-specific CD4 T cells in retinas of mice with uveitis to the CD4 T cells recognizing hybrid insulin peptides from pancreas of NOD mice with diabetes is likely due to the lack of central tolerance in the thymus in both cases. Moreover, in the study of Mitchell et al.,⁴⁴ transfer of CD4 T cells recognizing an insulin C chain-chromogranin A hybrid peptide induced diabetes, whereas transfer of CD4 T cells recognizing insulin B chain peptides were much less able to do so. Similarly, we found that TCR-transgenic IRBP P2-specific CD4 T cells spontaneously induced uveitis at an early age, but only if the mice also had a genetic defect in Aire.¹⁶ Thus, although there apparently are differences in the checkpoints that must be breached for spontaneous organ specific autoimmunity to develop in NOD mice versus in Aire^GW/+Lyn^−/− mice, there also may be a commonality that a lack of central tolerance to key autoantigenic epitopes is an important predisposing factor.

Human non-infectious uveitis is a heterogeneous disease that can be either standalone, or part of a systemic syndrome. The former can perhaps be explained by primary autoimmunity to retinal antigen(s), but the latter may be more “autoinflammatory” in nature and is accompanied by inflammatory damage to various tissues outside the eye including skin, joints, or intestine.⁴⁵ In the case of C57BL/6 Aire^GW/+Lyn^−/− mice, the organ-specific autoimmunity is confined to the retina and surrounding uvea,⁹ as is seen in standalone uveitis. The unique tissue-specificity can be explained by absence of IRBP extraocularly, which limits post-thymic tolerance on the one hand, while restricting the antigenic target tissue to the eye.^46,47

In summary, the observations presented here indicate that the Aire^GW/+Lyn^−/− mouse model of spontaneous autoimmune disease has considerable potential for uncovering mechanisms of autoimmune disease initiation as the key autoantigen has been identified and the autoreactive cells can readily be followed and analyzed. Moreover, the key checkpoint in going from autoimmune cell activation to disease onset appears to differ from the key checkpoint in the most highly studied animal model of spontaneous autoimmune disease, the NOD mouse model. Future studies can be envisioned to further test the hypothesis proposed above that fluctuations between animals in the small numbers of IRBP-specific CD4 T cells exiting the thymus and the fractions of them that become Tregs can tip the balance between immune control and autoimmune disease initiation in these mice engineered to have genetic susceptibility to autoimmune disease. Moreover, these mice will permit a more detailed tracking of the autoreactive CD4 T cells in the time before disease starts, which also promises to provide additional novel insights into how a small expansion of autoreactive CD4 T cells can lead to severe organ-specific tissue destruction or conversely can be controlled. Thus, the results presented here provide important new insights, and also raise significant new questions for defining the checkpoints that can prevent the development of spontaneous autoimmune disease in genetically susceptible individuals.

Author contributions

M.Y. and A.L.D. conceived and designed the study; M.Y. performed all aspects of the experimental work and analysis of single cell sequencing; K.J.H. provided help with the CyTOF analysis; I.P. provided advice regarding analysis of IRBP-specific T cells; J.C. and Y.M.H. genotyped mice; R.R.C. provided purified bovine IRBP protein and feedback on interpretations of the results. M.Y., C.A.L., and M.S. contributed new reagents and/or analytic tools. M.S.A. discussed the results, analysis and interpretations with M.Y. and A.L.D., M.Y. wrote the initial draft of the manuscript; and all authors reviewed drafts of the manuscript.

Mianmian Yin (Conceptualization [Equal], Data curation [Lead], Formal analysis [Lead], Investigation [Lead], Methodology [Lead], Project administration [Lead], Resources [Lead], Software [Lead], Supervision [Lead], Validation [Lead], Visualization [Lead], Writing—original draft [Lead], Writing—review & editing [Lead]), Kamir J Hiam-Galvez (Methodology [Supporting], Writing—review & editing [Supporting]), Irina Proekt (Methodology [Supporting], Writing—review & editing [Supporting]), Jackie Chan (Methodology [Supporting], Writing—review & editing [Supporting]), Yongmei Hu (Methodology [Supporting], Writing—review & editing [Supporting]), Clifford A. Lowell (Methodology [Supporting], Writing—review & editing [Supporting]), Rachel R. Caspi (Methodology [Supporting], Resources [Supporting]), Matthew H. Spitzer (Funding acquisition [Supporting], Methodology [Supporting], Writing—review & editing [Supporting]), Mark Anderson (Funding acquisition [Supporting], Resources [Supporting], Writing—review & editing [Supporting]), and Anthony L. DeFranco (Conceptualization [Lead], Data curation [Supporting], Formal analysis [Supporting], Funding acquisition [Lead], Investigation [Equal], Methodology [Supporting], Project administration [Supporting], Resources [Supporting], Software [Supporting], Supervision [Lead], Validation [Supporting], Visualization [Supporting], Writing—original draft [Supporting], Writing—review & editing [Equal])

Supplementary material

Supplementary material is available at The Journal of Immunology online.

Funding

This work was supported by NIH (National Institutes of Health) grants AI138479, R01AI097457, R01DE032033, and DP5OD023056. The CyTOF instrument used in this study was purchased with assistance by NIH award S10 OD018040.

Conflicts of interest

M.H.S. is founder and shareholder of Pro Biosciences and Teiko.bio, has received a speaking honorarium from Fluidigm Inc., Kumquat Bio, and Arsenal Bio, has been a paid consultant for Five Prime, Ono, January, Earli, Astellas, and Indaptus Therapeutics, and has received research funding from Roche/Genentech, Pfizer, Valitor, and Bristol Myers Squibb. The other authors have declared that no conflict of interest exists.

Data availability

Data for single cell sequencing have been deposited at GEO and are publicly available as of the date of publication. Accession numbers is GSE272733. This article does not report the original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.