Impaired Activated/Memory Regulatory T Cell Clonal Expansion Instigates Diabetes in NOD Mice

NOD mice spontaneously develop autoimmune diabetes due to the destruction of insulin-producing β-cells by autoreactive CD4⁺ and CD8⁺ T cells (1). Diabetes susceptibility in NOD mice is attributed to genetic defects encoded by the insulin-dependent diabetes loci including >40 genes, among which the MHC class II and interleukin-2 (IL-2) genes have major contributions to the pathophysiological immune response leading to diabetes (2). The unique I-A^g7 MHC class II molecule expressed by NOD mice has been reported to bind certain peptides with low affinity, thus impairing the thymocyte negative selection process (3,4). Reduced IL-2 production by activated effector T cells (Teffs) was found in NOD mice and linked to an impairment of regulatory T cell (Treg) numbers and functionality (5–7). An increased diabetes incidence was observed in NOD mice bearing an IL-2α receptor subunit with reduced affinity for IL-2, which was associated with a Treg but not a Teff defect (8). Likewise, the administration of low-dose IL-2 to NOD mice promotes Treg expansion and activation, prevents diabetes onset, and even cures diabetic mice (9,10). Thus, the altered tolerance promoting diabetes in NOD mice could be the result of: 1) a defect in thymic selection of Treg and Teff T-cell receptor (TCR) repertoires, and/or 2) an altered Treg fitness due to IL-2 deprivation.

Previous studies of the TCR repertoires of NOD mice and patients with type 1 diabetes yielded inconsistent results. Some studies showed a restricted TCR β-chain (TRB) V and TRBJ chain usage by CD4⁺ T cells from pancreatic islets and pancreatic lymph nodes (PLN), indicating that overrepresented islet-specific TCRs preferentially express certain TRBVJ genes (11–13). Conversely, others showed a rather diverse gene usage among islet-infiltrating CD4⁺ T cells in NOD mice (14), including islet-infiltrating memory CD4⁺ T cells (15,16), and no common highly used TRBVBJ combinations in T cells from PLN of patients with type 1 diabetes (17). Moreover, NOD Tregs were found to have a restricted TCR α-chain (TRA) repertoire and limited overlap with thymic CD4⁺ T cells (18,19), while we found similar TRB diversity between NOD peripheral Tregs and Teffs (20). These somehow conflicting results can be explained by the fact that: 1) most of these studies were performed on unsorted CD4⁺ T cells and thus did not distinguish between regulatory and effector populations, and/or 2) most Treg repertoire analyses were done on a limited fraction of the repertoire by using TCR transgenic NOD mice or by focusing on few rearrangements (18,19). More importantly, most studies have overlooked the existence of Treg subsets.

Indeed, we previously showed that Tregs could be phenotypically divided into CD4⁺Foxp3⁺CD44^lowCD62L^high naive Tregs (nTregs) that have a low turnover and CD4⁺Foxp3⁺CD44^highCD62L^low activated/memory Tregs (amTregs) that have a very rapid turnover due to their interaction with their cognate antigens (21). In the healthy C57BL/6 (B6) mouse background, we showed that amTregs are enriched in deep tissue–draining lymph nodes and are characterized by unique clonotype expansions (22). Moreover, we showed little overlap between amTreg and nTreg TRB repertoires (22). Thus, studying amTreg and nTreg TCR repertoires without separating them may be confounding.

In this study, we compared the TCR repertoire of splenic amTregs, nTregs, and Teffs of prediabetic NOD and normal B6 mice to investigate whether and how an altered TCR repertoire contributes to diabetes. We found that NOD amTregs have an unexpectedly more diverse repertoire than that of B6 mice, due to the absence of expanded clonotypes. Treatment of prediabetic NOD mice with IL-2, which prevents diabetes occurrence, restored the expansion of clonotypes that are preferentially diabetes-reactive. Thus, the TCR repertoire of amTregs is altered by an IL-2 insufficiency, which instigates diabetes development in NOD mice, and is amenable to therapeutic recovery.

Given the rarity of some of the studied cell subsets, we pooled the spleens of six to eight young mice matched for genetic background, age, and sex. Six (three female and three male) nontreated C57BL/6 or NOD pools of mice were used to study the repertoire at homeostasis. Six young female NOD mice were treated with IL-2. Sample size was determined for statistical significance and experimental feasibility. Investigators were not blinded to the allocations during experiments and analyses.

Eight- to 10-week-old male and female C57BL/6- and NOD-Foxp3-EGFP transgenic mice expressing GFP under the control of the Foxp3 gene promoter were, respectively, provided by B. Malissen (Luminy, Marseille, France) and V. Kuchroo (Brigham and Women’s Hospital, Boston, MA). All animals were maintained at the Sorbonne Université Centre d’Expérimentation Fonctionnelle animal facility under specific pathogen-free conditions in agreement with the current European legislation on animal care, housing, and scientific experimentation (agreement number A751315). All procedures were approved by the local animal ethics committee (Paris, France).

Recombinant AAV8 vectors were produced as described previously (9). Six female NOD-Foxp3-EGFP mice were injected once intraperitoneally with 10¹¹ of recombinant AAVs diluted in PBS at 6 weeks of age. Serum was collected 2 weeks postinjection, and IL-2 levels were measured using a mouse IL-2 ELISA (eBioscience) according to the manufacturer’s recommendations. Mice were sacrificed at 9 weeks of age.

Spleens were collected from euthanized mice and splenocytes were stained with anti–Ter-119-biotin (BD Biosciences), anti–B220-biotin (eBioscience), and anti–CD11c-biotin (eBioscience) antibodies for 20 min at 4°C and labeled with anti-biotin magnetic beads (Miltenyi Biotec) for 15 min at 4°C. B cells, erythrocytes, and monocytes were depleted on an AutoMACS separator (Miltenyi Biotec) following the manufacturer’s procedure. Enriched T cells were stained for 20 min at 4°C with the following monoclonal antibodies at predetermined optimal dilutions: CD3-APC (BD Biosciences), CD4-Horizon-V500 (BD Biosciences), CD8-Alexa-700 (BD Biosciences), CD44-PE (BD Biosciences), and CD62L-eFluor-450 (eBioscience) and sorted on an FACSAria II cytometer (BD Biosciences) with a purity >95% into the following subsets: CD4⁺FoxP3⁺CD62^lowCD44^high (amTregs), CD4⁺FoxP3⁺CD62^highCD44^low (nTregs), and CD4⁺FoxP3⁻ (Teffs). A total of 1 × 10⁵–5 × 10⁶ sorted cells were stored in RNAqueous kit lysis buffer (Invitrogen) at −80°C.

RNA was extracted using the RNAqueous Total RNA Isolation Kit (Invitrogen). TRB libraries were prepared on 100 ng of RNA with the SMARTer Mouse TCR a/b Profiling Kit (Takara Bio) and sequenced with HiSeq 2,500 single read (300 bp) (Illumina) plus 10% PhiX at the LIGAN Genomics platform (23).

Raw data in FASTQ format were processed for TRB sequence annotation using MiXCR software (v.3.0.3) (24), which ensures PCR and sequencing error corrections and provides for each sample a list of unique amino acid TRBs and their corresponding counts. Annotated sequences were then analyzed using an in-house workflow to eliminate long amino acid CDR3 sequences (length higher than the mean + 8). The resulting data sets are summarized in Supplementary Table 2.

Sample overlap was tested by computing the Jaccard distance (25) using the “vegan” R package and plotted into a heat map using the “pheatmap” package (26). Clustering analysis was performed using the “Ward” method. Principal component analysis was performed using the “stats” R package. Rényi entropy (27) was calculated using “vegan” and represents the distribution of clonal expansions as a function of the parameter α. Probability density plots were plotted using the “ggplot2” R package (28).

For each cell population, the following method was adopted: the top most frequent 1,000 CDR3 amino acid sequences from each of the 6 B6 or NOD mice were pooled, and the mean frequency was calculated for every sequence. The list was then ordered decreasingly, and the first 1,000 sequences were selected. A matrix of pairwise Levenshtein distances (LDs) between the CDR3s was computed and used to construct the networks. In these networks, vertices (CDR3 sequences) are connected by edges (LD = 1) only if they differ by one amino acid (insertion/deletion/substitution). A connected cluster is defined as a set of two or more vertices connected by edges. The clustering coefficient of a node in a network quantifies how close its neighbors are to being a clique (a closed cluster). The maximum value is 1 if every neighbor connected to the node A is also connected to every other node within the neighborhood. The node size represents the mean frequency of each CDR3 in the network.

Analyses were performed using R packages: “stringdist” (29) was used to calculate LDs and “Igraph” (30) to calculate network properties. Graphics visualization was done using Cytoscape (31) and “ggplot2” for Figs. 3 and 5, respectively. In the latter, only clustered nodes are represented and edges are not shown.

In each sample, the first (Q1) and third (Q3) quartiles and the interquartile range (IQR) were computed according to a negative binomial distribution fitted in the sequence read counts, while excluding private clonotypes with a count of 1. Expanded clonotypes are defined as those with counts greater than Q3 + 1.5 × IQR.

Mouse CDR3βs associated with diabetes (n = 52), experimental autoimmune encephalomyelitis (EAE; n = 50), systemic lupus erythematosus (SLE; n = 49), and cancer (n = 52) were collected from Friedman’s published database (32). These sequences were derived from CD4⁺ T cells and identified by different methods including reactivity assays, tetramer staining, selection of diabetes-specific TCRs following diabetes induction, and sequencing of tumor T-cell infiltrates in the context of cancer studies.

Statistical analyses were performed using the nonparametric Wilcoxon test with Holm multiple-testing correction using “ggpubr” R package (33). A P value of ≤0.05 was considered statistically significant: *P ≤0.05; **P ≤0.01; ***P ≤0.001; ****P ≤0.0001; not significant is P > 0.05. Mean percentage comparisons of diabetes, EAE, SLE, and cancer CDR3s were done using the Wilcoxon paired test with diabetes as a reference group.

Source data are provided in the manuscript and the Supplementary Material. Fastq data were deposited in the National Center for Biotechnology Information Sequence Read Archive repository under the BioProject identification number PRJNA635928.

We analyzed and compared the TRB repertoire of splenic CD4⁺Foxp3⁺CD44^lowCD62L^high nTregs, CD4⁺Foxp3⁺CD44^highCD62L^low amTregs, and Foxp3⁻CD4⁺ Teffs in B6 and NOD mice.

Compared with B6 mice, flow cytometry showed a significant decrease in the NOD Treg population (P ≤ 0.05) associated with a decrease of amTreg proportions (P ≤ 0.01) (Supplementary Table 1). Cell numbers for each sorted population and their corresponding number of TRB sequences and unique amino acid clonotypes are summarized in Supplementary Table 2.

Principal component analysis of TRBVBJ usage (i.e., the frequency of each combination in the repertoire irrespective of the clonotype abundance) clearly separated the genetic backgrounds and the cell populations (Fig. 1A). The first two components showed different TRBVBJ usage between B6 and NOD mice, with the first component (PC1) that separates the two genetic backgrounds explaining >68% of the variability. There was also a clear difference between the cell populations from the same background, with the two Treg subsets being close, but quite apart from Teffs. This reflects differences in the combined generation frequency and survival through the thymic selection and peripheral expansion of each generated TRBVBJ recombination.

To further investigate these differences, we assessed the clonotype (i.e., unique amino acid TRBV-CDR3-TRBJ sequence) overlap between samples using the Jaccard distance in which the higher the value, the more dissimilar the pair of samples compared (Fig. 1B). Hierarchical clustering on the Jaccard scores accurately separated B6 from NOD mouse samples. Surprisingly, within the latter cluster, amTregs are intermingled with nTregs and Teffs, while, as expected from our previous study (22), amTregs clustered separately in the B6 background. Overall, NOD mice display a peculiar repertoire structure compared with B6 mice both at the TRB gene and clonotype level.

We next explored the overall repertoire diversity at the clonotype level. We compared cell populations and mouse backgrounds by calculating the Rényi entropy, which evaluates diversity as a function of an α parameter (Fig. 2A). The higher the α, the higher is the weight put on frequent clonotypes. The higher the Rényi entropy, the more diverse is the repertoire.

In B6 mice, nTregs have the most diverse repertoire, while amTregs have the least diverse one, and Teffs fall in between. In contrast, while nTregs in NOD mice have a repertoire with a diversity similar to that in B6 mice, the diversities of the Teff and amTreg repertoires are increased compared with B6 mice (Fig. 2A). At α = 0, for which the Rényi index represents clonotype richness, no differences were observed between the B6 and NOD backgrounds (Fig. 2B, left). This indicates that the overall richness of each cell subset is not affected in NOD mice. At α = 8, which gives more weight to abundant clonotypes, Rényi indices were significantly higher in NOD amTregs and Teffs than in the corresponding subsets in B6 mice (Fig. 2B, right).

In view of these differences, we looked at the representativeness of the 100 most predominant clonotypes within each repertoire. The cumulative percentages of the 100 most predominant clonotypes in B6 amTregs and Teffs were significantly higher than their NOD counterparts (Fig. 2C). This is due to the presence, in the B6 cell populations, of highly frequent clonotypes depicted by blue lines, for which thickness is proportional to their percentage in the repertoire (Fig. 2D). In contrast, there were no such clonotypes in B6 nor in NOD nTregs.

Collectively, these results reveal that a healthy nTreg repertoire is diverse with no clonotype expansions. Conversely, the healthy amTreg repertoire, and to a lesser extent the Teff repertoire, is enriched in clonotypes with high frequency, likely resulting from antigen-driven expansions. In diabetes-prone NOD mice, such expansions are severely reduced in amTregs and Teffs.

Next, we examined the architecture of the repertoires. To this end, the 1,000 most frequent CDR3s per population were used to build similarity networks in which each CDR3 (node) has a size proportional to its frequency and is connected by edges to other CDR3s that differ to it by only one amino acid (i.e., having an LD = 1). This distance was shown to connect CDR3s that are likely to bind the same peptide (34).

The structure of the nTreg repertoire appeared similar for B6 and NOD mice, made of highly clustered CDR3s with low frequency (Fig. 3A). In contrast, B6 amTreg repertoire was mostly composed of individual or lightly connected with high frequencies, reflecting polyclonal self-antigen–driven expansions (22,35). In accordance with the previous observations (Figs. 1 and 2), these expansions are much reduced in amTregs from NOD mice. Similarly, node sizes of their Teffs appeared smaller than in B6 mice (Fig. 3A, middle).

Node frequency showed no differences for nTregs, a modest decrease for Teffs, and a marked decrease for amTregs in NOD mice compared with their counterparts in B6 mice (Fig. 3B). The clustering coefficient of each CDR3 within each network, which gives an indication of the importance of these nodes in the global repertoire architecture, showed no significant difference between NOD and B6 nTregs (Fig. 3C). However, significantly higher clustering coefficients were observed in Teffs and amTregs from NOD compared with B6 mice.

Overall, the structural analysis of these networks highlights a higher sequence similarity of Teff and amTreg CDR3s in NOD mice. As IL-2 production is notoriously low in NOD mice and Tregs rely primarily on it for proliferation, we hypothesized that the reduced amTreg expansions may result from an IL-2 shortage.

As administration of IL-2 to NOD mice specifically enables the expansion of Tregs (9), we investigated whether the abnormal repertoire diversity and structure of prediabetic NOD mice could be restored by an IL-2 treatment. NOD mice were injected with an adeno-associated virus coding for IL-2 (AAV-IL-2) that allows the continuous production of IL-2 at low dose and prevents the occurrence of diabetes (9). As expected, total Treg proportions increased significantly after IL-2 treatment (Fig. 4A). Importantly, within the Treg subsets, amTreg proportions showed a significant increase, at the expense of nTregs (Fig. 4A and Supplementary Table 3).

We next analyzed the expansions at the clonotype level after IL-2 treatment. To this end, we identified expanded clonotypes as the ones showing counts greater than the Q3 + 1.5 × IQR value, in which Q3 is the third quartile and IQR the IQR of sequence count distribution (see 2research design and methods and Supplementary Table 4). The abundance of the expanded clonotypes from each sample was plotted in Fig. 4B (left plots). Each circle represents a clonotype for which size is proportional to its frequency in the repertoire. amTregs and nTregs from AAV-IL-2–treated NOD mice were significantly enriched in expansions compared with amTregs and nTregs from B6 or NOD mice. This is confirmed by the density plots, which show that IL-2 treatment induces expansions of nTregs and amTregs compared with their B6 and NOD counterparts, but not of Teffs (Fig. 4B, right). Thus, IL-2 administration restores amTreg clonotype expansions that were absent in NOD mice.

We next examined whether the expanded clonotypes could be linked to diabetes. We queried a published catalog of TCRs with defined specificity (32). We identified a total of 203 mouse CDR3β amino acid sequences from CD4⁺ T cells associated with diabetes (n = 52), EAE (n = 50), SLE (n = 49), and cancer (n = 52). We then searched for these sequences within the expanded clonotypes and calculated their mean enrichment in each cell subset (Fig. 5A). Interestingly, diabetes-specific sequences showed higher expansions compared with nondiabetes-specific sequences in nTregs and amTregs from IL-2–treated NOD mice. There was also a modest enrichment of these sequences in B6 amTregs. To further analyze the specificity pattern toward diabetes, we connected disease-specific CDR3s by LD = 1 to the pool of expanded amTreg and nTreg CDR3s. The number of neighbors with LD = 1 to diabetes-related sequences showed a significant increase compared with other disease-specific CDR3s (Fig. 5B).

Noteworthily, clonotypes identified as specific for islet antigens such as insulin and GAD by tetramer staining were detected in large clusters of IL-2–expanded clonotypes (Fig. 5C and Supplementary Table 5).

In this study, we report an altered amTreg TCR repertoire in NOD mice that can be restored to normal by an IL-2 treatment. IL-2–expanded Tregs were enriched in islet-antigen–specific TCRs and were associated with protection from diabetes. These results have both heuristic and applied implications.

In normal B6 mice, a control model that does not spontaneously develop any inflammatory or autoimmune diseases, nTreg repertoires are very diverse, even more than those of Teffs, with most clonotypes forming dense networks. These are the attributes of an unmodified and tightly connected postthymic selection repertoire (36). In contrast, amTreg repertoires have a reduced diversity and are enriched in frequent clonotypes with few connections. This illustrates the responses to distinct self-antigens, as expected from a cell population that is constantly activated to control autoimmunity in numerous tissues. Such similarity structures have also been described in the context of immunization and aging (34).

In addition, amTreg repertoires show a higher overlap with each other than with nTregs or Teffs, indicating that the self-antigens to which amTregs respond are shared between individuals. This is also in agreement with our previous report of a low sequence overlap between the two Treg subsets in the peripheral lymph nodes (22). The comparison with Treg TRB repertoires from the NOD autoimmune-prone mouse strain strengthens this conclusion. Reduction of amTreg major expansions is associated with disease susceptibility and their recovery with disease prevention.

Collectively, these results identify amTregs as the major subset involved in the pathogenesis of autoimmunity and warrant separate analysis of amTregs and nTregs when investigating their roles in health or disease.

The severe reduction in frequent clonotypes found in NOD mice, and their appearance under IL-2 provision, highlight a major role of IL-2 in supporting amTreg survival.

NOD mice have a genetically determined IL-2 insufficiency, which is also reported in human type 1 diabetes as well as in SLE and rheumatoid arthritis (37,38). It is thus likely that the observed defect in amTreg fitness contributes to the pathophysiology of these diseases. If so, the good news is that disease-relevant amTregs can be regenerated by IL-2 treatment. We show in this study that one injection of recombinant AAV-IL-2 in prediabetic mice was enough to expand NOD amTregs at the cell and repertoire levels. In our experiments, the repertoire was analyzed 21 days after the initiation of treatment. As the approximate duration of T-cell differentiation in the thymus is 28 days (39), newly generated thymocytes are unlikely to contribute to the expanded clonotypes in the amTreg population. Thus, the IL-2–induced amTregs most probably originate from peripheral amTregs and/or nTregs that received their activation signal in the presence of IL-2. The diabetes-related TCR expansions in both nTregs and amTregs further support that both populations contribute to the restoration of a more balanced and complete disease-relevant amTreg repertoire.

The existence of public databases containing TCRs with known peptide specificities allowed us to investigate the presence of antigen-specific sequences in our repertoires. Diabetes-related CDR3s identified from the database were overrepresented in IL-2–treated NOD Tregs and most connected in LD = 1 networks, as shown by their high number of neighbors. Linking the CDR3 similarity networks to annotated TCR data sets was previously described by Madi et al. (34) and showed that CDR3s with similar specificity tend to cluster together. This indicates that diabetes-specific CDR3s could have core motifs engaged in pancreatic antigen recognition and shows the potential of such a tool in identifying novel pathogenic TCRs. Indeed, repertoire analysis of islet-infiltrating T cells revealed an accumulation of TCRs that were reactive in vitro to uncharacterized autoantigens, distinct from the well-known GAD and insulin antigens (13).

These observations highlight a relationship between the specific expansion of pancreatic antigen-related TCRs and the prevention of diabetes post–IL-2 treatment.

It was hypothesized that diabetes development in NOD mice is associated with the I-A^g7 molecule that they express. Coexpression with a different type of MHC or mutations within the I-A^g7 molecule was shown to prevent insulitis and diabetes (4,40). In contrast, substitution of non-MHC regions from the C57BL/10 into the NOD strain was sufficient to prevent insulitis or diabetes, despite the expression of the I-A^g7 molecule (41). Our studies highlight the importance of non-MHC genes in diabetes onset in the NOD mouse model. Indeed, we report in this study that IL-2 administration to prediabetic NOD mice, previously shown to prevent the disease outcome, improves amTreg fitness. This would argue that NOD Treg repertoires have the full potential to control autoimmune responses, but fail to do so because an IL-2 deficiency impairs the amTreg repertoire that can control islet autoreactivity.

However, our results do not exclude a potential enrichment of autoreactive T cells in the NOD mice repertoire that would take the lead in the periphery where Tregs fail to expand. In this regard, NOD mice have an overrepresentation of CDR3β sequences carrying hydrophobic amino acid doublets at positions 6 and 7 that have been shown to confer high self-reactivity (42). Further studies of the TCR repertoire selection process in the thymus of NOD mice should help address this important issue.

Altogether, our results reveal a link between a reduced expansion of clonotypes within the activated Treg TCR repertoire and the development of an autoimmune disease. IL-2 treatment not only activates Tregs, but also restores a repertoire of TCRs with relevant specificities. Thus, our work indicates that fit amTregs are key to protection from autoimmune diseases and highlights the value of studying these cells separately from the whole Treg population. Importantly, amTreg repertoire abnormalities and restoration are detected in the spleen, which affords a readout that is more representative of the circulation than PLN and thus has a translational value. Our results should guide further studies to elucidate the dynamics of pancreatic antigen-specific amTregs in diabetic versus nondiabetic NOD mice, including in mice outside of a pathogen-free environment, and identify TCR-based biomarkers of diabetes.

Acknowledgments. The authors thank F. Jebbawi and L. Barientos for the help in sample collection and cell sorting, M. Mendez and N. Papaya for the help in RNA extraction and library preparations, and K. El Soufi for the help in network analyses. The authors also thank the staff of the animal care facility for the excellent handling of mice and the UMR 8199 LIGAN-PM Genomics platform (Lille, France), which belongs to the “Federation de Recherche” 3508 Labex European Genomics Institute for Diabetes (ANR-10-LABX-46) and was supported by the ANR Equipex 2010 session, for performing the sequencing of our libraries.

Funding. This work was funded by the Training & Research Involving Polymer Optical Devices European Research Council Advanced EU (322856) and Recherche Hospitalo-Universitaire iMAP (ANR-16-RHUS-0001) grants to D.K., with additional support from the Sorbonne Université, INSERM, and Assistance Publique-Hôpitaux de Paris.

Duality of Interest. H.P.P was an employee of ILTOO Pharma. D.K. is an investor in a patent related to the use of IL-2 in the treatment of autoimmune diseases, owned by his public institution and licensed to ILTOO Pharma, in which D.K. has interests. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. V.M., G.F., P.B., V.Q., P.-G.R. F.B., B.G., G.C., and E.M.-F. performed the experiments. V.M. analyzed the data, with contributions from all authors. V.M. and H.P.P. performed the statistical analyses. V.M., A.S., E.M.-F., and D.K. conceived the experiments. V.M., A.S., E.M.-F., and D.K. wrote the manuscript, with input from all authors. D.K. conceptualized and supervised the study. D.K. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.