It was previously shown that Tregs developing during the neonatal period protect better from lethal autoimmunity upon adoptive transfer into AIRE-deficient NOD mice than cells differentiating in adult animals (6). We investigated the development of this population in NOD mice and compared it with that in other commonly used inbred mouse strains. Wyss and colleagues previously showed that newly developed Tregs expressing a GITR^highPD-1⁺ phenotype are more autoreactive than their GITR⁺PD-1⁻ counterparts (3). We therefore investigated the GITR vs. PD-1 phenotypes of Tregs newly developing in neonates (3 to 4 days old) of B6, BALB/c, C3H, CBA, DBA/1, FVB, NOD, and SJL mice. The proportions of GITR^highPD-1⁺ cells among CD4⁺CD8⁻TCR^highFoxp3⁺ Tregs (gated as shown in Supplementary Figure S1A ) varied dramatically, from 20.6% ± 2.1% in C3H mice to 49.4% ± 5.8% in NOD animals ( Figure 1A ). NOD mice also had particularly great proportions of GITR^highPD-1⁺ Tregs among CD4⁺CD8⁻TCR^high thymocytes (CD4SP, Supplementary Figure S1B ). These data suggested a genetic control of the development of strongly autoreactive GITR^highPD-1⁺ Tregs, paradoxically very robust in the NOD mouse.

Treg precursors are not entirely resistant to thymic deletion (22). It was therefore important to evaluate if the potentially highly autoreactive newly developed GITR^highPD-1⁺ Tregs can leave the thymus and populate the periphery. To assess this question, we first analyzed the expression of S1PR1, sufficient for the thymic export of T cells (23). Among thymus export-competent (S1PR1⁺) Tregs, we observed substantially more GITR^highPD-1⁺ cells in NOD mice than in B6 neonates ( Figure 1B , Supplementary Figure S1C ). We also observed much higher proportions of GITR^highPD-1⁺ cells among peripheral (splenic) Tregs in 4-day-old NOD than in age-matched B6 mice ( Figure 1C ).

Phenotypic differences between NOD and B6 Tregs have, to our knowledge, not been described. Moreover, it was reported that very autoreactive Tregs develop during the neonatal period, but not later on (24, 25). We therefore assessed the kinetics of GITR^highPD-1⁺ Treg development in NOD and B6 mice. Intriguingly, we observed higher levels of these cells in NOD than in B6 thymi only during the first week of age. Already at 1 week of age, the proportion of developing GITR^highPD-1⁺ Tregs had strongly dropped in NOD mice, and later on, they were similar in the two mouse strains ( Figure 1D ). Combined, these data indicate that, paradoxically, particularly high proportions of potentially strongly autoreactive GITR^highPD-1⁺ Tregs develop during the neonatal period in the NOD thymus and populate peripheral lymphoid organs.

The difference in neonatal GITR^highPD-1⁺ Treg development between NOD and B6 mice may be controlled by genetic factors acting within developing T cells and/or provided by the thymic microenvironment. To investigate this issue, we generated reconstituted fetal thymus organ cultures (rFTOC, Supplementary Figure S1D ). We depleted fetal NMRI thymi of hematopoietic cells and we seeded these emptied thymic lobes with T-cell precursors from 8-week-old B6 or NOD mice. To avoid the potential contribution of dendritic cells and B lymphocytes developing from thymic T-cell precursors, we used FACS-sorted thymocytes at the DN3 stage of development (CD4⁻CD8⁻TCR⁻CD44⁻CD25⁺), known to only have T-cell precursor potential (26). Whereas 36.6% ± 7.2% of Tregs developing from B6 DN3 in such rFTOC expressed PD-1 and high levels of GITR, 49.4% ± 7.8% of NOD Tregs had this phenotype ( Figure 1E ). These results indicated that part of the difference in GITR^highPD-1⁺ Treg development between B6 and NOD mice was due to factors acting within the T-cell compartment. Since the precursor cells seeded into the cultures were of adult origin, our observations also showed that the difference is not caused by factors intrinsic to newborn precursors.

It was paradoxical to find particularly robust development of GITR^highPD-1⁺ and, therefore, presumably highly autoreactive, neonatal Tregs in the T1D-prone NOD mouse. However, the highly autoreactive nature of GITR^highPD-1⁺ Tregs was described for B6 mice (3) and may or not be valid for NOD Tregs. Also, for technical reasons, we did not include in our study the CD25 marker used by Wyss and colleagues, which might bias the interpretation of our results. We therefore compared the autoreactivity of GITR^highPD-1⁺ vs. GITR⁺PD-1^– Tregs in NOD and B6 mice. Neonatal GITR^highPD-1⁺ Tregs from B6 and NOD mice expressed significantly lower levels of TCR and higher levels of Nur77 and CD5 than GITR⁺PD-1^– cells ( Figures 2A–C , Supplementary Figures S2A, B ). These data suggested a similarly higher autoreactivity of GITR^highPD-1⁺ than of GITR⁺PD-1^– Tregs in NOD and B6 mice (27–29), thus confirming and extending previously published data on B6 animals (3). Moreover, in NOD but not in B6 mice, thymic stromal cells express a superantigen encoded by the endogenous mammary tumor virus-3 (Mtv-3). When presented by the MHC class II molecule I-A^g7, the Mtv-3 superantigen strongly activates all T cells expressing the TCR Vβ3 variable segment. Strongly autoreactive Vβ3-expressing T-cell precursors are therefore deleted in the NOD thymus (30). Given that superantigens can also induce the selection of reactive Tregs (31), we investigated the development of Vβ3-expressing Tregs, highly autoreactive in Mtv-3-expressing NOD but not in B6 mice which lack Mtv-3. We found many more Vβ3-expressing cells among GITR^highPD-1⁺ than among GITR⁺PD-1^– Tregs in NOD but not in B6 neonates ( Figure 2D , Supplementary Figure S2C ). Combined, these data suggested that, as in B6 mice, in NOD animals GITR^highPD-1⁺ Tregs were substantially more autoreactive than GITR⁺PD-1^– cells.

To formally assess the particularly (most probably auto-) reactive nature of NOD GITR^highPD-1⁺ Tregs, we next analyzed their activation in the periphery. Upon leaving the thymus, a large proportion of Tregs is activated in the periphery (32). The expression levels of the T-cell activation marker CD44 were noticeably higher on GITR^highPD-1⁺ than on GITR⁺PD-1^– Tregs ( Figure 2E , Supplementary Figure S2D ). We also found many more cells expressing the proliferation marker Ki67 among GITR^highPD-1⁺ than among GITR⁺PD-1^– Tregs ( Figure 2F , Supplementary Figure S2E ). In Rag2-Gfp mice, GFP accumulates in developing T cells up to the extinction of Rag2 expression upon positive selection and then degrades with an in-vivo half-life of 56 h (33). The proliferation of recent thymic emigrants, which still express detectable levels of GFP, will lead to the dilution of GFP. At the age of 4 days, Tregs just start to leave the mouse thymus (34). In the spleen of 4-day-old mice, GITR^highPD-1⁺ recent thymic emigrant Tregs displayed markedly lower GFP fluorescence levels than GITR⁺PD-1^– cells ( Figure 2G ), indicating a higher proliferation rate of GITR⁺PD-1⁺ cells compared with their GITR⁺PD-1^– counterparts. Upon their activation in the periphery, Tregs recirculate back to the thymus (21, 35). If GITR^highPD-1⁺ are more activated than GITR⁺PD-1^– cells, the former cells should preferentially recirculate to the thymus. Consistent with this premise, we found that the proportions of GITR^highPD-1⁺ cells among recirculating GFP⁻ Tregs in the thymus were much higher than those among Tregs in the spleen of B6 newborns and, to an even greater extent, of NOD mice ( Figure 2H ). These data indicate that GITR^highPD-1⁺ Tregs preferentially recirculate back to the thymus. Combined, our data confirm and extend the particularly (probably auto-) reactive nature of GITR^highPD-1⁺ Tregs (3). Importantly, we obtained very similar differences between GITR^highPD-1⁺ vs. GITR⁺PD-1^– Tregs in B6 and NOD mice. Taken together, these data further indicate that a particularly high proportion of strongly autoreactive Tregs develops in the neonatal NOD thymus.

The very diverse levels of GITR^highPD-1⁺ Tregs we found in distinct inbred mouse strains ( Figure 1A ) suggested that their development is modulated by genetic factors. To more formally address this possibility, we next performed a genetic analysis. We crossed NOD mice to B6 animals and analyzed neonatal Treg development. In the thymus of F1 mice, we found proportions of GITR^highPD-1⁺ cells among Tregs similar to those observed in NOD mice ( Figure 3A ). The B6 phenotype therefore appears recessive. We then backcrossed F1 mice to B6 animals. These “backcross 1” (BC1) animals displayed a large diversity in the proportions of GITR^highPD-1⁺ cells among Tregs ( Figure 3A ). These results formally confirmed genetic control of the development of GITR^highPD-1⁺ Tregs and indicated its polygenic nature.

The I-A^g7 allele expressed by NOD mice is a major T1D susceptibility locus (“Idd1”) and regulates non-cognate negative selection in the thymus (36). Given the highly autoreactive nature of GITR^highPD-1⁺ Tregs, we postulated that I-A^g7 may play an important role in their differentiation. To assess this possibility, we identified homozygous I-A^b/b and heterozygous I-A^b/g7 mice among the BC1 neonates and compared the proportions of GITR^highPD-1⁺ cells among newly developing (i.e., GFP⁺) Tregs. We did not observe any difference between I-A^b/b and heterozygous I-A^b/g7 mice ( Supplementary Figure S3 ), excluding a role for the I-A^g7 allotype in GITR^highPD-1⁺ Treg development.

More than 40 other T1D susceptibility loci have been identified (37). To investigate the potential implication of these loci in the development of GITR^highPD-1⁺ Tregs, we first determined the Idd genotypes of the BC1 neonates by PCR and then compared the proportions of GITR^highPD-1⁺ Tregs among GFP⁺ Thy1.1⁺ CD4SP Tregs ( Supplementary Figure S4 ). Several Idd appeared involved in the development of these cells. Thus, the heterozygous presence of the NOD alleles of Idd5 and Idd26 (both localized on chromosome 1, “Chr. 1”) conferred increased proportions of GITR^highPD-1⁺ Tregs ( Figure 3B ). By contrast, the heterozygous presence of the NOD allele of Idd8 (Chr. 14), Idd12 (Chr. 14), and Idd22 (Chr. 8) unexpectedly conferred reduced proportions of GITR^highPD-1⁺ Tregs ( Figure 3B ). These data indicate that T1D susceptibility loci genetically control the development of GITR^highPD-1⁺ Tregs.

To comprehensively identify genomic loci controlling the development of GITR^highPD-1⁺ Tregs, we submitted genomic DNA from the BC1 animals to genome-wide, high-density single nucleotide polymorphism (SNP) profiling ( Figure 3C ). We thus identified loci on chromosomes (Chr.) 1 and 8 that control the difference in GITR^highPD-1⁺ Treg development between B6 and NOD mice ( Figures 3D, E ). On Chr. 1, two loci may be involved ( Figure 3D ). In BC1 mice heterozygous (B6/NOD) for these two loci, significantly more GITR^highPD-1⁺ Tregs developed than in homozygous (B6/B6) pups ( Figures 3F, G ). These observations confirm the dominance of the NOD allele(s) which favour(s) the development of GITR^highPD-1⁺ Tregs. However, the difference between mice homozygous and heterozygous for these loci was substantially smaller than the difference between B6 and (B6xNOD)F1 mice (cf. Figures 3A, F, G ), again indicating the polygenic control of the GITR^highPD-1⁺ Treg phenotype. In BC1 neonates heterozygous (B6/NOD) for the locus identified on Chr. 8, significantly less GITR^highPD-1⁺ Tregs developed than in homozygous (B6/B6) pups ( Figure 3H ). This observation indicates the dominance of the NOD allele(s) which, in contrast to the locus on Chr. 1, inhibits the development of GITR^highPD-1⁺ Tregs. The NOD alleles of the loci on Chr. 1 and 8 therefore control GITR^highPD-1⁺ Treg development in an opposite fashion.

The identified loci on chromosome 1 largely overlap with the diabetes susceptibility locus Idd5 (38) ( Figure 3D ). To assess the potential implication of Idd5 in neonatal GITR^highPD-1⁺ Treg development, we analyzed congenic NOD neonates carrying the Idd5 locus of B10 origin (NOD.B10 Idd5 mice, strain R974 (39)). Among Foxp3⁺ CD4SP Tregs in the thymi of NOD.B10 Idd5 mice, we found lower proportions of GITR^highPD-1⁺ cells than among NOD Tregs ( Figure 3I ). These data thus show that the particularly robust development of GITR^highPD-1⁺ Tregs in NOD mice is, in part, controlled by the T1D susceptibility locus Idd5.

B6, BALB/c, C3H, CBA, DBA/1, and FVB mice were purchased from Janvier labs (Le Genest-Saint-Isle, FRA), and NOD and SJL were from Charles River (Wilmington, USA) Laboratories. Foxp3-Thy1^a Rag2-Gfp-mutant B6 and NOD mice were previously described (21). NOD.B10 Idd5 congenic R974 mice (39) were from JAX (Bar Harbor, USA) Laboratories. All experiments involving animals were performed in compliance with governmental and institutional guidelines (ethical approval APAFIS#4151-201602171 0481496.v6).

We used the following monoclonal antibodies (mAbs) and secondary reagents: APC-Cy7- and Pacific Blue-labeled anti-CD4 (GK1.5), biotin-labeled anti-CD4 (RM4.4), PE-CF594-labeled anti-CD8α (53.6.7), BV510-labeled anti-CD8β (H35-17.2), PE-Cy7- and PE-labeled anti-CD25 (PC61), PECF-BV421- and biotin-labeled anti-TCRβ (H57-597), BV421- and APC-labeled anti-Thy1.1 (OX-7), AF647-labeled anti-Ki67 (B56), PE-labeled anti-Nur77 (12.14), PE-labeled anti-CD5 (53-7.3), PE-labeled anti-I-Ab (AF6-120.1), biotin-labeled anti-RT1B (OX6) for the detection of I-Ag7, biotin-labeled anti-TCR Vβ3 antibody (KJ25) (all from BD (Franklin Lakes, USA) Biosciences); PECy7-labeled anti-GITR (DTA-1), PerCP-ef710-labeled anti-PD-1 (J43), biotin-labeled anti-Thy1.1 (HIS51), PE-labeled anti-TCRβ (H57-597), PerCP-Cy5.5-labeled or APC-ef780-labeled anti-CD44 (IM7), APC-labeled anti-CD62L (MEL-14), ef450-BV605-PE-Cy7- or PE-labeled streptavidin, PE-labeled anti-H-2K^d (SF1-1.1), biotin-labeled anti-H-2K^b (AF6-88.5), PE-labeled anti-CCR7 (4B12), ef450- and ef660-labeled anti-Foxp3 (FJK16S) (all from eBioscience (thermofisher) (Waltham, USA)); and BV605-labeled anti-CD73 (TY/11.8) (from Biolegend (San Diego, USA)); anti-S1P1/EDG-1 antibody (713412, R&D (Minneapolis, USA) Systems); and biotin-SP (long spacer) AffiniPure F(ab’)₂ Fragment Donkey Anti-Rat IgG (H+L) (polyclonal, Jackson immuinoresearch (West Grove, USA)).

Sample preparation and staining were essentially performed as previously described (51). FACS data were acquired using an LSRII or a Fortessa flow cytometer (BD Biosciences, San Jose, CA, USA) and analyzed using FlowJo software (Tree Star, Ashland, OR, USA). Doublets and dead cells were excluded from the analysis by using appropriate FSC/SSC gates.

Thymic lobes from NMRI mice were collected and individually cultured on transwell inserts (0.4μm pores, polycarbonate membrane, Corning (Corning, USA)) at the air–liquid interface using a complete medium (10% FBS) containing 1.35 mM of 2-deoxyguanosine (2dGUO, Sigma (Burlington, USA)). Six days later, lobes were intensely washed and cultured for 24 h in hanging drops containing 50,000 DN3 (CD4⁻CD8⁻TCR⁻CD25⁺CD44^-) cells FACS-sorted from thymocytes (CD8-depleted using mAb 31M and rabbit complement) from 8-week-old NOD or B6 mice. Lobes were then cultured on transwells for 12 days (cf. Supplementary Figure S1D ). Lobes were collected and digested in RPMI containing 4 μg/ml of Liberase and 0.1 μg/ml of DNAse I (Roche (Basel, Switzerland)) for 5 min at 37°C with vigorous pipetting every 2 min. Cells were then washed, counted, and stained for flow cytometry analysis.

DNA was prepared from tail clips of 3-day-old NOD, B6, F1, and BC1 mice using standard procedures. Idd genotyping was performed by PCR using published primers (52).

DNA was prepared from the tail clips of 3-day-old NOD, B6, and BC1 (n = 94) mice using standard procedures, quality-controlled using a fragment analyzer Agilent (Santa Clara, USA)), quantified using PicoGreen, amplified, and hybridized to Illumina GGP Mouse GIGA-MUGA Arrays according to the supplier’s instructions (Infinium HTS Assay Protocol Guide, 15045738 Rev.A October 2013, Manual Protocol). Arrays were scanned using an iScan System (Illumina (San Diego, USA)).

Thus, the obtained raw data were quality-controlled and analyzed using GenomeStudio Software 2010 with the Genotyping module (Illumina). After optimization of the cluster file and exclusion of uninterpretable SNPs, we retained 34,003 SNPs polymorphic between NOD and B6 (out of the 143,446 SNPs on the arrays). LOD scores were calculated and genome regions graphed with a customized script under R programming language using the “qtl2” package (53). The analysis was performed using the “backcross” parameter taking as inputs the genomic map, the physical map (mm10), and the phenotype table (% of GITR^highPD1⁺ cells among thymic CD4⁺CD8⁻TCR^highGFP⁺Thy1.1⁺ Tregs). The “error probabilities” parameter was arbitrarily fixed to 0.002. The script and tables are available at https://github.com/arielgalindoalbarran/Idd5NOD.