Cooperative but Distinct Role of Medullary Thymic Epithelial Cells and Dendritic Cells in the Production of Regulatory T Cells in the Thymus

This article is featured in Top Reads, p.1625

Autoimmune disease is a pathological condition in which the immune system turns on our own tissues and/or cells by as yet unknown mechanisms. A coherent understanding of the mechanisms underlying the development of autoimmune disease has been one of the major challenges in immunology (1, 2). The breakdown of self-tolerance in the thymus, in the periphery, and both is a key event for the disease, and understanding the pathogenesis of this process is crucial for developing a therapeutic approach to the disease. In the thymus, pathogenic T cells autoreactive to self-Ags are eliminated by negative selection, and thymic epithelial cells (TECs) play a pivotal role in expressing a set of self-Ags encompassing, in principle, all the self-Ags expressed by peripheral parenchymal organs (3). Supporting this idea, Aire in medullary thymic epithelial cells (mTECs) has been demonstrated to control the process for the expression of self-Ags, and deficiency of Aire results in the development of organ-specific autoimmune disease due, at least in part, to the defect in the expression of self-Ags (4–6). In addition to negative selection, self-tolerance is maintained by another mechanism involving regulatory T cells (Tregs), which prevent CD4⁺ T cell–mediated, organ-specific autoimmune diseases (7, 8). The significance of Tregs in the maintenance of self-tolerance has been demonstrated by the fact that elimination of Tregs leads to the development of organ-specific autoimmune diseases in otherwise normal mice (9). Although Foxp3 has been identified as a key regulator for the development of Tregs in the thymus (10–12), factors controlling the production of Tregs in both Foxp3-dependent and Foxp3-independent mechanisms remain incompletely understood (13, 14).

One interesting feature of tolerance induction in the thymus is that negative selection and production of Tregs are not independent events. Instead, the two processes are interconnected in many aspects (15). For example, T cells reacting with self-Ags at a relatively high affinity are negatively selected, whereas those reacting with self-Ags at a moderate affinity develop into Tregs, proposing an avidity model for the generation of a self-tolerant T-cell repertoire (16). Consistent with this model, many autoimmune animal models due to the defect in the negative selection also showed the concomitant defect in the Treg production as exemplified by Aire-deficient mice (17). The exact mechanisms linking the two tolerance mechanisms, however, remain unknown. Although it is not easy to dissect which defect (i.e., defect in the negative selection or defect in the Treg production) plays a predominant role during the development of a given autoimmune disease, there is one report suggesting that multiorgan autoimmunity was due to the dysfunctional Treg generation rather than the failure of negative selection in mice lacking NF-κB-inducing kinase selectively in TECs (18).

Given that Tregs play an essential role in the establishment and maintenance of self-tolerance and are inevitably self-reactive in their Ag specificities, it is important to know how the Treg precursors recognize self-Ags in the thymic microenvironment. Because two types of APCs, TECs of stromal origin and dendritic cells (DCs) with phagocytosis activity, are present, one immediate question would be how the two types of APCs cooperate and/or divide their labor of presenting the self-Ags for the induction of Tregs. As for the cooperative side, it has been demonstrated that self-Ags produced by TECs are transferable to DCs through a process known as Ag transfer (19–21). The exact mechanisms underlying Ag transfer, however, have not been fully characterized. Besides the reactivities with self-Ags, cytokine and costimulatory signals are important factors that control the production of Tregs (7, 8). In this regard, the distinct roles played by the two types of APCs for nursing the Tregs are yet to be characterized.

We have approached these issues by depleting the mTECs and DCs individually using mice expressing diphtheria toxin (DT) receptors (DTRs) driven by Aire (22) and CD11c promoters (23), respectively. We found that the depletion of mTECs showed a more profound effect on Treg production both quantitatively and qualitatively compared with that of DCs. To understand the exact role of two APCs in the production of Tregs, we first focused on the cooperative action of Ag transfer between the two. We especially focused on what types of TECs are the major source of self-Ags. We further investigated how the differential effect on the production of Tregs by mTECs and DCs arose, and we showed that mTECs have a better ability to recruit non-Tregs as well as Tregs into the thymus (i.e., recirculation). Recruitment of non-Tregs into the thymus contributed to the provision of IL-2, an important cytokine required for Treg expansion, in the thymic microenvironment. Our studies have illuminated the cooperative but distinct role of mTECs and DCs in the production of Tregs and ultimately in the protection against the development of the broad spectrum of organ-specific autoimmune diseases.

β5t/GFP-IRES-human AIRE (hAIRE) knock-in (KI) mice (β5t/GFP-KI) were generated by homologous recombination in embryonic stem cells established from C57BL/6 mice. A KI cassette containing GFP-IRES-hAIRE-polyA/frt-Neo^r-frt was inserted into exon 1 of the psmb11 locus to disrupt the coding sequence of β5t. After targeted cells were injected into morula-stage embryos, the resulting chimeric male mice were mated with C57BL/6 females (CLEA Japan) to establish germline transmission. Neo^r was removed by crossing the mice with a transgenic (Tg) line expressing the general delete flippase. Aire/DTR-GFP-KI was reported previously (22, 24). Aire-deficient mice were reported previously (25). CAG-CAT-eGFP Tg mice (kindly provided by Dr. H. Kawamoto) were reported previously (26). Aire/human AIRE-GFP-Flag tag-KI were reported previously (22). The following mouse strains were purchased from The Jackson Laboratory: CD11c-DTR/GFP-Tg; B6.FVB-1700016L21RikTg (Itgax-DTR/EGFP)57Lan/J, Foxp3/EGFP-KI; B6.Cg-Foxp3tm2Tch/J, Foxn1-Cre-KI; B6(Cg)-Foxn1^{tm3(cre)Nrm/J}. C57BL/6 mice were purchased from CLEA Japan. All mice were maintained under pathogen-free conditions and handled in accordance with the Guidelines for Animal Experimentation of Tokushima University School of Medicine.

To abrogate of mTECs, mice were injected i.p. with 600 ng of DT every day for a total of five times. For the depletion of DCs, mice were injected i.p. with 600 ng of DT every other day for a total of three times. For analysis of autoimmunity, mice were injected i.p. with 600 ng of DT every 3 d for a total of nine times. For blocking the maturation of mTECs in vivo, mice were injected i.p. with 100 μg anti-RANKL mAb every 3 d for a total of two times.

TECs were prepared as previously described (25). In brief, the thymus was mechanically disrupted and digested with Liberase and DNase1 in RPMI 1640 medium supplemented with 10% FCS, 2 mM l-glutamine, 100 U/ml penicillin, 100 μg/ml streptomycin, and 20 mM HEPES, referred to as R10, for 30 min at 37°C. To isolate thymic DCs, the thymus was mechanically disrupted and digested with collagenase D and DNase1 in R10 for 30 min at 37°C. mAb against CD45 (clone 30-F11), epithelial cell adhesion molecule (EpCAM; clone G8.8), Ly51 (clone 6C3), CD80 (clone 16-10A1), CTLA-4 (clone UC10-4B9), CD4 (clone RM4-5), CD8a (clone 53-6.7), CD25 (clone PC61), CD11b (clone M1/70), CD11c (clone N418), CD73 (clone TY/11.8), OX40 (clone OX-86), FR4 (clone 12A5), GITR (clone DTA-1), neuropilin1 (clone 3E2), c-Rel (clone 1RELAH5), PD-1 (29F.1A12), TCRVβ2 (clone B20.6), TCRVβ5.1/5.2 (clone MR9-4), TCRVβ6 (clone RR4-7), TCRVβ7 (clone TR310), TCRVβ8.1/8.2 (clone MR5-2), TCRVβ9 (clone MR10-2), TCRVβ11 (cloneMR9-4), TCRVβ12 (clone MR11-1), and TCRVβ13 (clone QA19A88) were purchased from BioLegend. mAb against I-A/I-E (clone M5/114.15.2) and Foxp3 (clone FJK-16s) were purchased from Invitrogen. mAbs against Sirpα (clone P84), CD36 (clone CRFD-2712), TCRVβ3 (clone KJ25), and TCRVβ10b (clone B21.5) were purchased from BD Biosciences. Ulex europaeus agglutinin 1 (UEA-1) was obtained from Vector Laboratories. A rat anti-mouse Aire mAb (clone RF33-1) was prepared in our laboratory.

Thymi were fixed with 4% paraformaldehyde in PBS and immersed in a graded series of sucrose solutions (10%, 20%, and 30%). Thymi were then embedded in OCT compound (Sakura Finetek), frozen at −80°C, and sectioned into ∼6–7-μm-thick slices. Samples were blocked with 2% BSA in PBS for 1 h and stained with anti-GFP polyclonal Ab labeled with Alexa Fluor 488 (Invitrogen), anti-CD11c mAb labeled with Alexa Fluor 594 (BioLegend), anti-EpCAM mAb labeled with Alexa Fluor 647 (BioLegend), and anti-mouse Aire mAb (produced in our laboratory). For the detection of anti-GFP autoantibodies in GFP-expressing Tg animals, thymi from Aire/DTR-KI were harvested, embedded in OCT compound, and frozen at −80°C, followed by sectioning into ∼6–7-μm-thick slices. The thymic sections were blocked with 2% BSA in PBS for 1 h and incubated with sera from Aire/DTR-KI or Aire/AIRE-GFP-Flag tag-KI (1:300 diluted) for 1 h at room temperature. The slides were then incubated with anti-mouse IgG-Alexa Fluor 594 (Invitrogen) (1:400 dilution) for 30 min at room temperature for the detection of autoantibodies against GFP. For the immunocytochemical analysis, CD45⁺CD11c^highCD11b⁻GFP⁺ DCs were sorted from Foxn1-GFP mice. Sorted cells were attached to the glass slide, dried, and fixed with 4% paraformaldehyde. Cells were costained with Alexa Fluor 594–labeled anti-Lamp1 mAb and DAPI. Fluorescence images were visualized using the All-in-One Fluorescence Microscope BZ-X810 (Keyence). To obtain high-resolution images without the blurring caused by out-of-focus fluorescent signals, optical sectioning was employed.

Tregs and conventional T cells (Tconvs) were sorted from the thymi of Foxp3/EGFP-KI mice. Total RNA was extracted from Tregs and Tconvs using the RNeasy Mini Kit (Qiagen) and converted to cDNA using the Sensiscript Reverse Transcription Kit (Qiagen) in accordance with the manufacturer’s instructions. Real-time PCR was performed using the Quantitect Probe PCR kit (Qiagen) on the Thermal Cycler Dice Real Time System (Takara). The following primers and probes were used: IL-2-primers: forward: 5′-CTCCTGAGCAGGATGGAGAATT -3′; reverse: 5′-CCGCAGAGGTCCAAGTTCAT-3′; IL-2-probe: 5′-[FAM]CCCAAGCAGGCCACAGAATTGAAAGATCT[TAMRA]-3′; Ccr6-primers: forward: 5′-TCCATCATCATCTCAAGCCCTCA-3′; reverse: 5′-AGGGGTGAAGAACCCAAAGAACA-3′; Ccr6-probe: 5′-[FAM]GTCCAGTCCCATACCCAGCAGCTTCCACG[TAMRA]-3′; Gapdh primers: forward: 5′-CAAGCTCATTTCCTGGTATGACAA-3′; reverse: 5′- TTGGGATAGGGCCTCTCTTGC-3′; Gapdh probe: 5′-[FAM]GTGGTGGACCTCATGGCCTACATGGCC[TAMRA]-3′.

Bone marrow (BM) chimeras were generated as previously described (25). In brief, BM cells were harvested from mouse femurs. Following RBC lysis, T and B cells were depleted of BM cells. Recipient mice were irradiated with two split doses of 4.0 Gy and 4.5 Gy and were reconstituted with 1 × 10⁷ BM cells. The mice were analyzed 2 mo after BM transfer.

For in vitro suppression assay, irradiated wild-type (WT) and Aire/DTR-KI were reconstituted with BM cells from Foxp3/EGFP-KI. Irradiated WT were also reconstituted with CD11c/DTR-Tg crossed with Foxp3/EGFP-KI. CD25⁺Foxp3⁺ Tregs were isolated from the spleen of BM chimeras that had been treated with DT. CD4⁺ naive T cells isolated from the WT spleen were labeled with CellTrace Violet (Thermo Fisher). CD4⁺ naive T cells were cocultured with Tregs at a 10:1 or 2:1 ratio in the presence of Dynabeads Mouse T-Activator CD3/CD28 (Thermo Fisher). The cell divisions were assessed 72 h later.

For DNA methylation assay, irradiated WT and Aire/DTR-KI were reconstituted with BM cells from Foxp3/EGFP-KI mice. Irradiated WT were also reconstituted with CD11c/DTR-Tg crossed with Foxp3/EGFP-KI. Genomic DNA was isolated from sorted Tregs (CD25⁺GFP⁺CD4⁺ thymocytes) from BM chimeras treated with DT. After sodium bisulfite treatment with the Fast Bisulfite Conversion Kit (Abcam), bisulfite-converted genomic DNA was amplified by PCR and subcloned into the pMD20-T vector (Takara). The 40 colonies in each region were sequenced. The following primers were used for PCR amplification of bisulfite-converted genome DNA: Eos intron 1b: forward: 5′-TAAGAAATTGGGTGTGGTATATGTA-3′; reverse: 5′-TTTCCCCTACTAAAACTCCTTAAAC-3′; Ctla-4 exon 2: forward: 5′-TGGTGTTGGTTAGTAGTTATGGTGT-3′; reverse: 5′-AAATTCCACCTTACAAAAATACAATC-3′; GITR exon 5: forward: 5′-GAGGTGTAGTTGTTAGTTGAGGATGT-3′; reverse: 5′-AACCCCTACTCTCACCAAAAATATAA-3′.

Chimeric mice were treated with DT every 3 d for a total of nine times. Mice were analyzed 2 wk after the DT treatment. Formalin-fixed tissue sections were subjected to H&E staining. Histological changes were scored as none (score 0), weak to mild (score 1), and severe (score 2).

Statistical significance for the comparison of the two groups was analyzed with an unpaired, two-tailed Student t test. One-way ANOVA coupled with Tukey’s multiple comparison test was used for the comparison of three groups. Dunn’s Kruskal-Wallis multiple comparison test was used for the comparison of nonparametric pathological scores. Differences were considered significant at p < 0.05.

To examine the role of two types of APCs in the thymus, mTECs and DCs, in the production of Tregs, we depleted the mTECs and DCs using mice expressing DTRs under the control of Aire and CD11c promoters, respectively. We used Aire/DTR-KI mice, in which mTECs expressing Aire can be ablated upon injection of DT (22, 24). To exclude the effect of the ablation of nonstromal Aire-expressing cells, extrathymic Aire-expressing cells, we generated BM chimeras in which BM cells from WT mice were transferred into Aire/DTR-KI (Aire/DTR-KI chimera). After five DT injections (Supplemental Fig. 1A), Aire/DTR-KI chimeras showed a significant reduction of not only mature MHC class II high (MHC-II^high) mTECs (mTEC^high), including Aire⁺ mTECs, but also immature MHC-II^low mTECs (mTEC^low) (Supplemental Fig. 1B) because of the Aire expression from the transit amplifying cells with active cell cycling (27, 28). As expected, the composition of the DCs in the thymus was not affected, because they were derived from WT mice (Supplemental Fig. 1C). Thus, we were able to deplete a large proportion of mTECs, including both immature and mature mTECs, in this model. To deplete the DCs in the thymus, we transferred the BM cells from CD11c-DTR/GFP-Tg into WT mice (CD11c-DTR chimera). Following the multiple DT injections (Supplemental Fig. 1D), more than ∼90–95% of CD8α⁺ conventional DCs (cDC1) and Sirpα⁺ DCs (cDC2) in the absolute cell numbers were successfully depleted without affecting the numbers and the composition of mTECs (Supplemental Fig. 1E, 1F). When we generated BM chimeras by injecting BM cells from CD11c-DTR/GFP-Tg into lethally irradiated Aire/DTR-KI (CD11c-DTR→Aire/DTR-KI chimera), both mTECs and DCs were depleted upon DT treatment, as expected (Supplemental Fig. 2A). Thus, we could successfully achieve the depletion of mTECs and DCs in vivo with the use of Aire/DTR-KI chimeras and CD11c-DTR chimeras, respectively.

We first analyzed the de novo production of Tregs in the thymus from each chimera after excluding the CD73^high Tregs that were recirculating from the periphery into the thymus (29). Although both Aire/DTR-KI chimeras and CD11c-DTR chimeras showed no alterations in the proportion of CD4 and CD8 single-positive thymocytes (Fig. 1A), we found that both chimeras showed the altered production of Tregs compared with the control WT chimeras injected with DT as follows. First, both chimeras showed reduced percentages and the absolute numbers of the mature CD25⁺Foxp3⁺ Tregs and CD25⁻Foxp3⁺ Treg precursors (Fig. 1B). Second, reduced percentages and the absolute numbers of the CD25⁺Foxp3⁻ Treg precursors were observed only in the Aire/DTR-KI chimeras. Third, the degree of reduced production of mature CD25⁺Foxp3⁺ Tregs in the percentages, as well as the absolute numbers, was more profound in Aire/DTR-KI chimeras compared with CD11c-DTR chimeras. Thus, although both mTECs and DCs were individually required for the de novo production of Tregs in the thymus, the depletion of mTECs showed a more profound effect on the Treg production in the thymus quantitatively (Fig. 1B). Because CD11c-DTR→Aire/DTR-KI chimeras showed a greater degree of the reduction of both Treg precursors and mature Tregs compared with Aire/DTR-KI chimeras or CD11c-DTR chimeras (Supplemental Fig. 2B), the results further supported our view that mTECs and DCs were considered to play a nonredundant role in Treg generation.

We also evaluated the phenotypes of Tregs that developed in the absence of either mTECs or DCs to understand the distinct role of each APC in the production of Tregs. Although expression levels of Foxp3 and CD25 from mature CD25⁺Foxp3⁺ Tregs remained unchanged in both chimeras, CD25⁺Foxp3⁺ Tregs from Aire/DTR-KI chimeras showed decreased expression levels of GITR, OX40, and c-Rel and increased expression of FR4 and CTLA-4. In contrast, CD25⁺Foxp3⁺ Tregs from CD11c-DTR chimeras showed decreased GITR expression alone (Fig. 1C). Furthermore, CD25⁺Foxp3⁻ Treg precursors from Aire/DTR-KI chimeras showed decreased expression of GITR, OX40, and neuropilin 1 and increased expression of FR4. In contrast, CD25⁺Foxp3⁻ Treg precursors from CD11c-DTR chimeras showed increased OX40 expression alone (Fig. 1C). As for the CD25⁻Foxp3⁺ Treg precursors, Aire/DTR-KI chimeras showed increased expression of FR4 and CTLA-4 and decreased expression of OX40, whereas CD11c-DTR chimeras showed no obvious changes in these molecules (Fig. 1C). Thus, the lack of mTECs and DCs showed distinct phenotypic changes in the expression of molecules characteristic of Tregs. Nonetheless, in vitro suppression assay using the splenic Tregs did not show any obvious difference between Aire/DTR-KI chimeras compared with CD11c-DTR chimeras (Supplemental Fig. 2C).

Foxp3⁺ thymic Tregs have been divided into two functionally distinct subsets: GITR^highPD-1^highCD25^high (Triple^high) Tregs and GITR^lowPD-1^lowCD25^low (Triple^low) Tregs. Triple^high Tregs express more self-reactive TCRs than Triple^low Tregs, suggesting that these two subsets express distinct TCR repertoires (30). We examined whether the depletion of mTECs and DCs affects the composition of Tregs in terms of Triple^high and Triple^low. Depletion of mTECs significantly changed the composition of Triple^high and Triple^low Tregs, whereas depletion of DCs did not show the changes in the composition of Triple^high and Triple^low Tregs (Fig. 1D). Thus, depletion of mTECs showed a more profound effect on the Treg production both quantitatively and qualitatively compared with that of DCs.

To better understand the distinct effect of the depletion of mTECs and DCs on the production of Tregs, we have evaluated epigenetic changes in the gene characteristics for Tregs in both chimeras. It has been demonstrated that most of the Treg-specific signature genes are progressively demethylated during the development of Tregs (31, 32). We focused on the signature genes altered at the protein level, such as Ctla-4 and Gitr shown in Fig. 1C together with Eos. Tregs from Aire/DTR-KI chimeras showed decreased DNA demethylation in the intron 1b of Eos (Fig. 1E). The Ctla-4 exon 2 segment in the Tregs from Aire/DTR-KI chimeras showed higher demethylation than that from CD11c-DTR chimeras (Fig. 1E). In contrast, Tregs from CD11c-DTR chimeras showed a methylation status of Eos and Ctla-4 similar to that from WT chimeras. We did not see differences in the demethylation pattern in exon 5 of Gitr among the three chimeras (Fig. 1E). Taken together, the depletion of mTECs showed a more profound effect on Treg production through epigenetic changes in the signature genes for the Tregs compared with that of DCs. Indeed, the depletion of mTECs and DCs shaped the different TCRβ usages of the thymocytes (Fig. 2A). Consequently, depletion of mTECs and DCs after repeated injection of DT (Fig. 2B) resulted in the development of a distinct spectrum of autoimmunity in the hosts: Aire/DTR-KI chimeras showed severer lesions in the salivary gland and pancreas than did the CD11c-DTR chimeras (Fig. 2C). Consistent with the pathological changes in the pancreas, Aire/DTR-chimeras, but not CD11c-DTR chimeras, showed higher blood glucose levels than WT chimeras (Fig. 2D). In contrast, the lung lesion was more severe in the CD11c-DTR chimeras (Fig. 2C). Development of the distinct target organ specificity by the depletion of mTECs and DCs was further confirmed by the significant body weight loss only in Aire/DTR chimeras, but not in CD11c-DTR chimeras (Fig. 2E). Together, the results suggested that mTECs and DCs are responsible for the deletion of autoreactive T cells and/or production of the Tregs with different Ag specificity.

Agonistic stimuli by self-Ags play a pivotal role in the development of Tregs in the thymus (7, 8, 16). Because self-Ags produced by mTECs are transferable to DCs through a process known as Ag transfer (19–21), mTEC depletion may also affect the repertoire of the self-Ags presented by DCs, and this could be responsible, at least in part, for a more profound effect on the production of Tregs in Aire/DTR-KI chimeras compared with that in CD11c-DTR chimeras. We therefore visually monitored how self-Ags produced by the mTECs can be transferred to DCs using GFP as a surrogate molecule of self-Ag. We first used GFP reporter mice in which GFP is expressed in all the TECs, including both mTECs and cortical thymic epithelial cells (cTECs), by crossing CAG-CAT-eGFP Tg with Foxn1-Cre KI (Foxn1-GFP mice): GFP expression was observed from more than 80% of both mTECs and cTECs and ∼10% of CD45⁻EpCAM⁻ stromal cells with flow cytometric analysis (Fig. 3A). In contrast, we did not see GFP expression from T cells, if any. Immunohistochemical analysis also demonstrated GFP expression from both mTECs and cTECs, although GFP expression from mTECs was more obvious than that from cTECs (Fig. 3B). We then analyzed the expression of GFP from DCs that were obtained from TECs through Ag transfer. We have previously divided DCs in the thymus into three subsets based on their expression of CD11c and CD11b: immature CD11c^intCD11b⁻, CD11c^highCD11b⁻, and CD11c^highCD11b^int (Fig. 3C, right) (25). CD11c^highCD11b⁻ DCs and CD11c^highCD11b^int DCs corresponded to cDC1 (CD8α⁺ DCs) and cDC2 (Sirpα⁺ DCs), respectively. CD11c⁻CD11b^high cells and CD11c^intCD11b^high cells were also identified as neutrophils and eosinophils by their Gr-1 and Siglec-F expression, respectively (25). CD11c^highCD11b⁻ DCs showed higher GFP expression than CD11c^intCD11b⁻ DCs and CD11c^highCD11b^int DCs, and GFP expression was much lower in the eosinophils than in DCs (Fig. 3D, 3E). Because splenic CD11c⁺ DCs did not express GFP by flow cytometric analysis (Fig. 3F; compare with Fig. 3D), GFP detected from thymic DCs was considered to be derived from TECs. We also detected the signals of the GFP from thymic DCs classified into CD8α⁺ DCs (cDC1) and Sirpα⁺ DCs (cDC2) (Figs. 3G, 3H): The former showed a higher GFP expression than the latter. Of note, GFP expressed in Aire-GFP reporter mice behaved as a proxy “self” because the mice did not produce any autoantibodies against the GFP expressed from the thymic medulla (Supplemental Fig. 2D, 2E). Collectively, these data suggested that self-Ags expressed from the whole TECs are effectively transferred to both types of DCs, cDC1 and cDC2, in vivo.

Although the data described above clearly indicated that GFP expressed from the whole TECs can be transferred to DCs, which type of TECs (i.e., mTECs or cTECs) is providing the GFP remains unknown: Foxn1-GFP mice expressed GFP from both mTECs and cTECs (Fig. 3A). Because our flow cytometric analysis examined the GFP expression from DCs, regardless of their location (i.e., medulla or cortex), it would be especially important to know which types of TECs can be the source of self-Ag, GFP in this case. Indeed, DCs were present in both the medulla and cortex by immunohistochemical analysis (Fig. 4A). We therefore prepared another GFP reporter strain in which only cTECs can express GFP by the use of cTEC-specific promoter of β5t (β5t-GFP mice) (Supplemental Fig. 3A, 3B; also see Materials and Methods). Immunohistochemical analysis of the thymus from β5t-GFP mice revealed that GFP expression was not observed from the medulla, and, instead, it was confined to the cortex as expected (Fig. 4B). Flow cytometric analysis demonstrated GFP expression from cTECs, although the percentages of GFP from cTECs were lower than those in Foxn1-GFP mice (85.3% in Fig. 3A versus 40.1% in Fig. 4C). We additionally observed low levels of GFP expression from mTECs, probably because of the promiscuous expression of β5t from mTECs. We then examined the expression of GFP signals from DCs. In contrast to the case of Foxn1-GFP mice, we observed only scarce levels of GFP signals from three subsets of DCs with flow cytometric analysis (Fig. 4D; compare with Fig. 3D). Immunohistochemical analysis also showed no GFP expression from DCs in the cortex from β5t-GFP mice (Fig. 4E, right), unlike DCs in the medulla from Foxn1-GFP mice (see below). Thus, the results suggested that Ags expressed from the cTECs were scarcely subjected to the Ag transfer in contrast to those expressed from the mTECs.

Knowing that mTECs are the major source of self-Ag transfer, we next asked which types of mTECs, immature mTECs or mature mTECs, can provide self-Ags using Foxn1-GFP mice. GFP expression was observed from both immature mTEC^low and mature mTEC^high in this strain (Fig. 5A). Because it has been demonstrated that mature Aire⁺ mTECs can provide self-Ags to DCs efficiently (33), we were particularly interested in whether self-Ags expressed from immature mTEC^low can also become the source of self-Ags. For this purpose, we depleted mature mTEC^high by the treatment with anti-RANKL mAb while maintaining immature mTEC^low almost intact. Foxn1-GFP mice were treated with anti-RANKL mAb to block the differentiation of mTECs from the immature stage to the mature stage (22, 34). Ten days after the two injections of anti-RANKL mAb (Fig. 5B), mTEC^high were reduced to almost one-third compared with the control PBS-treated mice, and GFP-expressing mTECs resided predominantly in the immature mTEC^low (Fig. 5C). Despite the significant reduction of mTEC^high, GFP expression from CD11c^highCD11b⁻ remained unchanged (Fig. 5D), suggesting that remaining GFP expression from mTEC^low could compensate for the loss of GFP expression from mTEC^high. Consistent with this finding, WT mice treated with anti-RANKL mAb showing the reduced mTEC^high with a concomitant increase in mTEC^low (Fig. 5E) had no alteration in the production of CD25⁺Foxp3⁻ and CD25⁻Foxp3⁺ Treg precursors with only minimal reduction in the mature CD25⁺Foxp3⁺ Tregs (Fig. 5F). One possible reason for the reduction in the mature CD25⁺Foxp3⁺ Tregs in this case might be due to factors other than the reduced self-Ag transfer, such as reduced costimulatory signals and/or cytokines from mTEC^high required for the full maturation of Tregs (35, 36).

We also examined whether Aire itself is involved in Ag transfer from mTECs to DCs because it has been demonstrated that CD8α⁺ DCs were required for the Aire-dependent T cell selection (33). Aire⁺ mTECs from Foxn1-GFP mice expressed GFP by immunohistochemical analysis (Supplemental Fig. 3C), and Aire deficiency did not affect the GFP expression from mTECs (Supplemental Fig. 3D). In this experimental setting, we found that the lack of Aire did not affect the GFP expression from DCs (Supplemental Fig. 3E), suggesting that Aire was dispensable for the Ag transfer in this visual model using GFP that was expressed from the whole cells of TECs.

We then studied the mode of the transfer of cytoplasmic Ag of GFP expressed from TECs in Foxn1-GFP mice. Immunohistochemical analysis of CD11c⁺ DCs showed the localization of transferred GFP in their cytoplasm (Fig. 6A). Of note, GFP in the cytoplasm was overlapped with the endosomal pathway containing Lamp1 (CD107a) (Fig. 6B) by immunocytochemical analysis, suggesting that DCs are acquiring the mTEC-derived GFP through phagocytosis.

Because it has been reported that a scavenger receptor CD36 is critical for the transfer of cell surface Ags from mTECs to CD8α⁺ DCs (20), we examined the expression levels of CD36 from various DC subsets. We found that CD8α⁺ DCs (cDC1: CD11c^highCD11b⁻ DCs) expressed higher levels of CD36 than Sirpα⁺ DCs (cDC2: CD11c^highCD11b^int DCs) (Fig. 6C). Interestingly, expression levels of CD36 were well correlated with those of GFP signals among various DC subsets (Fig. 6D, 6E), further suggesting that CD8α⁺ DCs are superior in acquiring the Ags from TECs to Sirpα⁺ DCs (Fig. 3G, 3H), and CD36-mediated phagocytosis is one major route of transfer for the Ags (e.g., GFP) that are present in the cytoplasm of mTECs.

Depletion of mTECs showed an effect on the Treg production, both quantitatively and qualitatively, more profound than that of DCs, and Ag transfer from mTECs to DCs, at least in part, accounts for this because the production of Tregs requires agonistic stimuli by self-Ags as discussed above. Additionally, mTECs are required for generating the thymic microenvironment that can nurse the developing Tregs by providing several important cytokines (35–37). We therefore thought that reduced provision of the cytokines may also account for the more profound reduction of Tregs after the depletion of mTECs than the depletion of DCs. In the aforementioned experiments, we intentionally excluded the CD73^high recirculating Tregs to evaluate the “de novo production” of Tregs. However, it is also important to note that CD73^high recirculating CD4⁺ T cells, including both recirculating CD25⁺Foxp3⁺ Tregs and CD25⁻Foxp3⁻ Tconvs, are important sources of the Treg-nursing cytokines in the thymus. Accordingly, it is possible that the depletion of mTECs resulted in the disturbing thymic niche by reducing the cytokines from the recirculating T cells that are required for the production of Tregs. We therefore focused on the CD73^high recirculating CD4⁺ T cells in both chimeras to test this possibility. We found that both frequencies and the absolute numbers of CD73^high recirculating total CD4⁺ T cells in the thymus were significantly reduced in Aire/DTR-KI chimeras but not in CD11c-DTR chimeras when compared with the control mice (Fig. 7A, upper). Furthermore, reduced numbers of CD73^high recirculating CD25⁺Foxp3⁺ Tregs and CD73^high recirculating CD25⁻Foxp3⁻ Tconvs were also observed in Aire/DTR-KI chimeras but not in CD11c/DTR-KI chimeras (Fig. 7A, lower). Impaired recirculation of mature T cells into the thymus was also evident in untreated Aire-deficient mice (Supplemental Fig. 4), suggesting that Aire itself is responsible for the recirculation of both Tregs and Tconvs. Of note, although neither de novo produced Tregs (CD73^lowCD25⁺Foxp3⁺ CD4⁺ T cells) nor recirculating Tregs (CD73^highCD25⁺Foxp3⁺ CD4⁺ T cells) produced IL-2 (Fig. 7B, left), both recirculating Tconvs (CD73^highCD25⁻Foxp3⁻ CD4⁺ T cells) and nonrecirculating Tconvs (CD73^lowCD25⁻Foxp3⁻ CD4⁺ T cells) produced IL-2, suggesting that recirculating Tconvs serve as an important source of IL-2 for the Tregs to develop and that Aire-expressing mTECs play an important role to recruit CD4⁺ T cells into the thymus from the periphery.

To investigate how Aire-expressing mTECs recruit Tconvs into the thymus to let them provide IL-2 as an important source of Treg nursing, we examined the expression of CCR6. Recirculating CD73^high Tregs expressed Ccr6 at a higher level than the de novo produced CD73^low Tregs (Fig. 7B, right), consistent with the previous notion that CCR6⁺ Tregs are recirculating from the periphery into the thymus with the use of the CCL20–CCR6 axis (38). In contrast, both recirculating CD73^high and nonrecirculating CD73^low Tconvs did not express Ccr6 (Fig. 7B, right), suggesting that Tconvs recirculate into the thymus through Aire⁺ mTEC-dependent but CCL20-CCR6-independent mechanisms. Collectively, these data suggested that mTECs, in addition to the provision of self-Ags to cooperate with DCs, are contributing to the promotion of the de novo Treg production through recruiting the non-Tregs to make the thymic niche appropriate for Treg development independently from the DCs.

We have suggested that mTECs and DCs present distinct sets of self-Ags, thereby playing a nonredundant role in the establishment of self-tolerance. However, there should be significant numbers of overlapping self-Ags because of the equipment with the Ag transfer system from mTECs to DCs. Although mTECs are probably superior to DCs in terms of the spectrum of the self-Ags due to their inherent properties for promiscuous gene expression (39), the fact that depletion of DCs in the CD11c-DTR chimeras resulted in the development of lung lesions clearly indicated that mTECs alone cannot generate the full spectrum of the self-tolerant T cell repertoire. Instead, DCs are responsible for eliminating the autoreactive T cells that mTECs cannot cover. This may be due to the existence of the blood-borne Ags that can be effectively captured by DCs but not by mTECs (40, 41). The exact nature of self-Ags that distinguish transferable self-Ags from nontransferable self-Ags between the mTECs and DCs awaits further study.

Depletion of mTECs and DCs by the injection of DT into the chimeras was not complete in Aire/DTR-KI chimeras and CD11c-DTR chimeras, respectively. Total mTEC numbers were reduced by four- to fivefold, whereas total DC numbers were reduced by five- to sixfold, on average. Although significant, the reduction of Tregs was approximately twofold at most, which could be due to the incomplete depletion of mTECs or DCs after DT injection. However, CD11c-DTR→Aire/DTR-KI chimeras showed a greater degree of reduction of both Treg precursors and mature Tregs than Aire/DTR-KI chimeras or CD11c-DTR chimeras, suggesting that mTECs and DCs play a nonredundant role in Treg generation. Alternatively, it remains possible that other thymic cells besides mTECs and DCs might be involved in the production of Tregs. In fact, one recent report suggested that thymic fibroblasts play an important role in thymic T cell selection (42).

It would be important to know which thymic DC subsets are relevant to the self-Ag transfer. It has been reported that the selection of TCRs for many Aire-dependent Tregs was dependent on Batf3-dependent CD8α⁺ DCs, suggesting that Aire-dependent self-Ags were preferentially transferred to CD8α⁺ DCs (33). In contrast, another report suggested that the development of Tregs specific for Aire-dependent prostate Ag was achieved by the BM-derived APCs (BM-APCs) other than CD8α⁺ DCs (19). We have recently reported that both CD8α⁺ DCs and Sirpα⁺ DCs showed impaired Ag transfer in Aire-deficient mice when it was assessed with Y-Ae mAb that recognizes the Eα peptide derived from MHC-II (I-E^d) of the BALB/c strain that is complexed with MHC-II (I-A^b) from the C57BL/6 strain in the BM chimeras (25). Here, we also showed that GFP from mTECs was transferred to both CD8α⁺ DCs and Sirpα⁺ DCs, although it was more efficient in the former than in the latter. Thus, both DC subsets in the thymus have the ability to receive the self-Ags handled by mTECs, thereby contributing to the efficient cooperation between mTECs and DCs for establishing self-tolerance.

Because GFP expressed from the mTECs was effectively transferred to CD8α⁺ DCs, we visually investigated the cellular process underlying the Ag transfer. Our immunocytochemical analysis revealed that intracellular localization of the acquired GFP in CD8α⁺ DCs was partially overlapped with Lamp1, strongly suggesting that CD8α⁺ DCs acquired cytoplasmic Ag of GFP by phagocytosis. Consistent with this idea, we found a good correlation between GFP acquisition in CD8α⁺ DCs and expression levels of CD36, a scavenger receptor implicated in phagocytosis of apoptotic cells through its ability to bind with phosphatidylserine (43). However, other mechanisms for Ag transfer, such as trogocytosis (44–46) and the secretion of mTEC-derived exosomes (47), might also be involved in the Ag transfer, depending on the kind of Ags (e.g., cytoplasmic versus nuclear Ag, soluble versus particulate Ag, the molecular size of Ags). Further studies are required to reveal how DCs acquire various Ags from mTECs besides the phagocytosis suggested in the present study.

We have demonstrated that GFP expressed from a broad subset of mTECs (i.e., immature and mature mTECs) was effectively transferred to DCs, whereas GFP expressed from cTECs was not. Our results were somewhat unexpected because a previous study suggested that thymocytes interact with DCs in the cortex for receiving the signals for positive selection, including the self-peptide–MHC complex (48). In this regard, the differential requirement of self-Ags for the positive and negative selection needs to be considered. The negative selection in the medulla requires broad sets of self-Ags for the deletion of autoreactive T cells. Similarly, the production of Tregs depends on the agonistic stimuli by broad sets of self-Ags. Thus, the tolerogenic process in the medulla occurs in an Ag-specific fashion of one to one in principle. In contrast, self-Ags required for positively selecting the MHC-reactive T cells in the cortex can be much fewer: Even a single self-peptide–MHC complex could select a broad spectrum of thymocytes (49, 50). Although our study using the mice expressing GFP only from cTECs showed inefficient transfer of GFP to DCs, a trace amount of self-Ags might be sufficient for the positive selection. Alternatively, self-Ags expressed from DCs themselves and blood-borne Ags captured by the DCs in the cortex might be sufficient for the process. The exact reason why Ag transfer occurs more efficiently in the medulla than in the cortex, as well as the unique microenvironment of the thymic medulla for the tolerance induction, awaits further study [e.g., ectopically expressed Aire in the cortex did not contribute to the thymic tolerance (51)].

Ag transfer is not the only interplay between mTECs and DCs. We have previously demonstrated that when WT BM cells were transferred into irradiated NF-κB-inducing kinase mutant mice, levels of CD80 and CD86 expression were significantly reduced in thymic DCs, whereas these phenotypes were not observed in splenic DCs, suggesting that thymic stroma affects the function of DCs in the thymus (52). Because the costimulatory signal is indispensable for Tregs to develop (53), altered expression of CD80/CD86 results in the impaired production of Tregs. Indeed, we have demonstrated that Aire-deficient mTECs ectopically expressed CTLA-4 that captured CD80/CD86 on DCs, resulting in the reduced production of Tregs due to the defective CD80/86-CD28 signals for the Treg precursors (25). Thus, like canonical thymic crosstalk between mTECs and thymocytes, crosstalk between mTECs and DCs in the thymus should also be considered, and this could be another reason why the depletion of mTECs showed a more profound effect on Treg production than the depletion of DCs.

Besides TCR/CD28 signals, cytokine signals as exemplified by IL-2 are also critical for the development of Tregs whereby TCR signals result in the expression of proximal IL-2 signaling components for the induction of Foxp3 (35). It also has been reported that IL-2 regulates the positioning of the pioneer factor SATB1 in CD4⁺ thymocytes and controls genome-wide chromatin accessibility of Tregs (54). Because recirculation of CD73^highCD25⁻Foxp3⁻ Tconvs producing IL-2 was reduced in the thymus from Aire/DTR-KI chimeras, this could contribute to the impaired production of the Tregs due to the reduced availability of IL-2. The exact mechanisms underlying the recruitment of Tconvs into the thymus that is dependent on the mTECs need to be revealed. Interestingly, it has been reported that recirculation of CCR6⁺ Tregs into the thymus was Aire dependent, and Aire controlled the expression of CCL20, a ligand for CCR6, from mTECs (38). In this case, thymic crosstalk between mTECs and thymocytes through the CCL20–CCR6 axis was an important element for recruiting the Tregs into the thymus. Thus, it is reasonable to speculate that this canonical thymic crosstalk was also disrupted by the depletion of mTECs in Aire/DTR-KI chimeras, resulting in a profound effect on the total numbers of Tregs in the thymus.

Finally, although we did not determine the exact TCR repertoires of Tregs and Tconvs shaped by the mTECs and DCs by employing high-performance TCR sequencing in the present study, one previous report focused on this issue. The authors of that report concluded that mTECs and BM-APCs played nonoverlapping roles in shaping the TCR repertoire by deletion and Treg selection of distinct TCRs (33). However, it is important to note that this study examined the effect of the lack of Aire in mTECs using Aire-deficient mTECs but not the depletion of mTECs that we employed in the present study. Furthermore, because of the infinity of the TCRs, the TCR repertoire was determined using the mice in which TCR diversity was limited by a Tg fixed TCRβ-chain. Accordingly, the pathological outcome of the altered TCR repertoire by the lack of Aire in mTECs or by the knockdown of MHC-II in BM-APCs (not by the depletion of DCs) was not assessed by the histological evaluation of mice (33). Thus, our studies have filled the gap between the altered TCR repertoire caused by the manipulation of mTECs or DCs and the development of autoimmunity that the previous study could not discern, thereby unveiling the distinct and unique features of mTECs and DCs in preventing autoimmunity.

The authors have no financial conflicts of interest.

We thank Drs. R. Kawakami, M. Arai, and S. Sakaguchi at Osaka University for their valuable advice on the CpG methylation analysis.