Epithelial and dendritic cells in the thymic medulla promote CD4⁺Foxp3⁺ regulatory T cell development via the CD27–CD70 pathway

The development of αβ T cells was examined in CD27-deficient mice (Hendriks et al., 2000) and CD70-deficient (Cd70^Cre/Cre) mice (Coquet et al., 2013). Flow cytometric analysis of CD27- and CD70-deficient mice identified a reduction in the frequency and absolute number of CD4⁺Foxp3⁺ T_reg cells in these mice (Fig. 1, A and B). In contrast, numbers of conventional, Foxp3⁻CD4⁺ T cells in the thymus and spleen were similar (Fig. 1 B). In addition, the sizes of the various immature thymocyte subpopulations and the size of the CD4⁻CD8⁺ populations in thymus and spleen were comparable between all genotypes (Fig. 1, B and C). Thus, in mice lacking CD27 or CD70, generation of T_reg cells was specifically impaired, whereas the development of conventional T cells appeared normal.

During T_reg cell development in the thymus, T_reg cell precursors contact self-antigens presented in MHC class II molecules with their TCR. This induces expression of the IL-2 receptor α chain (CD25), before Foxp3 expression enables further T_reg cell development (Burchill et al., 2008; Lio and Hsieh, 2008). Thymic CD4⁺Foxp3⁻CD25⁺ cells are referred to as pre-T_reg cells and include a large proportion of cells destined to become mature T_reg cells (Lio and Hsieh, 2008). CD27 signals were not required to generate this cell population, as the frequency of pre-T_reg cells was normal in the thymus of Cd27^−/− mice (Fig. 1 D and Fig. S1 A). Generation of pre-T_reg cells as well as mature T_reg cells depends on TCR signals, as well as CD28 co-stimulation (Lio et al., 2010; Vang et al., 2010). We determined whether CD28 co-stimulation still made a contribution to T_reg cell development in the absence of CD27. For this purpose, Cd27^−/− mice were administered neutralizing antibodies to CD80 and CD86 over a period of 10 d, after which T_reg cells were enumerated. Anti-CD80/CD86 treatment drastically reduced the frequency and absolute numbers of CD4⁺Foxp3⁺ T_reg cells in the thymus of Cd27^−/− mice (Fig. 1 E and Fig. S1 B), and in the spleen (not depicted), demonstrating that the CD28–CD80/86 and the CD27–CD70 co-stimulatory pathways played nonredundant roles in T_reg cell development. Although absence of CD27 or CD70 did not affect the generation of pre-T_reg cells, blockade of CD28 signaling did (Fig. 1 E), strongly suggesting that CD27 co-stimulation acts subsequent to CD28 co-stimulation in thymic T_reg cell development.

To examine the impact of the CD27–CD70 pathway on thymic T_reg cell development, fetal thymic organ cultures (FTOCs) were performed using embryonic day (E) 15 fetal thymic lobes. The addition of a blocking antibody directed at CD70 (FR70) reduced the output of CD4⁺Foxp3⁺ T_reg cells from these cultures (Fig. 2 A), whereas the addition of an agonistic soluble form of CD70 (FcCD70; Peperzak et al., 2010b) enhanced T_reg cell output (Fig. 2 A). This suggested that CD27 stimulation on thymic progenitors by CD70 promoted T_reg cell development. To further confirm the impact of CD27–CD70 co-stimulation on T_reg cell development in the thymus, we enumerated T_reg cells in neonatal mice. Directly after birth, T_reg cell levels are very low because the thymic medullary niches that are essential for T_reg cell development are incompletely formed (Liston and Rudensky, 2007). In the thymus of Cd27^−/− neonates, the frequency and number of thymic T_reg cells was reduced as compared with WT, whereas numbers of CD4⁺Foxp3⁻ cells were normal (Fig. 2 B), confirming our earlier results. T_reg cells were virtually absent in the periphery of neonatal mice (not depicted), making it highly unlikely that the reduction in thymic T_reg cell numbers observed in Cd27^−/− mice was caused by a recirculation defect of mature peripheral T_reg cells.

To address whether CD27 played a cell-intrinsic role in T_reg cell development, BM chimeras were created. WT mice were lethally irradiated, reconstituted with a 1:1 mixture of BM from WT and Cd27^−/− mice and analyzed after 8 wk for the development of T_reg cells and conventional CD4⁺ T cells (Fig. 2 C). Within CD4⁺ cells, the percentage of Foxp3⁺ cells originating from WT precursors was much higher than those originating from Cd27^−/− precursors (Fig. 2, C and D). This difference was also reflected in the total number of T_reg cells originating from WT or Cd27^−/− precursors. In contrast, conventional CD4⁺Foxp3⁻ T cell development in the thymus and spleen was comparable (Fig. 2 D). These results indicate that CD27 expression by thymic precursors is important for T_reg cell development but not required for development of conventional CD4⁺ T cells.

To determine whether CD27 co-stimulation affected the functional programming of T_reg cells, their suppressive activity was assessed. T_reg cells were purified from WT and Cd27^−/− mice and co-cultured for 3 d with CFSE-labeled conventional CD4⁺CD25⁻ responder T cells. WT and Cd27^−/− T_reg cells dampened responder CD4⁺ T cell proliferation in a highly similar, dose-dependent fashion (Fig. 3 A). This finding indicates that the CD27–CD70 pathway is not important for acquisition of suppressive function by T_reg cells, as assessed in this in vitro assay. Furthermore, Foxp3 was expressed at a similar level in T_reg cells from WT, Cd27^−/−, and Cd70^Cre/Cre mice (Fig. 1 A and not depicted), suggesting that, contrary to CD28 co-stimulation (Tai et al., 2005), CD27 co-stimulation did not regulate Foxp3 expression in developing T_reg cells.

We next determined whether CD27–CD70 co-stimulation affected the development and survival of T_reg cells. For this purpose, pre-T_reg cells were purified from the thymus of WT and Cd27^−/− mice on basis of a CD69⁺HSA⁺CD25⁺CD4⁺CD8⁻ phenotype (Fig. S1 C) and cultured in vitro with or without IL-2, as described previously (Lio and Hsieh, 2008). In the absence of IL-2, the pre-T_reg cells hardly gave rise to Foxp3⁺ cells (Fig. 3 B) and most cells died (Fig. 3 C). In the presence of IL-2, pre-T_reg cells developed into Foxp3⁺ T_reg cells, with Cd27^−/− precursors giving rise to significantly less Foxp3⁺ cells than WT precursors (Fig. 3 B). Moreover, significantly more Annexin-V⁺ apoptotic cells were observed in the Cd27^−/− cultures as compared with the WT cultures (Fig. 3 C), suggesting that CD27 provided important survival signals to developing T_reg cells. Direct ex vivo analysis of Annexin-V binding also indicated that CD4⁺CD25⁺ cells from Cd27^−/− mice underwent more apoptosis than their WT counterparts (Fig. 3 D and Fig. S1 D). This effect was selective for T_reg cells because there was no difference in the proportion of apoptotic cells within the conventional CD4⁺CD8⁻CD25⁻ or CD8⁺CD4⁻ populations of WT and Cd27^−/− mice when analyzed directly ex vivo (Fig. 3 D).

To gain molecular insight into survival signaling by the CD27–CD70 system, we determined the mRNA expression level of a comprehensive set of key apoptosis regulators in WT and Cd27^−/− CD4⁺CD25⁺ and CD4⁺CD8⁻CD25⁻ thymocytes. Although antiapoptotic Bcl-2 family members were similarly expressed, the proapoptotic Bcl-2 family members Bad and Bak were expressed at significantly higher levels in Cd27^−/− CD25⁺CD4⁺ thymocytes than in the WT population, while Puma showed a similar trend (Fig. 3 E). These proapoptotic molecules were not differentially expressed in CD25⁻CD4⁺ thymocytes from WT or Cd27^−/− mice (Fig. 3 E), suggesting that CD27 specifically counteracts the mitochondrial apoptosis pathway in developing T_reg cells. To further assess whether CD27 signaling promoted the survival of developing T_reg cells by inhibiting the mitochondrial apoptosis pathway, we examined whether deliberate expression of the antiapoptotic Bcl-2 protein could rescue the survival of Cd27^−/− T_reg cells. CD45.2⁺ Cd27^−/− BM cells were transduced with a vector encoding Bcl-2 and GFP in an IRES configuration or with the empty vector encoding GFP only. Each population was mixed in a 1:1 ratio with CD45.1⁺ WT BM cells that had been transduced with empty GFP vector and injected into lethally irradiated CD45.2 WT recipients. 8 wk later, the thymus was harvested and analyzed by flow cytometry for the contribution of Cd27^−/− cells (CD45.2⁺) among gated GFP⁺CD25⁺CD4⁺ T_reg cells (Fig. 3 F). Cd27^−/− BM cells contributed only ∼10% of the total GFP⁺ T_reg cell population when they had been transduced with empty GFP vector, confirming the superiority of WT precursors over Cd27^−/− precursors to develop into T_reg cells. However, the proportion of Cd27^−/− cells contributing to the total GFP⁺ T_reg cell pool was significantly enhanced to almost 50% when Cd27^−/− BM cells had been transduced with the Bcl-2_IRESGFP vector (Fig. 3 G). The contribution of Bcl-2 to T_reg cell development from Cd27^−/− BM cells was also reflected in absolute numbers of T_reg cells that were significantly higher in the Bcl-2–transduced population than in the empty vector–transduced population (Fig. 3 H). Importantly, Bcl-2 gene transduction did not affect the development of conventional CD25⁻CD4⁺ cells from Cd27^−/− BM cells (Fig. 3 H). The fact that thymic Cd27^−/− T_reg cell numbers but not conventional CD4⁺ T cell numbers increased upon apoptosis inhibition by Bcl-2 lends further support to our findings that CD27 deficiency enhances apoptosis of developing T_reg cells but not conventional CD4⁺ T cells.

In the mouse, CD70 is expressed on DEC205⁺ cells in the thymic medulla (Tesselaar et al., 2003). DEC205 marks both epithelial cells and DCs (Jiang et al., 1995), but we suggested that CD70 was expressed by mTECs because Cd70 mRNA was found in murine mTECs (Derbinski et al., 2005) and CD70 was detected on human TEC lines (Hintzen et al., 1994). To precisely define CD70 expression on epithelial cells in the mouse thymus, we here performed immunohistochemistry combined with confocal laser-scanning microscopy (CLSM), using thymi from Cd70^Cre/Cre mice as negative controls. It was confirmed that CD70 localizes exclusively to the thymic medulla, as defined by nuclear staining with DAPI and detection of mTECs with anti–Keratin-5 mAb (Fig. 4, II and III; Lomada et al., 2007) or ER-TR5 mAb (Fig. 4, I, II, VII, and VIII; van Ewijk et al., 1988). Using both mTEC markers in conjunction with nuclear staining, we consistently found that a large proportion of Keratin-5⁺ and ER-TR5⁺ mTECs expressed CD70, typically at different locations in the same cell (Fig. 4, I–IV). This is consistent with CD70 residing in late endosomal compartments and the cell surface (Keller et al., 2007) and Keratin-5 forming a network within the cytoplasm. Interestingly, CD70 was found on Aire⁻ mTECs but also on the great majority of (if not all) Aire⁺ mTECs (Fig. 4, V and VI). These data firmly establish that CD70 is expressed on mTECs, including those that express Aire and are implicated in T_reg cell selection (Aschenbrenner et al., 2007; Lei et al., 2011).

To address whether CD27–CD70-dependent survival signals were required for positive selection of thymic T_reg cells, we used a model of superantigen (sAg)-driven T_reg cell selection in C57BL/6 (B6) versus DBA/2 hosts (Ribot et al., 2006). Both strains express sAgs encoded by endogenous mouse mammary tumor virus 8 (Mtv-8) and Mtv-9. In DBA/2 mice, these sAgs bind to I-E^d MHC class II molecules and form high-affinity ligands for Vβ5 TCRs. In B6 mice, however, the sAgs bind to I-A^b MHC class II molecules and form low-affinity ligands for Vβ5 TCRs (Luther and Acha-Orbea, 1997).

Based on this, we created BM chimeras in DBA/2 and B6 mice by irradiating mice and reconstituting them with a 1:1 mixture of WT and Cd27^−/− BM. In both DBA/2 and B6 recipients, CD27 co-stimulation significantly promoted the development of T_reg cells, without affecting conventional CD4⁺ T cell development (Fig. 5 A), thus corroborating the aforementioned results. To assess whether CD27 promoted positive selection of sAg-specific T_reg cells, Vβ5 expression was analyzed in the T_reg cell compartment (Fig. 5 B). In DBA/2 recipient mice, the frequency of WT Vβ5⁺ T_reg cells was significantly higher than that of Cd27^−/− Vβ5⁺ T_reg cells (Fig. 5 C). In B6 mice, the frequencies were not significantly different. The frequencies of control Vβ4⁺ T_reg cells, which do not recognize these sAgs (Luther and Acha-Orbea, 1997), were similar among WT and Cd27^−/− T_reg cells (Fig. 5 C), Furthermore, the frequencies of Vβ5⁺ and Vβ4⁺ conventional CD4⁺ T cells were comparable between WT and Cd27^−/− cells in B6 and DBA/2 mice (not depicted). These results suggest that CD27 co-stimulation promotes the positive selection of Vβ5⁺ T_reg cells.

To determine whether CD70 expression by epithelial cells alone sufficed to promote positive selection of T_reg cells, irradiated DBA/2 mice were reconstituted either with a 1:1 mixture of WT (Cd27^+/+) and Cd27^−/− BM cells or with a 1:1 mixture of Cd27^+/+;Cd70^Cre/Cre and Cd27^−/−;Cd70^Cre/Cre BM cells. In this way, we created environments in which CD70 was expressed by both TECs and hematopoietic cells or by TECs only. In these DBA/2 chimaeras, the positive selection of Vβ5⁺ T_reg cells was significantly impaired by CD27 deficiency, both when CD70 was present or absent on hematopoietic cells (Fig. 5 D). These data suggest a scenario in which T_reg cell precursors recognize MHC class II–sAg complexes on TECs and are supported in their positive selection by concomitant CD27–CD70 co-stimulation. Intriguingly, we also noted that total T_reg cell development in this system was most efficient when both hematopoietic and radio-resistant cells expressed CD70 (Fig. 5 E). This suggested that CD70 on hematopoietic cells also played a role in thymic T_reg cell selection.

The BM chimera studies in DBA/2 mice suggested that not only CD70 on epithelial cells but also CD70 on hematopoietic cells contributed to T_reg cell selection. Previously, it has been shown that DCs play a role in thymic T_reg cell development (Proietto et al., 2008; Lei et al., 2011). Because CD70 is expressed on activated DCs in the periphery, we examined whether it was present on DCs in the thymus. For this purpose, thymus sections were stained with antibodies to CD70 and the DC marker CD11c, in conjunction with nuclear staining (not depicted) or staining for CD8α (Fig. 6 A). Both analyses indicated that a small fraction of medullary DCs expressed CD70. CD70 localized primarily intracellularly, consistent with its storage in late endosomal MHC class II compartments in mature peripheral DCs (Keller et al., 2007). CD70 was not readily detectable on thymic DCs by ex vivo flow cytometry (not depicted), and therefore, we confirmed its expression at the transcriptional level. Cd70 gene transcription was analyzed by RT-PCR on thymic CD11c⁺B220^lo DC purified from WT and Cd70^Cre/Cre mice on the basis of CD8α and SIRPα expression (Fig. 6 B). Cd70 mRNA was detected in both CD8α⁺ and SIRPα⁺ WT DCs but not in Cd70^Cre/Cre DCs (Fig. 6 C). To specify which thymic DCs may offer CD70 to promote T_reg cell development, we performed in vitro experiments. CD8α⁺ and SIRPα⁺ CD11c^hiB220^lo cDCs and CD11c^intB220^int plasmacytoid DCs (pDCs) were purified from WT and Cd70^Cre/Cre thymus and co-cultured with purified WT CD4⁺CD25⁻ thymocytes. After 5 d, CD4⁺ cells were analyzed for Foxp3 and CD25 expression. In cultures with WT SIRPα⁺ and CD8α⁺ cDCs, but not in cultures with pDCs, T_reg cell development was observed after 5 d (Fig. 6, D and E), consistent with published data (Proietto et al., 2008). Importantly, CD70 deficiency significantly reduced the capacity of CD8α⁺ cDCs to induce T_reg cells, whereas Cd70^Cre/Cre SIRPα⁺ cDCs could induce T_reg cells equally well as their WT counterparts (Fig. 6, D and E). These data suggest that CD8α⁺ cDCs in the thymic medulla, which are thymus-resident cells (Li et al., 2009), present CD70 and promote the development of T_reg cells.

To test whether CD70 on hematopoietic cells supports T_reg cell development in vivo, we created BM chimeras. Cd70^Cre/Cre mice were irradiated and reconstituted with a 1:1 mixture of WT and Cd27^−/− BM. In this setting, CD70 was present on hematopoietic cells including DCs but not on epithelial cells of the thymus. After 8 wk, the number of CD4⁺Foxp3⁺ and CD4⁺Foxp3⁻ cells was determined in thymus and spleen. Although the number of CD4⁺Foxp3⁻ cells of WT and Cd27^−/− origin was similar, CD4⁺Foxp3⁺ cells were more efficiently generated from WT donor cells than from Cd27^−/− donor cells (Fig. 7 A). This indicates that interaction between CD27 on T_reg cell precursors and CD70 on a BM-derived cell type (most likely the DCs identified above) contributes to T_reg cell development.

To test the relative contributions of CD70 on BM-derived versus epithelial cells in T_reg cell development, we performed a comparative series of BM chimera experiments. As a control, WT BM was transferred into WT mice. To exclude a role for CD70 on BM-derived cells, Cd70^Cre/Cre BM was transferred into WT mice. To exclude a role for CD70 on epithelial cells, WT BM was transferred into Cd70^Cre/Cre mice. The fourth setting, in which Cd70^Cre/Cre BM was introduced into Cd70^Cre/Cre mice, excluded CD70 on both epithelial and hematopoietic cells. Analysis of the thymus and spleen after reconstitution revealed that elimination of CD70 on either BM-derived or epithelial cells each slightly reduced T_reg cell output but not to a significant extent (Fig. 7 B). When CD70 was lacking on both cell types, however, T_reg cell development was significantly impaired. These data indicate that CD70 on both DCs and epithelial cells contributes to T_reg cell development by engaging CD27 on developing T_reg cells.

CD4⁺ T cell precursors with specificity for self-antigens undergo selection processes in the thymus that prevent autoimmunity. They can be negatively selected (clonally deleted), which is the basis of recessive tolerance. Alternatively, they can be positively selected into the T_reg cell lineage that mediates tolerance by suppressing the remaining autoreactive T cells in the periphery. Despite the expression of various co-stimulatory molecules in the thymic medulla, conventional CD4⁺ and CD8⁺ T cells appear to develop normally in their absence, and the consequence of defective co-stimulation is only apparent in peripheral T cell responses. In contrast, thymic T_reg cell development is quantitatively impaired by lack of co-stimulatory input, as shown previously for co-stimulation by CD28 (Kumanogoh et al., 2001; Tai et al., 2005; Proietto et al., 2008; Spence and Green, 2008) and in this study for co-stimulation by CD27.

Although cortical T_reg cell development has been described (Liston et al., 2008), it is generally accepted that T_reg cell precursors find the optimal signals to differentiate into mature T_reg cells in the thymic medulla. The niche for T_reg cell development is easily saturable because T_reg cell precursors most efficiently develop into mature T_reg cells when they are present at a low frequency (Bautista et al., 2009). Thus, even within the thymic medulla, the possibility for a T_reg cell precursor to become a mature T_reg cell is limited. It is becoming increasingly clear in mouse and human that subtle interplay between mTECs and DCs creates the appropriate niches for deletion of autoreactive thymocytes and induction of T_reg cells. Aire-expressing mTECs that have been implicated in T_reg cell selection (Aschenbrenner et al., 2007) can directly present a great variety of tissue-restricted antigens, but thymic DCs can also cross-present these (Gallegos and Bevan, 2004). In the human thymus, clusters of terminally differentiated mTECs called Hassall’s corpuscles produce thymic stromal lymphopoietin (TSLP), which activates DCs to express high levels of CD80 and CD86 and induce T_reg cell development (Watanabe et al., 2005). More recently, XCL-1 production by mTECs in mice was shown to attract XCR1⁺ thymic DCs into clusters within the thymic medulla, and this XCL-1 production by mTECs was shown to be Aire dependent. Mice lacking XCL-1 or Aire displayed a twofold reduction in T_reg cell numbers in the thymus, strongly suggesting that the clusters of mTECs and DCs are important for T_reg cell development (Lei et al., 2011).

Our data strongly support the scenario that mTECs and DCs in the thymic medulla jointly create the appropriate environment for T_reg cell development and implicate CD70 on both of these cell types in T_reg cell development. CD70 was exclusively found in the thymic medulla, with evident expression on a large proportion of mTECs. Most, if not all Aire⁺ mTECs expressed CD70, but Aire⁻ mTECs also expressed CD70. It is known that mTECs develop from intrathymic progenitor cells and gradually mature to terminally differentiated cells that express high levels of Aire, CD80, and MHC class II (Gray et al., 2007; Yano et al., 2008). Intermediate developmental stages express CD80 in the absence of Aire, and therefore CD70⁺Aire⁻ mTECs are most likely the precursors of CD70⁺Aire⁺ mTECs. CD70 expression on thymic DCs was less evident, but incidental DCs with primarily intracellular CD70 expression were detected by CLSM. CD70 expression on the surface of thymic DCs was not detectable by flow cytometry. This is perhaps not surprising because CD70 expression in DCs is tightly regulated, both at the transcriptional level and at the level of intracellular transport (Keller et al., 2007). CD70 is stored in MHC class II compartments and is brought to the cell surface in the immune synapse upon cognate interaction between T cells and DCs. Possibly, it is regulated in a similar manner within the thymus and only brought to the cell surface when developing T_reg cell contact the appropriate DCs.

T_reg cell generation in vitro relied on CD70 in co-cultures with CD8α⁺ cDCs but not with SIRPα⁺ cDCs. CD8α⁺ cDCs express high levels of XCR1 (Dorner et al., 2009), which may serve to attract DCs to Aire⁺ mTECs (Lei et al., 2011). This fits in with our hypothesis that mTECs and DCs form niches to promote T_reg cell development, in which CD80/86 and CD70 play an important role. However, SIRPα⁺ cDCs could also induce T_reg cell development in vitro, in a CD70-independent manner. SIRPα⁺ cDCs are a circulatory population and express higher levels of CD80, CD86, and MHC II than thymus-resident CD8α⁺ cDCs (Proietto et al., 2008; Li et al., 2009; Lei et al., 2011). The ability of SIRPα⁺ cDCs to promote T_reg cell development independently of CD70 may therefore be caused by their high expression of other co-stimulatory molecules. Alternatively, SIRPα⁺ cDCs may present different autoantigens in MHC class II than CD8α⁺ cDCs and positively select a different pool of T_reg cell precursors.

Interestingly, it was previously demonstrated that like CD70, CD80 and/or CD86 on both TECs and DCs contributes to T_reg cell development (Proietto et al., 2008). Furthermore, CD40 on epithelial cells or hematopoietic cells can also contribute to T_reg cell development (Spence and Green, 2008). It is likely that CD40 engagement promotes T_reg cell development in part by inducing the expression of CD70 on the surface of mTECs or thymic DCs because it does so on DCs in lymphoid organs (Tesselaar et al., 2003; Sanchez et al., 2007).

We found that CD28 co-stimulation promoted T_reg cell development in the absence of CD27. During the priming of naive T cells, CD28 and CD27 also perform complementary functions. CD28 amplifies the TCR signal and lowers the threshold for entry into cell cycle (Acuto and Michel, 2003), whereas CD27 provides survival signals throughout successive cell cycles (Hendriks et al., 2003). During T_reg cell development, however, CD28 does not merely act as a signal amplifier for the TCR. It contributes to Foxp3 induction over a range of TCR affinities, rather than supporting the development of cells in which Foxp3 is already switched on (Tai et al., 2005). Specifically, CD28 signaling via the Lck tyrosine kinase, but not via the PKB survival pathway, drives pre-T_reg cell development to the next stage. In contrast, CD27 deficiency did not impact pre-T_reg cell numbers, Foxp3 induction, or programming of suppressive function in T_reg cells. Rather, CD27 signaling counteracted the mitochondrial apoptosis pathway in developing T_reg cells, as demonstrated by the increased exposure of phosphatidylserine and increased expression of proapoptotic Bcl-2 molecules in CD27-deficient T_reg cells, as well as the capacity of ectopic Bcl-2 to rescue the development of CD27-deficient T_reg cells. Several lines of evidence argue that T_reg cells rely on survival signals for their thymic development. First, CD4⁺CD25⁺ thymocytes are more resistant to negative selection than CD4⁺CD25⁻ thymocytes (van Santen et al., 2004; Taylor et al., 2007). Furthermore, deficiency in the proapoptotic Bcl-2 protein Bim leads to a large increase specifically in T_reg cell numbers (Ouyang et al., 2010). Our data lend further support to the idea that T_reg cell development relies on signals that inhibit the intrinsic apoptosis pathway and implicate the CD27–CD70 pathway in providing such signals, as it does in peripheral T cells (Peperzak et al., 2010a).

The question remains as to why CD27 signals enhance the numerical output of T_reg cells without affecting that of conventional αβ T cells. Thymic γδ T cells are the only other cells that have been shown to benefit from CD70 in the thymus (Ribot et al., 2009). Both T_reg and γδ T cells express CD27 at approximately twofold higher levels than conventional αβ T cells (not depicted), and perhaps they profit more from CD27–CD70 co-stimulation for this reason. Higher expression of CD27 on γδ T cells and T_reg cells may be explained by the higher-affinity TCR signals they receive in the thymus (Ciofani and Zúñiga-Pflücker, 2010).

Another critical factor in T_reg cell development is signaling by the common γ chain of cytokine receptors, as provided largely through the IL-2 receptor, which stabilizes Foxp3 expression via STAT-5 signaling (Burchill et al., 2007, 2008). IL-2 is provided to T_reg cell precursors in a paracrine fashion by other thymocytes. Accordingly, the co-transfer of WT precursors can correct the inability of IL-2–deficient precursors to become T_reg cells (Fontenot et al., 2005b; Tai et al., 2005). Importantly, CD27 as well as CD28 contribute to T_reg cell development independently of paracrine IL-2 because co-transfer of WT donor cells together with CD27- or CD28-deficient donor cells in BM chimeras did not correct their defect in T_reg cell generation (Tai et al., 2005; this study).

T_reg cell numbers in Cd27^−/− or Cd70^Cre/Cre mice or chimeras thereof are reduced to maximally about half of control numbers in the periphery (Fig. 1 B). We have not observed signs of spontaneous autoimmunity in these mice. The same applies to Cd28^−/− mice, in which T_reg cell numbers are reduced to about one third of control (Tai et al., 2005). In Cd28^−/− mice, there are also no reported signs of autoimmunity. Therefore, it is possible that an ∼50–70% reduction in T_reg cell numbers does not affect self-tolerance. Alternatively though, self-tolerance may not be affected in these mice because conventional T cells also lack CD27 or CD28 and therefore have a higher threshold for responsiveness.

It was recently shown that immune surveillance to tumors with CD70⁺ infiltrating lymphocytes was greater in Cd27^−/− mice than in WT mice. This was caused by impaired expansion and survival of peripheral T_reg cells (Claus et al., 2012). Therefore, the regulation of T_reg cell numbers via the CD27–CD70 pathway can play an important physiological role. Furthermore, we have recently shown that Cd27^−/− and Cd70^Cre/Cre mice develop more severe EAE. The CD27–CD70 pathway functionally disabled T-helper 17 cells that play a dominant role in this disease (Coquet et al., 2013). Possibly, apart from increased numbers of T-helper 17 cells, reduced expansion of peripheral myelin-specific T_reg cells also contributed to enhanced EAE in Cd27^−/− and Cd70^Cre/Cre mice. The use of MHC II tetramers or TCR sequencing technology to detect antigen-specific T_reg cells in cancer or autoimmune models may help to shed light on the role of the CD27–CD70 pathway in peripheral T_reg cell responses.

In conclusion, although T_reg cell development abides by many of the rules applicable to conventional αβ T cell development, T_reg cells are unique in their requirement for co-stimulation and IL-2 in dedicated niches in the thymic medulla. Our data demonstrate that the CD28 and CD27 co-stimulatory pathways make specific and nonredundant contributions to T_reg cell development. TCR and CD28 signals kick start the developmental process by induction of Foxp3 expression. Thereafter, IL-2/IL-2 receptor signaling stabilizes T_reg cell differentiation, whereas CD27–CD70 co-stimulation rescues developing T_reg cells from apoptosis and thereby allows them to take part in the peripheral T_reg cell population.

B6 WT, Cd27^−/−, Cd70^Cre/Cre, DBA/2 mice, and congenic CD45.1 mice were bred at the animal facilities of the Netherlands Cancer Institute (NKI) and Instituto de Medicina Molecular (IMM). Cd27^−/− mice were made on a 129J/Ola background and backcrossed to B6 for eight generations (Hendriks et al., 2000). Cd70^Cre/Cre mice were generated as described below. All animal experiments were performed according to national and institutional guidelines and were approved by the respective experimental animal committees of the NKI and IMM.

Cd70^Cre/Cre mice are deficient for CD70 expression because exon I of the Cd70 locus has been replaced by coding sequences of the Cre recombinase. The genetic modification was performed in E14 (129P2/Ola) embryonic stem cells (Coquet et al., 2013). Using speed congenics, germline competent chimeras were crossed for five generations onto a B6 background (>98% B6 contribution) and subsequently maintained as inbred strain. Note that we have thus far no indication that the Cre gene is efficiently expressed under control of the endogenous Cd70 gene promoter.

Mice were irradiated with two doses of 5 Gy, 3 h apart, and the next day injected intravenously with a total of 10⁷ whole BM cells in 200 µl PBS. The hematopoietic compartment was allowed to reconstitute for 8 wk before organs were harvested for analysis by flow cytometry. In Fig. 5 (D and E), CD27 was used as a marker for T_reg cell generation from Cd70^Cre/Cre donor cells because both Cd70^Cre/Cre and Cd27^−/−;Cd70^Cre/Cre donor cells were of CD45.2⁺ origin. In the Bcl2 transduction experiments, c-Kit⁺ BM cells were enriched by magnetic cell sorting (Miltenyi Biotech) and infected overnight with pMig.IRES-GFP retroviral empty vector or containing Bcl2 (cloned from Addgene plasmid no. 8750) in the presence of 0.8 mg/ml polybrene (Sigma-Aldrich). c-Kit⁺GFP⁺ cells were sorted by flow cytometry, and appropriate mixes of a total 10⁵ cells were injected intravenously in irradiated hosts.

Cd27^−/− mice were injected i.p. every 2 d, for 2 wk, with 100 µg of each anti-CD80 (16-10A1) and anti-CD86 (GL1) mAb or corresponding isotype controls.

To isolate lymphocytes from thymus and spleen, organs were passed through 100-µM nylon mesh (BD), and red blood cells were lysed in 0.14 M NaCl and 0.017 M Tris-HCl, pH 7.2, for 1 min at room temperature (rT). To isolate BM, femurs were removed from mice and flushed with PBS using a 25-G needle. Red blood cell lysis was subsequently performed for 1 min at rT. Cells from spleen, thymus, or BM were resuspended in PBS containing 2% bovine serum albumin and counted on a CASY cell counter (Schärfe).

Cells were stained with the following fluorochrome-conjugated mAbs: anti-CD4 (GK1.5), anti-CD8 (53-6.7), anti-CD25 (PC-61), anti-CD44 (IM7), anti-CD11c (N418), anti-CD45RA (RA3-6B2), anti-Vβ4 (KT4), and anti-Vβ5 (MR9-4) from BD; and anti-CD45.2 (104), anti-CD27 (LG.3A10), anti-SIRPα (P84), and anti-Foxp3 (FJK-16s) from eBioscience. Intracellular staining for Foxp3 was performed using the staining buffer set from eBioscience. Cells were analyzed on Cyan (Dako) or FACSCanto (BD) flow cytometers. Cells were purified by flow cytometric sorting on either a FACSAria (BD) or MoFlow (Dako). Sorted cells were always at least 98% pure. Data were analyzed using FlowJo software (Tree Star).

Mouse thymus was harvested and roughly diced into small pieces using scissors in 25 ml IMDM containing 1 mg/ml collagenase type IV (Worthington) and 25 µg/ml DNase (Sigma-Aldrich) and incubated at 37°C for 1 h. Liberated cells were filtered through 100-µM nylon mesh and incubated with CD11c microbeads (Miltenyi Biotech) for 30 min before being run through a MACS column (Miltenyi Biotech). Enriched thymic DCs were then stained for expression of CD11c, CD45RA, CD8α, and SIRPα and sorted by flow cytometry. RNA was isolated from purified thymic DCs using the RNeasy kit (QIAGEN), and cDNA was transcribed using random hexamers. Primers to detect CD70 and HPRT message were previously described (Tesselaar et al., 2003).

Cells were cultured in either IMDM or RPMI containing 8% FCS, penicillin, streptomycin, and β-mercaptoethanol (tissue culture medium). For co-cultures of thymic DCs and T cells, 10⁴ sorted thymic DCs were cultured with 3 × 10⁴ sorted thymic CD4⁺CD8⁻CD25⁻ cells in 96-well U-bottom plates for 5 d. Thereafter, cells were harvested and stained for the surface markers CD4 and CD25 before being stained intracellularly for Foxp3 and analyzed by flow cytometry. For T_reg cell suppressor assays, CD4⁺CD25⁺ cells (T_reg cells) were sorted from WT or Cd27^−/− mice and CD4⁺CD25⁻ responder T cells were sorted from WT mouse spleen. U-bottom 96-well plates were coated with 2 µg/ml anti-CD3 mAb 145.2C11 for 2 h at 37°C. Responder T cells were labeled with 5 µM CFSE in PBS supplemented with 2% FCS for 5 min at 37°C. T_reg and responder T cells were cultured in tissue culture medium at various ratios as indicated in coated U-bottom plates for 3 d in the presence of irradiated splenocytes. After 3 d, cells were harvested and analyzed for dilution of CFSE by flow cytometry.

FTOCs were maintained at 37°C, 5% CO₂ in FTOC medium, which is RPMI-1640 with 10% FCS (STEMCELL Technologies), 50 µM β-mercaptoethanol, 2 mM l-glutamine, 100 U penicillin, and 100 µg/ml streptomycin. B6 fetal thymic lobes at E15 were cultured for 14–16 d in FTOC medium on Nuclepore filters (Whatman) and then analyzed by flow cytometry. Cultures were provided with fresh FTOC medium at day 7. For CD27 stimulation, a fusion protein of the extracellular domain of mouse CD70 and the Fc portion of human IgG1 (FcCD70; Peperzak et al., 2010b) or human IgG1 control antibody was added to the cultures. To disrupt CD27–CD70 interactions, the anti-CD70 mAb FR70 (provided by H. Yagita, Juntendo University School of Medicine, Tokyo, Japan; Oshima et al., 1998) or a control rat IgG2b isotype control (eBioscience) was added to culture. Cultures containing antibodies were rested for 24 h in fresh medium before analysis.

Mouse thymus was embedded in Cryo-Block medium (LTI) and frozen at −80°C, after which sections were made, fixed in acetone, air dried, and stored at −20°C. Sections were rehydrated in PBS for 30 min at rT, immersed in PBS, 5% BSA for 30 min at rT, and stained with antibodies in PBS, 1% BSA for 1 h at rT. Primary antibodies used were rat anti–mouse CD70 mAb FR70 (homemade purified Ig), rat anti–mouse mAb ER-TR5 (provided by P.J. Leenen, Erasmus University, Rotterdam, Netherlands; Van Vliet et al., 1984), rabbit anti–mouse Keratin-5 polyclonal PRB-160P (Covance), Alexa Fluor 488–conjugated rat anti–mouse Aire mAb 5H12 (eBioscience), FITC-conjugated Armenian hamster anti–mouse CD11c mAb N418 (eBioscience), and Allophycocyanin-conjugated rat anti–mouse CD8α mAb 53-6.7 (BD). Second step antibodies were Alexa Fluor 488– or Alexa Fluor 568–conjugated anti–rat IgG or anti–rabbit IgG (Invitrogen). In case two rat mAbs were used sequentially, sections were incubated with normal rat serum as a blocking step in between. After antibody staining, sections were incubated as indicated with DAPI (Sigma-Aldrich) at 0.15 µg/ml in PBS for 20 min at rT. Stained sections were mounted using Vectashield (Vector Laboratories) or FLUORO-GEL with TES buffer (Electron Microscopy Sciences) and observed under a TCS NT CLSM (Leica), using an HCX PL APO CS objective lens with 40× magnification and 1.3 aperture. Signals were acquired with LAS AF Lite software (Leica), and Fiji software was used for image processing.

Thymic CD4⁺CD25⁺ and CD4⁺CD25⁻ cells were sorted, and RNA was isolated from pelleted cells using the RNeasy mini kit (QIAGEN). 5 ng of RNA was used for MLPA (MRC-Holland) using the RM002-B1 mouse apoptosis probe mix as previously described (Eldering et al., 2003). Three separate experiments were performed, and data for Cd27^−/− cells were normalized to WT levels in each experiment.

Fig. S1 shows flow cytometry gating schemes for pre-T_reg cells, CD80/CD86 blockade, CD69 and HSA staining on pre-T_reg cells, and Annexin-V staining.

We thank H. Veiga-Fernandes, R.G. Domingues, I. Barbosa, and S. Madeira (IMM) for help with the Bcl2 transduction experiments, H. Yagita for FR70 antibody, P.J. Leenen for the ER-TR5 antibody, E.A.M. Veraar for FcCD70 production, and the staff of the animal and flow cytometry facilities of our institutes for expert technical assistance.

This work was supported by grants NKI 2004-3087 and NKI 2008-2023 from the Dutch Cancer Society to J. Borst, European Molecular Biology Organization long-term fellowships to J.M. Coquet and J.C. Ribot, a Rubicon (The Netherlands Organisation for Scientific Research [NWO]) fellowship to J.M. Coquet, and a Fundação para a Ciência e Tecnologia project grant (PTDC/SAU-MII/104158/2008) and a European Research Council starting grant (StG260352) to B. Silva-Santos. J.F. Neves is funded by the Fundação para a Ciência e Technologia of Portugal; D.J. Pennington is funded by the Wellcome Trust.

The authors declare no commercial affiliations or competing financial interests.