Costimulation via the tumor-necrosis factor receptor superfamily couples TCR signal strength to the thymic differentiation of regulatory T cells.

Regulatory T cells (Treg cells) express members of the tumor-necrosis factor (TNF) receptor superfamily (TNFRSF), but the role of those receptors in the thymic development of Treg cells is undefined. We found here that Treg cell progenitors had high expression of the TNFRSF members GITR, OX40 and TNFR2. Expression of those receptors correlated directly with the signal strength of the T cell antigen receptor (TCR) and required the coreceptor CD28 and the kinase TAK1. The neutralization of ligands that are members of the TNF superfamily (TNFSF) diminished the development of Treg cells. Conversely, TNFRSF agonists enhanced the differentiation of Treg cell progenitors by augmenting responsiveness of the interleukin 2 receptor (IL-2R) and transcription factor STAT5. Costimulation with the ligand of GITR elicited dose-dependent enrichment for cells of lower TCR affinity in the Treg cell repertoire. In vivo, combined inhibition of GITR, OX40 and TNFR2 abrogated the development of Treg cells. Thus, expression of members of the TNFRSF on Treg cell progenitors translated strong TCR signals into molecular parameters that specifically promoted the development of Treg cells and shaped the Treg cell repertoire.

The development of T lymphocytes with randomly generated T cell receptors (TCRs) allows for diverse and productive immune responses, but also raises the risk for autoimmunity. The thymic medulla contains antigen presenting cells (APCs) that display self-peptide:major histocompatibility (MHC) complexes to immature CD4 and CD8 single positive (SP) thymocytes, and plays a critical role in limiting autoimmunity[1]. This occurs in part by deletion of a substantial fraction of autoreactive thymocytes via the process of negative selection. However, some autoreactive T cells escape deletion in the thymus, and thus alternative mechanisms to control autoreactivity have evolved including the production of thymic-derived regulatory T cells (Treg)[2]. The transcription factor Foxp3 plays a critical role in this process as inactivating mutations in Foxp3 abrogate Treg development and cause lethal multi-organ lymphoproliferative disease[3, 4]. More subtle perturbations in Treg development and/or function, such as reduced responsiveness to interleukin-2 (IL-2), are also associated with autoimmune diseases such as Type 1 Diabetes[5-7]. Treg development occurs via a multistep differentiation process that preferentially drives high affinity autoreactive T cells to express Foxp3. A key question is how thymocytes translate relative differences in TCR signal strength into distinct developmental programs that give rise to naïve CD4+ T cells or regulatory T cells, respectively.

Thymic Treg development occurs via a two-step process[8, 9]. An initial TCR-CD28-driven signaling event leads to expression of CD25 and CD122 (the α and β subunits of the IL-2 receptor) and c-REL-dependent chromatin remodeling at the Foxp3 locus[10-15]. A second, TCR-independent but cytokine-dependent step occurs when CD25+Foxp3− Treg progenitors receive an IL-2 signal, which via a STAT5-dependent process drives Foxp3 expression to complete the Treg differentiation process[8, 9, 16].

How high affinity TCR stimuli drive a molecular program that results in Treg differentiation remains poorly defined. A hallmark of Treg progenitor cells that might play a role in this process is their high expression of the TNFRSF member TNFRSF18, (called GITR hereafter). As GITR expression precedes the induction of Foxp3 in developing Treg cells, we hypothesized that costimulation via GITR may support conversion of Treg progenitors into mature Foxp3+ Treg cells in the thymus.

In addition to GITR, several other TNFRSF members are expressed on Treg cells[17]. We screened thymocytes by flow cytometry and found that GITR, TNFRSF4 (called OX40 herafter), and TNFRSF1B (called TNFR2 hereafter) are uniquely overexpressed on Treg progenitors when compared to conventional CD4SP thymocytes. Multiple APCs including dendritic cell subsets and medullary thymic epithelial cells express the corresponding ligands GITR-L, OX40-L, and TNF. We found that TNFRSF expression by Treg progenitors strongly correlates with TCR signal strength. Thus, CD4SP thymocytes encountering the highest affinity TCR signals most strongly upregulate GITR, OX40 and TNFR2. This process occurred via a TAK1-and CD28-dependent pathway as TAK1- and CD28-deficient Treg progenitors fail to express GITR, OX40, and TNFR2 and do not convert into mature Foxp3+ Treg cells. Stimulation of wild type Treg progenitors with either GITR-L or OX40-L promoted their conversion into mature Foxp3+ Treg cells at much lower doses of IL-2. In contrast, the addition of neutralizing antibodies to TNFSF members in neonatal thymic organ cultures markedly inhibited Treg development. Likewise, blocking signaling collectively through OX40, TNFR2 and GITR completely abrogated Treg development in vivo indicating that GITR, OX40 and TNFR2 function in a cell-intrinsic manner to program Treg differentiation. Finally, competition for TNFRSF costimulation skews the Treg repertoire toward higher affinity TCRs, as increasing GITR-L availability dose-dependently broadens the Treg repertoire by enriching for cells with reduced TCR signal strength. These findings support a model in which high-affinity TCR signals in CD4SP thymocytes are translated into a molecular program that entrains Treg development via increased expression of specific TNFRSF members. Ligation of GITR, OX40 or TNFR2 on Treg progenitors enhances the ability of Treg progenitors to compete for limiting amounts of IL-2 and thereby helps define the developmental niche for Treg cells in the thymus.

Results

Treg progenitors highly express GITR, OX40 and TNFR2

A prominent feature of Treg progenitors is their high-level expression of GITR, which is upregulated prior to Foxp3 during Treg development (refs 8, 9 and Fig. 1a). Studies to elucidate a role for GITR in Treg cells have primarily focused on the effects of costimulation on mature Treg cells in peripheral lymphoid organs[18-20]. However, whether GITR costimulation plays any role during thymic Treg development is unknown. Because of the large potential for redundancy within the TNFRSF, we sought to clarify which of these receptors are expressed during Treg development. Staining Foxp3-GFP thymocytes with specific antibodies to evaluate expression by conventional (non-Treg) CD4SP thymocytes, Treg progenitors, and mature Treg cells (gated as CD4+CD8−CD25−Foxp3−, CD4+CD8−CD25+Foxp3−, and CD4+CD8−CD25+Foxp3+, respectively) revealed that in addition to GITR, Treg progenitors also highly express OX40 and TNFR2 (Fig. 1a). Mature Foxp3+ Treg cells continue to express high amounts of these TNFRSF members in the thymus (Fig. 1a) and in peripheral lymphoid organs (data not shown). Treg progenitors do not globally overexpress TNFRSF members, as 4-1BB, CD30, and TNFR1 were not detectable above background in thymic Treg progenitors or Treg cells (Fig. 1a and data not shown). CD27, which is already expressed on DP thymocytes, is highly expressed by all CD4SP thymocytes but was not further upregulated by Treg progenitors or mature Foxp3+ Treg cells (Fig. 1a, Supplementary Fig. 1). Thus, GITR, OX40, and TNFR2 are highly upregulated on Treg progenitors during thymic Treg development.

Figure 1

Expression of GITR, OX40, and TNFR2 on thymic Treg progenitors and GITR-L, OX40-L, and TNF on APCs in the thymic medulla

(a) Thymocytes from adult (6-10 week old) female or male Foxp3-GFP mice were stained with antibodies against the indicated TNFRSF members in conjunction with antibodies against CD4, CD8, and CD25. In the left panel, conventional (non-Treg) CD4SP thymocytes were gated as CD4+CD8−CD25−Foxp3− (red lines), Treg progenitors as CD4+CD8−CD25+Foxp3− (green lines), and mature thymic Treg cells were CD4+CD8−CD25+Foxp3+ (blue lines). The histograms to the right display the expression of TNFRSF within the indicated populations as compared to isotype-stained controls (gray shaded histograms). Data are representative of at least three separate experiments (n ≥ 3). (b) Expression analysis of the specific TNFSF ligands GITR-L, OX40-L, and TNF was conducted by flow cytometry on collagenase- and DNase I-digested thymii harvested from 6-8 week old male or female mice, which were magnetically depleted with anti-CD45 for enrichment of mTEC or enriched with anti-CD11C for DCs. Histograms show the expression of GITR-L (top), OX40-L (middle), and total TNF (membrane + intracellular, bottom) within the indicated populations gated as follows: mTEC (CD45−EpCAM+UEA1+), SIRPα+ cDC (CD11c+B220−SIRPα+CD8α−), CD8α+ cDC (CD11c+B220−SIRPα−CD8α+), and pDC (CD11c+B220+). Data are representative of thymic DC or mTEC isolations pooled from 3-4 mice in at least three separate experiments (n ≥ 3). (c) Frozen thymic sections from 6-12 week old female or male mice were stained with anti-Keratin 5 (K5-FITC, green), anti-OX40-L or isotype control (PE, red), and corresponding images were merged (yellow) to evaluate the spatial distribution of TNFSF ligands within the thymus. (10X magnification, white bars represent 100 um; n=3, one of three independent experiments).

APCs in the thymic medulla express GITR-L, OX40-L and TNF

For signaling, homotrimeric TNFRSF members require ligation by specific TNFSF ligands[21]. Various APC subsets have been described to express TNFSF ligands upon activation in secondary lymphoid organs, but the constitutive expression of GITR-L, OX40-L, and TNF within the thymic APC subsets that actually mediate Treg differentiation has not been defined. Medullary thymic epithelial cells (mTEC; CD45−EpCAM+UEA1+BP-1lo) constitutively express GITR-L, OX40-L, and TNF (Fig. 1b). Both GITR-L and TNF are expressed by CD8α+ and SIRPα+ conventional DC subsets (CD11C+B220−), and by plasmacytoid DCs (CD11C+B220+), whereas OX40-L expression is restricted to mTEC. Since TNFR2 can bind both membrane-bound and soluble TNF [22, 23], we evaluated TNF expression by surface staining and by intracellular staining after fixation and permeabilization. We were able to detect membrane bound TNF on DCs (data not shown) albeit at much lower levels than total TNF (Fig. 1b).

We used immunofluorescence microscopy to define the pattern of TNFSF ligand expression within the thymic microenvironment. Co-staining slides with anti-Keratin 5, which marks the thymic medulla (the location of thymic Treg differentiation), showed a high degree of overlap with OX40-L expression (Fig. 1c). Similar results were obtained when staining for GITR-L and TNF (data not shown). Thus, the specific ligands that bind the TNFRSF members present on developing Treg cells are expressed by APCs at the site of thymic Treg maturation.

GITR, OX40 and TNFR2 correlate with TCR signal strength

When we gated on Treg progenitors and plotted the expression of GITR versus OX40 (Fig. 2a) or GITR versus TNFR2 (data not shown), we found that cells expressing the most GITR also express the most OX40 and TNFR2, and vice versa. This finding prompted us to ask the question if TNFRSF expression is related to TCR signal strength in Treg progenitors. For these experiments, we used Nur77-GFP BAC reporter mice, which provide a reliable readout of TCR signal strength via GFP expression[24]. We found a significant positive correlation between Nur77-GFP fluorescence and GITR, OX40 and TNFR2 expression in Treg progenitors (Fig. 2a,b). In contrast, expression of CD27, a TNFRSF member recently reported to drive thymic Treg development via the inhibition of apoptosis[25], did not correlate significantly with Nur77-GFP (Fig. 2b). Therefore, thymic Treg progenitors receiving the strongest TCR signals during development are engendered with the highest expression of GITR, OX40, and TNFR2. Moreover, mature Treg cells express substantially more OX40 and GITR than Treg progenitors (Fig. 1a) suggesting that Treg progenitors expressing relatively more of these TNFRSF members have a selective advantage in thymic Treg differentiation.

Figure 2

Treg progenitors express select TNFRSF members in direct proportion to TCR signal strength

(a) Six-to-twelve week old female or male Nur77-GFP reporter mice were used to evaluate the role of TCR signal strength on the expression of TNFRSF members. Treg progenitors were plotted on the basis of their expression of GITR and OX40 (left), and quartile gates representing the spectrum of expression of these receptors were evaluated for Nur77-GFP expression in the corresponding histograms (right). Data are representative of four separate experiments (n = 4). (b) Raw values for GITR, OX40, TNFR2, and CD27 on Treg progenitors were plotted versus Nur77-GFP expression. Lines represent the correlation between x and y variables and were used to calculate Pearson correlation coefficients (r). Correlation data for GITR are representative of 8 separate experiments (n = 8), for OX40 from 4 experiments (n = 4), for TNFR2 from 3 experiments (n = 3), and for CD27 from 3 experiments (n =3). (c) Histograms derived by gating on Treg progenitors from Cd4Cre × Tak1FL/FL mice (abbreviated as Tak1 KO, black histograms) and their wild type littermates (B6 background; gray shaded histograms; top panel) or Cd28−/− mice (black histograms) and their wild type littermates (BALB/c; gray shaded histograms; lower panel) are plotted demonstrating the expression of GITR, OX40, TNFR2, and CD27. (d) Cumulative data show the relative expression of GITR, OX40, TNFR2, and CD27 on Treg progenitors from TAK1-deficient and CD28-deficient mice in comparison to wild type littermates. Raw MFI values obtained from each mouse after gating on CD25+Foxp3− Treg progenitors were divided by the average MFI obtained from all wild type mice for normalization amongst separate experiments (mean ± SEM, for Tak1 KO experiments n ≥ 4 from at least two separate experiments, and for Cd28−/− n ≥ 3 from one experiment on the BALB/c background and n ≥ 3 from one experiment on the B6 background, see Supplementary Fig. 2, * P < 0.001).

TAK1 and CD28 govern GITR, OX40 and TNFR2 expression

To identify the pathways downstream of the TCR that entrain expression of specific TNFRSF members we used mice that lack the MAPK kinase kinase TAK1 (also known as MAP3K7) specifically in T cells ( mice; Tak1 KO). TAK1 links TCR signaling to NF-κB activation[26] and thus is a key signaling intermediate downstream of the TCR. Previous studies have documented that TAK1-deficiency profoundly blocks thymic Treg differentiation but did not examine effects on Treg progenitors as these studies preceded the identification of this cell population[26, 27]. We found that Treg progenitors could be detected in Cd4 × Tak1 mice although their numbers are reduced (data not shown). These Treg progenitors fail to express GITR, OX40, and TNFR2 on their surface and do not convert into mature Foxp3+ Treg cells (Fig. 2c,d and data not shown). In contrast, CD27 expression was not affected by TAK1-deficiency. Thus, TAK1 is required for TCR-induced expression of GITR, OX40, and TNFR2 on Treg progenitors.

We also examined the surface expression of TNFRSF in Treg progenitors from CD28-deficient mice, as CD28 plays an important role in Treg development[28-30]. Cd28−/− mice showed a significant reduction in GITR and OX40 expression on Treg progenitors relative to BALB/c controls; a more modest trend toward reduced TNFR2 expression was also observed (Fig. 2c,d, bottom panels). In contrast, CD27 surface expression on Treg progenitors was unaffected by the absence of CD28. Similar results were obtained with Cd28AYAA knockin mice (data not shown), which lack the distal proline-rich motif that was previously shown to be required for Treg development[13]. This was not a strain-specific effect as we found a similar reduction in the expression of GITR and OX40 on Treg progenitors from Cd28 mice on the C57Bl/6 background (Supplementary Fig. 2). Thus, CD28 signals are also required for upregulation of GITR and OX40 expression (and likely TNFR2) on Treg progenitors.

GITR, OX40 or TNFR2 costimulation drives Treg maturation

To determine if signaling through GITR, OX40 or TNFR2 directly augments Treg development by promoting maturation of Treg progenitors, we utilized an assay that models step 2 of Treg development in vitro (i.e., the IL-2-dependent conversion of Treg progenitors into mature Foxp3+ Treg cells). Purified Treg progenitors sorted from Foxp3-GFP reporter mice were incubated in culture with low dose IL-2 with or without agonist TNFSF ligands. Conversion was assessed as the frequency of cells that acquired Foxp3-GFP expression during the course of the incubation. The addition of GITR-L–Fc, OX40-L–Fc, or TNF plus IL-2 enhanced conversion to Foxp3+ Treg cells by ~2-3 fold when compared to treatment with IL-2 alone (Fig. 3a,b). In contrast, three independent CD27 agonists (CD70–Fc, soluble CD70, and an agonist monoclonal antibody to CD27) did not enhance Treg development (Fig. 3a,b and data not shown). Treg progenitors do not express 41BB or CD30, and as expected, stimulation with 41BB-L or CD30-L did not affect Treg development (data not shown).

Figure 3

TNFRSF agonists enhance conversion of Treg progenitors to Foxp3+ Treg cells

(a,b) Thymii from six- to eight-week old female or male Foxp3-GFP mice were harvested, pooled, and depleted of CD8SP, DP, and non-T cells prior to sorting purified CD4+CD25+Foxp3-GFP− Treg progenitors. Sorted cells were incubated in culture with or without IL-2 (1 U/mL) and 100 nM of the indicated TNFRSF agonists for 72 hours prior to flow cytometric analysis to determine the fraction of cells that acquired Foxp3-GFP during the course of the incubation (mean ± SEM from three independent experiments, n = 3, * P < 0.05). (c) The expression of CD25 amongst Foxp3-GFP− Treg progenitors after stimulation with IL-2 and GITR-L–Fc is plotted as fold change in geometric mean fluorescence intensity (gMFI) over IL-2 treatment alone (mean ± SEM derived from seven independent experiments, n = 7, * p < 0.05). (d) CD25 expression was measured as gMFI in the Foxp3-GFP+ gate after a 72h stimulation with IL-2 and GITR-L–Fc (mean ± SEM from three independent experiments, n = 3, * P < 0.05). (e) The efficiency of Treg progenitor maturation with or without TNFRSF costimulation in response to a range of IL-2 concentrations is shown (mean ± SEM from three independent experiments, n = 3). (f) Intracellular staining for p-STAT5 was performed after stimulating sorted Treg progenitors with IL-2 and/or TNFRSF agonists for 24 hours. Data are representative of one of two independent experiments (n = 2). (g) Shown are the percentages of sorted Treg progenitors that converted to Foxp3-GFP+ Treg cells during a 72h incubation with IL-2 and GITR-L–Fc or IL-2 and the pan-caspase inhibitor, z-VAD-FMK, using DMSO (v/v) as a vehicle control (mean ± SEM derived from three independent experiments, n = 3, * P < 0.05).

One potential mechanism by which TNFRSF agonists promote Treg development is to enhance sensitivity to IL-2. To directly test this, we examined CD25 expression on both Foxp3-GFP− Treg progenitors and Foxp3-GFP+ mature Treg cells that had been stimulated with IL-2 ± GITR-L–Fc. CD25 expression was significantly enhanced on both Treg progenitors and mature Treg cells (Fig. 3c,d). To examine the relative effect of TNFRSF co-stimulation on IL-2-dependent Treg differentiation we carried out a dose-response experiment. Stimulation of Treg progenitors with 1 U/mL IL-2 plus GITR-L or OX40-L resulted in 50% maximal conversion to mature Treg cells; in the absence of GITR-L or OX40-L co-stimulation, almost 10-fold more IL-2 was required to achieve the same effect (Fig. 3e). Agonist GITR-L–Fc, OX40-L–Fc, or TNF did not promote Treg maturation in the absence of IL-2 (data not shown). Finally, we observed that GITR-L–Fc costimulation enhanced IL-2-dependent STAT5 phosphorylation in Treg progenitors, suggesting that GITR-L acts directly on the IL-2R signaling pathway (Fig. 3f). These data demonstrate that GITR, OX40, or TNFR2 costimulation augments IL-2 responsiveness in Treg progenitors and allows them to effectively respond to low and likely physiologically relevant levels of IL-2.

An alternative mechanism by which TNFRSF costimulation enhances Treg development is through the induction of anti-apoptotic proteins. To address this, we determined whether blocking apoptosis via the addition of the pan-caspase inhibitor, z-VAD-fmk, had a similar effect in promoting Treg development to GITR-L–Fc. The addition of z-VAD-fmk significantly increased total cell recovery (p<0.02, data not shown); however, z-VAD-fmk did not enhance IL-2-dependent conversion of Treg progenitors into mature Treg cells (Fig. 3g). Thus, costimulation via GITR, OX40, and TNFR2 provide more than just pro-survival signals to developing Treg cells.

GITR- or OX40-deficiency modestly blocks Treg development

We next sought to determine if TNFRSF members drive Treg differentiation in vivo in a cell-intrinsic manner. To examine this we created bone marrow chimeras in which a 50-50% mixture of CD45.1/SJL wild type congenic marrow was mixed with CD45.2 Gitr −/− (also known as Tnfrsf18) bone marrow, depleted of mature lymphocytes, and engrafted into sublethally irradiated Rag2−/− recipients. After 10-18 weeks of reconstitution, chimeric recipients were euthanized and we assayed Treg development by flow cytometry. As shown in Fig. 4b, Gitr cells gave rise to conventional CD4+CD25−Foxp3− thymocytes equally as well as their wild type counterparts, but were outcompeted by wild type cells by approximately 30% in generating mature CD25+Foxp3+ thymic Treg cells. Treg progenitor populations were also slightly reduced in GITR-deficient cells, although the reduction was significantly less than observed in mature Foxp3+ Treg cells. Similar results were observed in splenic Treg cells (data not shown). We carried out similar studies using Ox40 −/− (also known as Tnfrsf4) mice; we observed a 50% reduction in the ability of Ox40 −/− -derived cells to give rise to Foxp3+ Treg cells (Fig. 4b). The effect of TNFR2 on thymic Treg development has not been examined as closely but as TNFR2-deficient mice are viable it is unlikely that they are devoid of Treg cells. Thus, deletion of any one TNFRSF member alone imposes only a modest cell-intrinsic Treg developmental block.

Figure 4

GITR- or OX40-deficiency imposes a modest cell-intrinsic thymic Treg developmental block

Bone marrow was harvested from 8-12 week old female or male (a) Gitr −/− or (b) Ox40 −/− mice (CD45.2; also known as Tnfrsf18 and Tnfrsf4, respectively) and age- and sex-matched congenic recipients and depleted of mature lymphocytes. Donor cells were mixed prior to engraftment into sublethally irradiated adult Rag2 −/− mice by IV injection. Ten to 18 weeks later thymii and spleen were analyzed for Treg development and homeostasis by flow cytometry. (a,b) Shown are the percentages of CD45.2+ vs. CD45.1+ cells occupying gates that identify conventional T cells (CD25−Foxp3−), Treg progenitors (CD25+Foxp3−), and mature thymic Treg cells (CD25+Foxp3+) normalized to the ratio of CD45.2+ vs. CD45.1+ within total CD4SP thymocytes. Results in panel a are from 10 individual chimeras from two independent experiments; results in panel b are from 12 individual chimeras from three separate experiments (mean ± SEM, P < 0.05 values calculated by ANOVA with Bonferroni’s multiple comparison test).

TNFRSF blockade inhibits Treg development and maturation

We next sought to determine if TNFRSF function redundantly to drive Treg development by blocking multiple ligands simultaneously. Thymic organ cultures (TOCs) are a widely utilized technique to study thymocyte development in vitro with minimal manipulation[31]. We harvested thymic lobes from 1 day-old Foxp3-GFP reporter mice and subjected them to pair-wise comparison in TOCs for 14 days supplemented with either irrelevant isotype control antibodies or neutralizing anti-GITR-L, OX40-L, and CD70 or anti-GITR-L, OX40-L, CD70 and TNFR2. Neutralization of either three or four distinct TNFRSF members resulted in a dramatic reduction in the percentage of mature CD25+Foxp3+ Treg cells within CD4SP thymocytes (Fig. 5a,b).

Figure 5

Antibody neutralization of TNFSF members in thymic organ cultures inhibits Treg development and the acquisition of maturation markers

Neonatal Foxp3-GFP thymic lobes from one day-old pups were separated and plated in organ cultures with isotype controls or neutralizing antibodies to GITR-L, OX40-L, and CD70, or the combination of anti-GITR-L, OX40-L, CD70, and TNFR2. After 14 days, lobes were dissociated and Treg development was analyzed by flow cytometry. (a) A representative comparison of CD4SP thymocytes from isotype- vs. antibody-treated thymic lobes is shown. (b) Shown are the percentages of CD25+Foxp3+ Treg cells amongst total CD4SP thymocytes (mean ± SEM, n=21 over four experiments for isotype controls, n=16 over three experiments for anti-GITR-L, OX40-L, and TNF; n=5 over two experiments for anti-GITR-L, OX40-L, CD70, and TNFR2; * p < 0.05 calculated by ANOVA with Bonferroni’s multiple comparison test). (c,d) The expression of CD25, Foxp3, CD73, and FR4 expression in CD4+Foxp3+ cells from TOCs treated with isotype control antibodies- (solid gray) or TNFSF neutralizing Abs treated (open black) is displayed in the histograms and corresponding bar graphs below (n=16 from three independent experiments, mean ± SEM). P-values were generated using paired, two-tailed T tests ( * P < 0.0001).

We also evaluated the acquisition of Treg maturation markers by Foxp3+ cells from TOCs with or without neutralizing antibodies to GITR-L, OX40-L and CD70 and found that inhibition of this costimulatory pathway markedly reduced not only the efficiency of Treg generation, but also the expression of CD25 and Foxp3 (Fig. 5c,d). We also stained for the extracellular AMP nucleotidase, CD73, and the folate receptor, FR4, both of which mark highly functional Treg cells[32, 33], and found significant reductions in the expression of both in Foxp3+ cells when costimulation via GITR-L, OX40-L, and CD70 were blocked (Fig. 5c,d). Although the numbers of Treg cells recovered were lower in cultures treated with neutralizing antibodies to GITR-L, OX40-L, CD70 and TNFR2 the same effects on Treg maturation markers were observed. Thus, neutralization of select TNFSF members inhibits thymic Treg development and maturation.

TNFRSF members collectively drive Treg development

We next sought to determine whether inhibiting multiple TNFRSF members blocked Treg development in vivo in a cell-intrinsic manner. For these studies we focused on OX40, GITR, and TNFR2 as their expression correlated with TCR signal strength while CD27 did not (Fig. 2b). One difficulty with these studies is that the genes encoding GITR and OX40 are spaced a mere ~10 kb apart on mouse Chromosome 4. Tnfr2 is encoded nearby on Chromosome 4 as well. This overall pattern is the same in humans, except that GITR, OX40, and TNFR2 are located on Chromosome 1. In mice, Tnfsf18 (encoding GITR-L) and Tnfsf4 (OX40-L) are also located next to each other on Chromosome 1. Because of this genomic organization, conventional double- or triple-knockout strategies to account for biologic redundancy in this pathway were not feasible. Instead, we exploited the fact that TNFRSF members must multimerize to function and generated tailless, dominant negative forms of GITR and TNFR2.

Dominant negative GITR (dnGITR) and dominant negative TNFR2 (dnTNFR2) were linked to IRES-GFP (pMIGR vector) and IRES-Thy1.1 (pMITR vector) reporter constructs and administered to Ox40 bone marrow (CD45.2) via retroviral transduction. Transduced cells were mixed with congenically marked wild type marrow (CD45.1 or CD45.1 × CD45.2), engrafted into sublethally irradiated Rag2 −/− recipients, and reconstituted for 10-12 weeks. This strategy produced chimeric mice in which 5 populations of bone marrow cells could compete to reconstitute the thymus: (1) wild type: CD45.1+, (2) OX40-deficient: CD45.2+, (3) OX40 plus GITR-deficient: CD45.2+GFP+, (4) OX40 plus TNFR2-deficient: CD45.2+Thy1.1+, and (5) OX40 plus GITR plus TNFR2-triple deficient (CD45.2+GFP+Thy1.1+) (Fig. 6a,b). We then plotted the relative percentages of conventional CD4SP, Treg progenitors, and Treg cells within each of the gates shown in Figure 6b corresponding to increased expression of dnGITR, dnTNFR2, or both. Treg progenitors were not significantly affected by reduced GITR, OX40, and TNFR2 signaling (Supplementary Fig. 3). In contrast, superimposition of GITR- or TNFR2-deficiency in Ox40−/− cells led to a striking dose-dependent abrogation of Treg development of >20 fold (Fig 6c,d). Importantly, we noted a further significant decrease in Treg cells derived from cells deficient in all three TNF receptors. In fact, Ox40−/− cells expressing the highest levels of both dnGITR and dnTNFR2 completely lacked mature thymic Treg cells (Fig. 6c,d, and for a summary of all statistical comparisons, see Supplementary Table 1). Thus, costimulation via GITR, OX40, and TNFR2 is collectively required for Treg differentiation in vivo.

Figure 6

GITR, OX40, and TNFR2 redundantly drive Treg development in vivo

(a) Bone marrow from 8-12 week old male or female CD45.2+ Ox40 −/− mice was harvested and prestimulated with IL-3, IL-6, and SCF prior to retroviral transduction with vectors encoding dominant negative GITR (dnGITR) and TNFR2 (dnTNFR2). Cells were mixed with congenically marked CD45.1+ control marrow and engrafted into adult sublethally irradiated Rag-deficient hosts for 10-12 weeks prior to analysis by flow cytometry. Representative flow cytometry plots derived by gating on CD45.1 WT or CD45.2 (Ox40−/−) cells which were negative for GFP and Thy1.1 (untransduced) show the percentages of CD25−Foxp3− conventional CD4SP cells, CD25+Foxp3− Treg progenitors, and CD25+Foxp3+ Treg cells amongst CD4SP thymocytes. (b) To evaluate dose-dependent effects of inhibiting multiple TNFRSF on Treg development, Ox40−/− (CD45.2+) thymocytes were plotted for their expression of dnGITR (GFP) and dnTNFR2 (Thy1.1) and gates were drawn on cells expressing increasing levels of GFP, Thy1.1, or both. Gates are labeled 1 through 4 and colored in blue, green, orange, and red, respectively. (c) Cells derived from gates in panel (b), representing increasing expression of dnGITR, dnTNFR2, or both, were evaluated for CD25 and Foxp3 expression to determine the frequencies of conventional CD4SP, Treg progenitors, and mature Treg cells within each gate (d) Cumulative data from one of three independent experiments are shown in the scatter plot below (mean ± SEM, n=6, * P < 0.05 as determined by 1-way ANOVA using Bonferroni’s multiple comparison test). For a summary table of all statistical comparisons see Supplementary Table 1.

TNFRSF costimulation shapes Treg repertoire selection

The observation that TCR signal strength correlates with the expression of GITR, OX40, and TNFR2 in Treg progenitors suggests that these receptors may play an important role in biasing the Treg repertoire toward higher-affinity. To test this hypothesis, we generated Foxp3-RFP × Nur77-GFP reporter mice. We then sorted Treg progenitors and stimulated them with IL-2 and increasing doses of agonist GITR-L–Fc. Increasing GITR-L concentration led to the development of a Treg population with reduced expression of Nur77-GFP (Fig. 7a). This was not due to a loss of Treg cells with high TCR signal strength, but rather inclusion of cells with lower Nur77-GFP expression. We found a clear negative correlation between GITR-L concentration and TCR signal strength within newly developed Treg cells (r 2 = 0.81, Fig. 7b). These results demonstrate that altering the abundance of a TNFSF ligand directly affects selection of the Treg repertoire.

Figure 7

Excess TNFSF ligand broadens the Treg repertoire to contain a greater fraction of cells with a lower affinity for self

(a) Treg progenitors from 6-8 week old female or male Foxp3-RFP × Nur77-GFP dual reporter mice were isolated by cell sorting and stimulated in culture for 72h with 1 U/ml IL-2 and increasing concentrations of agonist GITR-L–Fc. Shown are stacked histograms displaying Nur77-GFP levels amongst converted Foxp3-RFP+ Treg cells. (b) Cumulative data from three independent experiments (n = 3) show a dose-dependent relationship between GITR-L–Fc concentration (x-axis) and the Nur77-GFP MFI (y-axis) within converted Foxp3-RFP+ Treg cells (as normalized to the value obtained with IL-2 treatment alone; mean ± SEM, n=3).

A CD25−Foxp3lo alternative Treg progenitor population has also been described[34]. A key question is whether this putative Treg progenitor is also affected by TNFRSF costimulation. We found that CD25−Foxp3lo Treg progenitors also express GITR and OX40 in direct proportion to TCR signal strength (Supplementary Fig. 4a-d). Stimulation of sorted CD25−Foxp3lo Treg progenitors in the presence of low dose IL-2 plus GITR-L–Fc dose-dependently enhanced maturation to CD25+Foxp3+ Treg cells (Supplementary Fig. 4e). Furthermore, as GITR-L concentration was increased, the average Nur77-GFP expression decreased in cells that differentiated into CD25+Foxp3+ Treg cells (Supplementary Fig. 4f). Thus, TNFRSF costimulation plays a critical role in the differentiation of all potential Treg progenitors into mature Treg cells, and competition for TNFSF ligands regulates TCR repertoire selection during Treg development.

Discussion

Treg development is a finely tuned process that is critical for preventing autoimmune disease. An intriguing feature of Treg development is that this process discriminates between relative differences in TCR signal strength. Treg cells with TCRs that recognize self-ligands with high-affinity predominate, although Treg cells with lower-affinity TCRs can be found[35, 36]. The molecular program that accounts for this skewing process remains to be fully elucidated. One recent mechanism that has been proposed is TCR-induced expression of the NR4A family of transcription factors including NUR77, NOR1, and NURR1[37]. Expression of these transcription factors correlates with TCR signal strength[24] and these three transcription factors bind the Foxp3 gene and are redundantly required for Treg development[37]. This suggests that one potential mechanism to link TCR signal strength to Treg development involves this transcription factor family. However, it is unlikely that this is the only mechanism that links TCR signal strength to Treg development. For example, Treg progenitors have clearly induced expression of NUR77 as documented by Nur77-GFP reporter mice yet they do not express Foxp3. In addition, in Bim−/− mice, many CD4SP thymocytes that have escaped deletion have high expression of NUR77 yet they do not differentiate into Foxp3+ Treg cells[38]. TCR signals give rise to graded levels of NUR77 expression in thymocytes, yet Foxp3 is expressed in relative quantum increments with essentially no expression in Treg progenitors and high levels in mature Treg cells. If the effect of NUR77 family members on Foxp3 transcription was truly linear then one might expect a continuum of Foxp3 expression in Treg progenitors culminating in that seen in mature Treg cells; that is not observed experimentally suggesting that other mechanisms must also contribute to the process of linking TCR signal strength to Treg differentiation.

We propose that the TNFRSF members GITR, OX40 and TNFR2 also play a critical role in converting strong TCR signals into molecular pathways that drive Treg differentiation. In support of this hypothesis we observed that Treg progenitors express much greater amounts of GITR, OX40, and TNFR2 relative to conventional CD4+CD25−Foxp3− thymocytes and that the level of expression of these receptors correlates directly with TCR signal strength as read out using Nur77-GFP reporter mice. This results in a competitive advantage as mature Foxp3+ Treg cells express even greater amounts of GITR and OX40. Blocking signaling via these receptors either by neutralizing antibodies or through combinations of gene knockouts and dominant negative receptors completely abrogates Treg development. Thus, signaling through these receptors is collectively required for thymic Treg development in vivo. Finally, the relative abundance of TNFSF ligands directly affects the TCR repertoire in developing Treg cells. Treg progenitors expressing the highest affinity TCRs, and hence the highest amounts of GITR, OX40, and TNFR2 can more effectively compete for their respective TNFSF ligands, and are thereby more likely to differentiate into mature Treg cells.

The mechanism by which GITR, OX40 and TNFR2 promote Treg development appears to primarily have effects on cell differentiation and not survival. GITR-L and OX40-L costimulation resulted in an ~10-fold increase in sensitivity to low doses of IL-2 that likely mimic what is available in vivo. This most likely reflects a direct effect on IL-2R signaling, as downstream STAT5 phosphorylation was also enhanced by costimulation with TNFSF agonists. Since GITR, OX40, and TNFR2 are known to signal via the NFκB pathway, we examined the expression of two NFκB-regulated genes that affect IL-2R signaling. We observed increased expression of CD25 on both Treg progenitors and mature Treg cells following costimulation with GITR-L, OX40-L, or TNF. In contrast, Socs1 expression (which is induced by IL-2 signaling, and negatively regulated by NFκB via mIR-155[39]) was unaffected by GITR-L costimulation of Treg progenitors (data not shown). When evaluating the effects of costimulation on mature Treg cells, however, we observed that GITR-L reduced the Socs1 expression induced by IL-2 stimulation by approximately 50% (data not shown). Thus, the predominant effect of GITR-L costimulation on enhancing Treg progenitor sensitivity to IL-2 appears to be via CD25 induction.

An additional mechanism by which GITR, OX40, and TNFR2 could influence Treg development is through the induction of anti-apoptotic genes[19, 21, 40, 41]. Such a mechanism has been proposed to account for the modest effect of CD27-CD70 on Treg development[25]. We did observe that GITR, OX40, and TNFR2 costimulation weakly induced Bcl2 protein in Treg progenitors (data not shown). However, blocking apoptosis with caspase inhibitors did not mimic the effect of GITR-L costimulation. Preliminary results with mice overexpressing Bcl2 indicate that ectopic anti-apoptotic gene expression does not prevent the defect in Treg development imposed by TNFRSF blockade (data not shown). Thus, unlike CD27, which promotes cell survival, the induction of anti-apoptotic gene expression by GITR, OX40, or TNFR2 costimulation is not the dominant mechanism by which these TNFRSF drive Treg differentiation.

It is unclear why three distinct TNFRSF members are required for Treg differentiation. One possible explanation is that partial redundancy limits the chances of defects in Treg development. A similar biological redundancy is observed for the NR4A transcription factor family in which neither NUR77, NOR1, or NURR1 is individually critical, but loss of all three stringently prevents Treg development[37]. Alternatively, it is possible that selection for multiple TNFRSF members on mature Treg cells is important for their later function. For example, ligation of OX40 or GITR on mature Treg cells has been shown to limit their suppressor function[18, 20]. In this regard, enhanced selection of Treg progenitors that express GITR and OX40 in the thymus could function as a possible safeguard to shut down their suppressive function during inflammatory responses to pathogens. Other reports have suggested that expression of TNFRSF members is required to maintain survival of mature Treg cells during inflammatory conditions[42, 43]. Finally, it is possible that higher expression of GITR and OX40 allows Treg cells that most avidly recognize self-ligand to preferentially compete for IL-2 needed to sustain these cells in peripheral lymphoid organs.

In conclusion, our data support a model of Treg differentiation in which developing thymocytes that bind MHC-II:self-peptide ligands with higher-affinity express greater amounts of GITR, OX40 and TNFR2. This elevated expression of select TNFRSF members on highly self-reactive Treg progenitors results in a selective advantage that allows those cells to preferentially undergo maturation. Thus, upregulation of GITR, OX40, and TNFR2 by Treg progenitors is a mechanism by which high-affinity TCR signals are translated into physical parameters that allow for more effective competition for IL-2 in a tightly regulated developmental niche. In agreement with previous reports[36], our model still allows for cells bearing TCRs with lower affinity for MHC-II:self-peptide complexes to emerge in the thymic Treg population, but strongly favors Treg differentiation in thymocytes receiving strong TCR signals since these cells most strongly upregulate TNFRSF receptors. This precisely matches the observed TCR affinity distribution in mature Treg cells[24]. Thus, TNFSF ligand availability is an additional level of control by which thymic APCs define the Treg developmental niche and couple TCR affinity to Treg development.

Online Methods

Mice

All mice were housed in specific-pathogen free facilities at the University of Minnesota and experiments were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee. Gitr −/− mice were previously described[19] and provided by Drs. Carlo Riccardi and Ethan Shevach. Ox40 −/−, Foxp3-GFP, Foxp3-RFP, Rag2 −/−, and Cd4Cre mice were purchased from the Jackson Laboratory (Bar Harbor, ME, stock numbers 012839, 006772, 008374, 008449, and 013234 respectively). CD45.1 (B6.SJL) mice were purchased from the NCI (Bethesda, MD). Nur77-GFP BAC reporter mice[24], Tak1 and Bcl2-Tg mice were previously described[44, 45]. Cd28−/− and Cd28 knock-in mice were previously described[46]. All mice used were on the C57Bl/6 background, except for Cd28−/− and Cd28AYAA mice and their wild type littermates, which were on a BALB/c background. Mice were randomly selected for experiments from our mouse facility in age-matched cohorts. The investigators were not blinded to genotype during data acquisition.

Tissue preparation and cell isolation

For analysis of thymocyte and Treg development thymii and peripheral lymphoid organs were mechanically dissociated into PBS with 2% FBS and 2 mM EDTA, pH 7.4, using frosted glass slides. Cell suspensions were passed through 70 μm filters and washed prior to staining. For analysis of thymic DC and epithelial cells (TEC), thymii were dissected into 5 mL complete RPMI supplemented with Collagenase D (120 Mandl U/mL) and DNase I (3 U/ml, both from Roche Applied Sciences, Indianapolis, IN) and diced into ~20 pieces and placed in a 37°C incubator for 30 minutes. Pieces were further homogenized by pipetting and the addition of EDTA (20 mM final concentration) which was immediately diluted 10-fold with buffer and the suspension filtered through 70 μm mesh. Thymic DC were magnetically enriched by labeling with anti-CD11C FITC (N418, eBioscience, San Diego, CA) and secondarily with anti-FITC microbeads (Miltenyi Biotec, Auburn, CA). Thymic epithelial cells were enriched by magnetic depletion of leukocytes using biotinylated anti-CD45 antibody (30F11, eBioscience) and streptavidin-conjugated microbeads (Miltenyi Biotec).

Flow cytometry and antibodies

All analysis was conducted in the Center for Immunology Flow Cytometry Core Facility using BD LSR II and Fortessa cytometers (BD, San Jose, CA). For surface staining, cells were stained for 20 minutes with 1:100 dilutions of fluorochrome-conjugated antibodies prior to washing and analysis or intracellular staining. Anti-mouse CD4 (GK1.5 or RM4-4), CD8 (53-6.7), CD25 (PC61.5), CD122 (TMb1), Foxp3 (FJK-16s), GITR (DTA-1), OX40 (OX86), TNFR2 (TR75-32), CD27 (LG.7F9), 4-1BB (17B5), GITR-L (YGL386), OX40-L (RML134L), CD70 (FR70), CD45 (30-F11), MHC Class II (I-A/I-E, M5/114.15.2), BP1 (6C3), Thy1.2 (30-H12 or 53.21), CD11C (N418), FR4 (eBio 12A5), CD73 (TY/11.8), CD103 (2E7), KLRG1 (2F1), and CD45R/B220 (RA3-6B2) were from eBioscience (San Diego, CA). Fluorescein labeled Ulex Europaeus Agglutinin I (UEA I) was purchased from Vector Laboratories (Burlingame, CA). FITC Anti-GFP was from Rockland Immunochemicals (Gilbertsville, PA). Intracellular detection of Foxp3 and P-STAT5 was performed as previously described[9].

Immunofluorescence microscopy

Tissues were frozen in 2-methylbutane surrounded by dry ice. Frozen blocks were cut into 7 μm thick sections and fixed in acetone. Nonspecific binding was blocked for 1 hour with 5% bovine serum albumin in PBS. Sections were then stained with hamster anti-CD11C conjugated-FITC; rabbit A488 conjugated anti-K5; UEA-1 conjugated-FITC; goat anti-OX40-L; rat anti-TNF PE (eBioscience); or rat anti-GITR-L PE (eBioscience). Signal from PE-conjugated anti-TNF and anti-GITR-L or irrelevant isotype controls was amplified with rabbit anti-PE (NB120-7011; Novus Biologicals) followed by staining with donkey anti-Rabbit Cy3 (Jackson Immunoresearch). Jackson Immunoresearch secondary antibodies conjugated to various fluorochromes were used for staining of unconjugated antibodies including Cy3–conjugated donkey anti-rat, Alexa-488-conjugated donkey anti-rabbit, and Cy3-conjugated donkey anti-goat. Sections were incubated for 1 hour in the dark at room temperature, and then washed with PBS 3 times prior to the next stain. After the final wash, slides were stained with DAPI (4,6-diamidino-2-phenylindole; Invitrogen) for 10 minutes before Fluoromount G was added with a cover slip. Tiled images were acquired with an automated Leica DM5500B microscope and analysis was done with Adobe Photoshop.

Treg progenitor conversion assays

Treg progenitors were isolated as previously described[9]. Briefly, 6-8 week old Foxp3-GFP thymii were dissected, dissociated, and pooled (up to 8 mice per experiment). CD4SP cells were enriched by magnetically depleting with biotinylated anti-CD8, CD11B, CD11C, CD19, CD45R/B220, MHC class II, NK1.1, and Ter119 antibodies (eBioscience) followed by secondary labeling with streptavidin-conjugated microbeads (Miltenyi Biotec). Enriched CD4SP cells were stained with fluorchrome-conjugated anti-CD4, CD25, and streptavidin prior to sorting CD4+CD25+GFP− cells using a BD Facs Aria sorter (BD Biosciences). Purified Treg progenitors were incubated in complete RPMI and supplemented with human IL-2 (1 Unit/mL, from the NIH repository at Frederick National Laboratory) with or without 100 nM recombinant GITR-L–Fc, OX40-L–Fc (both from Imgenex, San Diego, CA), recombinant murine TNF (Peprotech, Rocky Hill, NJ), recombinant CD70-Fc (Sino Biologicals, Beijing, China), soluble CD70 (Enzo Life Sciences, Farmingdale, NY), agonist CD27 mAb (RM27-3E5)[47], or recombinant mouse CD30-L and 4-1BBL (the latter two of which were crosslinked using 5 ug/ml anti-poly His antibody, both from R&D Systems, Minneapolis, MN). After 72 hours, cells were harvested, stained with anti-CD4 and CD25 antibodies and analyzed by flow cytometry for the percentage of cells that express GFP after incubation.

Thymic organ cultures

Neonatal thymic organ cultures were performed as previously described[31]. Briefly, thymic lobes were harvested from 1-day-old Foxp3-GFP mice and set up in pairwise comparisons using isotype control or anti-TNFSF blocking antibodies. Neutralizing anti-OX40-L (RM134L) and anti-CD70 (FR70) were obtained from Bio X Cell (West Lebanon, NH). Neutralizing anti-GITR-L (Clone 337122) was from R&D Systems (Minneapolis, MN). Neutralizing anti-TNFR2 (Clone TR75-54) was previously described[48]. Thymic lobes were bisected and placed atop Millipore Cellulose Ester Gridded 0.45 μm filters (CAT HAWG01300) sitting on Gelfoam absorbable gelatin sponges (Pfizer, NY, NY). These filters were placed in 6 well plates with 2.1 mL of complete RPMI per well supplemented with 20 μg/mL of the indicated antibodies, and up to 6 bisected lobes placed on each filter. Medium and antibodies were replenished on day 7 and bisected lobes were collected and stained with primary antibodies for 20 minutes, washed, and analyzed by flow cytometry on day 14.

Mixed bone marrow chimeras

Rag2−/− recipient mice were sublethally irradiated 24 hours prior to engraftment using an X-ray irradiator (Radsource RS2000 Biological Irradiator, Radsource, Suwanee, GA) and administering two doses of 250 rads separated by a two-hour rest. Bone marrow was harvested from tibias and femurs of donor mice, dissociated into single cell suspensions, and magnetically depleted of contaminating mature T and B cells by labeling with biotinylated anti-CD3, CD4, CD8, CD19, and CD45R antibodies (eBioscience) and secondarily with streptavidin-conjugated microbeads (Miltenyi Biotec) prior to passage through magnetic columns. Cell suspensions were washed in PBS and injected intravenously via the tail vein. Chimeric recipients were engrafted for 10-18 weeks prior to analysis.

Preparation of dominant negative retroviral vectors

Dominant negative retroviral vectors were cloned into pMIGR[49] or a modified version of this vector in which Thy1.1 replaces GFP, hereafter called pMITR. Sequences encoding truncated TNFRSF members were prepared by amplifying the extracellular and transmembrane domains of GITR and TNFR2 from a murine cDNA library and inserting stop codons immediately after the transmembrane domain. Specific primers utilized were as follows: for GITR dominant negative (Forward CTGCTCGAGCTGACCATGGGGGCATGGGCCAT, Reverse CTGGCGGCCGCCTGTCACAGGTCCTCCTCTGAGATCAGCTTCTGCATTGATGCCATTTGCCTCCTCAGCTGCCATATG), and for TNFR2 dominant negative (Forward CTGCTCGAGCTGACCATGGCGCCCGCCGC, Reverse CTGGCGGCCGCCTGTCATTTGTCATCGTCATCCTTGTAGTCGGAGGGCTTCTTTTTCCTCTGCACC). Linear products were trimmed with XhoI and NotI and ligated into pMIGR or pMITR prior to transformation of chemically competent E. coli DH5α cells and plating on carbenicillin. LB starter cultures were prepared, digested, and picked for large culture. Mini- and maxipreps were performed using PureLink® kits according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA). Direct-Sanger sequencing was performed by the University of Minnesota Genomics Core facility and base calling was done by hand.

Bone marrow retroviral transduction

Subconfluent 293FT cells (ATCC, Manassas, VA) were transfected using Effectene reagent (Qiagen, Valencia, CA) with pMIGR-dominant negative GITR (dnGITR) or pMITR-dominant negative TNFRF2 (dnTNFR2) constructs in parallel with pHit60 and pHit123 packaging vectors for retrovirus production[50]. After 16 hours, sodium butyrate was added to the culture medium to a final concentration of 10 mM to enhance viral production. Eight hours later, medium was replaced to begin collecting retroviral particles. Supernatants were collected and medium replaced at 24 and 48 hours. After filtration through 0.45 μm filters, viral particles were centrifuged onto Retronectin-coated 6 well plates (Clontech Laboratories, Mountain View, CA) prior to overlaying Ox40−/− bone marrow cells at a density of 3 million cells per well. Bone marrows used for transduction had been isolated the previous day, depleted of T and B cells, and stimulated with 10 ng/mL IL-3 and IL-6, and 50 ng/ml SCF (Peprotech) to induce cell division. Medium and cytokines were replenished at the first and second retroviral transductions. Twenty-four hours after the second transduction, bone marrow was collected and Ox40−/− cells were mixed 80%/20% with congenically marked wild type bone marrow that had been stimulated identically in culture with cytokines. The mixture was washed in PBS and intravenously injected into sublethally irradiated Rag1−/− recipients and thymic Treg was analyzed 10-12 weeks later. Bone marrow chimeric recipient mice that died due to a failure of engraftment were excluded from analysis.

Statistical analysis

P-values were generated using two-tailed T-tests; paired T-tests were used in the case of mixed bone marrow chimeras. Multiple sample comparisons were analyzed by one-way ANOVA. P-values in figure 2 were calculated based on Pearson correlation coefficients. All values were calculated using Prism (GraphPad Software, LaJolla, CA). Sample sizes were selected to ensure a power of ≥ 80% for the detection of two-fold differences in figure 2d, 3b, 3c, 3g, 5a, and 6d. Sample sizes were selected to ensure a power of ≥ 80% for the detection of ≥ 20% differences in figure 3d, 4a, 4b, and 5d.