Gene Expression in the Gitr Locus Is Regulated by NF-κB and Foxp3 through an Enhancer

Glucocorticoid-induced TNFR (Gitr; Tnfrsf18) and Ox40 (CD134, Tnfrsf4) are members of the TNFR superfamily (1) and are expressed on T cells. Expression of both receptors is upregulated by activation and, in particular, kept constitutively high on CD4⁺CD25⁺Foxp3⁺ regulatory T cells (Treg) (2–4), cells essential to the prevention of autoimmune disease and other forms of immunopathology (5–7). The development and function of Treg are regulated by the transcription factor Foxp3, which represses many genes by associating with other transcription factors, including NFAT (8), the p65 subunit of NF-κB (9), and Runx1 (10). However, a number of other genes such as Ctla4, Il35, Cd25, Ox40, and Gitr are upregulated in Foxp3⁺ Treg (2, 11), with all of these gene products playing essential roles in Treg function (12). Signaling through Gitr and Ox40 has been shown to neutralize the suppressive effects of Treg in vitro (3, 13). Additionally, Ox40 signaling inhibits induced Treg (iTreg) development in periphery by blocking TGF-β/TCR signal–mediated induction of Foxp3 (14, 15). The Gitr and Ox40 genes are particularly interesting as models for Foxp3-mediated transcriptional upregulation, because they are located within a short 15.1-kb stretch of the mouse genome, suggesting control by a common regulatory region. We have previously shown a similar clustering of MS4A gene family members, also upregulated by Foxp3 (16), where we postulated the existence of a comparable Foxp3-associated regulatory region at that gene locus. We selected the Gitr/Ox40 rather than the MS4A gene locus for study because the latter is far larger, extending over 600-kb, and is therefore more challenging for analysis.

Not only are Gitr and Ox40 influential in Treg, but they also act as costimulatory receptors for effector T cells (Teff) (1, 17–19). Gitr and Ox40 are unique molecules playing seemingly different roles in Teff and Treg, with the distinct functions possibly determined by expression levels of these receptors on each T cell subset (low on Teff and high on Treg). In either case, the mechanisms underlying gene expression remain poorly understood, although we know that Ox40 expression is regulated by constitutively active transcription factors and upregulated by NF-κB in activated Teff (20). To define the molecular mechanisms responsible for upregulation of Gitr and Ox40 in more detail, in this study we sought clues within the Gitr/Ox40 gene locus in both Teff and Treg.

We identify an enhancer located downstream of the Gitr gene. Histone H4 molecules in this region are highly acetylated both in activated T cells and Foxp3⁺ Treg. We show that enhancer activity is regulated by NF-κB in activated Teff, and by NF-κB in conjunction with Foxp3 in Treg. We propose that Foxp3 stabilizes binding of the p50 subunit of NF-κB to the enhancer. Although p50 does not contain a transactivation domain, the p50/Foxp3 complex on the enhancer interacts with other members of the NF-κB and IκB family (e.g., p65, c-Rel, B cell lymphoma 3 [Bcl-3]) to supply transactivation domains to the complex. This may explain how Foxp3 can suppress expression of many genes while also upregulating others.

EL4 subclones LAF and BO2 cells were cultured in IMDM with l-glutamine and 25 mM HEPES (Cellgro) and 5% FBS. T cells were cultured in RPMI 1640 with l-glutamine (Cellgro) and 10% FBS. CD4⁺ T cells and T cells were isolated from mouse spleen using a CD4⁺ T cell isolation kit (Miltenyi Biotec) and a Pan T cell isolation kit (Miltenyi Biotec), respectively. CD4⁺CD25⁺ Treg cells were purified by a cell sorter using anti-CD4 (RM4-5, eBioscience) and anti-CD25 (PC61.5, eBioscience) from preisolated CD4⁺ T cells, isolated CD4⁺CD25⁺ T cells were analyzed by an internal Foxp3 staining kit (eBioscience), and Foxp3⁺ T cells were >95%. When required, T cells were stimulated with plate-coated anti-CD3 (KT3, 5 μg/ml in PBS). To generate iTreg, CD4⁺CD25⁻ T cells were isolated from wild-type (WT) and Foxp3-GFP reporter mice (The Jackson Laboratory) and stimulated with TGF-β (5 ng/ml) (PeproTech), anti-CD28 (1 μg/ml), and anti-CD3 (plate-coated). To isolate Foxp3⁺ iTreg, the stimulated cells were sorted with GFP. Splenocytes and thymocytes were isolated from Nfκb p50-deficient mice (The Jackson Laboratory) and WT mice.

The preparation of bone marrow–derived dendritic cells has been described previously (21). Briefly, bone marrow cells from BALB/c mice were cultured for 7 d with GM-CSF (5 ng/ml). When required, bone marrow–derived dendritic cells were stimulated with LPS (10 μg/ml).

EL4 and CD4⁺ T cells were cultured with NF-κB activation inhibitor (NAI; EMD Biosciences) in DMSO. These cells were pretreated with NAI for 30 min before stimulation and then cultured with anti-CD3–coated plates. Ox40 and Gitr expression was analyzed by quantitative RT-PCR (EL4), and by FACS (CD4⁺ T cells, gated by 7-aminoactinomycin D− and CD4⁺) with anti-Ox40 (OX-86, eBioscience) and anti-Gitr (DTA-1, eBioscience).

Total RNA was isolated from EL4 cells and T cells using an RNeasy Mini kit (Qiagen), and cDNA was prepared with an iScript cDNA synthesis kit containing random and oligo(dT) primer mixture (Bio-Rad). Quantitative RT-PCR was performed by using SsoFast EvaGreen supermix (Bio-Rad). PCR primers used were as follows: Gitr, forward, 5′-GACCCTCAGT GCAAGATCTGC-3′, reverse, 5′-CCTCAGCTGACAACTGCACCTC-3′; Ox40, forward, 5′-GTAGACCAGGCACCCAACC-3′, reverse, 5′-GGCCAGACTGTGGTGGATTGG-3′; Gapdh, forward, 5′-TGGTGAAGGTCGGTGTGAACGGATTT-3′, reverse, 5′-TGTGCC GTTGAATTTGCCGTGAG-3′; 18S rRNA, forward, 5′-CTTAGAGGG ACAAGTGGCG-3′, reverse, 5′-ACGCTGAGCCAGTCAGTGTA-3′. Gitr, Ox40, and Gapdh expression levels were normalized to 18S rRNA levels.

A DNA fragment (1.22 kb) of the Gitr promoter and its deletion mutants were amplified by PCR, cloned into the pGL4 basic vector (Promega), and DNA sequences of the all inserted fragments were determined to remove deformed fragments generated by PCR errors. To construct enhancer luciferase reporter plasmids, Pro8 Gitr promoter luciferase reporter plasmid was used. Potential enhancer DNA fragments (1.6-kb fragment [containing HS1 site], 1.15-kb fragment [containing HS2 site], and 0.28-kb fragment [containing HS3 site]) were PCR amplified and inserted into SalI site located downstream of the luciferase gene. To construct Ox40 promoter-enhancer plasmid, the enhancer core fragment was inserted into the SalI site located downstream of the luciferase gene of the Ox40 promoter reporter plasmid (carrying 1.97-kb DNA fragment as Ox40 promoter) shown in a previous publication (20). The enhancer deletion mutant fragments were PCR amplified and integrated in to the SalI site of the Gitr Pro8 reporter plasmid. The +39, +136, +183, and WT core enhancer fragments contain the same 3′ end (+286), and +71, +108, +193, and WT core enhancer fragments contain the same 5′ end (+1). These positions are shown in Supplemental Fig. 3A. Mutations in the enhancer sequence (Fig. 3A) were introduced into the enhancer fragment by PCR assemble procedure, and the mutated fragments were integrated in to the SalI site in the Gitr Pro8 promoter reporter plasmid. DNA sequences of the all inserted fragments were determined to remove defective fragments generated by PCR errors.

For the luciferase assay, 5 × 10⁶ EL4 LAF (Figs. 2, 3) and EL4 BO2 (Fig. 8) cells were transfected by Gene Pulser Xcell (Bio-Rad) with 5 μg luciferase reporter plasmids and 2 μg phRL-TK (Promega) as an internal control plasmid and cultured for 24 h in six-well plates. When required, cells were activated with plate-coated anti-CD3 (5 μg/ml in PBS). Cells were harvested, and promoter or enhancer activity was analyzed by the Dual-Luciferase reporter assay system (Promega). Foxp3, NF-κB p50, p65, and c-Rel expression plasmids were constructed using pMF-neo vector (expression is regulated by EF1α promoter). To remove the IκB homologous from p50, p50 cDNA encoding 1–423 aa was PCR amplified and cloned into pMF-neo vector. For cotransfection experiments, 4 μg Gitr promoter-enhancer plasmid was cotransfected with 4 μg each expression plasmid indicated in Fig. 8. Total DNA amount was adjusted with the empty pMF-neo vector. EL4 BO2 subclone was used for this assay.

Isolated nuclei were treated with DNase I (0–25 U/ml) at 25°C for 5 min in 60 mM KCl, 15 mM NaCl, 5 mM MgCl₂, 0.1 mM EGTA, 15 mM Tris-HCl (pH 7.4) 0.5 mM DTT, 5% glycerol, and 10% sucrose, and DNAs were isolated and digested with KpnI, SphI, EcoRI, XbaI, XhoI, or SalI. DNase I hypersensitive sites were analyzed by Southern blot hybridization, and probe positions (for KpnI digestion) are indicated in Supplemental Fig. 2.

EMSA was performed with ³²P-labeled probes and 2 μg nuclear extracts in 20 μl EMSA reaction buffer (2 μg poly(deoxyinosinic-deoxycytidylic) [poly(dI-dC)]⋅poly(dI-dC) acid, 20 mM HEPES [pH 7.9], 1 mM MgCl₂, 40 mM KCl, 0.1 mM EDTA, 1 mM DTT, and 12% glycerol). Nuclear extracts were prepared from nonactivated or CD3-activated (for 1.5 and 24 h) EL4 cells as described previously (20). To perform competition assays, 100-fold excess of unlabeled competitor oligonucleotides was added to EMSA reaction mixture. To perform supershift assay, nuclear extracts in EMSA reaction buffer were incubated for 15 min with anti-Sp1 (Santa Cruz Biotechnology, PEP2), anti-Sp3 (Santa Cruz Biotechnology, D-20), anti–NF-κB p50 (Santa Cruz Biotechnology, D-17), and anti–NF-κB p65 (Santa Cruz Biotechnology, C-20), and probes were added into the reaction mixture.

Chromatin immunoprecipitation (ChIP) assay was performed using Pan T cells, CD4⁺CD25^– T cells, TGF-β–induced iTreg (93% purity), and CD4⁺CD25⁺Foxp3⁺ nTreg cells (purity >95%) as described previously (20). These cells were fixed (for 10 min at room temperature in 1% formaldehyde, 4.5 mM HEPES [pH 8.0], 9 mM NaCl, 0.09 mM EDTA, and 0.045 mM EGTA) and sonicated (Bioruptor) in lysis buffer (1% SDS, 10 mM EDTA, and 50 mM Tris-HCl [pH 8.0]) with proteinase inhibitor (Sigma-Aldrich, P8340). Precleared lysates were incubated overnight at 4°C with polyclonal anti–acetyl histone H4 (Millipore), anti–NF-κB p50 (Abcam, anti–p105/p50-ChIP grade), or control rabbit IgG (Santa Cruz Biotechnology). DNA fragments were isolated from the immunoprecipitated chromatin and analyzed by real-time PCR with SsoFast EvaGreen Supermix (Bio-Rad). PCR primers for ChIP used were as follows: κB1/κB2, forward, 5′-TTACACTGGAAACACCACAGGTGG-3′, reverse, 5′-TGCTGGCTTCAAGGCAAGGATACA-3′; κB3, forward, 5′-TGCATTCCAC TCACGTCCAC-3′, reverse, 5′-GGGCACTGTCCCTCAGCTAC-3′.

A 101-bp DNA fragment containing the κB1 and κB2 sites was amplified by PCR using biotinylated primers and purified using QIAEX II gel extraction system (Qiagen). Purified probes (500 ng) were mixed with nuclear extract (50 μg) in 400 μl binding buffer (60 mM KCl, 12 mM HEPES [pH 7.9], 4 mM Tris-HCl [pH 8.0], 1 mM EDTA, 1 mM EGTA, and 12% glycerol) containing 1.66 mM DTT, 0.06% BSA, and 20 μg poly(dI-dC) acid at 4°C for 2 h. When required, 100 pmol Sp1 binding oligo from CD40 promoter (22) and 500 pmol indicated competitor oligonucleotides were added. Then, precleared streptavidin-agarose beads (Life Technologies) were mixed with the DNA–nuclear extract mixture for 2 h. The streptavidin-agarose beads were then washed five times with 1 ml binding buffer, and then 2× SDS sample buffer was added. The samples were analyzed by immunoblotting. PCR primers used were as follows: κB, forward, 5′-biotin-GAAACACC ACAGGTGGGACA-3′, reverse, 5′-CACACCCATCAGCCGCCCACA-3′; Gitr promoter, forward, 5′-biotin-TGGGAGAGGCATGTAGGGGTTAGA-3′, reverse, 5′-TTTC CGGCAGACATCTGAGGT-3′.

A nuclear extract was prepared from a p50 (1–423 aa)–Foxp3 (FLAG-tagged) transfectant. Nuclear extract (10 μg) was precleared with protein G (Invitrogen) for 3 h in 5% glycerol, 12 mM HEPES (pH 7.9), 4 mM Tris (pH 8.0) 60 mM KCl, 0.1 mM EDTA, 1 mM DTT, poly(dI-dC)⋅poly(dI-dC) acid 20 μg, and 4 μl proteinase inhibitor (Sigma-Aldrich, P8340) with or without 92-bp DNA (16–108, Fig. 3). Anti-p50 (2 μg; Santa Cruz Biotechnology, C-19) or control goat IgG was added and incubated at 4°C overnight, and protein G was added and incubated for 2 h at 4°C. Protein G was washed with the binding buffer and analyzed by immunoblotting.

Immunoblotting was performed with anti-p50 (Santa Cruz Biotechnology, C-19), anti-p65 (Santa Cruz Biotechnology, C-20), anti–c-Rel (Santa Cruz Biotechnology, C), anti-Bcl-3 (Acris, SP7024P), anti-Foxp3 (Cell Signaling Technology, D608R), and anti-FLAG (Sigma-Aldrich). Ab binding was detected by using Luminata Forte Western HRP substrate (Millipore).

Significance was determined with a Student t test.

Because CD4⁺CD25⁺Foxp3⁺ Treg are normally present in relatively low abundance, and because primary T cells are not ideal for transfection-based procedures such as luciferase reporter assays, we adopted the following strategy to identify regulatory regions in the Gitr/Ox40 locus in activated Teff and Treg. First, we identified regulatory regions such as promoters and enhancers in both activated T cells (using TCR cross-linking by anti-CD3) and the EL4 T cell line (which is known to retain many features of T cells), and we characterized the regulatory region in activated T cells. We then determined whether that regulatory region was functional in both TGF-β iTreg (23, 24) and nTreg and thereafter proceeded to investigate the molecular basis for the upregulation in Foxp3⁺ Treg. Because Foxp3 gene expression is maintained by transcription factors induced by TCR-CD3 signaling, these factors should translocate into the nuclei of Foxp3⁺ Treg. Knowing that Foxp3 can associate with these activated factors in Treg (8–10, 25), it was then conceivable that the Foxp3-responsive regulatory region coincides with those regions involved in CD3 activation.

Ox40 gene expression can be upregulated by CD3 activation in both T cells and EL4 cells (20), and as shown in Fig. 1A, Gitr gene expression is also upregulated in both cell types, suggesting that expression of these genes in primary T cells and the EL4 cell line is regulated by the same regulatory mechanisms. Regulation of Foxp3 gene expression in T cells and our EL4 subclone was also comparable (26). These data suggest that the EL4 T cell line can be used to probe Gitr and Ox40 gene expression and that this information is relevant to primary Teff and iTreg.

Using this scheme, we analyzed promoter activity in activated cells. We had previously identified an NF-κB binding site in the Ox40 promoter (20) that was involved in the response to CD3. We used the same approach in the present study to investigate the Gitr promoter by using a luciferase reporter assay with a series of deletion mutants (Pro1 to Pro8 in Fig. 1B). Gitr promoter activity was detected in both nonactivated and CD3-activated EL4 cells, but no CD3 activation response element was identified within the 1.22-kb Gitr promoter. Regardless of activation, promoter activity was reduced to basal levels by deleting 39 bp from −90 (Pro3) to −51 (Pro2). Binding of the transcription factor NFI, which is known to be constitutively and ubiquitously expressed (27), to this region was detected using the EMSA (Supplemental Fig. 1). Gitr basal promoter activity seems, therefore, to be constitutively regulated through this region, suggesting that this gene expression is upregulated by other regulatory regions containing CD3 activation response elements in activated T cells.

To identify the regulatory regions involved, DNase I hypersensitive assays were performed. Because DNase I hypersensitive sites (HSs) were generated by disruption of nucleosome chromatin structure by additional factor bindings, some of the HSs are observed in promoters and enhancers. Regulatory regions containing CD3 response elements might be located near CD3-activated T cell–specific HSs. We therefore compared positions of DNase I HSs in nonactivated and CD3-activated CD4⁺ T cells, EL4 cells, and bone marrow–derived dendritic cells (Fig. 1C, Supplemental Fig. 2). Many common HSs were detected in the Gitr/Ox40 gene locus (Fig. 1D). However, three additional sites were observed only in CD3-activated cells (Fig. 1C, 1D, Supplemental Fig. 2), which we term HS1, HS2, and HS3 (HS1 was detected only in CD3-activated CD4⁺ T cells).

We next examined enhancer activities in regions containing HS1, HS2, or HS3 using a luciferase reporter assay. As shown in Fig. 2A, DNA fragments containing these HSs were integrated downstream of the luciferase gene in the Gitr promoter reporter plasmid (Fig. 1B, Pro8). Promoter activity was enhanced with the fragment containing HS3 in both orientations (Fig. 2A, 2B) in CD3-activated cells but not in nonactivated cells. The enhancer activity was also analyzed using the Ox40 promoter (1.97 kb). Unlike the Gitr promoter, Ox40 promoter activity is itself increased 2-fold in CD3-activated cells (Fig. 2C, Pro) by NF-κB, as previously described (20), but this promoter activity was upregulated far more by this enhancer (Fig. 2C), suggesting strong enhancer activity. Enhancer activity was further analyzed by luciferase assays (Fig. 2A, 2D, 2E). Searching the transcription factor database, we found three potential NF-κB binding sequences in this enhancer region (referred to henceforth as κB1, κB2, and κB3) (Fig. 2D, 2E, gray boxes). Because the enhancer activity was only detected in CD3-activated cells but not resting T cells, we hypothesized that enhancer activity might be regulated by activated NF-κB through these sites. This possibility was investigated by using the 5′ (Fig. 2D) and 3′ (Fig. 2E) deletion mutants of the enhancer. Luciferase activity using the WT enhancer was reduced by deletions from +1 to +39 and from +39 to +136 (Fig. 2D). The deleted regions +1/+39 and +39/+136 contain κB1 and κB2 sites, respectively (Fig. 2D). Although the 5′ deletion mutant (+183) contained one potential NF-κB site (κB3), the enhancer activity was similar to that of the negative control (Fig. 2D, +183 and no enhancer). To investigate whether the κB3 site is simply not functional, or whether the κB sites cooperate to regulate enhancer activity, we performed luciferase assays using the 3′ deletion mutants (Fig. 2E). Luciferase activity was reduced by the deletion containing the κB3 site (Fig. 2E, +193), suggesting that this 3′ region containing κB3 cooperatively regulates enhancer activity as well. Such cooperative regulation was also deduced from the 3′ deletion mutant +71 containing κB1 but not κB2 and κB3 (Fig. 2E), as the enhancer activity of this deletion mutant was similar to the negative control. These results strongly suggest that the three NF-κB binding sequences, identified using the transcription factor database, are strong candidates as regulatory elements within this enhancer. We then analyzed histone H4 acetylation by ChIP as a marker for open chromatin in the κB1 plus κB2 (κB1 and κB2 cannot be analyzed separately by this assay because these two sites are too close to each other) and κB3 regions (Fig. 2F). H4 molecules in T cells became highly acetylated within 24 h of activation, suggesting that the chromatin of these regions opens up after CD3 activation. Histone H4 in these regions was also highly acetylated in nTreg but not in naive CD4⁺CD25⁻ T cells (Fig. 2G), suggesting that this regulatory region also functions in nTreg.

As shown in Fig. 2, the 286-bp enhancer sequence (Fig. 3A) encodes three potential NF-κB binding sequences. Transcription factor binding to these sites was analyzed by EMSA in Figs. 3B and 4. To assess the contribution of these potential NF-κB sites and to investigate the existence of other regulatory elements in the enhancer, luciferase reporter assays were performed using mutations in the potential NF-κB binding sequences in the Gitr promoter/enhancer reporter plasmid (Fig. 3A, 3C). No factor binding to the mutated sequences could be detected by EMSA using a nuclear extract from CD3-activated cells (Fig. 3B). Mutation of any of the potential NF-κB binding sites reduced enhancer activity, and mutating all three sites eliminated it (Fig. 3C), suggesting that all three sites regulate enhancer activity. The κB1 site was identified as the strongest regulatory element in this enhancer because the enhancer activity was reduced to <20% by mutation of the κB1 sequence. However, similar reductions were observed with double mutations at the κB2 and κB3 sites. These results indicate that enhancer activity is cooperatively regulated by the transcription factors binding to κB1, κB2, and κB3 sites.

Transcription factor binding to each κB site was detected by a simple EMSA using a nuclear extract from CD3-activated cells (Fig. 3B). Although all probes encode potential NF-κB binding sequences, binding patterns were not exactly the same. We therefore investigated transcription factor binding in more detail (Fig. 4). First, we performed EMSA using nuclear extracts from nonactivated and CD3-activated cells (Fig. 4A). We detected both constitutive and induced binding to the κB1 and κB2 probes, but only induced binding to the κB3 probe (Fig. 4A). As shown in Supplemental Fig. 3, the constitutively active transcription factors binding to κB1 and κB2 were Sp1 and Sp3, which bind to sequences overlapping the NF-κB binding sites. Binding of the same induced factors to all probes was suggested by inhibition of binding to the ³²P-labeled κB3 probe with excess of the “cold” κB1 and κB2 competitors (Fig. 4B). Because histone H4 acetylation in this enhancer region was induced by CD3 activation (Fig. 2F), binding of the induced transcription factors is likely to be the key to regulation. To identify binding of NF-κB to the κB1 and κB2 sites, excess cold competitor encoding an Sp1 binding sequence in the CD40 promoter (22) was added to block Sp1 and Sp3 binding to the labeled probe. We confirmed NF-κB p50 binding to these sites by supershift EMSA using anti–NF-κB p50. All complexes of transcription factors and the probes were supershifted with anti–NF-κB p50, and some of complexes were supershifted with anti–NF-κB p65, but unmoving complexes were still detected with this anti-p65 (Fig. 4C). This suggests that all complexes contain p50; in other words, all κB sites were bound by p50 homodimers, p50/p65 heterodimers, and/or p50 heterodimers with other transcription factors including other NF-κB members (as shown later, binding of c-Rel and Bcl-3 was detected by a DNA pull-down assay).

Any contribution of Sp1 and Sp3 to this enhancer activity is likely to be minimal, as they are present constitutively. However, it could be that they are bound in nonactivated cells but then replaced by NF-κB on activation, as described for Foxo1 and Foxp3 binding to Foxp3 gene (28).

As shown above, enhancer activity is regulated by NF-κB. Because the EMSA result (Fig. 4C) suggests that NF-κB p50 is a main component of the binding factors (p50 homodimer and heterodimer with other transcription factors), we further analyzed p50 binding in T cells by ChIP. p50 binding to the κB1 plus κB2 and κB3 regions was increased in T cells 24 h after CD3 activation (Fig. 5A). This information, together with previous evidence that NF-κB p50 and p65 bind to the Ox40 promoter (20), indicates that gene expression in the Gitr/Ox40 locus is predominantly regulated by NF-κB p50. To test this, we analyzed Gitr and Ox40 expression in CD3-activated CD4⁺ T cells isolated from the spleens of mice lacking the Nf-κb p50 gene (Fig. 5B). Gitr and Ox40 expression was upregulated by CD3 activation on CD4⁺ T cells from both mice, but the Gitr^high (36.8%) and Ox40^high (39.3%) cell populations from WT mice were less than half (Gitr^high, 15.2%; Ox40^high, 18.7%) of values observed from p50-deficient mice. This result suggests that both Gitr and Ox40 expression are regulated by NF-κB p50. This conclusion was not matched by the observation that a significant number of Gitr^high and Ox40^high cells could be detected in p50-deficient mice, despite NF-κB p50 preferably binding to these NF-κB sites in the enhancer (Fig. 4C) and in the Ox40 promoter (20). We propose that the lack of NF-κB p50 is compensated by binding of a very similar family member, NF-κB p52, as previously suggested using NF-κB p50-, p52-, and p50/p52-deficient mice (29, 30). Unfortunately, the p50/p52 double-deficient mice would not be suitable to test this hypothesis, as these mice have a major problem in bone development and exhibit a profound immunodeficiency (29, 30). To overcome this problem, we analyzed Gitr and Ox40 expression by using NAI. Expression of Gitr and Ox40 RNAs was strongly inhibited by NAI in CD3-activated EL4 cells in a dose-dependent manner (Fig. 5C, 5D), in contrast to the housekeeping gene control (Gapdh) (Fig. 5E). Gitr and Ox40 cell surface expression on CD3-activated T cells was also inhibited by NAI (Fig. 5F). These results indeed suggest that Gitr/Ox40 expression depends on NF-κB.

Upregulation of Gitr and Ox40 expression in Foxp3⁺ Treg has been previously demonstrated (2–4). The enhancer we have identified in the present study might be involved in this process. To assess whether this enhancer functions in iTreg and nTreg, we examined NF-κB p50 binding to this region by ChIP assay in these cells. p50 binding to the enhancer (κB1 plus κB2 and κB3 regions) was detected in nTreg, but not in naive CD4⁺CD25^– T cells (Fig. 6A). iTreg were generated using CD4⁺CD25⁻ T cells from Foxp3-GFP reporter mice by stimulation with TGF-β plus anti-CD3 plus anti-CD28; 37% cells were Foxp3⁺ iTreg after 48 h, and these were isolated (93% purity) using cell sorting and GFP fluorescence (Fig. 6B). p50 was bound to the enhancer in these Foxp3⁺ iTreg, but not in CD4⁺CD25⁻ T cells (Fig. 6B).

We also examined Gitr expression in CD4⁺Foxp3⁺ Treg isolated from p50-deficient mice. Because Ox40 promoter activity is regulated by NF-κB, and NF-κB p50 binding to the promoter is detectable by ChIP assay (20), we investigated only Gitr expression to avoid any confusion. Splenocytes and thymocytes from these mice had a larger proportion of CD4⁺Foxp3⁺ T cells with low Gitr expression than did those from control WT mice (Fig. 6C). This supports the argument that p50 is involved in upregulating Gitr expression in Treg. We speculate, therefore, that those Gitr^highCD4⁺Foxp3⁺ T cells in p50-deficient mice are generated by compensatory NF-κB p52. Taken together, we conclude that Gitr expression is upregulated by the enhancer in conjunction with p50 molecules in both nTreg and iTreg.

Because the enhancer seems to be the key regulatory region in the locus, it is likely that Gitr and Ox40 expression is further upregulated in Treg by Foxp3-mediated mechanisms operating through the enhancer. We noticed that within the enhancer the κB1 binding sequence is followed by a potential Foxp3 binding sequence, analogous to the NFAT-Foxp3 binding site in the IL-2 promoter (8) (Fig. 7A, Supplemental Fig. 3D). p50 binding to the κB2 site was weak, yet this site also possesses a similar arrangement (Supplemental Fig. 3A). We therefore analyzed Foxp3 binding to the κB1 plus κB2 region using a DNA pull-down assay and nuclear extracts from CD4⁺CD25^– T cells stimulated for 72 h with TGF-β plus anti-CD3 plus anti-CD28. Foxp3 was coprecipitated with the biotinylated κB probe but not a control probe of the same length from the Gitr promoter (Fig. 7B). To confirm Foxp3 binding to the potential site (3′ side of the p50 binding site) (Fig. 7A), we performed a competition assay with κB1 and Foxp3 competitors (i.e., nonbiotinylated double-stranded oligonucleotides shown in Fig. 7A). Stimulation-specific NF-κB p50 and Foxp3 binding to the κB probe was detected (Fig. 7C), and both binding to the enhancer probe was completely inhibited by the κB1 competitor (containing the κB1 binding site and the potential Foxp3 binding site, Fig. 7A; Fig. 7C, κB1), suggesting that p50 and Foxp3 bind to the competitor. Foxp3 binding was also inhibited by a Foxp3 competitor (lacking the κB1 binding site but containing the potential Foxp3-binding site, Fig. 7A; Fig. 7C, Foxp3); this indicates that Foxp3 binds to the overlapping region containing the potential Foxp3 binding site (Fig. 7A). Because Foxp3 binds to both NFAT and the IL-2 promoter (Supplemental Fig. 3D), we examined whether Foxp3 binds directly to enhancer DNA and/or in association with p50. To examine this, p50 binding to the enhancer was inhibited in the presence of a competitor containing only a p50 binding sequence from the CD40 promoter (CD40–NF-κB competitor) (22) (Fig. 7A, 7D), but Foxp3 binding was minimally affected (Fig. 7D); this suggests that Foxp3 binds to DNA in the enhancer. However, p50 binding to the enhancer was not completely inhibited by a 500-fold excess of the CD40–NF-κB competitor (containing only NF-κB site) (Fig. 7D), but was completely inhibited by the κB1 competitor (containing NF-κB plus Foxp3 binding sites) (Fig. 7C). This suggests that a p50/CD40–NF-κB competitor complex binds to Foxp3 on the probe (i.e., a protein–protein interaction), as shown in Supplemental Fig. 3E. This possibility was supported by a coimmunoprecipitation experiment where FLAG-tagged Foxp3 was coprecipitated with anti-p50 without DNA (Fig. 7E). Foxp3 seems therefore to bind to both p50 and DNA in the enhancer (Fig. 8D).

We confirmed that p50/Foxp3 binds to the enhancer in a manner analogous to NFAT/Foxp3 binding to the IL-2 promoter (but note that Foxp3 represses IL-2 expression). We investigated the functional relevance of this binding using a luciferase reporter assay. Hori et al. (31) have demonstrated that Gitr expression is upregulated by ectopically expressed Foxp3 in CD3-activated CD4⁺CD25⁻ T cells. We therefore coexpressed Foxp3 in CD3-activated EL4 cells and performed the luciferase assay with the Gitr promoter/enhancer reporter plasmid. Because NF-κB p50 and p65 bind to the enhancer (Fig. 4C) and NF-κB c-Rel regulates Treg development (32–36), we additionally coexpressed p50 and/or p65 (Fig. 8A) and p50 and/or c-Rel (Fig. 8B). Enhancer activity was increased by coexpressed p50, p65, or p50 plus p65 (Fig. 8A) and by coexpressed p50, c-Rel, or p50 plus c-Rel (Fig. 8B). Importantly, any p50-mediated upregulation was further increased by Foxp3 coexpression. In contrast, p65- (Fig. 8A) and c-Rel–mediated (Fig. 8B) upregulation was reduced by Foxp3, and indeed repression of p65 and c-Rel by Foxp3 was previously described (9, 37). This inhibitory effect of Foxp3 was neutralized by p50 coexpression (p50 plus p65 and p50 plus c-Rel expression). Taken together, these findings suggest that p50 and Foxp3 cooperate to upregulate the Gitr promoter by binding to the enhancer. To confirm this, the Gitr promoter/enhancer reporter was cotransfected with a fixed amount of the p50 expression plasmid, but with different amounts of the Foxp3 expression plasmid. Enhancer activity was increased by Foxp3 coexpression in a dose-dependent manner (Fig. 8C). A cooperative effect of p50 and Foxp3 on the enhancer was also observed using the Ox40 promoter/enhancer reporter plasmid (Supplemental Fig. 3G) operating through the enhancer only, or through both the promoter and enhancer.

How might Foxp3 upregulate gene expression using a p50 that possesses no transactivation domain? Usually, NF-κB p50 positively regulates gene expression by associating with other NF-κB and IκB family members that contain transactivation domains (e.g., through a p50/p65 heterodimer, p50/c-Rel heterodimer, or p50 homodimer/Bcl-3). In support of this, we detected p65, c-Rel, and Bcl-3 binding to the κB probe (Fig. 7C). We then asked what additional necessary part Foxp3 might play. We analyzed p50 binding to the κB probe with or without Foxp3. We considered using a nuclear extract from CD3-activated T cells (no Foxp3), but the amount of p50 in this nuclear extract is different from that in Foxp3⁺ T cells (stimulated with TGF-β). We therefore trapped Foxp3 protein in the same nuclear extract by adding a Foxp3 competitor (Fig. 7C) and performed a DNA pull-down assay using the κB probe. p50 binding to the enhancer probe was reduced as a result of inhibition of Foxp3 binding to the probe (Fig. 7C), suggesting that Foxp3 stabilizes p50 binding by itself binding to both p50 and DNA (Fig. 8D), thereby enabling accumulation of p50/p65 and p50/c-Rel heterodimer (Fig. 8E) and/or p50/p50 homodimer/Blc-3 (Supplemental Fig. 3F) to the enhancer. To be able to understand the whole picture of the Foxp3-mediated transcriptional upregulation, further investigation is obviously required. However, it is clear that the activation function of the transactivation domains of p65 and c-Rel do not seem to be inhibited by such indirect binding of Foxp3 mediated via p50. This would be consistent with the enhanced luciferase activity seen when all components (Fig. 8A, p50, p65, and Foxp3; Fig. 8B, p50, c-Rel, and Foxp3) were coexpressed.

We have identified a strong enhancer within the Gitr/Ox40 locus that is regulated by NF-κB and Foxp3. Gitr gene expression is upregulated by NF-κB through this enhancer in activated Teff, and further upregulated by NF-κB in conjunction with Foxp3 in Treg. It is well established that Foxp3 can bind to other transcription factors and inhibit activities of these factors (as shown in Fig. 8A, 8B, the activity of the p65 and c-Rel subunits of NF-κB was inhibited by coexpression of Foxp3). However, the enhancer activity was upregulated by coexpression of the p50 subunit of NF-κB with Foxp3. Foxp3 seems to stabilize binding of the NF-κB p50 to this enhancer. To explain how Foxp3 might stimulate p50-mediated transcription while lacking a transactivation domain (38), we suggest that p50 associates with other family members (e.g., p65 and c-Rel) to exploit their activation domains (38). Indeed, we have also shown that p65 and c-Rel bind to κB sites with p50 in the enhancer (Figs. 4C, 7C). Foxp3 could bind to the p50 molecule and DNA without altering the p50/p65 and p50/c-Rel complex formation (Fig. 8E) or inhibiting the transactivation domain of p65 and c-Rel. Although nuclear NF-κB levels are tightly regulated by negative feedback in activated T cells, we detected NF-κB p50, p65, and c-Rel in 72 h–stimulated T cells with TGF-β plus anti-CD3 plus anti-CD28. Nuclear translocation of these transcription factors sustained in Treg through TGF-β signaling and/or a Foxp3-mediated pathway. In CD3-activated T cells, nuclear p65 levels peak 2–6 h after activation (39, 40). Although we detected p65 binding to the κB probe using a nuclear extract from 72 h–stimulated T cells, p65 levels may have diminished from those found earlier after activation. c-Rel levels seem to remain high even after p65 decreased (40). The Foxp3 promoter activity is regulated by c-Rel (34), and nTreg development is also regulated by c-Rel through CNS3 in the Foxp3 gene (36). Taken together, p50/c-Rel heterodimers may be a key component in upregulating Gitr gene expression. Alternatively, transactivation domains might also be supplied by an IκB family member Bcl-3, which can bind to the p50 homodimer (41, 42) (Supplemental Fig. 3F). Gene expression upregulated by the p50 homodimer/Bcl-3 complex seems to depend on NF-κB sites (the p50 homodimer without Bcl-3 inhibits transcription through some NF-κB sites). Because Bcl-3 does not bind to DNA, it is not clear how Bcl-3 selects a p50 homodimer to be bound. We imagine that the p50 homodimer and Bcl-3 complex may be stabilized in conjunction with other factors that can bind close to the p50 homodimer binding site, in the way Foxp3 appears to do so in this study. In our case, Bcl-3 and p65 binding to the κB site was not detected and c-Rel binding was reduced unless Foxp3 was also bound to the enhancer (Fig. 7C). This may be because less p50 can bind to the κB site in the absence of Foxp3, or that NF-κB p50/p65, p50/c-Rel, and/or p50 homodimer/Bcl-3 formation is stabilized by Foxp3.

It is generally considered that promoters are upregulated by communication between the promoter and an enhancer via loop formation (43). The identified enhancer in the present study is located ∼5 and 20 kb downstream of the Gitr and Ox40 promoters, respectively. These distances are within the range found with other well-characterized enhancers and promoters. For example, the IgH (44) and Bcl-2 (45) promoter interacts with the 3′ enhancer over ∼100 kb. These findings suggest that Ox40 expression in Treg is also upregulated by Foxp3 through this NF-κB p50–mediated enhancer, but we cannot rule out a role for NF-κB p50 acting through the Ox40 promoter. Even so, the strong enhancer activity driven by the combination of p50 and Foxp3 seems to regulate Treg-specific upregulation within Gitr/Ox40 gene locus, by analogy with the Il4, Il5, and Il13 genes that are located over 140 kb within the Th2 cytokine-specific gene locus (46).

Recent analyses of DNase I HS sequencing and Foxp3 ChIP sequencing in Foxp3⁺ Treg and CD4⁺Foxp3^– T cells (28) have shown that Foxp3 predominantly utilizes functional enhancers that are present in Foxp3^– T cells during Treg development. Taken with our own findings, we now propose the following sequence of events during the activation of Gitr and Ox40 in Teff and Treg: 1) NF-κB p50/p65 and/or p50/c-Rel bind to the enhancer in activated Teff at an early stage of Treg development, leading to chromatin remodeling and Gitr and Ox40 gene expression; 2) induced Foxp3 in Treg stabilizes binding of the p50/p65 and/or p50/c-Rel heterodimers and p50-homodimer/Bcl-3 to the enhancer; and 3) these complexes then further upregulate Girt and Ox40 expression in Treg.

We suggest that other important Treg genes such as Ctla4, Il35, Cd25, and MS4A members may be regulated by Foxp3 by similar mechanisms. We therefore offer this scheme as a solution to how Foxp3 may upregulate certain genes enforcing regulation while simultaneously being able to damp so many others promoting inflammation.

We thank Drs. J. Kaye, W. Sabbagh, C. Norbury, and P. Cook for review of the manuscript.

The authors have no financial conflicts of interest.