Novel Foxo1–dependent transcriptional programs control T_reg cell function

Abstract

Regulatory T (T_reg) cells, characterized by expression of the transcription factor forkhead box P3 (Foxp3), maintain immune homeostasis by suppressing self-destructive immune responses^1–4. Foxp3 operates as a late-acting differentiation factor controlling T_reg cell homeostasis and function⁵, whereas the early T_reg-cell-lineage commitment is regulated by the Akt kinase and the forkhead box O (Foxo) family of transcription factors^6–10. However, whether Foxo proteins act beyond the T_reg-cell-commitment stage to control T_reg cell homeostasis and function remains largely unexplored. Here we show that Foxo1 is a pivotal regulatorof T_reg cell function. T_reg cells express high amounts of Foxo1 and display reduced T-cell-receptor-induced Akt activation, Foxo1 phosphorylation and Foxo1 nuclear exclusion. Mice with T_reg-cell-specific deletion of Foxo1 develop a fatal inflammatory disorder similar in severity to that seen in Foxp3-deficient mice, but without the loss of T_reg cells. Genome-wide analysis of Foxo1 binding sites reveals ~300 Foxo1-bound target genes, including the pro-inflammatory cytokine Ifng, that do not seem to be directly regulated by Foxp3. These findings show that the evolutionarily ancient Akt–Foxo1 signalling module controls a novel genetic program indispensable for T_reg cell function.

Among the three Foxo genes expressed in T cells, the transcript of Foxo1 is specifically upregulated in mature thymocytes⁸. To genetically mark cells expressing Foxo1, we engineered a Foxo1 allele (Foxo1^tag) encoding an in-frame fusion protein tag containing green fluorescent protein (GFP), Flag and a biotin-labelling peptide that enables Foxo1 biotinylation in birA-transgenic mice that express the bacterial biotin ligase BirA (Fig. 1a and Supplementary Fig. 1). T-cell expression of CD127 (also known as Il7r), the product of a previously identified Foxo1 target gene^11,12, was unaffected in Foxo1^tag/tag mice (Supplementary Fig. 1d), indicating that the protein tag does not alter Foxo1 function. Using GFP as a reporter, we found that immature thymocytes expressed low amounts of Foxo1 (Fig. 1a). Foxo1 was more strongly upregulated in T_reg cells than in conventional CD4⁺ T cells in the thymus, whereas the peripheral T cells expressed similar amounts of Foxo1 (Fig. 1a).

Figure 1. Foxo1 expression and TCR-induced Foxo1 nuclear exclusion in T_reg cells.

a, Foxo1 fusion protein with a tag containing GFP, Flag and a biotin-labelling peptide (top). Flow cytometric analysis of Foxo1–GFP expression in thymic CD4⁺CD8⁺ double-positive T cells (Thy DP), CD4⁺Foxp3⁻ single-positive T cells (Thy SP) and CD4⁺Foxp3⁺ T_reg cells (Thy T_reg), and splenic CD4⁺Foxp3⁻ conventional T cells (CD4) and CD4⁺Foxp3⁺ T_reg cells from Foxo1^tag/tag mice (T_reg) (bottom). Cells from wild-type mice were used as a control. b, c, TCR-induced Foxo1 nuclear clearance in CD4⁺RFP⁻ conventional (T_con) and CD4⁺RFP⁺ T_reg cells from Foxo1^tag/tag mice. Cells were stimulated with 0.01 μg ml⁻¹ (low) or 0.1 μg ml⁻¹ (high) anti-CD3 antibody (Ab) in the presence of 1.0 μg ml⁻¹ anti-CD28. b, Quantification of Foxo1 nuclear clearance. Nuclear exclusion leads to a decrease in clearance ratio. c, Representative images of Foxo1 nuclear clearance of T_con and T_reg cells. d, Phosphorylated (p)-Foxo1 Thr 24, total Foxo1, p-Akt Ser 473, total Akt, p-Erk1/2 and total Erk1/2 proteins in T_con and T_con cells stimulated with low anti-CD3 and anti-CD28 were determined by immunoblotting. The relative amounts of p-Foxo1, p-Akt and p-Erk1/2 proteins normalized to total Foxo1, Akt and Erk1/2 proteins are shown. Error bars represent s.d.; all experiments were repeated at least three times with similar results.

Foxo1 resides in the nucleus of quiescent T cells, and relocates to the cytosol after T-cell receptor (TCR) stimulation. To investigate the kinetics of Foxo1 translocation, we performed live-imaging experiments on CD4⁺ T cells from Foxo1^tag/tag mice on a Foxp3-internal ribosome entry site (IRES)-red fluorescent protein (RFP)-transgenic background in which T_reg cells are marked by the expression of RFP. Whereas high-dose CD3 antibody induced Foxo1 nuclear clearance in all T cells (Fig. 1b, c and Supplementary Video 1), low-dose CD3 antibody triggered Foxo1 translocation in conventional T cells but not in T_reg cells (Fig. 1b, c and Supplementary Video 2). Attenuated Foxo1 nuclear clearance in T_reg cells was further revealed by the immunostaining of T cells from wild-type mice (Supplementary Fig. 2a, b). Foxo1 nuclear export is regulated by Akt-induced Foxo1 phosphorylation. Low-dose anti-CD3 induced robust Akt and Foxo1 phosphorylation in conventionalT cells, which was markedly decreased in T_reg cells (Fig. 1d). The phosphorylation defects of Foxo1 and Akt were less profound in T cells stimulated with high-dose CD3 antibody (Supplementary Fig. 2c). Anti-CD3-induced Erk1 and Erk2 (also known as Mapk3 and Mapk1, respectively) phosphorylation was not compromised at the late time points in T_reg cells (Fig. 1d and Supplementary Fig. 2c), suggesting that the Akt pathway might be specifically modulated, probably through high expression of the Akt phosphatase Phlpp1 (ref. 13). Thus, compared to conventional T cells, T_reg cells are resistant to TCR-induced Akt activation and Foxo1 nuclear clearance.

To investigate the function of Foxo1 in T_reg cells, we crossed mice carrying floxed Foxo1 alleles (Foxo1^fl/fl) with Foxp3^cre mice¹⁴. Foxo1 protein was barely detectable in T_reg cells from Foxp3^cre Foxo1^fl/fl mice, whereas conventional CD4⁺ T cells from Foxp3^cre Foxo1^fl/fl mice expressed comparable amounts of Foxo1 to wild-type T cells (Fig. 2a). In the absence of Foxo1 in T_reg cells, mice developed a hunched posture associated with lack of mobility, in addition to crusting of the ears, eyelids and tail (Fig. 2b) before they became moribund within 35 days of birth. Such a fulminant phenotype was comparable in severity to that of mice with the scurfy mutation of the Foxp3 gene (Foxp3^sf; Fig. 2b), or mice depleted of T_reg cells^15,16. In addition, Foxp3^cre Foxo1^fl/fl mice showed splenomegaly and lymphadenopathy (Fig. 2c), and had a dense infiltrate of leukocytes in the salivary glands, lung interstitia, liver sinusoids, pancreas acini, stomach, and colon mucosa (Fig. 2d).

Figure 2. An essential role for Foxo1 in the control of T_reg cell function.

a, Foxo1 protein in CD4⁺Foxp3⁻ conventional (T_con) and T_reg cells from wild-type (WT) and Foxp3^creFoxo1^fl/fl mice was determined by immunoblotting. β-actin was used as a loading control. b, Images of 21-day-old wild-type, Foxp3^sf and Foxp3^creFoxo1^fl/fl mice. Arrows indicate the scurfy tails of Foxp3^sf and Foxp3^creFoxo1^fl/fl mice. c, A representative picture of spleens and peripheral lymph nodes from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice. d, Haematoxylin and eosin staining of salivary gland, lung, liver, pancreas, stomach and colon sections from 21-day-old wild-type and Foxp3^creFoxo1^fl/fl mice. e, CD4⁺ and CD8⁺ T-cell numbers from spleens and peripheral lymph nodes (pLN) of 17–21-day-old wild-type and Foxp3^creFoxo1^fl/fl mice (n=8 per genotype). f, Flow cytometric analysis of CD44 and CD62L expression in peripheral lymph node CD4⁺ and CD8⁺ T cells from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice. g, Peripheral lymph node T cells from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice were stimulated with phorbol myristate acetate (PMA) and ionomycin for 4 h. The expression of IFN-γ in CD4⁺ and CD8⁺ T cells was determined by intracellular staining. h, Flow cytometric analysis of Foxp3 expression in peripheral lymph node CD4⁺ T cells from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice (left). Peripheral lymph node T_reg cell numbers from 17–21-day-old wild-type and Foxp3^creFoxo1^fl/fl mice (right; n=8 per genotype). The P values between the two groups are shown. Asterisk indicates a statistically significant difference. Results represent at least three independent experiments.

The severe immunopathology was associated with the expansion of T-cell populations expressing high amounts of the cell-proliferation marker Ki67 (Fig. 2e and Supplementary Fig. 3). T cells displayed an effector/memory phenotype (Fig. 2f and Supplementary Fig. 4a) and produced increased amounts of the pro-inflammatory cytokine interferon (IFN)-γ (Fig. 2g and supplementary Fig. 4b). Deletion of Foxo1 from CD4⁺CD8⁺ immature T cells results in compromised T_reg cell differentiation^8,10. However, thymic and splenic T_reg cell numbers were unaffected in 12-day-old Foxp3^cre Foxo1^fl/fl mice (Supplementary Fig. 5). By the age of day 20, T_reg cell numbers in the peripheral lymphoid tissues were increased in Foxp3^cre Foxo1^fl/fl mice (Fig. 2h and Supplementary Fig. 4c). These findings imply that the lymphoproliferative disease is caused by a loss of T_reg cell function rather than a loss of T_reg cells.

The observation that T_reg cells were less sensitive to TCR-induced Foxo1 nuclear exclusion implies that Foxo1-dependent transcriptional regulation is crucial for T_reg cell function. However, cytosolic Foxo1 regulates autophagy through transcription-independent mechanisms¹⁷. To differentiate the activities of nuclear Foxo1 and cytosolic Foxo1 induced by Akt activation, we inserted a complementary DNA fragment coding for a haemagglutinin (HA)-tagged human Foxo1 mutant in which amino acids at the Akt phosphorylation sites were substituted with alanines¹⁸ (HA-hFoxo1(AAA)), preceded by a loxP-flanked `neo-STOP' cassette into the Rosa26 locus (Supplementary Fig. 6). Mice carrying the mutant allele (Foxo1^AAA) were crossed with oestrogen receptor-Cre (creER) mice. Wild-type T cells or T cells from creER Foxo1^AAA/+ mice previously treated with tamoxifen to induce HA-hFoxo1(AAA) expression were stimulated with CD3 and CD28 antibodies. As expected, whereas wild-type Foxo1 translocated from the nucleus to the cytosol, the HA-hFoxo1(AAA)-expressing mutant was defective in nuclear export (Supplementary Fig. 6c). We further crossed Foxo1^AAA/+ mice with Foxp3^cre mice, and confirmed specific HA-hFoxo1(AAA) expression in T_reg cells (Fig. 3a). Ectopic expression of HA-hFoxo1(AAA) did not affect T_reg cell differentiation or homeostasis (Supplementary Fig. 7a). Upon breeding the Foxo1^AAA allele to Foxp3^cre Foxo1^fl/fl mice, the lethal inflammatory phenotype was completely rescued (Fig. 3b and supplementary Fig. 7), indicating that the T_reg cell defects are caused by a loss of nuclear Foxo1 activity.

Figure 3. Foxo1-dependent transcriptional programs in T_reg cells.

a, A schematic representation of the mouse line expressing HA-hFoxo1(AAA) specifically in T_reg cells. A cassette containing a floxed stop sequence, a cDNA coding for a HA-tagged human Foxo1(AAA) mutant, and IRES-GFP proteins was inserted into the Rosa26 locus. Mice with the knock-in allele were crossed with Foxp3^cre mice (top). The expression of HA-FOXO1(AAA) was determined by immunoblotting with anti-HA. β-actin was used as a loading control (bottom). b, Survival of wild-type, Foxp3^creFoxo1^fl/fl and Foxp3^creFoxo1^fl/flFoxo1^AAA/+ mice (n=6 to 10 per genotype). c, The overlapping set of peaks shared between antibody and biotin ChIP-seq samples was associated with functional annotations based on the genomic location of the summits within each peak, which revealed enrichment in gene promoters (P=0.04) and 5′ UTRs (P<1×10⁻¹⁰) above the background rate using randomly generated peaks. d, Foxo1-bound target genes differentially expressed between wild-type and Foxp3^creFoxo1^fl/fl T_reg cells, and whose expression was corrected in Foxp3^creFoxo1^fl/flFoxo1^AAA/+ T_reg cells.

For global identification of Foxo1 DNA-binding sites in T_reg cells, we performed chromatin immunoprecipitation coupled to high-throughput sequencing (ChIP-seq) experiments using Foxo1 antibody precipitation of chromatin purified from CD4⁺CD25⁺ T cells from C57BL/6 mice, or streptavidin pull-down of chromatin from T_reg cells of Foxo1^tag/tag birA mice. To define a ChIP-seq peak as significant, we used the model-based analysis of ChIP-seq (MACS) peak calling software and applied an empirical false-discovery rate of 0.01 at which the noise ratio of peak detection was less than 0.005 (Supplementary Fig. 8a). Using this criterion, we identified 3,431 enriched genomic loci that were shared between the antibody and the biotinylated Foxo1 samples (Supplementary Fig. 8b), and were designated as putative Foxo1 binding sites (Supplementary Table 1). Among the Foxo1binding peaks, we could identify previously characterized binding sites of Foxo1 target genes, including an intronic binding site in the Foxp3 locus⁸, an Il7r enhancer element^11,12, as well as binding sites in the proximal promoter regions of the Fbxo32 and Bcl2l11 genes^19,20 (Supplementary Fig. 9).

Foxo1 binding sites were mostly enriched in the promoters and the 5′ untranslated regions (UTRs), whereas peaks mapped to the intergenic regions and 3′ UTRs were substantially under-represented (Fig. 3c). Further analysis of the peaks showed that Foxo1 preferentially bound to regions within 500 base pairs of the transcription start sites (Supplementary Fig. 8c). Foxo1 binding peaks were enriched for the high-affinity Foxo1 binding motifs, and de novo motif prediction from the top 500 ranked binding peaks revealed a conserved Foxo1 recognition site (Supplementary Fig. 8d, e). In addition, approximately 73% of Foxo1 binding sites were mapped to the conserved genomic regions of mammalian species (Supplementary Table 1). These observations support an evolutionarily conserved function for Foxo1 in classical transcriptional regulation of target-gene expression.

To identify the cellular functions directly regulated by Foxo1 transcriptional control, we performed gene-expression profiling on T_reg cells from wild-type, Foxp3^creFoxo1^fl/fl, Foxp3^creFoxo1^AAA/+ and Foxp3^creFoxo1^fl/fl Foxo1^AAA/+ mice. Comparison of data from wild-type and Foxp3^creFoxo1^fl/fl T_reg cells revealed 942 and 1,155 genes downregulated or upregulated, respectively, by more than 1.5 fold, ~80% of which were corrected in Foxp3^creFoxo1^fl/fl Foxo1^AAA/+ T_reg cells (Supplementary Table 2). By cross-referencing this data set with the data set of Foxo1-bound genes, we identified 310 putative Foxo1 direct target genes, among which 240 and 70 genes were activated or repressed, respectively, by Foxo1 (Fig. 3d and Supplementary Table 3).

T_reg cell homeostasis and function are dependent on the T_reg-cell-specific transcription factor Foxp3 (ref. 5). ChIP and genome tiling array studies have revealed Foxp3-bound genes in murine T cells^21,22. Approximately 6.8% or 9.6% of Foxo1-bound genes were occupied by Foxp3 in T_reg cells or in a T-cell line that overexpressed Foxp3 (Supplementary Fig. 10a). Comparison of the putative direct target genes of Foxo1 and Foxp3 revealed that ~90% of activated genes and ~99% of repressed genes were specifically regulated by Foxo1 or Foxp3 (Supplementary Fig. 10b, c and Supplementary Table 3). Indeed, expression of the Foxp3 target gene Il2ra (coding for CD25) was decreased in T cells transcribing a Foxp3 null allele that encodes a GFP reporter²³, but not in Foxo1-deficient T_reg cells (Fig. 4a and Supplementary Fig. 10d). A recent study showed that Foxo1 binds to the promoter region of the Foxp3 target gene Ctla4 (ref. 10). ChIP-seq experiments revealed weak Ctla4-promoter binding of Foxo1 in antibody but not biotin samples (data not shown). In the absence of Foxo1, CTLA4 was marginally downregulated, whereas the amount of Ctla4 transcript was approximately fourfold lower in the absence of Foxp3 (Fig. 4a and Supplementary Fig. 10e). Foxp3 and CTLA4 are essential for T_reg cell inhibition of conventional T-cell proliferation in vitro^23,24. By contrast, as previously reported in ref. 11, Foxo1-deficient T_reg cells had normal suppressive activity in such assays (Fig. 4b). These observations imply that the Foxp3-dependent program is largely intact in the absence of Foxo1, and the loss of suppressive activity of Foxo1-deficient T_reg cells is probably independent of CTLA4.

Figure 4. Restoration of Foxo1-deficient T_reg cell function in the absence of IFN-γ.

a, Flow cytometric analysis of Foxp3, CD25 and CTLA4 expression in lymph node T_reg cells from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice. b, Suppression of wild-type naive CD4⁺ T cells, labelled with the cytosolic dye carboxyfluorescein diacetate succinimidyl ester (CFSE) (responding T cells (T_resp)), by wild-type and Foxo1-deficient (Foxo1^−/−) T_reg cells. T_resp cell division was assessed by CFSE dilution at the indicated ratios of cell numbers between T_reg cell and T_resp cells. c, Foxo1-bound regions in the Ifng locus. d, Foxo1 binding to the ChIP-seq peak was confirmed by ChIP-quantitative (q)PCR, and is presented relative to enrichment by immunoprecipitation with isotype-matched control antibody. e, Expression of IFN-γ mRNA in T_reg cells from wild-type and Foxp3^creFoxo1^fl/fl mice was determined by qPCR. f, Peripheral lymph node T cells from 20-day-old wild-type and Foxp3^creFoxo1^fl/fl mice were stimulated with PMA and ionomycin for 4 h. The expression of IFN-γ in T_reg cells was determined by intracellular staining. g, h, Naive T cells (N) were transferred to Rag1^−/− mice alone or in combination with wild-type, Foxo1^−/− or Foxo1^−/− Ifng^−/− T_reg cells. Changes in body weight of Rag1^−/− recipients over time are shown (n=4) (g). Haematoxylin and eosin staining of colon sections from Rag1^−/− mice at 4 weeks after the transfer (h). Error bars represent s.d.; results are representative of at least three independent experiments.

To gain insights into the general functional features of the Foxo1-regulated program, we analysed the Gene Ontology and BioCarta pathway association of Foxo1 target genes²⁵. The putative Foxo1 direct target genes were strongly associated with cell communication, signal transduction, transcription and metabolism, and were specifically enriched for genes involved in the Jak-STAT, TCR and insulin-signalling pathways (Supplementary Table 4). A salient characteristic of T_reg cells is their inability to produce pro-inflammatory cytokines such as IFN-γ. Foxo1 was recruited to a regulatory element located 22 kilobases upstream from the transcriptional start site of the Ifng gene²⁶ in T_reg cells (Fig. 4c, d). In the absence of Foxo1, IFN-γ messenger RNA and protein were highly induced in T_reg cells (Fig. 4e, f and Supplementary Fig. 11a). However, similar to wild-type T_reg cells, Foxo1-deficient T_reg cells were unable to produce interleukin (IL)-2 (Supplementary Fig. 11b), which is probably regulated by Foxp3 (ref. 22). To determine whether Foxo1 had a cell-intrinsic role in inhibiting IFN-γ expression, we co-cultured wild-type T_reg cells and T_reg cells from the disease-free creER Foxo1^fl/fl mice previously treated with tamoxifen to acutely delete Foxo1. A sizable fraction of Foxo1-deficient but not wild-type T_reg cells produced IFN-γ under the condition of IL-12 and IFN-γ stimulation (Supplementary Fig.12a). Importantly, wild-type and Foxo1-deficient conventional CD4⁺ T cells produced similar amounts of IFN-γ (Supplementary Fig. 12b), revealing a specific function for Foxo1 in repressing IFN-γ expression in T_reg cells.

To investigate whether IFN-γ production by Foxo1-deficient T_reg cells contributed to their loss of suppressive function, we crossed Foxp3^creFoxo1^fl/fl mice to the IFN-γ-deficient background. Ablation of IFN-γ partially corrected the lethal inflammatory phenotype as well as the lymphoproliferative disease of Foxp3^creFoxo1^fl/fl mice (Supplementary Fig. 13). To assess the specific function of IFN-γ produced by Foxo1-deficient T_reg cells, we used a transfer model of colitis. Recipient mice deficient in recombination-activating gene 1 (Rag1^−/−) did not develop colitis upon receipt of wild-type, Foxo1^−/− or Foxo1^−/− Ifng^−/− T_reg cells (Supplementary Fig. 14). However, recipients of naive T cells developed a wasting disease and colitis within 4 weeks, which was prevented by wild-type but not Foxo1-deficient T_reg cells (Fig. 4g, h and Supplementary Fig. 14b). Importantly, the suppressive activity was largely restored in Foxo1^−/− Ifng^−/− T_reg cells (Fig. 4g, h and Supplementary Fig. 14b), supporting Ifng as a critical Foxo1 target gene in the control of T_reg cell function.

Foxo proteins control Foxp3 expression and T_reg cell differentiation^8–10. Our new data reveal that the Akt–Foxo1 signalling pathway is modulated in mature T_reg cells to maintain high nuclear Foxo1 activity; this is crucial for T_reg cell function, in part through Foxo1 inhibition of IFN-γ expression. IFN-γ produced by T_reg cells may activate conventional T cells or antigen-presenting cells compromising T_reg cell function, but the precise mechanisms remain to be determined. Diminished TCR-induced Akt activation has been observed in human T_reg cells²⁷, suggesting that Foxo1 controls human T_reg cell function. Under conditions of strong TCR stimulation, nuclear Foxo1 is translocated to the cytosol in T_reg cells, raising the possibility that the Foxo1-dependent program is subject to regulation. Intriguingly, T_reg cells from mice infected with Toxoplasma gondii produce IFN-γ²⁸. Increased frequencies of IFN-γ-secreting T_reg cells are also present in multiple sclerosis and diabetes patients^29,30. Future studies will determine whether the loss of Foxo1 function accounts for T_reg cell plasticity under these pathological conditions, and whether the Akt–Foxo1 signalling pathway can be manipulated for the treatment of T_reg-cell-associated immunological disorders.

METHODS

Mice

To generate a Foxo1 knock-in mouse model (Foxo1^tag), two genomic DNA fragments of the Foxo1 gene were isolated from a C57BL/6 bacterial artificial chromosome library (Genome Systems). A cDNA coding for GFP, Flag and biotin-labelling peptide was inserted into the genomic DNA fragment containing the second exon of Foxo1 before the stop codon. Targeting vector was constructed by cloning the fragments into pEasy-Flox plasmid. To generate a Rosa26-hFoxo1^AAA knock-in mouse model (Foxo1^AAA), the cDNA coding for a HA-tagged human Foxo1 (AAA) mutant was inserted into a Rosa26 targeting vector. Linearized targeting vectors were transfected into embryonic stem cells derived from the C57BL/6 strain. Homologous recombinants were identified by Southern blotting analysis, and were implanted into foster mothers. Chimaeric mice were bred to C57BL/6 mice, and the F₁ generation was screened for germline transmission. The neo gene in Foxo1^tag mice was removed by breeding F₁ mice with a strain of cytomegalovirus-promoter-driven cre-transgenic mice (The Jackson Laboratory). Mice containing floxed Foxo1, Foxp3-IRES-RFP, Foxp3^Cre, Foxp3^gfpko (Foxp3 null allele coding for a GFP reporter) and birA alleles were previously described^{11,14,23,31,32}. C57BL/6, Foxp3^sf, Ifng^−/−, Rag1^−/− and UBC-cre-ERT2 (creER) mice were purchased from the Jackson Laboratory. Mice with intact or floxed Foxo1 alleles were used as wild-type controls. Mice with T_reg-cell-specific deletion of Foxo1 were generated by crossing Foxo1 floxed mice with Foxp3^cre mice. To mark T_reg cells with RFP, mice were bred with Foxp3-IRES-RFP mice. To label Foxo1 with biotin, Foxo1^tag mice were bred with birA-transgenic mice. All mice were maintained under specific pathogen-free conditions, and animal experimentation was conducted in accordance with institutional guidelines.

Antibodies and immunoblotting

Anti-p-Foxo1 Thr 24 (no. 9464), anti-Foxo1 (C29H4), anti-p-Akt Ser 473 (D9E), anti-Akt (C67E7), anti-p-Erk1/2, anti-Erk1/2 and anti-HA (6E2) were purchased from Cell Signaling. Anti-β-actin (AC-15) was obtained from Sigma. To determine Foxo1 expression in T_reg cells, CD4⁺Foxp3⁺ T_reg cells were purified by FACS. Total protein extracts were dissolved in SDS sample buffer. To assess Foxo1 translocation, T cells were stimulated with anti-CD3 and anti-CD28, and the cytosolic and nuclear extracts were prepared. To analyse Akt, Foxo1 and Erk phosphorylation, conventional CD4⁺ T cells and CD4⁺Foxp3⁺ T_reg cells were purified by FACS. T cells were stimulated with anti-CD3 and anti-CD28, and the total protein extracts were prepared. Protein extracts were separated on 8% SDS–PAGE gels and transferred to polyvinylidene difluoride membrane (Millipore). The membranes were probed with antibodies and visualized with the Immobilon Western Chemiluminescent HRP Substrate (Millipore).

Live imaging of Foxo1 translocation

CD4⁺ cells were isolated from Foxo1^tag/tag Foxp3-IRES-RFP mice, and stimulated with plate-bound anti-CD3 and anti-CD28 on glass surfaces. GFP, RFP and bright-field images were acquired every 2 min after addition of cells using a fluorescence videomicroscope (Olympus IX-81) fitted with a 40X objective lens. Nuclear Foxo1 clearance was quantified by analysing two orthogonal linescans through each cell. For each linescan, the average intensity of three equally spaced pixels inside the cell was divided by the average intensity of the cell edge on each side to determine a `clearance ratio'. Nuclear exclusion leads to a decrease in clearance ratio.

Immunofluorescence microscopy

CD4⁺ T cells were isolated from spleen and peripheral lymph nodes of C57BL/6 mice with anti-CD4 microbeads (Miltenyi Biotec Inc.), stimulated with plate-bound anti-CD3 (low dose, 0.01 μg ml⁻¹; high dose, 0.1 μg ml⁻¹) and anti-CD28 (1 μg ml⁻¹) on chamber slides at 37 °C for 20 min. After fixation with 4% paraformaldehyde, cells were permeabilized with Foxp3 Fixation/Permeabilization buffer (eBioscience) according to the manufacturer's instructions. After blocking with Permeabilization buffer and 3% BSA, cells were incubated with 1:250 diluted rabbit anti-Foxo1 (C29H4, Cell Signaling) and rat anti-Foxp3 (FJK-16 s, eBioscience), followed by allophycocyanin (APC)-anti-rabbit and fluorescein isothiocyanate (FITC)-anti-rat secondary antibodies in Permeabilization buffer and 1% BSA. Slides were mounted with gold anti-fading mounting buffer (Invitrogen). Images were acquired with a Leica TCS SP5-II confocal microscope. For quantitative analysis, five fields were selected randomly and total cells in the field were manually counted and grouped with Volocity software (PerkinElmer Inc.), on the basis of their Foxo1 nuclear or cytosolic localization and Foxp3 expression.

Flow cytometry

Fluorescent-dye-labelled antibodies against cell-surface markers CD4, CD8, TCR-β, CD44, CD62L, CTLA4 and CD25 were purchased from eBiosciences. Thymic, splenic and lymph node cells were depleted of erythrocytes by hypotonic lysis. Cells were incubated with specific antibodies for 30 min on ice in the presence of 2.4G2 monoclonal antibody to block FcγR binding. All samples were acquired and analysed with an LSR II flow cytometer (Becton, Dickinson) and FlowJo software (TreeStar). Intracellular Foxp3, CTLA4 and Ki67 stainings were carried out with kits from eBiosciences. To determine cytokine expression, splenic and lymph node T cells were stimulated with 50 ng ml⁻¹ phorbol 12-myristate 13-acetate (Sigma), 1 μM ionomycin (Sigma) and GolgiStop (BD Biosciences) for 4 h. After stimulation, cells were stained with cell-surface marker antibodies, fixed and permeabilized, and stained with anti-Foxp3 in combination with specific antibodies against IFN-γ and IL-2 (eBiosciences).

ELISA

T_reg cells were purified from wild-type and Foxp3^cre Foxo1^fl/fl mice by FACS and stimulated with CD3 and CD28 antibodies in the presence of IL-2 for 24 h. Cytokine amounts in tissue-culture supernatants were assayed with ELISA antibody pair for IFN-γ (eBioscience, 88-8312) in accordance with the manufacturer's recommendations.

In vitro T-cell differentiation

CreER-mediated deletion of floxed Foxo1 alleles was induced by intraperitoneal injection of 2 mg of tamoxifen (Sigma) emulsified in 200 ml of corn oil (Sigma) every day for 5 days. CD4⁺Foxp3⁺ T_reg cells or CD44^loCD4⁺Foxp3⁻ conventional T cells from CD45.1⁺ wild-type and CD45.2⁺ tamoxifen-treated creER Foxo1^fl/fl (Foxo1^−/−) mice were isolated by FACS and were co-cultured at 1:1 ratio with irradiated splenocytes, anti-CD3 (1 μg ml⁻¹), anti-CD28 (2 μg ml⁻¹) and IL-2 (200 U ml⁻¹) in the presence or absence of IL-12 (5 ng ml⁻¹) and IFN-γ (10 ng ml⁻¹) for 3 days.

In vitro T_reg cell suppression

CD44^loCD4⁺Foxp3⁻ conventional T cells labelled with CFSE were used as responder cells. Responder T cells (5 × 10⁴) were cultured for 72 h with irradiated splenocytes (1 × 10⁵) and anti-CD3 (2 μg ml⁻¹) in the presence or absence of various numbers of T_reg cells. The division of responder T cells was assessed by dilution of CFSE.

Transfer model of colitis

Naive T cells sorted by FACS (CD4⁺CD44^loCD62L^hi, 4 × 10⁵) from CD45.1⁺ C57BL/6 mice were transferred into Rag1^−/− mice alone or in combination with CD45.2⁺ T_reg cells (CD4⁺Foxp3⁺, 2 × 10⁵). After T-cell reconstitution, mice were weighed weekly and monitored for signs of disease. Mice were killed at 4 weeks after T-cell transfer and their colons were used for histopathological analysis. The following grades were used to evaluate their disease severity: 0, normal colonic crypt architecture; 1, mild inflammation: slight epithelial cell hyperplasia and increased numbers of leukocytes in the mucosa; 2, moderate colitis: pronounced epithelial hyperplasia, significant leukocyte infiltration, and decreased numbers of goblet cells; 3, severe colitis: marked epithelial hyperplasia with extensive leukocyte infiltration, significant depletion of goblet cells, occasional ulceration, or cryptic abscesses; 4, very severe colitis: marked epithelial hyperplasia with extensive transmural leukocyte infiltration, severe depletion of goblet cells, many crypt abscesses and severe ulceration.

Histopathology

Tissues from killed animals were fixed in Safefix II (Protocol) and embedded in paraffin. 5-μm sections were stained with haematoxylin and eosin.

ChIP-seq

T_reg cells were fixed for 10 min at 25 °C with 10% formaldehyde. After incubation, glycine was added to a final concentration of 0.125 M to `quench' the formaldehyde. Cells were pelleted, washed twice with ice-cold PBS and lysed. The lysates were pelleted, re-suspended and sonicated to reduce DNA length to 300–500 base pairs (bp). The chromatin prepared from T_reg cells of C57BL/6 mice was incubated with protein-A-anti-Foxo1 (ab39670, Abcam) or an iso-type control antibody overnight. The chromatin prepared from T_reg cells of Foxo1^tag/tagbirA Foxp3-IRES-RFP or control birA mice was incubated with streptavidin overnight. The immune complexes were washed, and eluted in 500 ml of elution buffer containing 50 μM Tris, 10 mM EDTA and 1.0% SDS. Precipitated ChIP DNA and input DNA were incubated at 65 °C to reverse the crosslinking. After digestion with RNase and proteinase K, the ChIP and input DNA were purified with phenol/chloroform extraction and ethanol precipitation. The purified DNA was repaired, ligated with an adaptor, and amplified by PCR for 15–20 cycles. The amplified DNA was purified by gel extraction and used for sequencing. SR-36 sequencing was done at the Genome Center of Cold Spring Harbor Laboratory.

Visualizing read densities and identifying binding sites

Uniquely mapped sequence reads (36 bp) were aligned to the mouse genome (July 2007, NCBI37/mm9) using the Bowtie³³ short-read alignment software (version 0.12.7). Candidate binding sites were predicted using the MACS³⁴ peak detection software (version 1.4.1). In brief, MACS determines regions of enrichment by building sequencing-density profiles for sense and antisense reads, with respect to the reference, and pairing them such that the average distance between pairings can be used to model the approximate library fragment length (d). Reads from both strands were then repositioned by shifting a distance half the estimated fragment length, ½d, downstream to capture the most likely binding location at the fragment midpoint rather than skewing to the 5′ or 3′ ends. Enrichment was then determined by scanning the shifted, genome-wide read profile with a window twice the size of the estimated fragment length, where the number of shifted reads falling within this 2 × d window was assumed to follow a Poisson distribution, with a mean parameter determined from the local background in the input sample. The empirical false-discovery rate (eFDR) was fine-tuned using the noise ratio to maximize the effectiveness of the peak-calling strategy. The noise ratio was defined as the number control IgG peaks overlapping control BirA peaks (control overlaps) divided by the number of Foxo1 antibody peaks overlapping Foxo1 biotin peaks (ChIP overlaps). A noise ratio threshold of 0.005 was chosen as it corresponded to a large reduction in noise. To achieve this ratio, peak-calling was performed at an eFDR of 0.01. As defined in ref. 34, false background peaks were enumerated as the number of peaks called when swapping the input and ChIP sample. The eFDR was then calculated as the number of peaks detected divided by the number of background peaks detected. We, thus, determined a suitable threshold of P < 2.5 × 10⁻⁹ and P < 5.0 × 10⁻⁶, yielding 19,904 and 61,180 significantly enriched loci for antibody- and biotin-based ChIP-seq of Foxo1, respectively. To visualize read-density profiles, we aggregated counts at each genomic position for uniquely mapped reads that were extended to the estimate d. Counts were then smoothed with a moving average within a 200-bp sliding window across the genome. Smoothed average counts were visualized near to genes of interest using the R software package (http://www.R-project.org).

Associating peaks with gene features

To further refine our enriched loci, we corroborated peaks detected in antibody-based ChIP-seq with those from our biotinylated-Foxo1 tag-based method. To accomplish this, we distinguished high-confidence peaks by selecting for all `antibody' peaks overlapped by one or more `biotin' peaks and extending the 3′ and 5′ most coordinates to encompass all overlaps, so as to reduce any redundancies. The merged coordinates yielded 3,431 high-confidence, unique, merged regions, which we deemed putative Foxo1 binding sites. To associate peaks with functional regions at high precision and resolution, we intersected antibody peak summits (position of highest read density within a candidate peak region) falling within high-confidence corroborated regions with the set of genomic features (for example, 5′ UTR, 3′ UTR, exons, introns, 5-kilobase promoter, intergenic space) obtained from the UCSC Table Browser³⁵ (http://genome.ucsc.edu/). We also compared our regions with conserved phastCons scores and elements (Euarchontoglire subset)³⁶. Over-representation of these features was estimated empirically by generating 1,000 background sets each containing 3,431 random peaks, maintaining the same distribution of chromosomes and widths as our high-confidence set. Peaks from each background set were assigned to appropriate functional categories. The significance of the observed number of each annotation from our high-confidence summits was then inferred using a Z-test with respect to these randomly generated background counts.

DNA motif analysis

We first extracted nucleotide sequences±250 bp flanking all antibody-based ChIP-seq summits associated with peaks corroborated by biotinylated-Foxo1 tag-based ChIP-seq enrichment. These DNA sequences were screened for over-representation of known consensus motifs, using both the JASPAR³⁷ and TRANSFAC³⁸ databases, and the most enriched known motifs were noted. To examine to what extent the consensus Foxo1 TRANSFAC motifs, M00473 and M00474 were enriched, we ranked our putative binding site candidates by P value and assessed the cumulative occurrence rate one peak at a time, from most to least enriched. This was done to illustrate the expectation that the cumulative motif occurrence rate approaches the number of peaks with a motif divided by the total number of peaks but would be significantly increased in more enriched subset of binding candidates. Occurrence was determined using the STORM motif analysis tool available in the CREAD software package³⁹ at a threshold of P<1×10⁻⁴. We further substantiated the over-representation of consensus Foxo1 motifs by performing de novo motif prediction from the top 500 ranked candidate binding site sequences using the GADEM⁴⁰ motif-discovery algorithm, compared to shuffled sequences with similar di-nucleotide composition as background. Statistical similarity of predicted de novo motifs was estimated using the Tomtom⁴¹ motif comparison software available as part of the MEME software suite.

Gene-expression profiling

T_reg cells were isolated from the spleens and peripheral lymph nodes of Foxp3^creFoxo1^fl/fl, Foxp3^creFoxo1^AAA/+, Foxp3^creFoxo1^fl/fl Foxo1^AAA/+ and wild-type mice by FACS. RNA was prepared with the miRNeasy Mini Kit according to the manufacturer's instructions (Qiagen). Two rounds of RNA amplification, labelling and hybridization to M430 2.0 chips (Affymetrix) were done at the Core Facility of Memorial Sloan-Kettering Cancer Center.

Microarray data analysis

Microarray data was analysed using the R statistical software package. To reduce technical variations from array to array, chip normalization was performed using robust multi-array analysis available via the affy BioConductor package and differential gene expression defined by 1.5-fold change with a false-discovery rate of <0.01 was determined using the limma package^42,43. When multiple probes for a given gene were significantly differentially expressed, their log₂-fold changes were averaged.

qPCR

mRNA amounts of Ifng, Il2ra, Ctla4 and Actb were determined by qPCR with the following primers: Ifng, 5′-gcgtcattgaatcacacctg-3′ and 5′-tgagctcattgaatgcttgg-3′; Il2ra, 5′-gagacttcctgccccataac-3′ and 5′-gccactgctaccttatactcc-3′; Ctla4, 5′-tggactccggaggtacaaag-3′ and 5′-aaacggcctttcagttgatg-3′; Actb, 5′-ttgctgacaggatgcagaag-3′ and 5′-acatctgctggaaggtggac-3′. The mRNA amounts were normalized to those of Actb. Chromatin amounts of the Foxo1 binding site in the Ifng locus were determined by qPCR with the following primers: 5′-gacctgcacttctgtgagca-3′ and 5′-tctccttcctgtggatcacc-3′.

Heat-map

To effectively visualize Foxo1-dependent gene expression, we gated relative expression values on wild-type versus Foxp3^creFoxo1^AAA/+ values; that is, the log₂-fold change between wild-type versus Foxp3^creFoxo1^AAA/+ was used to define a wild-type value equal to½log₂-fold-change and a Foxp3^creFoxo1^AAA/+ value equal to ½ log₂-fold change. The remaining Foxp3^creFoxo1^fl/fl and Foxp3^creFoxo1^fl/fl/Foxp3^creFoxo1^AAA/+ fold-change values were then similarly normalized with respect to this new wild-type value. We focused on the range of genes within±2 log₂-fold change; values exceeding this range at the extremes were coloured blue or yellow. A selection of genes was depicted in which the Foxp3^creFoxo1^fl/fl phenotype was partially or completely rescued by Foxp3^creFoxo1^AAA/+, and putative Foxo1 binding sites, as determined by ChIP-seq experiments, resided nearby.

Gene-set enrichment analysis

Gene set and pathway analysis was performed using the FatiGO⁴⁴ tool via the Babelomics²⁵ software suite (http://babelomics.bioinfo.cipf.es) on Foxo1-bound target genes showing differential gene expression rescued by Foxp3^creFoxo1^AAA/+. Significantly over-represented categories were defined by an adjusted P value of 0.05 or less. We highlighted a selection of Gene Ontology categories and BioCarta pathways made available on the Babelomics web-based tool in our analysis.

Statistical analysis

Student's t-test was used to calculate statistical significance for difference in a particular measurement between groups. A P value of <0.05 was considered statistically significant.