Regulatory T cells (Treg cells), whose differentiation and function are controlled by transcription factor Foxp3, express the closely related family member Foxp1. Here we explored Foxp1 function in Treg cells. We found that a large number of Foxp3-bound genomic sites in Treg cells were occupied by Foxp1 in both Treg cells and conventional T cells (Tconv cells). In Treg cells, Foxp1 markedly increased Foxp3 binding to these sites. Foxp1 deficiency in Treg cells resulted in their impaired function and competitive fitness, associated with markedly reduced CD25 expression and interleukin-2 (IL-2) responsiveness, diminished CTLA-4 expression and increased SATB1 expression. The characteristic expression patterns of CD25, Foxp3 and CTLA-4 in Treg cells were fully or partially rescued by strong IL-2 signaling. Our studies suggest that Foxp1 serves an essential non-redundant function in Treg cells by enforcing Foxp3-mediated regulation of gene expression and enabling efficient IL-2 signaling in these cells.
Regulatory T cells (Treg cells) play an essential role in limiting immune response-induced inflammation and associated pathology during infection and tissue injury1. The X-linked transcription factor Foxp3 is essential for establishment and maintenance of the Treg cell lineage and suppressor function. Loss-of-function mutations in the Foxp3 gene cause Treg cell deficiency and consequent fatal multi-organ autoimmune and inflammatory disease in humans and mice2,3,4.
Foxp3 is a member of the Foxp transcription factor subfamily (Foxp1−4), whose members regulate differentiation of a variety of cell types. All Foxp proteins contain a zinc finger domain and a leucine zipper motif amino (N)-terminal to the conserved DNA-binding forkhead domain5. The Foxp transcription factors are capable of homo- and heterodimerization, which is thought to enhance their binding to a relatively low-affinity consensus motif6,7,8.
Besides the well-established role for Foxp3 in Treg cell biology, another Foxp family member, Foxp1, plays an important role in the adaptive immune system by facilitating early B cell development, opposing follicular helper T cell (TFH cell) differentiation, and maintaining quiescence of naïve T cells9,10,11,12. Foxp1 is highly expressed in naïve T cells and undergoes downregulation on their activation. Foxp1 forms homodimers and can also heterodimerize with Foxp3, which forms large protein complexes in Treg cells that include Foxp113,14.
These observations prompted us to investigate whether Foxp3 might be binding predominantly to the sites occupied by Foxp1 in naïve T cells, replacing Foxp1 during Treg cell differentiation. This scenario would imply that Foxp1 plays a redundant role in Treg cells. Alternatively, it is possible that Foxp1 has a non-redundant function in Treg cells through cooperative binding to sites occupied jointly with Foxp3 and involvement in Foxp3-mediated regulation of gene expression.
We found that Foxp1-bound sites in naïve CD4+ T cells were largely overlapping with those found in Treg cells, where the vast majority of Foxp1-bound sites were jointly occupied by Foxp3. Treg cell-restricted Foxp1 deficiency resulted in a genome-wide reduction of Foxp3 binding and altered gene expression of a number of Treg cell signature genes. Although Treg cell-specific deficiency in Foxp1 did not result in severe autoimmune disease, it led to Treg cell dysfunction, driven at least in part by sharply diminished CD25 expression and resultant impairment in IL-2 signaling. These results suggest that Foxp1 has a non-redundant role in Treg cell fitness and suppressor function.
Foxp1 and Foxp3 share a majority of binding sites
To determine whether Foxp3 and Foxp1 bound similar sites in T cells, we performed ChIP-seq in Treg and naïve Tconv cells (Supplementary Tables 1 and 2). Foxp1 and Foxp3 ChIP-seq analyses identified 3,071 and 7,147 sites significantly bound by Foxp1 and Foxp3 in Treg cells, respectively (FDR-adjusted P < 0.01, Fig. 1a). 1980 peaks were significantly bound by both Foxp1 and Foxp3, whereas 1,101 significantly bound Foxp3 peaks did not overlap with any Foxp1 peaks called at a relaxed threshold, and only 54 significantly bound Foxp1 peaks did not overlap with any Foxp3 peaks called at a relaxed threshold (Fig. 1a; see Methods). In naïve Tconv cells, we identified 1,088 significantly bound Foxp1 peaks (false discovery rate (FDR)-adjusted P < 0.01, Fig. 1b), over 95% of which were also bound by Foxp1 in Treg cells at a relaxed threshold (Fig. 1b; see Methods). While more than half of Foxp3-bound sites were in intronic and intergenic regions, less than 30% of Foxp1-bound sites were in such regions. Rather, Foxp1 binding was primarily at promoters in Treg cells (Fig. 1c). The remarkable overlap of Foxp3 and Foxp1 peaks is noteworthy, as it is highly improbable for two unrelated ChIP-seq profiles to highly correlate with each other15.
The ChIP signal for Foxp3 in Treg cells at the summits of shared peaks was slightly stronger than the ChIP signal at peak summits bound only by Foxp3, while the opposite was true for Foxp1 (Fig. 1d,e and Supplementary Fig. 1). To identify motifs enriched within the Foxp3 and Foxp1 peaks, we performed de novo motif discovery using HOMER16. The canonical forkhead motif was most significantly enriched in Foxp1 ChIP-seq peaks, while in Foxp3 peaks Ets and Runx motifs were more significantly enriched than the forkhead motif, in line with our previous study17 (Supplementary Fig. 1). Specifically, 12.6% of Foxp3 peaks and 17.7% of Foxp1 peaks contained the canonical forkhead motif (Fig. 1f,g). The forkhead motif occurred only once in the vast majority of peaks in which it was found. By accounting for all individual occurrences of the forkhead motif around peak summits using a relaxed motif hit threshold, we found 1,625 Foxp3 peaks out of 7,174 with at least one forkhead motif occurrence. However, only 102 Foxp3 peaks (1.4% of all) contained two or more occurrences of the forkhead motif, which is not more than expected by chance if the forkhead motifs were randomly placed among peaks (see Methods). Of these 102 peaks, 57 peaks contained the two forkhead motifs at a distance 20 base pairs (bp) or larger. Thus, it is unlikely that both forkhead domains of Foxp3 homodimers bind simultaneously to DNA at two closely adjacent forkhead motifs18, although they may mediate long-range chromatin interactions by bridging two separate DNA sites, which would require only one forkhead motif per regulatory element19. Finally, it is also possible that one dimer member mediates the interaction with DNA at a given time. Foxp1 and Foxp3 peaks contained Ets family motifs (26.7% and 30.8%, respectively) and 17.7% of Foxp3 peaks contained Runx motifs (Fig. 1f,g). The observation of a relatively small percentage of Foxp1 and Foxp3 peaks with a forkhead motif supports previous reports that other transcription factor(s) present in the Foxp3 protein complexes may contribute to DNA binding20,21. Thus, Foxp1 binds predominantly to the sites bound by Foxp3.
Foxp3 DNA binding is reduced in the absence of Foxp1
We considered that loss of Foxp1, and the resulting inability of Foxp3 to form heterodimeric protein complexes with Foxp1, could lead to a reduction in Foxp3 binding specifically at sites occupied by both Foxp3 and Foxp1. Alternatively, the reduced abundance of total Foxp (‘1 + 3’) protein due to loss of Foxp1 could lead to reduced Foxp3 binding at all genomic sites due to redistribution of Foxp3 protein. To distinguish between these two possibilities, we generated a novel conditional Foxp1 allele and combined it with the previously described Foxp3YFP-Cre allele to obtain mice with Treg cell-restricted Foxp1 deficiency (Supplementary Fig. 2). Foxp3 ChIP-seq analysis of Foxp1-deficient and -sufficient Treg cells using background noise normalization in all replicates for multiple sample comparison of ChIP signal (see Methods) revealed that the number of sites significantly bound by Foxp3 was reduced in Treg cells lacking Foxp1 (Fig. 2a). Furthermore, Foxp3 ChIP signal was predominantly reduced in Foxp1-deficient Treg cells at all Foxp3-bound sites and not only at sites bound by Foxp1 (Fig. 2b). This result suggested that Foxp1 deletion in Foxp3-expressing cells may lead to a redistribution of Foxp3 genome wide (Fig. 2c). Foxp1 expression in T cells is induced at the double negative 2 stage, and we observed that Foxp1 expression was high in Treg precursor cells (Fig. 2d).
To exclude the possibility that decreased Foxp3 genome occupancy might be due to a change in Foxp3 protein abundance, we assessed Foxp3 protein content in Foxp1-deficient and -sufficient resting CD62Lhi Treg cells and activated CD44hi Treg cells using flow cytometry. We observed a modest reduction in Foxp3 protein abundance in Foxp1-deficient naïve Treg cells and unaltered Foxp3 protein amounts in Foxp1-deficient activated Treg cells, compared with their Foxp1-sufficient counterparts (Fig. 2e,f). Foxp3 messenger RNA abundance was also reduced in Foxp1-deficient Treg cells, suggesting that the decrease of Foxp3 protein in Foxp1-deficient cells may be due to altered transcriptional regulation of Foxp3 (Supplementary Table 3). Foxp3 protein abundance and genome occupancy in highly activated Treg cells was increased in comparison with resting Treg cells (12,961 versus 7,497 Foxp3 peaks22; J. van der Veeken, Y.P. and A.Y.R, unpublished observations) while Foxp1 protein abundance was decreased. It is noteworthy that a larger percentage of Foxp1-deficient Treg cells exhibited an activated phenotype compared with Foxp1-sufficient Treg cells, and thus the majority of Foxp1-deficient Treg cells expressed more Foxp3 protein on a per cell basis than Foxp1-sufficient Treg cells. However, in contrast to Foxp1-sufficient activated Treg cells, Foxp1-deficient Treg cells exhibited fewer significant Foxp3 peaks than Foxp1-sufficient counterparts (3,935 versus 7,174 peaks). These observations are consistent with our proposed ‘dilution’ model for diminished Foxp3 binding in the absence of Foxp1.
Foxp1-dependent regulation of gene expression in Treg cells
To assess the impact of Foxp1 loss on Treg gene expression, we performed RNA-seq of YFP+ CD62Lhi naïve or CD44hi activated Treg cells (Supplementary Table 3). Since we observed that Foxp3YFP-CreFoxp1fl/fl mice exhibited moderate immune activation, we performed RNA-seq analysis of Treg cells isolated from healthy heterozygous Foxp3YFP-Cre/+ Foxp1fl/fl or Foxp3YFP-Cre/+ Foxp1+/+ female mice harboring both Foxp1-deficient and -sufficient Treg cells as the result of random X-chromosome inactivation. We performed k-means clustering of genes that were bound by Foxp1 in Treg cells and that were differentially expressed between Foxp1-deficient and Foxp1-sufficient Treg cells (FDR-adjusted P < 0.01) and performed gene ontology pathway analysis for the six clusters identified (Fig. 2g, Supplementary Fig. 3 and Supplementary Table 4). Cluster 3 included genes that were downregulated in the absence of Foxp1 in both naïve and activated Treg cells, including Treg signature genes Il2ra and Ikzf4. Cluster 1 was enriched for genes associated with skeletal and heart muscle, reproductive organ development and adrenal cortex formation. Foxp1 is known to play an essential role in these developmental processes23,24,25,26. Accordingly, we found a number of genes linked to development of these organs, including Arntl, Cited2, Hdac4, Nr4a1, Chd7, Dnajb6, H3f3b and Spo11, differentially expressed in Foxp1-deficient Treg cells and directly bound by Foxp1. Foxp1 binding to these gene loci have not been previously described, with the exception of Cited227. These findings suggest that Foxp1 may directly regulate expression of these genes and thereby affect both embryonic development and Treg cell function. Cluster 5 included genes that were more highly expressed in activated Treg cells and were downregulated in Foxp1-deficient Treg cells compared with Foxp1-sufficient Treg cells. This cluster included T cell activation genes Cd83, Ctla4, Ikzf3, Hivep3, Nfatc1, Nfkb2, Prdm1, Relb, Rora and Smad3, and was enriched for immune system development, leukocyte activation, regulation of gene expression and positive regulation of apoptotic signaling pathways.
Genes that were differentially expressed between Foxp1-sufficient and -deficient naïve and activated Treg cells and that were bound by Foxp3 more prominently in Foxp1-sufficient Treg cells included Itga6, Smad7 and Satb1 (Fig. 3a and Supplementary Tables 2 and 3). Most of these genes were also bound by Foxp1. Using a previously published Treg gene signature that identified 407 overexpressed and 196 underexpressed genes, we found 136 genes overexpressed (‘Treg up’) and 51 genes underexpressed (‘Treg down’) that were associated with Foxp1 or Foxp3 ChIP-seq peaks (Fig. 3b,c and Supplementary Table 5)28. The ‘Treg up’ genes were significantly more highly expressed in Foxp1-sufficient Treg cells, while the ‘Treg down’ genes were more highly expressed in Foxp1-deficient Treg cells (Fig. 3b). In Treg cells, expression of the genome organizer SATB1 is repressed in a Foxp3-dependent manner and this repression is important for their function29. Ectopic expression of SATB1 led to a loss of suppressive function and production of effector T cell cytokines30. Foxp1-deficient Treg cells expressed more SATB1 at the mRNA and protein level than wild-type Treg cells (Fig. 3c-e). To determine whether Foxp1 regulates Satb1 expression in Tconv cells, we measured SATB1 expression in Foxp1-deficient Tconv cells from Cd4Cre Foxp1fl/fl mice using flow cytometric analysis and found it to be unchanged (Supplementary Fig. 5). We found that Foxp1 and Foxp3 were both bound to two alternative forkhead motifs containing Satb1 promoters in Treg cells, and Foxp3 binding at these promoters was markedly diminished in Foxp1-deficient Treg cells (Fig. 3f). Together these data suggest that, in addition to Foxp3, Foxp1 plays a specific non-redundant role in repression of Satb1 in Treg cells.
Foxp1-deficient Treg cells produce effector cytokines
Treg cell-restricted Foxp1 deficiency resulted in mild lymphadenopathy (Fig. 4a), which was in contrast to the reduced cellularity of spleens and lymph nodes of Cd4CreFoxp1fl/fl T cells (Supplementary Fig. 4)11. The proportion of CD4+ T cells that were Foxp3+ in lymph nodes was reduced approximately twofold (Fig. 4b,c). Foxp1-deficient Treg cells were strongly activated on the basis of CD44 expression and exhibited higher proliferative activity than their Foxp1-sufficient counterparts, as measured by Ki-67 expression (Fig. 4d,e). Foxp1-deficient Treg cells also produced more interferon-γ (IFN-γ) on phorbol 12-myristate 13-acetate (PMA) and ionomycin stimulation in vitro in comparison with control Treg cells (Fig. 4f). These characteristics of activation were similar to those observed in Foxp1-deficient CD4+ and CD8+ T cells (Supplementary Fig. 4)11.
Activation of Foxp1-deficient CD4+ and CD8+ T cells was linked to elevated expression of IL-7 receptor (IL7R) due to derepression of the Il7ra locus and heightened IL-7-dependent proliferation11. Foxp3 is also known to regulate expression of IL7R α-chain (IL-7Rα). Thus, we investigated whether Foxp1 deficiency in Treg cells resulted in increased cell surface IL7R expression and proliferation in response to IL-7. We found that neither was affected by Foxp1 deficiency in Treg cells, probably because Foxp3 binding to the Il7ra locus was not reduced in the absence of Foxp1 (Supplementary Fig. 5). Thus, while Foxp1 deficiency results in activation of Tconv and Treg cells, the mechanisms underlying their activation appear to be distinct.
Foxp1 deficiency causes Treg cell functional impairment
The gene expression analysis of Foxp1-deficient Treg cells revealed that expression of several functionally important Treg cell signature genes, including Ctla4, Il2ra, Nt5e and Entpd1, was markedly reduced in comparison with Foxp1-sufficient Treg cells. At the protein level, Foxp1-deficient Treg cells in Foxp3YFP-CreFoxp1fl/fl mice expressed markedly lower amounts of CD25 and CTLA-4 (Fig. 4g). Notably, CD25 induction on T cell antigen receptor (TCR) stimulation was intact in Foxp1-deficient Tconv cells (Supplementary Fig. 5).
Thus, we investigated the suppressive function of Foxp1-deficient Treg cells. Indeed, CD4+ and CD8+ T cells were more activated and more proliferative in mice harboring Foxp1-deficient Treg cells than in control animals (Fig. 5a,b). We also observed increased production of IFN-γ and IL-2, but not IL-4 or IL-17, by CD4+ T cells and IFN-γ by CD8 T cells (Fig. 5c,d; data not shown). Next, we adoptively co-transferred Foxp1-deficient or -sufficient Treg cells with the T cells isolated from Foxp3DTR mice harboring diphtheria toxin receptor-expressing Treg cells into T cell-deficient recipients and treated them with diphtheria toxin to deplete the Foxp3DTR Treg cells. Mice that received Foxp1-deficient Treg cells along with Foxp3DTR cells lost weight similarly to mice that received only Foxp3DTR T cells, while mice that received Foxp1-sufficient Treg cells did not lose weight (Fig. 5e). The Tconv and CD8+ T cells from the former produced more IFN-γ, IL-2 and TNF on stimulation by PMA plus ionomycin in comparison with those from control recipients expression (Supplementary Fig. 6). We found that fewer Foxp1-deficient Treg cells remained at day 22 relative to Foxp1-sufficient Treg cells, although they maintained Foxp3 (Supplementary Fig. 6). We also observed markedly delayed growth of a syngeneic breast carcinoma orthotopically transplanted into the mammary fat pads of female Foxp3YFP-Cre Foxp1fl/fl versus littermate control mice (Fig. 5f). Collectively, these results suggest that Foxp1 is required for Treg cell suppressive function in vivo.
Cell-intrinsic effects of Foxp1 deficiency in Treg cells
To distinguish between cell-intrinsic and -extrinsic effects of Foxp1 deficiency in Treg cells, we analyzed Foxp3YFP-Cre/+ Foxp1fl/fl heterozygous female mice. Due to random X-chromosome inactivation, approximately half of the Treg cells in these mice were expected to express YFP. Indeed, when Foxp3YFP-Cre/+ Foxp1+/+ heterozygous female mice were examined, 40% of the Treg cells were YFP+, indicating that expression of the Foxp3YFP-Cre protein confers a slight disadvantage to Treg cells in a competitive setting (Fig. 6a). When we analyzed Foxp3YFP-Cre/+ Foxp1fl/fl mice, we found that Foxp1-deficient YFP+Foxp3+ thymocytes were not competitively disadvantaged relative to Foxp1-sufficient Foxp3+YFP+ thymocytes. However, when we analyzed Treg cells in the periphery we observed that Foxp1-deficient cells were severely outcompeted by Foxp1-sufficient ones (Fig. 6a,b). To gain insight into the mechanism underlying the reduced numbers of Foxp1-deficient Treg cells in a competitive setting we assessed activation, proliferation and apoptosis of YFP+ Foxp1-deficient Treg cells in these mosaic mice. Foxp1 deficiency was shown to lead to cell-intrinsic activation of CD4+ Tconv cells and CD8+ T cells12. In contrast, Foxp1-deficient and -sufficient Treg cells in heterozygous mice were activated to the same extent (Fig. 6c; data not shown). Next we assessed expression of key molecules implicated in Treg suppressor function in Foxp3YFP-Cre/+ Foxp1fl/fl females. We observed diminished expression of CTLA-4 and CD25 in these mice (Fig. 6d). However, contrary to comparable CD39 and CD73 expression by Foxp1-sufficient and Foxp1-deficient Treg cells in male Foxp3YFP-Cre Foxp1fl/fl and littermate control mice, their expression was noticeably reduced in Foxp1-deficient Treg cells in comparison with Foxp1-sufficient Treg cells in these mosaic mice (Fig. 6d). These experiments suggest that the aforementioned activated Treg phenotype observed in Foxp3YFP-Cre Foxp1fl/fl mice was cell extrinsic and driven by increased activation of CD4+ and CD8+ T cells in response to impaired Treg suppression.
Despite being comparably activated, YFP+ Foxp1-deficient Treg cells in Foxp3YFP-Cre/+ Foxp1fl/fl heterozygous females were more proliferative than Foxp1-sufficient Treg cells (Fig. 6e). Next, we examined the rate of apoptosis of YFP+ Foxp1-deficient and Foxp1-sufficient Treg cells in Foxp3YFP-Cre/+ Foxp1fl/fl heterozygous female mice using annexin V staining. The percentages of annexin V-positive but 7-amino-actinomycin D (7-AAD)-negative YFP+ Treg cells were increased in the absence of Foxp1, while frequencies of apoptotic cells within YFP– Treg cell subsets were comparable (Fig. 6f,g; data not shown). Thus, Foxp1 deficiency in Treg cells has both cell-intrinsic and -extrinsic effects.
IL-2 signaling is impaired in Foxp1-deficient Treg cells
IL-2 signaling is critical for Treg cell suppressive function and for competitive fitness of Treg cells31, both of which are impaired in the context of Foxp1 deficiency, suggesting that a major functional consequence of Foxp1 loss in Treg cells is dysregulation of CD25 expression. The diminished competitive fitness of Foxp1-deficient Treg cells was consistent with markedly diminished expression of CD25 in comparison with Foxp1-sufficient ones. The lower expression of CD25 was also observed in YFP+ Treg cells from Foxp3YFP-Cre/+ Foxp1fl/fl female mice, indicating that it was cell intrinsic (Fig. 7a,b). We also observed that Foxp1 and Foxp3 shared a binding site within the Il2ra locus. Although Foxp3 remained significantly bound to this site in the absence of Foxp1, Foxp3 binding at several other sites within the Il2ra locus was reduced or lost in Foxp1-deficient Treg cells (Fig. 7c).
To assess whether lower CD25 expression by Foxp1-deficient Treg cells resulted in lower IL-2 responsiveness, we stimulated Treg cells with varying concentrations of IL-2 and assessed STAT5 phosphorylation (Fig. 7d). We found that at low doses of IL-2 STAT5 was significantly less phosphorylated in Foxp1-deficient Treg cells but that this difference was absent at high doses (Fig. 7d). IL-2 receptor-dependent activation of the transcription factor STAT5 has an essential role in the suppressor function of Treg cells31. The STAT5 and TCR signaling pathways control largely distinct sets of genes and aspects of Treg cell suppressor activity31,32. We analyzed the expression of the STAT5 activation-dependent gene set in our RNA-seq datasets and found that genes upregulated on STAT5 activation were more highly expressed by Foxp1-sufficient Treg cells, whereas genes downregulated in a STAT5 activation-dependent manner were more highly expressed by Foxp1-deficient ones (Fig. 7e and Supplementary Table 5)31. Together, these results demonstrated that IL-2 signaling was severely impaired in Foxp1-deficient Treg cells. Next, we tested whether saturating amounts of IL-2 could also restore key Treg cell features affected by Foxp1 deficiency and observed a full restoration of CD25 and a pronounced partial rescue of Foxp3 and CTLA-4 expression in Foxp1-deficient Treg cells. In contrast, we did not observe a rescue of SATB1 expression pattern in these cells (Supplementary Fig. 7). Thus, Foxp1 has an essential non-redundant function in Treg cells by enforcing Foxp3-mediated regulation of expression of several characteristic Treg lineage genes that can be partially rescued by provision of potent IL-2 signaling.
Treg cells were first identified as a CD25+CD4+ T lymphocyte subset33. Expression of Foxp3 during Treg cell differentiation is associated with acquisition of distinct phenotypic and functional features, including high levels of CD25 and CTLA-4, and suppressor capacity and fitness2,3,34,35,36. Foxp3 is a member of the Foxp subfamily of transcription factors, which emerged as the result of two sequential ancestral gene duplication events during the invertebrate-to-vertebrate transition with a common ancestor, potentially giving rise to Foxp1 and the other three Foxp proteins37. Foxp1 plays an essential role in heart, central nervous system and lung development, in tumor progression in multiple cancer types23,26,38,39,40,41,42,43 and in immune cell differentiation10,11,12,44.
In this study we demonstrated that the close relative of Foxp3, Foxp1, exerts a non-redundant function in Treg cells. Our experiments show that persistent expression of Foxp3 in resting and activated Treg cells is not sufficient to supplant the function of Foxp1. Foxp1 and Foxp3 ChIP-seq analyses of Treg and Tconv cells revealed that Foxp1 was bound to a large subset of sites occupied by Foxp3. Foxp1 binding in Treg and Tconv cells was highly overlapping and predominantly observed at promoter regions of target genes, and very few Foxp1-occupied sites were not bound by Foxp3 in Treg cells. The remarkable co-occurrence of Foxp1 binding sites in both Tconv and Treg cells with those occupied by Foxp3 binding was in contrast to a previously reported loss of Foxo1 from a subset of Foxo1-bound sites in Tconv cells and gain of Foxp3 binding to these sites in Treg cells17. These results suggest that Foxp1 probably acts cooperatively and synergistically with Foxp3 in Treg cells rather than acting like Foxo1 as a ‘predecessor’ of Foxp3. Consistently, several reports describe antagonistic roles of Foxo1 and Foxp1 at certain loci, including Il7ra and Il912,44.
Despite the fact that Foxp1 was bound only to a subset of Foxp3-bound sites, Foxp1 loss resulted in diminished genome-wide occupancy of Foxp3. Foxp1 therefore maintains Foxp3 occupancy at its target genes, probably by binding to a subset of its targets and enabling Foxp3 complexes lacking Foxp1 to bind elsewhere. Thus, it seems reasonable to assume that loss of Foxp1 in Treg cells results in an overall diminution in Foxp(1 + 3) protein concentration resulting in Foxp3 complex ‘redistribution’ among the ‘Foxp1/3 shared’ and ‘Foxp3-only’ bound sites. This assumption is consistent with the observation that Foxp1 ‘assists’ Foxp3 binding to DNA, potentially through heterodimerization, and the observation that sites occupied by Foxp1 in naïve T cells remain bound by both Foxp1 and Foxp3 in Treg cells.
Although the DNA-binding domains of Foxp1 and Foxp3 are highly conserved, these proteins have pronounced sequence divergence at the N terminus; Foxp3 is also truncated at the C terminus in comparison with Foxp15. Foxp3 is a component of large and heterogeneous 400–800 kDa multiprotein complexes, which include Foxp1, Runx1 and its essential cofactor CBFβ and other DNA-binding transcription factors14,17. CBFβ deficiency in Treg cells had no effect on expression of CD25 or CTLA-4, but did lead to increased proportions of GITRhi, ICOShi and Ki-67+ cells, indicative of Treg cell activation45. In contrast to Runx1/CBFβ deficiency, Foxp1-deficient Treg cells had dysregulated expression of certain Foxp3-bound genes that constitute the core Treg cell signature, including CTLA-4 and CD25, and this dysregulation probably contributes to the observed impairment in Treg cell functionality.
Unlike the cell-intrinsic loss of quiescence observed on Foxp1 ablation in CD4+ and CD8+ T cells, activation of Foxp1-deficient Treg cells was not cell intrinsic. The activation state of Foxp1-sufficient and -deficient Treg cells was comparable in mosaic heterozygote Foxp3YFP-Cre/+ Foxp1fl/fl mice, as was IL7R expression. Since Foxp3-mediated repression of IL7R expression has been reported46,47, it is probable that Foxp3 is capable of repressing Il7ra expression in the absence of Foxp1. The observed Treg cell activation in the context of Foxp1 deficiency appeared to be in response to CD4+ and CD8+ T cell activation, which was probably caused by the impaired suppressor function of Foxp1-deficient Treg cells.
Cell-intrinsic effects of Foxp1 deficiency in Treg cells included increased apoptosis, decreased CD25, CTLA-4, CD73 and CD39 expression, and increased expression of the genome organizer SATB1. Previous studies have demonstrated an important role for CD25 and CTLA-4 expression and for SATB1 repression in functional competence of Treg cells: ablation of conditional Il2ra and Ctla4 alleles resulted in a loss of Treg cell function31,48. Furthermore, ectopic SATB1 expression in Treg cells resulted in a loss of suppressor function and increased pro-inflammatory cytokine production30. In agreement with these observations, Foxp1-deficient Treg cells relieved from SATB1 repression produced pro-inflammatory cytokines on TCR activation, potentially contributing to the observed functional impairment. Thus, Foxp1-mediated regulation of gene expression in Treg cells, including positive regulation of CD25 and IL-2 signaling, as well as of CTLA-4, CD39 and CD73, and negative regulation of SATB1 expression, probably contributes to its non-redundant role in Treg cell functionality.
Mice were bred and housed in the pathogen-free animal facility at Memorial Sloan Kettering Cancer Center and were used in accordance with institutional guidelines. Mouse experiments were approved by the Institutional Animal Care and Use Committee of Memorial Sloan Kettering Cancer Center. Unless otherwise noted, 8- to 10-week-old male and female sex-matched mice were used for all experiments. The Foxp1 conditional targeting construct was designed to use two site-specific recombinations in vivo: (1) Flp recombinase to delete the neor marker from the mouse germline, and (2) Cre recombinase to conditionally delete Foxp1 exon 11 flanked by two loxP sites. Foxp1fl/+ mice (C57BL/6/129 mixed background) were backcrossed with C57BL/6 mice for at least five generations. Genotypes of Foxp1fl/+, Foxp1fl/fl mice were determined by PCR amplification. The wild-type allele was identified by the production of a 376-bp PCR product and mutated Foxp1 allele was identified by the production of a 257-bp PCR product (Supplementary Table 6). Foxp1fl/fl mice were crossed with Foxp3YFP-Cre transgenic mice to generate Foxp3YFP-Cre/+Foxp1fl/+ female mice. These mice were bred to Foxp3+Foxp1λ male mice. Foxp3YFP-Cre/+Foxp1fl/fl and Foxp3YFP-Cre/+Foxp1+/+ female offspring and Foxp3YFP-CreFoxp1fl/fl and Foxp3YFP-CreFoxp1+/+ male offspring were used in experiments. In mammary tumor experiments, Foxp3YFP-Cre/YFP-CreFoxp1fl/fl mice were used as experimental mice and Foxp3YFP-Cre/+Foxp1fl/fl mice were used as control mice. Foxp1fl/fl mice were crossed with Cd4Cre transgenic mice to generate Cd4CreFoxp1fl/+ mice. These mice were crossed to generate Cd4CreFoxp1fl/fl experimental and Cd4CreFoxp1+/+ control mice. Tcrb–/–Tcrd−/− mice and C57BL/6 Ly5.1 mice were purchased from Jackson Laboratories and bred in-house. No blinding was performed during the mouse experiments.
Tissue lymphocyte isolation
For lymphocyte isolation from lung, mice were perfused with PBS and tissues were digested with collagenase A (1 mg ml−1) and DNase I (1 unit ml−1) for 45 min at 37 °C. For isolation of lamina propria lymphocytes, small and large intestines were first incubated with 1 mM EDTA and 1 mM dithiothreitol in PBS for 15 min at 37 °C. Samples were washed and incubated with collagenase A and DNase I for 40 min at 37 °C. Digested cell suspensions were filtered and lymphocyte-containing fractions were isolated using differential Percoll centrifugation.
Single-cell suspensions of cells were prepared from thymi, spleens, pooled axillary, brachial and inguinal lymph nodes, mesenteric lymph nodes, lungs and large and small intestines. For exclusion of dead cells, samples were first stained with Ghost Dye (Tonbo) cell viability reagent for 10 min in PBS on ice. Cells were then stained for 30 min on ice in FACS buffer (PBS, 2% fetal bovine serum, 2 mM EDTA) with antibodies against cell surface proteins. For downstream in vitro assays, cells were sorted into fetal bovine serum. For intracellular staining, cells were washed and fixed with Fix/Perm buffer (BD Biosciences) for 20 min on ice or with Foxp3-staining buffer (eBioscience) for 45 min on ice, washed and stained with antibodies against intracellular proteins (Supplementary Table 7). Stained cells were either analyzed on an LSRII flow cytometer (BD) or sorted using a FACSAria II cytometer (BD). Flow cytometry data were analyzed with FlowJo software (TreeStar).
Ex vivo stimulation
To measure T cell production of cytokines, single-cell suspensions were prepared from spleens, lymph nodes and mesenteric lymph nodes and cells were plated at 1 × 106 cells in complete RPMI in round-bottom 96-well plates and treated with PMA (50 ng ml−1; Sigma) and ionomycin (1 nM; Calbiochem) for 3 h in the presence of GolgiPlug (brefeldin A) and GolgiStop (monensin; BD Biosciences).
Ex vivo IL-2 responsiveness
To measure Treg cell responsiveness to IL-2, CD62L+ naïve Treg cells were isolated from lymphoid organs and plated at 1 × 105 cells in complete RPMI in round-bottom 96-well plates and treated with IL-2 for 15 min at 37 °C. Cells were prepared for flow cytometric analysis according to the BD Phosflow protocol for mouse splenocytes (BD Biosciences). Cells were stained with anti-STAT5(p-Y694) (BD Biosciences).
Ex vivo apoptosis
To measure T cell apoptosis, single-cell suspensions were prepared from spleens, lymph nodes and mesenteric lymph nodes and 1 × 106 cells were stained with annexin V and 7-AAD in Annexin V Binding Buffer (BD Biosciences).
CellTrace labeling and cell culture
YFP+ Treg cells or CD62L+CD4+ cells were sorted from pooled spleens and lymph nodes of Foxp3YFP-CreFoxp1+/+, Foxp3YFP-CreFoxp1fl/fl and Cd4CreFoxp1fl/fl mice, washed twice with PBS and incubated for 10 min at 37 °C at a density of 1 × 107 cells per ml in PBS with 1 μM CellTrace Violet (Invitrogen), then washed with complete RPMI. CellTrace-labeled cells were cultured for 7 days with or without recombinant mouse IL-7 (3 ng ml−1; R&D systems) in complete RPMI at 37 °C. CellTrace profiles, cell viability and maintenance of CD4 or Foxp3 expression were assessed on day 7 by flow cytometry.
IL-2 cell culture
Single-cell suspensions were prepared from spleens, and from pooled axillary, brachial and inguinal lymph nodes. Cells were cultured for 24 h with or without human IL-2 (500 U ml−1; NIH) in complete RPMI at 37 °C. Treg cell viability and expression of CD25, CTLA-4, Foxp3 and SATB1 were assessed by flow cytometry.
In vivo suppression assay
CD4+ T cells (4 × 105) isolated from Foxp3GFP-DTRLy5.1 were transferred together with YFP+ Treg cells (1 × 105) isolated from Foxp3YFP-CreFoxp1fl/fl or Foxp3YFP-Cre Foxp1+/+ mice into T cell-deficient recipients. Recipients were treated with dithiothreitol on day 0, day 1 and then every 4 days post-transfer for the duration of the experiment. Weight was monitored throughout the experiment. Treg cell frequencies in lymphoid and non-lymphoid tissues were analyzed at the experimental endpoint.
CD44hi or CD62Lhi YFP+ Foxp1+ or Foxp1– Treg cell populations were FACS sorted from Foxp3YFP-Cre/+Foxp1+/+ and Foxp3YFP-Cre/+Foxp1fl/fl mice directly into TriZol LS (ThermoFisher) and volume was adjusted post-sort. Samples were stored at −80 °C before RNA extraction. Total RNA was extracted and assessed for nucleic acid quantity and quality on the Agilent BioAnalyzer. SMARTer amplification with 12–16 cycles of PCR was followed by Illumina TruSeq paired-end library preparation following manufacturer’s protocols. Samples were sequenced on the Illumina HiSeq 2500 to an average depth of 30 million, 50-bp read pairs per sample.
Chromatin immunoprecipitation and sequencing
FACS-sorted cell populations (2 × 106 cells for Foxp3 ChIP-seq; 6 × 106 cells for Foxp1 ChIP-seq) were cross-linked in 1% paraformaldehyde in PBS for 10 min at 21 °C. The reaction was quenched by addition of 125 mM glycine. Cross-linked cells were lysed and digested nuclei were resuspended in nuclear lysis buffer containing 1% SDS. Cells were sonicated for three rounds of 10 × 30 s with a Bioruptor (Diagenode) on high setting. Following sonication, chromatin was diluted with nuclear lysis buffer to final SDS concentration of 0.1% and precleared for 1 h at 4 °C with rotation with protein A/G magnetic beads (Pierce). Precleared DNA (5%) was saved as input. Chromatin was incubated overnight at 4 °C with rotation with polyclonal antibodies against Foxp3 (Rudensky laboratory) or Foxp1 (EMD Millipore ABE68). Immunocomplexes were incubated with protein A/G magnetic beads for 1 h at 4 °C with rotation. Immunocomplexes were precipitated and washed. After washing, precipitated chromatin and input DNA was decross-linked overnight at 65 °C in the presence of proteinase K, and DNA fragments were isolated using Qiagen PCR-purification kit. Purified DNA was processed for sequencing on Illumina HiSeq 2500 to an average depth of 30 million, 50-bp read pairs per sample. The resulting ChIP-seq samples are summarized in Supplementary Table 1.
Mouse genome assembly mm10.GRCm38 was used for read alignments. Gene annotations from Ensembl (Mus_musculus.GRCm38.83) were used for all analysis. Unless otherwise noted, all analysis was done using R v3.4.0.
Reads were aligned to the mouse genome using HISAT2 v2.0.1-beta49 with default parameters including splice sites obtained from gene annotations. Uniquely aligned reads were extracted using grep with parameters ‘-v ‘NH:i:[2–9]’’ and SAMtools v1.250 with parameters ‘view -h -F 4 -q 20 -b’ and sorted and indexed using SAMtools.
For analysis of RNA-seq in naïve or activated Foxp1+ or Foxp1– Treg cell samples, reads aligned to genes were counted using Rsubread v1.22.351 in R v3.3.0. Read counting for RNA-seq in naïve and activated Tconv and Treg cells was done using R version 3.4.0, Rsubread v1.26.0.
Differential gene expression between any pair of cell states was assessed using DESeq2 v1.16.152, using default FDR adjustment of P values for multiple hypothesis testing. A total of 1,091 genes were detected as significantly differentially expressed between naïve Foxp1+ and Foxp1– Treg cells, and 2,353 genes were detected as significantly differentially expressed between activated Foxp1+ and Foxp1– Treg cells (FDR-adjusted P < 0.01), for a total of 2,881 differentially expressed genes in either comparison. For visualization in heat maps and for clustering, read counts in genes were log-transformed (using log2(count + 1)), and z-scores of the transformed counts per row were used. The log2fold change values were used to generate the cumulative distribution plot in Fig. 7e using STAT5-up and STAT5-down gene signatures listed in Supplementary Table 3. Significantly upregulated (163) or downregulated genes (108) in STAT5bCA Treg cells relative to their expression in control Treg cells were defined as expressed genes with a log2fold change in expression of at least 1, and a FDR-adjusted P ≤ 0.05 (GSE84553).
Transcription factor ChIP-seq datasets for Foxp3 and Foxp1 were generated for Foxp1+ and Foxp1– Treg cells and Foxp1+ naïve Tconv cells (described in ‘Chromatin immunoprecipitation and sequencing’ section).
Reads were aligned to the mouse genome using Bowtie2 v2.2.553 with parameters ‘--no-unal -X 500 --no-mixed --no-discordant’. Uniquely aligned reads were extracted using SAMtools v1.1 with parameters ‘view -h -bS -F 4 -q 20’ and sorted and indexed using SAMtools.
For peak calling, MACS2 v18.104.22.1686030954 was used with parameters ‘-f BAM -B --SPMR -g mm -p 0.1 --keep-dup ‘auto’ --call-summits’. For Foxp1, pooled Foxp1+ Treg cells and Foxp1+ Tconv cells ChIP-seq samples were used, with Foxp1− Treg cell samples as control, to obtain a list of 55,265 putative peaks on standard chromosomes (chr1–chr19, chrX, chrY, chrM), with multiple summits per peak. ChIP-seq fragment length was estimated by MACS2 as 198 bp (Supplementary Fig. 1a). To obtain peaks with significant coverage in Treg cells, ChIP-seq reads from each Foxp1+ Treg cell sample and from pooled control samples were counted in putative peaks using Rsubread v1.26.0, and DESeq2 v1.16.1 was applied to these counts, with FDR adjustment of P values for multiple hypothesis testing. To estimate sample size factors, DESeq2 was applied to read counts from the same samples in flanking regions of putative peaks, defined by shifting the putative peaks by 2 kilobases (kb) upstream and by 2 kb downstream and taking only those flanking regions that do not overlap with the original putative peaks. Values of log2fold change ChIP-seq for Foxp1 ChIP over control estimated using these sample size factors were used in plots in Figs. 1d,e, 2a and Supplementary Fig. 1b. This analysis resulted in 3,071 Foxp1 peaks in Treg cells at MACS2 q < 0.05 and DESeq2 FDR < 0.01 that contained 4,619 summits. The same procedure was applied to each Foxp1+ Tconv cell sample and pooled control samples and resulted in 1,088 Foxp1 peaks in Tconv cells. Out of 3,071 Foxp1 peaks identified in Treg cells, 881 peaks were also detected among 1,088 Foxp1 peaks in Tconv cells. However, out of these 3,071 Foxp1 peaks in Treg cells, 2,521 peaks were detected as Foxp1 peaks in Tconv cells at a relaxed threshold (MACS2 q < 0.05 and DESeq2 FDR < 0.1), while 550 were not. Out of 1,088 Foxp1 peaks in Tconv cells, 1,049 peaks were detected as Foxp1 peaks in Treg cells at a relaxed threshold (MACS2 q < 0.05 and DESeq2 FDR < 0.1), while 39 were not.
The same procedure for peak calling was applied to pooled Foxp3 ChIP-seq samples from Foxp1+ and Foxp1- Treg cells, using Foxp3 ChIP-seq in Foxp1+ and Foxp1– naïve Tconv cells as controls. This analysis resulted in 141,556 putative Foxp3 peaks on standard chromosomes, with estimated ChIP-seq fragment length also 198 bp (Supplementary Fig. 1a). The same procedure for significant ChIP-seq coverage assessment was applied to each Foxp3 ChIP-seq sample in Foxp1+ Treg cells and pooled control samples. The log2fold change ChIP-seq values of Foxp3 ChIP over control estimated using this procedure were used in plots in Figs. 1d,e, 2a and Supplementary Fig. 1b. This procedure resulted in 7,174 Foxp3 peaks in Foxp1+ Treg cells at MACS2 q < 0.05 and DESeq2 FDR < 0.01 that contained 13,997 summits. The same procedure applied to each Foxp3 ChIP-seq sample in Foxp1– Treg cells and pooled control samples resulted in 3,935 Foxp3 peaks in Foxp1– Treg cells.
Foxp3-only peaks were defined as those of 7,174 Foxp3 peaks in Foxp1+ Treg cells that did not overlap with any of the 55,265 putative Foxp1 peaks. This resulted in 1,101 Foxp3-only peaks that contained 1,956 summits. Foxp1-only peaks were defined as those of the 3,071 Foxp1 peaks in Treg cells that did not overlap with any of 141,556 putative Foxp3 peaks. This resulted in 54 Foxp1-only peaks that contained 66 summits.
Each peak was associated with the closest gene according to distance in genomic coordinates, if this distance was not more than 20 kb. Peaks within 2 kb from a transcription start site of any annotated transcript were classified as promoter peaks; the remaining peaks were classified as exonic if overlapping with any exon of any annotated transcript; the remaining peaks were classified as intronic if within a gene body of any annotated gene; the remaining peaks within 20 kb of a gene were classified as intergenic; all the remaining peaks were left unclassified in this classification.
All Foxp3 and Foxp1 peaks detected with the above procedure are listed in Supplementary Table 2.
We performed k-means clustering of RNA-seq in Foxp1-bound genes that were differentially expressed between naïve Foxp1+ and Foxp1– Treg cells or between activated Foxp1+ and Foxp1– Treg cells (total of 661 genes), using function kmeans() in R. Gene ontology enrichment analysis was run using GOstats v2.44.0 and GO.db v3.5.0 in R.
For significant Foxp3 peaks in Foxp1+ and Foxp1– Treg cells, we applied DESeq2 to read counts from all Foxp1+ and Foxp1– Treg cell samples. Library size factors were estimated using read counts in flanking regions of the peaks, as described above, resulting in balanced estimates (Supplementary Fig. 1c, left). Then these library size factors were used for each sample in the read count analysis in actual peaks (Supplementary Fig. 1c, right). At FDR < 0.1, we identified 38 peaks (associated with 20 unique genes) with significantly stronger Foxp3 ChIP-seq signal in Foxp1– Treg cell samples, and 94 peaks (associated with 53 unique genes) with significantly stronger Foxp3 ChIP-seq signal in Foxp1+ Treg cell samples. All significantly differentially expressed genes associated with any of these 94 peaks were used in the heat map in Fig. 3a. Cumulative distribution function plots of RNA-seq expression of genes associated with these differentially bound Foxp3 peaks between Foxp1+ and Foxp1– Treg cells are shown in Supplementary Fig 2b (first and third plots). The log2fold change Foxp3 ChIP-seq values of Foxp1– over Foxp1+ estimated as described here were used in plots in Fig. 2b. As a control in Fig. 2b, log2fold change Foxp3 ChIP-seq values in the flanking regions of the peaks were used. As an additional control, log2fold change Foxp3 ChIP-seq values in accessibility peaks were used. For this, we pooled all ATAC-seq samples obtained from GEO (GSE83315)22. We aligned them to the genome and called the peaks in the same manner as described above for ChIP-seq. This resulted in 101,224 accessibility peaks at MACS2 q < 0.01. A total of 98.9% of our significant Foxp3 peaks (in either Foxp1+ or Foxp1– Treg cells) overlapped with these ATAC-seq peaks.
Sequence motifs in 150 bp windows around peak summits were identified de novo with HOMER50, using script findMotifsGenome.pl with parameters ‘mm10 -p 1 -size given -len 6,8,10,12 -noknown -mset vertebrates’. For Foxp3, 20 motifs were identified as significantly enriched, and 17 motifs were identified as significantly enriched for Foxp1. The top 10 most significant motifs were shown in Supplementary Fig 1. Then, for the identified motifs, their positions in the peak summits were identified using FIMO55 with parameters ‘--no-qvalue --verbosity 1 --max-strand --thresh 0.01’. All motif occurrences of the most significant motifs (according to HOMER) with FIMO P < 2 × 10−4 are shown in Fig. 1. Runx family motifs were not detected as significantly enriched for Foxp1. We identified all individual occurrences of the forkhead motif in the Foxp3 peaks, and compared the number of peaks with more than two occurrences against a randomized number. For each randomization, we placed the same number of motif hits into the same number of peaks, placing each motif hit uniformly at random across peaks. We repeated this 10,000 times and calculated the expected number of peaks with two or more occurrences of the motif. The actual observed number of peaks with two or more occurrences of the forkhead motif was not larger than would be expected by chance.
All statistical analyses (excluding RNA-seq and ChIP-seq analyses) were performed using Prism7 (GraphPad Software). Differences between individual groups were analyzed for statistical significance using unpaired two-tailed Student’s t-tests. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001. If no *, the difference is not significant (P > 0.05). Error bars show s.e.m. centered on the mean.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The datasets generated during the current study are included in this resource (and its supplementary information files) or are available in the Gene Expression Omnibus repository with accession code GSE121279. Datasets are relevant to Figs. 1, 2, 3, 7 and Supplementary Figs. 1, 3, 5. The ATAC-seq data relevant to Fig. 2 can be found in GEO (GSE83315) and the STAT5BCA dataset relevant to Fig. 7 can be found in GEO (GSE84553). Detailed information on experimental design and reagents may be found in the Life Sciences Reporting Summary.
We thank the Memorial Sloan Kettering Cancer Center (MSKCC) integrated genomics operation for sequencing RNA-seq and ChIP-seq libraries and preprocessing data. We thank A.C. Scott and D. Tran for help with tissue lymphocyte isolation and experiment optimization and all members of the A.Y. Rudensky laboratory for helpful discussions and technical assistance. This study was supported by NIH grants R37 AI034206 (A.Y.R.), P30 CA008748 (A.Y.R.) and U01 HG007893 (A.Y.R. and C.S.L.), as well as the Hilton–Ludwig Cancer Prevention Initiative funded by the Conrad N. Hilton Foundation and Ludwig Cancer Research (A.Y.R.). A.Y.R. is an investigator with the Howard Hughes Medical Institute.