|Home | About | Journals | Submit | Contact Us | Français|
Regulatory T (Treg) cells, whose identity and function are defined by the transcription factor Foxp3, are indispensable for immune homeostasis. It is unclear whether Foxp3 exerts its Treg lineage specification function through active modification of the chromatin landscape and establishment of new enhancers or by exploiting a pre-existing enhancer landscape. Analysis of the chromatin accessibility of Foxp3-bound enhancers in Treg and Foxp3-negative T cells showed that Foxp3 was bound overwhelmingly to pre-accessible enhancers occupied by its cofactors in precursor cells or a structurally related predecessor. Furthermore, the bulk of Foxp3- bound Treg cell enhancers lacking in Foxp3− CD4+ cells became accessible upon T cell receptor activation prior to Foxp3 expression with only a small subset associated with several functionally important genes being exclusively Treg cell-specific. Thus, in a late cellular differentiation process Foxp3 defines Treg cell functionality in an “opportunistic” manner by largely exploiting the preformed enhancer network instead of establishing a new enhancer landscape.
Lineage-specifying transcription factors (TFs) are defined by their sufficiency and necessity to establish cell identity, coordinate cellular differentiation, and maintain developmentally established transcriptional programs. Differential use of regulatory elements defines most previously studied lineage specific gene expression programs (Odom et al., 2004; Heintzman et al., 2009; Heinz et al., 2010; Natoli, 2010; Thurman et al., 2012). Thus, it seems reasonable to suggest that lineage-specifying TFs establish distinct differentiated cell states by setting up novel enhancer repertoires (Mercer et al., 2011). On the other hand, some activation induced transcription factors such as the glucocorticoid receptor largely utilize pre-established enhancers to impart changes in gene expression (John et al., 2011). These considerations raise the question of whether a late-acting differentiation factor like Foxp3 exerts cell lineage specification function by actively remodeling the chromatin landscape and establishing a distinct new set of enhancers or by exploiting an enhancer landscape prepared in precursor cells by their earlier developmental history.
Foxp3, an X-chromosome encoded member of the forkhead TF family, controls differentiation and function of regulatory T (Treg) cells (Littman and Rudensky, 2010). This distinct and stable lineage of suppressive CD4+ T cells is characterized by a unique gene expression program and serves as a critical guardian of immune homeostasis (Josefowicz and Rudensky, 2009; Rubtsov et al., 2010). Treg cell depletion in normal adult mice results in a fatal lympho- and myeloproliferative disorder with widespread inflammatory lesions (Kim et al., 2007). Foxp3 is both necessary and sufficient to confer suppressor capacity to naïve CD4+ T cells (Fontenot et al., 2003; Hori et al., 2003; Khattri et al., 2003; Gavin et al., 2007). Foxp3 is induced during thymic differentiation or upon activation of peripheral CD4+ T cells in response to T cell receptor (TCR) stimulation in combination with several other signals including IL-2 and TGF-β. Furthermore, forced expression of Foxp3 confers suppressor function to Treg precursor cells and Foxp3 ablation in mature Treg cells results in loss of lineage identity and immunosuppressive phenotype (Fontenot et al., 2003; Williams and Rudensky, 2007). However, an understanding of how Foxp3 coordinates the differentiation of Treg cells and their distinct suppression program is lacking.
We examined chromatin accessibility of Foxp3 bound enhancers in Treg cells and Foxp3− CD4+ T cells, which serve as precursors during extra-thymic Treg cell generation. Genome-wide analysis of Foxp3 binding sites using chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) was combined with genome-wide analysis of enhancers using DNase I hypersensitive site sequencing (DNase-seq). We found that Foxp3 was bound overwhelmingly to enhancers already accessible in precursor CD4+Foxp3− T cells prior to Foxp3 expression with only 2% of all Foxp3 bound enhancers observed in Foxp3+ Treg cells, but not in resting Foxp3-negative T cells. However, even these seemingly Treg-specific sites were mostly established in a Foxp3-independent manner in response to TCR signaling except for a small subset of exclusively Treg-restricted enhancers found in several genes important for Treg cell function. Analysis of DNA sequences at Foxp3 binding sites identified a forkhead motif only in a small subset of these DNA regions suggesting cofactor contribution. High-resolution digital footprinting analysis revealed similar footprints in Foxp3 expressing Treg cells and Foxp3- negative CD4+ T cells for several Foxp3 cofactors supporting the notion that Foxp3 functions through pre-existing enhancers. Moreover, a related transcription factor Foxo1 appeared to serve as a predecessor at many Foxp3-binding loci in precursor cells and its displacement in Treg cells by Foxp3 resulted in downregulation of proximal genes.
Thus, Foxp3 does not substantially change the accessible chromatin landscape but rather binds at previously established enhancers with cofactors already present and establishes the Treg cell transcriptional and functional programs likely by modification of transcriptional activity of these enhancers and by recruiting additional nuclear factors. These results suggest that late-acting lineage specification transcription factors like Foxp3 can establish functionality and define identity of their corresponding cell type by exploiting a subset of enhancers pre-established in precursor cells and maintained in a poised state by a distinct set of cofactors and predecessors.
To test whether Treg cells exhibit a unique set of enhancers supporting their distinct function and phenotype we employed DNase-seq, which affords the most reliable assessment of genome-wide chromatin accessibility and regulatory element activity (Thurman et al., 2012). Thus, we isolated nuclei from FACS sorted purified Foxp3+Treg and Foxp3−CD4+ T cells and subjected them to DNase I digestion followed by high-throughput sequencing to find DNase hypersensitive sites (DHSs; Figures 1A, S1). We identified over 100,000 DNase I hypersensitive sites in Treg and CD4+Foxp3− T cells each corresponding to the most accessible regions in the genome (Figure 1B). Unexpectedly, the location and the extent of accessibility of over 99% of DHSs was similar between naïve CD4+ T cells and Treg cells implying that very few alterations in chromatin accessibility occur during differentiation of Treg cells from their precursors (Figures 1B, S1). DHSs are found at exons, promoters, CTCF-bound elements, and enhancers. Indeed, comparison of ChIP-seq datasets for a variety of histone marks vs. DNase-seq datasets showed extremely strong correlation between DHSs and histone marks that are traditionally associated with enhancers including H3K4me1 (ENCODE Consortium, 2012).
While most DHSs were comparable in precursor CD4+ T cells and Treg cells, a small, but likely important subset of DHSs (<1% of all DHSs; 679 total) were found in Treg cells, but not in naïve Foxp3−CD4+ T cells. These new sites were located near or within many Treg cell characteristic genes including Foxp3, Ctla4, and Helios (Figure 1B). In addition to those sites newly accessible in Treg cells, 250 DHSs prominent in naïve Foxp3−CD4+ T cells were markedly diminished in Treg cells. The observed differences in chromatin accessibility were associated with directionally consistent changes in the nearest gene’s expression (Figure 1D). In contrast to the nearly identical DHS landscape observed in Treg cells and their precursors, the genome-wide DHS landscape of B cells, a sister lineage of T cells, was markedly different from that of either T cell subsets (Figure 1C). Thus, these results suggest that Treg cells utilize predominantly T-cell specific enhancers present in precursor Foxp3-negative cells and very few novel enhancers emerge in Treg cells.
The observation that very few regulatory elements are accessible only in Treg cells raised the question as to whether Foxp3 binds preferentially to sites of accessible chromatin that are Treg-specific or to sites that are pre-accessible in Foxp3-negative precursor cells. To address this question we performed ChIP-seq analysis of the genome-wide binding sites of Foxp3 in purified Treg cells. Foxp3 was bound at over 2800 sites across the genome associated with more than 1400 genes (Figures 2A and S2). Foxp3 binding sites were highly enriched in the promoters and first introns of genes (p<10−77 and p<10−300, respectively; binomial test), as were DHSs (p<10−300 and p<10−300, respectively; binomial test), consistent with direct Foxp3 mediated transcriptional regulation of gene expression via binding to promoters and enhancers (Figure 2B, S3). Foxp3 bound loci were also highly enriched for genes differentially expressed in Treg cells relative to precursor CD4+ T cells (Figure 2C). We confirmed association of Foxp3 with up and down regulation of gene expression by stratifying the data across both Foxp3 peak ranks and the magnitude of gene expression changes (Figure S2). Foxp3 binding correlated with both up- and down-regulated transcripts in Treg cells known to be involved in T cell activation and regulation by GO ontology.
We next identified regulatory elements that are bound by Foxp3 and are similarly or differentially accessible between Foxp3−CD4+ T cells and Treg cells. To do this, we cross-referenced DNase-seq read counts with Foxp3 binding sites (Figure 2D). Consistent with the observed broad distribution of Foxp3 binding sites within the genome and overall similar chromatin architecture of Treg cells and their precursors, 98% of Foxp3 binding sites were observed to be already accessible, i.e. DHS positive in CD4+ T cells (Figure 2E). The remaining Foxp3 binding sites corresponded to Treg-specific enhancers including the previously identified CNS2 enhancer within the Foxp3 locus (Figure 2D) responsible for the auto-regulatory feed-forward loop for stabilizing Foxp3 expression (Zheng et al., 2010). A much larger proportion of Foxp3 binding was observed at T-cell specific DHS sites in contrast to Treg specific sites (29% T-cell specific, 2% Treg specific; Figures 2E,F). Upon cross-referencing our DHS-seq dataset and the recent analysis of histone H3 modifications (Wei et al., 2009) we observed significant overlap between the increased presence of H3K4me3 marks and newly accessible DHSs and between H3K27me3 marks and decreased chromatin accessibility (Figure S2L).
These results indicated that Foxp3 does not dramatically alter the chromatin accessibility landscape. Instead, it binds primarily to already accessible enhancers and promoters in order to coordinate the lineage-specific gene expression program of Treg cells consistent with the idea that Foxp3 exploits chromatin features established during differentiation of Treg precursor cells.
Given that Foxp3 predominantly binds to chromatin that is accessible in Foxp3−CD4+ T cells, we next wanted to know if proteins occupying these sites in precursor cells might serve as cofactors to facilitate Foxp3 recruitment to, or function at these sites. Foxp3 is known to physically interact with several transcription factors. To explore if any of these factors could associate with Foxp3 binding sites we examined Foxp3 binding regions for the presence of known and novel motifs. Significant enrichment for canonical ETS, RUNX, CNOT, and forkhead (FKHD) motifs (as defined by TRANSFAC and JASPAR) accounted for 58% of Foxp3 occupied sites (p < 10−270, comparison to flanking region; Fisher’s exact test), whereas the remaining peaks contained weaker motifs (Figure 3A, 3B, S4). As Foxp3 contains a forkhead domain that can bind to the canonical FKHD motif (Koh et al., 2009), it was surprising that ETS and RUNX motifs were substantially more enriched at Foxp3 binding sites than the FKHD motif (Figure 3B). These motifs were also enriched in the Foxp3 peaks with the highest rank by read count (Figure 3C). Furthermore, ETS and RUNX motifs were enriched at Foxp3 bound loci relative to genome-wide DHS sites (Figure S4B and S4C). It is noteworthy that our recent mass-spectrometric analysis identified ETS family members, Cnot3, and Runx1 and its cofactor Cbf-β as components of Foxp3 transcriptional complexes (Rudra et al., submitted).
Next, we employed ChIP-seq to test whether several transcription factors, whose DNA binding motifs were enriched at sites occupied by Foxp3, were bound to these sites. These experiments confirmed binding of ETS family members Ets-1 and Elf-1 as well as Runx/Cbf-β to Foxp3 occupied loci (Fig 3D); the binding was specifically enriched at Foxp3 sites with corresponding motifs (Figure 3E). These results suggested that Foxp3 and several transcription factors co-occupy a large portion of Foxp3 binding sites in Treg cells and raised the possibility that these cofactors are associated with the maintenance of preformed accessible sites in precursor cells and facilitate Foxp3 recruitment to these regulatory elements.
To better understand the role of DNA-binding cofactors in precursor cells, we wanted to determine if Foxp3 factors were directly associated with DNA already in naïve Foxp3−CD4+ T cells. Thus, we examined deeply sequenced DNase-seq libraries (~400 million reads each) for DHSs that contained cofactor motifs preferentially protected from DNase I cleavage, suggesting direct protein binding (Hesselberth et al., 2009). These digital genomic footprints (DGFs) of Foxp3 cofactors allowed genome-wide identification of specific protein-protected DNA footprints and provide evidence for specific DNA-protein interactions in situ.
Genome-wide DGFs containing a FKHD motif sorted by locus accessibility were observed to be enriched for Foxp3 binding sites (Figure 4A and S5). To gain further insights into these particular protein-DNA interactions the number of reads at individual nucleotides of accessible regions of the genome were aligned to the same motif and a corresponding map of contacts was observed consistent with conservation of the nucleotides and previously published crystal structures (Tahirov et al., 2001; Bandukwala et al., 2011). Footprints of distinct sizes and patterns were found for cofactor motifs including ETS, RUNX, and FKHD (Figure 4B).
We next wanted to know if these motifs were preferentially protected in Treg vs. Foxp3− CD4+ cells and whether Foxp3 presence was correlated with alterations of the footprint. To determine whether cofactors binding to these motifs were pre-bound to regulatory DNA elements prior to Treg cell differentiation and expression of Foxp3, DGFs were analyzed in precursor Foxp3−CD4+ cells (Figure 4B). Interestingly, similar footprints were observed for the Foxp3 binding sites containing ETS and RUNX as well as FKHD motifs in Treg cells and corresponding sites in Foxp3−CD4+ T cells suggesting that this group of Foxp3 binding sites is pre-bound and protected in precursor cells. Consistent with the results of digital footprinting, binding of Runx/Cbf-β and Ets family members to corresponding sites in precursor Foxp3−CD4+ cells was confirmed by ChIP-seq experiments, suggesting that Foxp3-independent cofactor binding was significantly enriched at sites of Foxp3 peaks when the corresponding cofactor motifs were present (Figure 4C). Furthermore, Elf1 and Foxp3 co-bound loci showed highly similar quantitative binding patterns of Elf1 in Treg and Foxp3−CD4+ cells (Figure 4D). Thus, at a large portion of its overall binding sites within the Treg cell genome Foxp3 binds indirectly likely via interactions with its cofactors pre-bound in precursor cells or directly via substantially weaker FKHD motif(s).
As Foxp3 binding sites are predominantly accessible in precursor cells and infrequent alterations of DNase footprints in Treg cells were observed, we hypothesized that Foxp3 cofactors might maintain the enhancer chromatin state to allow for Foxp3 binding during Treg cell lineage specification. Similar DGFs observed at Foxp3 binding sites containing FKHD motif in both Treg and Foxp3−CD4+ T cells pointed to the possibility that another forkhead transcription family member could serve as a “predecessor” at Foxp3 binding sites in Foxp3− CD4+ cells. Foxp3 in Treg cells could then displace this putative factor in a competitive manner leading to changes in gene expression. We considered the possibility that Foxo1, a forkhead TF family member, might serve a role of a Foxp3 predecessor bound to these sites in precursor cells because Foxo1 has been implicated in regulation of gene expression in both effector T cells and in Treg cells (Ouyang et al., 2009; 2010). Consistent with this idea ChIP-seq analysis of Foxo1 binding showed that in Foxp3−CD4+ T cells Foxo1 was bound to sites that were occupied by Foxp3 in Treg cells (Figure 5A). Furthermore, in Treg cells Foxo1 binding was preferentially decreased at Foxp3 bound, but not at unbound sites in comparison to Foxp3−CD4+ cells (Figure 5B). We also excluded the scenario that the observed decrease of Foxo1 at Foxp3 bound sites in Treg cells is due to an overall decrease in chromatin accessibility since Foxp3-bound Foxo1 sites were on average increased in accessibility (Figure S6).
Given that Foxo1 acts as a transcriptional regulator, its displacement could be an important component of the Foxp3-mediated program of gene repression. A corollary to this hypothesis is that corresponding genes in precursor cells should be either repressed or de-repressed and exhibit diminished or increased expression, respectively, in comparison to Treg cells, where Foxp3 displaces its putative FKHD-containing predecessor, Foxo1. Thus, we examined Foxp3 dependent gene expression changes by transcriptional profiling of Treg cells and their precursors expressing a functional (Foxp3GFP) and null Foxp3 reporter allele (Foxp3GFPKO), respectively. We found that Foxp3 bound gene loci with a decreased Foxo1 occupancy were significantly down-regulated in Treg cells in contrast to Foxp3 unbound genes or Foxp3 bound sites that have increased Foxo1 (Figure 5C).
The observed gene repression associated with displacement of Foxo1 by Foxp3 highlights its functional importance and implicates Foxo1 as a functionally relevant predecessor at Foxp3 direct binding sites in precursor cells. On a more general level, displacement of a transcriptional regulator by a structurally related TF with a similar DNA binding domain specificity represents a mode for implementation of gene expression programming during cellular differentiation.
We next wished to understand how apparently Treg-specific enhancers are established. We found that while an overwhelming majority of Foxp3 binding sites exhibited an open chromatin state in precursor CD4+ cells, the 679 loci identified by DNase-seq as newly accessible in Treg cells were enriched in genes known to be critical for Treg cell function. Additionally, Foxp3 binding to these new DHSs in Treg cells was significantly enriched in comparison to all DNase accessible loci genome-wide (6% and 2%, respectively).
We wished to examine if Foxp3 facilitates establishment of these new sites through recruitment of chromatin remodelers or if remodeling of these sites might precede Foxp3 expression and occur independently of it. To address these questions we first analyzed DNA sequence motifs present at Treg-specific DHSs. We found that they were highly enriched for the AP-1 septamer motif (Figure 6A) and the enrichment was prevalent in those that were directly bound by Foxp3. AP-1 and its binding partner NFAT are activated in T cells upon TCR signaling (Macian, 2005). Furthermore, thymic and extrathymic Treg cell differentiation and Foxp3 induction requires TCR stimulation by self or non-self antigens (Josefowicz et al., 2012). Thus, the presence of the AP-1 motif and necessity of TCR signaling for Treg cell differentiation suggests that TCR signaling may be playing an important role in the establishment of these enhancers in the Treg cell lineage even prior to Foxp3 expression. In contrast to AP-1 motif enrichment at Foxp3-bound sites newly accessible in Treg cells, sites with Treg cell-specific diminished accessibility were enriched for the RUNX motif along with a HMG motif associated with Lef1 and Tcf7 (Figure 6A).
In order to explore the role of TCR signaling in establishing Treg-specific accessible chromatin loci, we performed DNase-seq analysis of chromatin state in activated Foxp3−CD4+ T cells. For these experiments, activated Foxp3−CD4+ T cells were FACS purified from Foxp3DTR mice after human diphtheria toxin receptor (DTR)-expressing Treg cells were ablated upon DT injection. Massive T cell activation observed upon Treg cell ablation is dependent upon stimulation of TCR by self and environmental antigens (Kim et al., 2007). The vast majority of seemingly Treg-specific enhancers (>75%) acquired an increase in chromatin accessibility in activated cells suggesting that TCR dependent T cell activation without Foxp3 expression is sufficient to confer accessibility to these sites (Figure 6B,C). Accordingly, genes associated with Treg-specific DHSs were largely up-regulated in Treg “wannabe” cells expressing Foxp3GFPKO allele relative to Foxp3−CD4+ T cells (Figure 6D)(Gavin et al., 2007). These cells most likely received TCR signals and represent direct Treg cell precursors, reflected by Foxp3 locus activation, but lack of Foxp3 protein expression and suppressor function.
The Foxp3-independent expression of genes containing “Treg-specific” DHSs enriched for the AP-1 motif and chromatin accessibility at these sites in activated, but not resting Foxp3− CD4+ T cells suggested that activation of AP-1 most likely in cooperation with NFAT may account for these features. We next examined whether the impairment in NFAT activation upon ablation of calcineurin B1 in Foxp3CreCnb1fl/fl Treg cells disproportionally affected the genes containing “Treg-specific” DHSs. In the absence of calcineurin signaling, a prerequisite for NFAT activation and translocation to the nucleus, these genes had significantly decreased expression, particularly those that both contained an AP-1 motif and were Foxp3 bound (Figure 6E). Importantly, loss of NFAT activation in Treg cells led to a loss of their function and severe autoimmunity (data not shown).
The remaining subset of Foxp3 bound enhancers present in Treg cells, but not in precursor Foxp3−CD4+ or activated CD4+ T cells were found in a subset of genes encoding Ccr6, Lrrc32 (GARP), Foxp3, Itgb8, HOPX, Alk1, and PHLPP proteins whose functions in Treg cells include regulation of Foxp3 and TGF-β expression and activation, cell trafficking, and Akt signaling (Figure S7). Treg-specific DHS loci that are not more accessible in activated Foxp3− CD4+ T cells had markedly less AP-1 motif enrichment. In accordance with the presence of Treg-cell specific Foxp3-bound enhancers in these loci, these genes are expressed in a Foxp3- dependent manner in Treg cells.
In addition to the aforementioned genes with a known role in Treg cells, genes containing Treg-specific DHS sites included genes whose role in Treg cells is currently unknown. Together these genes may play an important role in Treg cell function under homeostatic and inflammatory conditions. Consistent with the latter idea, we found overlaps between Treg-specific DHS sites and several single nucleotide polymorphisms (SNPs) identified in genome-wide association studies as associated with a variety of autoimmune and inflammatory diseases (Figure S7). These include Il10 and Lrrc32 in ulcerative colitis, Ccr6 in rheumatoid arthritis and Rhoh in Graves’ disease (Stahl et al., 2010; Anderson et al., 2011; Chu et al., 2011). These results highlight the power of the datasets generated by combined DHS-seq and Foxp3 ChIP-seq analyses and might offer potential insights into the role of Treg cells in these diseases.
Taken together, our experiments suggest that over 98% of sites bound by Foxp3 in Treg cells are accessible in their precursors and are occupied by Foxp3 cofactors or Foxo1 serving as a Foxp3 “placeholder” and that TCR signaling is responsible for Foxp3-independent establishment of the remaining minor “Treg-specific” subset of Foxp3-bound enhancers. Thus, Foxp3 controls Treg cell differentiation and function by modulating gene expression upon binding to these preexistent enhancers without profound alterations in chromatin accessibility and enhancer repertoire.
Cell type specific gene expression and functional features of differentiated cells are established by genetically defined programs of specification, which employ epigenetic control, transcriptional and post-transcriptional regulation. Recent genome-wide studies of histone modifications indicate that cell identity is largely defined by permissive or repressive chromatin features present at enhancers (Heintzman et al., 2009; Mercer et al., 2011). We examined the late differentiation process of Treg lineage specification resulting from Foxp3 expression. Treg cell differentiation is known to occur at a late stage of thymocyte maturation and beyond the thymus as mature naïve Foxp3-negative CD4+ T cells acquire Foxp3 expression upon TCR stimulation under particular conditions that give rise to extrathymic Treg cells (Josefowicz et al., 2012). Our analysis of genome-wide DNase-seq and Foxp3 ChIP-seq datasets demonstrates that >99% of Foxp3 bound enhancers are accessible in precursor Foxp3-negative CD4+ resting or activated T cells. While there may be subtle undetected changes in chromatin accessibility imparted by Foxp3, DNase-seq technology offers the most accurate means of genome-wide enhancer analysis currently available (Boyle et al., 2008).
Interestingly, the lack of alterations in chromatin accessibility imparted by Foxp3 contrasts with early cellular differentiation processes, which are thought to rely on epigenetic modifications of chromatin at regulatory gene loci that can persist even after removal of the initiating factor (Cavalli and Paro, 1999). The establishment of heritable cell lineage-specific enhancer repertoires is critical for early tissue development (Cirillo et al., 2002; Heinz et al., 2010; Natoli, 2010; Mercer et al., 2011). Unlike early differentiation defined by lineage-specification factors, gene expression programs induced by some extracellular cues are driven by ligand- or signaling-dependent mobilization of latent transcription factors, which bind to preconditioned enhancers to alter gene expression without substantially altering the chromatin landscape (John et al., 2011). It must be noted that certain extracellular stimuli such as LPS that result in drastic changes in gene expression may be associated with considerable changes in chromatin states (Ghisletti et al., 2010; Smale, 2010).
Thus, features of late cellular differentiation exemplified by Foxp3-dependent generation of Treg cells are similar to those of responses triggered by extracellular stimuli like glucocorticoid steroid receptor ligand response in that both are associated with minimal chromatin remodeling (Biddie et al., 2011). Importantly, these results can explain the previously demonstrated need for continuous expression of Foxp3 for the maintenance of suppressive function and phenotypic features in fully differentiated Treg cells (Williams and Rudensky, 2007). This observation may also extend to the reported requirement for continuous expression for Pax5 in mature B cells, which may also be acting through modulation of preexisting enhancers instead of de novo heritable alterations in chromatin state (Cobaleda et al., 2007).
We also observed that while the vast majority of enhancers were accessible in precursor CD4+ cells, 2% of Foxp3 binding sites appeared to be accessible in a Treg-specific manner and were found at important Treg cell signature gene loci. However, a large majority of these sites emerged in activated T cells in a Foxp3 independent manner. These sites were highly enriched for a motif bound by AP-1, which is particularly interesting since AP-1 forms protein complexes with NFAT to jointly control expression of numerous genes downstream of TCR signaling (Macian, 2005). Thus, we propose that TCR activation driven mobilization of AP-1 likely in cooperation with NFAT facilitates chromatin remodeling at these “Treg-specific” enhancers in precursor cells and prepares sites for Foxp3 binding (Fig. 7c). Furthermore, X-ray crystallographic analysis has shown that Foxp3 forms protein complexes with NFAT and is thought to displace AP-1 in the NFAT complex in its DNA bound form (Wu et al., 2006), which suggests that Foxp3 might replace AP-1 at activation-dependent enhancers in Treg cells. Consistent with this idea we found altered expression of genes containing these enhancers in Treg cells subjected to ablation of a conditional Cnb1 allele resulting in impaired NFAT activation. In conjunction with a recent study showing that AP-1 maintains open chromatin, our data point to a general role for AP-1 as a pioneer factor capable of establishing an accessible chromatin state at regulatory elements in response to extracellular stimuli in diverse biological contexts including both cellular differentiation and activation (Biddie et al., 2011). We also note that T cell activation is able to substantially remodel chromatin at multiple functionally important gene loci in addition to TCR response elements which were previously characterized as primed to be pre-accessible for rapid T cell response (Barski et al., 2009).
Since Foxp3 seemed to not alter chromatin accessibility directly, we wished to better understand alternative mechanisms for how it implements the Treg cell lineage gene expression program. Analysis of sequence patterns showed that a canonical FKHD motif was present only in a minority of Foxp3 sites. In contrast, a majority of sites contained motifs for Foxp3 cofactors, including members of ETS and RUNX families of nuclear factors. Thus, Foxp3 may interact with DNA in large part indirectly, through protein-protein interactions. This result is consistent with the observed loss-of-function mutations of Foxp3 in the oligomerization leucine zipper domain in IPEX (immunodysregulation, polyendocrinopathy, enteropathy, X-linked) syndrome patients, which disrupts protein-protein interactions (Wildin et al., 2002; Lopes et al., 2006; van der Vliet and Nieuwenhuis, 2007). This is also consistent with Foxp3 DNA binding properties revealed by EMSA and crystallization studies (Koh et al., 2009; Bandukwala et al., 2011) and the large size of Foxp3 protein complexes, which contain Runx/Cbf-β and ETS family members among other cofactors (Rudra et al., submitted). Also, both Ets and Runx/Cbf-β play important roles in Treg cell differentiation and function that can now be explained in part by their cooperation with Foxp3 at its binding sites (Rudra et al., 2009; Mouly et al., 2010). Digital footprinting suggested that in precursor cells these cofactors are already present at sites of Foxp3 binding prior to expression of Foxp3 and this binding in precursor cells was confirmed by ChIP. Thus, Foxp3 may be recruited to these sites with prebound cofactors leading to conformational perturbation or recruitment of other factors and changes in gene expression (Fig. 7a).
In addition to co-binding and directly interacting cofactors, our results suggested a novel mode of transcriptional regulation through structural homologs of transcription factors acting as predecessors or placeholders. Analysis of Foxo1 and Foxp3 ChIP-seq and DNase-seq datasets indicated that Foxp3 in Treg cells displaced another family member Foxo1, which was bound to FKHD motif-containing enhancers in precursor cells (Fig. 7b). Importantly, transcripts of genes associated with these sites were predominantly down-regulated genome-wide in a Foxp3- dependent manner. Previously, a swap of structurally unrelated transcription factors at enhancers has been shown to cause a change in gene expression (Sun et al., 2002; Reichard et al., 2007). Thus, while Foxo1 and its relative Foxo3 enhance Treg cell function, possibly in part through Foxp3 induction (Ouyang et al., 2009; 2010), our results suggest an additional role for Foxo family members in establishing or maintaining enhancers for Foxp3. This mode of Foxp3 function raises an exciting possibility that members of a given family of transcription factors with distinct DNA binding specificity prepare or preserve enhancers in precursor cells during differentiation (or activation) to pass them on to another member of the same family, which through dislodgement of the predecessor imparts transcriptional activation or repression. This idea was supported by an earlier observation that the forkhead transcription factor family member FoxD3 binds at an enhancer within the Alb1 locus in embryonic stem cells and serves as a “placeholder” for FoxA1 and FoxA2 binding, which in turn promote chromatin remodeling and binding of GATA family members during definitive endoderm and hepatocyte differentiation (Xu et al., 2009). This “evolutionary” model, where a factor exploits highly similar structural domains for alternative means of transcriptional regulation, offers an intuitive mechanism for maintenance of enhancers in an “open” chromatin state and its engagement by distinct members of different families of transcription factors.
Finally, our datasets of Foxp3 targets and Treg cell-specific changes in enhancer repertoire identified a number of novel regulatory targets, which may inform further exploration of mechanisms and functions of Treg cells in human diseases. The potential importance of these regulatory elements and genes was suggested by the overlap of these datasets with SNPs associated with a variety of clinical inflammatory conditions.
In summation, our results suggest several modes for “opportunistic” control of Treg cell differentiation and function by Foxp3 through a network of preformed enhancers and cofactors operating in precursor cells. Foxp3 exploits the pre-existing enhancer landscape by binding to its cofactor-occupied enhancers accessible in precursor cells, by displacement of its putative predecessor forkhead family member Foxo1, and by binding to TCR stimulation dependent, but Foxp3-independent enhancers established during Treg cell differentiation. Thus, in contrast to early cellular differentiation characterized by alteration of chromatin accessibility at key enhancers, late differentiation relies overwhelmingly on a set of enhancers established during developmental history of precursor cells and on a minor set of enhancers established in precursor cells in response to extracellular cues promoting differentiation of a given cell type. However, given that even in early development, cells respond to extracellular cues from adjacent accessory cells and soluble ligand gradients, it may very well be the case that the mechanisms we describe here are also extensively employed during early developmental processes.
Foxp3GFPKO, Foxp3GFP-DTR, Foxp3YFP-Cre mice were previously described (Gavin et al., 2007; Kim et al., 2007; Rubtsov et al., 2008). All the mice were bred and housed in the specific pathogen-free animal facility at the Memorial Sloan-Kettering Cancer Center and used in accordance with institutional guidelines. Activated Foxp3-CD4+ T cells were sorted from spleens and lymph nodes of mice 10 days after two administrations of i.p. diphtheria toxin (DT) (Sigma). CD4 T cells were isolated by CD4 Dynabeads positive selection (Invitrogen) followed by sorting using an Aria2 flow cytometer (BD Biosciences).
Chromatin immunoprecipitation was performed as previously described (Zheng et al., 2007). Briefly, nuclei were isolated and lysed in 0.2% SDS followed by sonication to fragment DNA to 200–300 bp fragments (Branson). The chromatin was then incubated overnight with the appropriate antibody (polyclonal rabbit Foxp3 antibody (Zheng et al., 2007), Ets1, Elf1, and Cbf-β antibodies (all Santa Cruz clones C-20, C-20, FL-182). Precipitated chromatin was then washed, de-cross-linked, digested with proteinase K, and DNA was isolated using Qiagen PCR purification kit or phenol-chloroform extraction. ChIP was validated by qPCR for known targets of corresponding transcription factors. Sequencing libraries were prepared by ChIP-seq DNA prep kit and sequenced using a Genome Analyzer IIx or HiSeq (Illumina).
DNA reads generated in DNase-seq and Foxp3 ChIP-seq experiments were aligned to the UCSC mm9 genome using Bowtie allowing for 2 mismatches. Only uniquely aligning reads were analyzed. For ChIP experiments, all reads starting at an identical position were compressed to single reads to remove monoclonal reads. ChIP-seq peaks were called using SPP and peak-height was determined by number of reads that aligned to a 200bp window around peak center after strand-specific 75nt shift. Peaks with high input-control signal were excluded from subsequent analysis. This included peaks with input RPM greater than a threshold of 0.5RPM and peaks that were not enriched relative to input (p < 0.001) as determined by a Poisson distribution using a local estimate (200bp) of λ. Several total peak estimation methods were examined and many analyses were done using peak rank instead of discretized peak calls (Figure S2). The final number of peaks for Foxp3 was determined by the overlap in top 5000 peaks in both replicates, resulting in 2886 peaks. Peaks were assigned to genes by proximity to gene body defined by transcription start and end sites. When comparing multiple ChIP-seq experiments, peak heights were quantile normalized to account for potential differences in experimental data quality. When plotting tracks for multiple ChIP-seq experiments for a single transcription factor, a multiplicative factor was used to normalize for enrichment i.e. differing numbers of tags in peaks.
Treg and resting and activated Foxp3− T cells as well as B cells were assayed for chromatin accessibility using DNase-seq as described elsewhere (John et al., 2011). Briefly, intact nuclei were treated with DNase I, DNA was isolated following nuclear lysis, and fragments sized 300–1000bp were sequenced. DNase-seq data shown in scatter plots is in units of reads per million (RPM). Data were incremented by 0.3 (to smooth and avoid zeros), log-transformed, capped at 50RPM, and quantile normalized. DNase-seq peaks were called using the HotSpot algorithm (FDR 1%) (Sabo et al., 2004).
Differential accessibility was determined by an asymmetric cutoff for up- and down- regulation of accessibility based on an empirical 5% FDR derived from replicate-to-replicate differences and consistency in replicates (Figure S1A, B). These loci were then confirmed independently when comparing an additional replicate of Treg DNase-seq to two replicates of CD4 DNase-seq (Figure S1C). This analysis revealed differential sites that were ranked highly by rank expectation and negative binomial dispersion approaches (data not shown) (Robinson et al., 2010; Thomas et al., 2012).
Footprints were aggregated into heatmaps by aligning read start sites to specific motif instances (Figure S5). De novo footprints at individual loci were identified using previously described methods (Hesselberth et al., 2009; Neph et al., 2012). Conservation across placental mammals was analyzed using phyloP downloaded from UCSC genome browser.
For analysis of transcription factor binding motif enrichment, motifs were taken from JASPAR and TRANSFAC, whereas de novo motifs (AP-1 and HMG) were found by MDscan, cERMIT, and MEME motif discovery tools (Liu et al., 2002). To determine motif presence at a ChIP-seq or DNase-seq peak, motifs were scanned within 100bp regions around peak centers. Motif scores were defined as the log likelihood ratio of observing nucleotide frequency defined by the PSSM compared to genome-wide nucleotide frequencies. Motifs were considered present when the maximum PSSM score was greater than that found in 90% of the maximum scores of the flanking regions. This can be considered an empirical p-value of 0.1 on the distribution of maximum PSSM score over a given window. While this threshold may seem insufficiently stringent, when we examined regions considered positive for the motif, we found that nearly all sites (>90%) contained canonical 6mer binding motifs for the FKHD, RUNX, and ETS motifs. The few sites that lacked canonical sites had additional flanking nucleotides that increase statistical association.
GFP+ and GFP− CD4 T cells were isolated by CD4 positive selection (Invitrogen) followed by cell sorting using an Aria2 flow cytometer (BD Biosciences). To ensure high purity cells were sorted through consecutive rounds of sorting to attain >99% purity. Cells were resuspended in Trizol and RNA was isolated according to manufacturer instructions. cDNA libraries were amplified and hybridized to Affymetrix MOE 430 2.0 chips. Arrays were normalized using RMA and differentially expression was estimated using the limma package in Bioconductor. Genes were considered differentially expressed if they had a q-value <0.05 after Benjamini- Hochberg FDR estimation. Significance of differences in gene expression between sets of genes (e.g. those associated with regions of increased/decreased chromatin accessibility in Figure 1) was determined by a two-sample Kolmogorov-Smirnov (KS) test, where the background distribution was change in all expressed genes unless specified otherwise by dashed line emphasizing the two cumulative curves being compared.
All data are being deposited in the relevant public databases. Gene expression data will be uploaded to the GEO database and sequencing experiments to the SRA and ENCODE archives.
We would like to thank D. Rudra and A. Chaudhry for helpful discussions. This work was supported by NIH DK091968 grant, MSTP grant GM07739 (both R.M.S) and NIH 5R37AI034206 grant (A.Y.R.). A.Y.R. is an investigator with the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.