An NF-κB transcription factor-dependent, lineage specific transcriptional program promotes regulatory T cell identity and function

Canonical NF-κB signaling is crucial for Treg development

We explored the specific roles of the canonical NF-κB subunits c-Rel and p65 in natural (n)Treg and induced (i)Treg development by crossing mice with floxed Rel and Rela alleles with a Cd4^cre deleter strain resulting in loss of p65 and c-Rel expression (Figure S1A). We did not observe changes in development of CD4⁺ and CD8⁺ T cells in the thymus (data not shown). However, we observed a dramatic decrease in the proportion and number of CD4⁺CD8⁻FoxP3⁻CD25⁺GITR⁺ Treg precursors in mice lacking p65 and c-Rel (Figure 1A, B and S1B, C). There was a graded reduction in the frequency and number of FoxP3⁺ Treg cells upon deletion of Rel, Rela, or both. This demonstrated unique and partially redundant functions of both canonical NF-κB subunits in the development of Treg progenitors.

To assess whether NF-κB activation was required for the transition of Treg progenitors to FoxP3⁺ Treg cells, we isolated Treg progenitors and induced deletion of NF-κB subunits in vitro using TAT-CRE protein (Joshi et al., 2002; Lio and Hsieh, 2008). We observed a 3-fold reduction in Treg frequency in cells lacking Rel, or both Rela and Rel (Figure 1C and data not shown). Hence these results suggested an intrinsic, specific and non-redundant role for canonical NF-κB subunits in the specification of FoxP3⁻ Treg precursors and in the expression of FoxP3. Moreover, deletion of Rela alone led to a modest, but statistically significant, decrease in the proportion and numbers of Treg cells in both spleen and lymph nodes (LN), but not in other tissues (Figure 1D–F and S1D). Mice lacking Rel exhibited a dramatic decrease in Tregs frequency in all tissues. This was further amplified by the deletion of both p65 and c-Rel, demonstrating a partially redundant role of both NF-κB subunits in homeostasis of peripheral Treg cells. Finally, we assessed the potential role of each NF-κB subunit in iTreg induction in vitro. Upon stimulation with Interleukin 2 (IL-2) and Transforming Growth Factor-Beta (TGF-β), cells lacking Rela gave rise to normal proportions of FoxP3⁺ cells (Figure 1G). Naïve T cells lacking Rel exhibited a partial defect in iTreg induction that was rescued by increasing doses of TGFβ. However, full ablation of the NF-κB canonical pathway, by deletion of both Rela and Rel, almost completely abolished the in vitro differentiation of naïve T cells into iTreg cells. These results suggested that, although p65 and c-Rel partially compensated for one another, they also played discrete roles in multiple steps of both nTreg and iTreg development.

Treg-specific deletion of c-Rel leads to a late and mild inflammatory phenotype

To bypass the block imposed by loss of NF-κB on Treg development and assess the role of NF-κB subunits in the homeostasis and function of mature Treg cells, we deleted Rel in Tregs, but not Tconv cells, using the Foxp3^YFP-cre deleter strain (Figure S2A). We did not observe early lethality or systemic inflammation (Figure 2A and data not shown) in Foxp3^YFP-creRel^f/f mice. Although 6–8 week-old Foxp3^YFP-creRel^f/f mice did not show any overt signs of illness, lung and liver histology revealed a modest increase in lymphocyte infiltration relative to the WT controls (Figure 2B). Moreover, the cellularity of the peripheral LN, but not the spleen, was significantly increased (Figure 2C). The percentage and number of activated CD4⁺ and CD8⁺ T cells were similar between WT and Foxp3^YFP-creRel^f/f mice in spleen (Figure 2D–F), but there was a mild increase in CD8⁺ T cell activation in LN (Figure S2B). We observed a trend toward increased production of IL-17 and Interferon (IFN)- by splenic CD4⁺ T cells in Foxp3^YFP-creRel^f/f mice (Figure 2G, H). Also, production of IFNγ, but not IL-17, by CD4⁺ T cells in the colon lamina propria and skin was increased in Rel-deficient mice (Figure S2E, F). The proportion and number of Treg cells were unaffected by the absence of c-Rel (Figure 2I, J and S2C). This data showed that c-Rel in Tregs was mostly dispensable for the maintenance of immune tolerance despite its crucial role in thymic Treg development.

Although aged Foxp3^YFP-creRel^f/f mice did not exhibit any obvious morbidity, 8-month old mutant mice showed overt splenomegaly and lymphadenopathy (data not shown), and liver and lung histology revealed increased lymphocytic infiltrates (Figure S2G). We also observed an increase in the frequency of effector T cells with a skewing towards IFNγ secreting T helper 1 (Th1) cells in spleen and LN (Figure S2H). Finally, we tested the suppressive function of c-Rel deficient Treg in T cell transfer-induced colitis, in which naïve Tconv cells were transferred to Rag1^−/− recipients alone or with WT or mutant Tregs. In contrast with WT Tregs, Rel-deficient cells were completely unable to protect the recipient animals from weight loss or intestinal pathology (Figure 2K, L) suggesting a role for c-Rel in Treg suppressive function. Taken together, this data showed that c-Rel was largely dispensable for maintenance of immune tolerance but was important for regulation of Treg function in lymphopenic hosts.

Mice with Treg specific deletion of NF-κB p65 develop a severe inflammatory phenotype

To assess the potential role of p65 in Tregs, we crossed Rela-floxed mice with Foxp3^YFP-cre mice, which resulted in selective loss of p65 in Treg, but not Tconv cells (Figure S3A). Foxp3^YFP-creRela^f/f mice were smaller than their WT littermates (data not shown) and showed accelerated mortality, with all mice dying between 5–15 weeks after birth (Figure 3A). This was associated with lymphoproliferative disease including splenomegaly and lymphadenopathy, as well as lymphocytic infiltration of lung and liver (Figure 3B, C and data not shown). Dorsal skin sections from Foxp3^YFP-cre Rela^f/f mice revealed hyperplastic epidermis and a robust immune cell infiltration with significant epidermal thickening (Figure S3B, C). Moreover, there was a dramatic increase in frequencies and numbers of CD4⁺ and CD8⁺ T cells secreting IFN-γ and with an activated/memory phenotype (Figure 3D–H and S3D).

Despite the lymphoproliferative disease, the proportion and number of Treg cells were in fact increased in secondary lymphoid tissues (Figure 3I, J and S3E). p65-deficient Tregs displayed normal expression of CD25 and Helios, which are important factors for Treg homeostasis (Figure S3F). This suggested a defect in the suppressive function of Tregs lacking p65. Tregs from Foxp3^YFP-creRela^f/f mice were able to suppress proliferation of Tconv cells in vitro (Figure S3G). However, Rela^−/− Treg cells were completely unable to rescue recipient mice from colitis in an in vivo suppression assay (Figure 3K, L). This, together with the increased production of IFNγ and IL-17 by CD4⁺ T cells in multiple tissues of Foxp3^YFP-creRela^f/f mice (Figure S3H, I), demonstrated that p65 is essential for Treg-dependent maintenance of immune tolerance.

Complete ablation of canonical NF-κB in Tregs leads to a Scurfy-like phenotype

Our results (Figure 1) suggested that c-Rel and p65 exhibited partially redundant functions, and therefore to test this prediction we generated Foxp3^YFP-creRela^f/fRel^f/f mice (Figure S4A). Consistent with our hypothesis, these mice showed a profound Scurfy-like phenotype with robust cutaneous inflammation and severe runting (data not shown). Foxp3^YFP-creRela^f/fRel^f/f mice died between 2 and 4 weeks after birth (Figure 4A) and demonstrated pronounced lymphocytic infiltration in lungs and liver (Figure 4B), as well as severe splenomegaly and lymphadenopathy (Figure 4C). Foxp3^YFP-creRela^f/fRel^f/f mice had significantly higher numbers and frequencies of CD44^high central and effector memory CD4⁺ and CD8⁺ T cells (Figure 4D–F and S4B)). There was also an increase in the number and frequency of CD4⁺ Tconv cells from mutant mice that produced IFNγ (Figure 4G, H). In addition, we also detected an increase in antibodies recognizing double stranded DNA (anti-dsDNA antibodies) in Foxp3^YFP-creRela^f/fRel^f/f mice (Figure S4E), similar to observations in Scurfy mice (Aschermann et al., 2013). Thus, canonical NF-κB in Treg cells is required for cellular and humoral tolerance.

This phenotype observed in Foxp3^YFP-creRela^f/fRel^f/f mice suggested a profound loss of Treg homeostasis. Foxp3^YFP-creRela^f/fRel^f/f mice had decreased Treg frequency in all examined tissues (Figure 4I, J and S4C). Also, the Mean Fluorescence Intensity (MFI) of Helios and CD25 was decreased in NF-κB-deficient Treg cells (Figure S4D). We next investigated whether those Treg cells were functional in vivo. In contrast to WT cells, Rela^−/−Rel^−/− Tregs were completely unable to rescue Rag1^−/− mice from colitis induced by transfer of WT CD4⁺ Tconv cells (Figure 4K, L). Increased production of inflammatory cytokines in multiple tissues also confirmed the inability of Rela-Rel-deficient Treg cells to suppress effector and systemic inflammatory responses (Figure S4F, G). Thus, complete ablation of canonical NF-κB signaling impairs Treg homeostasis and function. This highlighted the crucial role of the canonical NF-κB pathway in the maintenance of immune tolerance.

Canonical NF-κB maintains the function and identity of mature Treg cells

We next investigated whether NF-κB was continuously required for mature Treg function. We induced conditional deletion of p65 and c-Rel in peripheral Tregs by administrating tamoxifen to Rela- and Rel-floxed mice crossed with Foxp3^{eGFP-Cre-ERT2}xRosa26^stop-eYFP mice (Rubtsov et al., 2010) (Figure 5A). In these mice, all Tregs express Green Fluorescent Protein (GFP), but tamoxifen treatment induces expression of Cre and Yellow Fluorescent Protein (YFP), allowing isolation of NF-κB deleted Treg cells by flow cytometry. We assessed Treg frequency seven days after initiating tamoxifen treatment. The percentages of FoxP3⁺ Tregs were significantly reduced upon deletion of Rel and Rel+Rela (Figure 5B), suggesting that NF-κB played a role in the maintenance of Tregs or Treg identity. Moreover, the MFI of staining for FoxP3, as well as CD25 and GITR, was decreased in Tregs following inducible deletion of Rel or both Rel and Rela, but not in cells in which only Rela was deleted (Figure 5C). We also observed a loss of Foxp3 mRNA expression in Rel+Rela deleted cells (Figure 5D). These data suggested that canonical NF-κB activity and, in particular c-Rel, was essential for the maintenance of the mature Treg population. We next tested the in vivo suppressive function of Treg cells with inducible deletion of Rel and Rela (induced-double knockout, iDKO Treg). Similar to Foxp3^YFP-creRela^f/fRel^f/f Tregs, iDKO (TMX-Foxp3^{eGFP-Cre-ERT2}Rosa26^stop-eYFPRela^f/fRel^f/f) Tregs failed to prevent development of colitis in Rag1^−/− mice (Figure 5E, F). This data established an essential role for NF-κB in Treg function in vivo and suggested the canonical NF-κB pathway may play an important role in the maintenance of Treg identity.

Canonical NF-κB maintains the Treg transcriptome

The finding that NF-κB was required for Treg suppressive function is consistent with NF-κB being a key mediator of transcriptional events downstream of TCR ligation. However, it is unclear how NF-κB fulfills this role in both Treg and Tconv cells. Our finding that loss of canonical NF-κB in Tregs resulted in a loss of Treg function led us to hypothesize that NF-κB controls a distinct, lineage specific, transcriptional program in Treg cells. To delineate the NF-κB-dependent transcriptomes in Tregs we used high-throughput RNA-sequencing (RNAseq) analysis to compare the global gene expression patterns between Tregs of various genotypes and WT Tconv cells stimulated or not with anti-CD3 and anti-CD28. Unsupervised hierarchical clustering (Figure 6A) of the data by genotype indicated a profound effect of NF-κB on the Treg transcriptome. As expected, based on the disease phenotype observed, single deletions of Rela or Rel induced largely distinct perturbations in gene expression. Rela^−/−Rel^−/− Tregs were distinct from WT Tregs, Rela^−/− and Rel^−/− Tregs, clustering as an outlier group. Furthermore, we found that stimulated and unstimulated Tregs clustered together, segregating by genotype rather than by presence or absence of stimulation. When additional published Treg and Tconv datasets were included in clustering (Wakamatsu et al., 2013), Rela^−/−Rel^−/− Tregs clustered together with WT Tconv cells, emphasizing the loss of Treg signature gene expression upon NF-κB deletion (Figure S5A). The distinct set of genes in Rela^−/−Rel^−/− containing loss of Treg signature genes was also observed in the clustering of our RNAseq dataset without public data, using the genes with ANOVA fold change higher than 1 (P < 0.05) (Figure S5B). Principal component (PC) analysis confirmed that Rela^−/−Rel^−/− Tregs exhibit a distinct transcriptome in comparison to either single knockout or WT Tregs (Figure S5C). While the first PC, accounting for 32.9% of the variance, established Rela^−/−Rel^−/− Tregs as having a divergent transcriptional profile, PC2 and PC3, together accounting for 31.2% of the variance, emphasized the transcriptional similarity between Rela^−/−Rel^−/− Treg and Tconv cells (Figure S5C). Consistent with both the PC analysis and clustering, we found that gene expression was equally impaired in unstimulated and stimulated Rela^−/−Rel^−/− Tregs (Figure 6B).

Although NF-κB can both activate and repress gene expression, c-Rel and p65 are generally considered positive regulators of transcription. However, almost equal numbers of genes were up- and down-regulated upon loss of these NF-κB subunits in Tregs (Figure 6C). We also found that the majority of genes differentially expressed in Rela^−/−Rel^−/− vs WT Tregs cells were not significantly altered in either Rel or Rela-deficient Tregs (Figure 6C). This, together with the partial phenotypic overlap resulting from Treg specific gene deletion (Figures 2 and 3), suggested considerable compensation between p65 and c-Rel. We found that numerous Treg signature genes were downregulated, and conversely expression of inflammatory cytokine transcripts increased, in Rela^−/−Rel^−/− Tregs compared to WT Tregs in both stimulated and unstimulated cells (Figure 6B, D). In particular, the expression of Ifng, Il17a, Il2, Il4 and Il13 was increased at the mRNA and protein levels in Rela^−/−Rel^−/− Tregs (Figure 6E, S5F). Gene set enrichment analysis (GSEA) revealed significant enrichment of Treg genes as well as chemokine, cytokine and transcription factor genes (Figure S5E). Within the Treg gene set, expression of Foxp3 as well as two transcription factors that act cooperatively with FoxP3, Ikzf2 (Helios) and Ikzf4 (Eos) were reduced in Tregs from Foxp3^YFP-creRela^f/fRel^f/f mice (Figure 6D, left panel, and S5F). Similarly, the Treg signature genes, Cd83, Gpr83 and Lrrc32 (GARP), were also downregulated in Rela^−/−Rel^−/− Tregs. Additional genes implicated in Treg homeostasis that were significantly altered in Rela^−/−Rel^−/− Tregs included chemokines and chemokine receptors, e.g. Ccr7 and Ccr6 (Figure S5E). These results suggested that deficiencies in Treg localization and recruitment may potentially explain why the defect in function of Treg lacking NF-κB manifested in vivo, but not in vitro. We also noted that several cytokines such as Il22, Il4 and Il13, which are not expressed in WT Treg cells, were upregulated in Rela^−/−Rel^−/− Tregs and further increased upon TCR stimulation (Figure 6B, D and Figure S5D–F) As expression of Foxp3 was down-regulated in Rela^−/−Rel^−/− Treg cells, we also assessed whether the impairment of Rela^−/−Rel^−/− Treg transcriptome could be the consequence of FoxP3 loss rather than a direct effect of NF-κB ablation. Comparison of FoxP3-dependent genes (Gavin et al., 2007) with Rela^−/−Rel^−/− -affected genes revealed only a partial overlap (Figure S5G). While some prototypic Treg genes with decreased expression in Rela^−/−Rel^−/− Tregs, such as Ctla4 or Ikzf2, were FoxP3-dependent, GSEA of the 2653 Rela^−/−Rel^−/− -affected FoxP3-independent genes also highlighted a loss of the Treg signature (Figure S5H). Thus, ablation of NF-κB led to a loss of Treg identity not only indirectly through the down-regulation of FoxP3 expression but also by directly affecting expression of NF-κB-dependent genes. As Foxp3^YFP-creRela^f/fRel^f/f mice develop an early systemic inflammatory syndrome, we then asked whether the altered gene expression in Rela^−/−Rel^−/− Tregs was due to the intrinsic absence of NF-κB in the cells, or due to inflammatory cytokines in the local environment. The modest but significant increase of the expression of Ifng, Il5 and Il22 in Rel-deficient Tregs was a first indication that loss of NF-κB subunits led towards a Teff-like phenotype independent of local inflammation as young Foxp3^YFP-creRel^f/f mice do not display any autoimmune phenotype (Figure 2). To further analyze the effect of NF-κB-ablation independently of the local milieu, we deleted floxed Rela and Rel in vitro, using TAT-Cre. Gene expression patterns of control and iDKO Tregs were then compared by RNAseq. As observed upon in vivo deletion of NF-κB, iDKO Tregs exhibited increased expression of various inflammatory cytokines such as Il17a, Il22, Ifng and Il4 (Figure 6F). Moreover, some hallmarks of Treg identity, such as Foxp3, Cd83 and Id3, were also down-regulated. This demonstrated that NF-κB exerts a cell-intrinsic effect to maintain Treg identity and prevent the expression of Tconv-specific genes. Taken together, this analysis suggested that NF-κB regulates distinct transcriptional programs in Treg and Tconv cells and demonstrated an important role for NF-κB in maintenance of a lineage specific transcriptome in Tregs.

NF-κB p65 directly coordinated the expression of Treg-signature genes

To characterize lineage specific, differential DNA binding patterns by NF-κB in Treg and Tconv cells, we used chromatin immunoprecipitation with massively parallel DNA sequencing (ChIPseq). After identifying putative p65 occupancy sites using the Irreproducible Disovery Rate (IDR) framework (Li, 2011) as outlined by ENCODE, we compared the number of peaks between populations. A fraction of peaks were detected both with and without stimulation, and these were mostly overlapping between Treg and Tconv (Figure 7A). To define a more rigorous set of overlapping regions putatively bound by p65 across all four conditions, we identified peaks in stimulated Treg cells that were also observed in stimulated Tconv cells, as well as in both unstimulated Treg and Tconv cells. We named this subset of 1571 peaks associated with 1297 genes “constitutive” because of the relatively high enrichment compared to input across all 4 samples (Figure 7B). GSEA on these “constitutive” p65-bound genes showed genes such as cell cycle-related genes (e.g. Cdk4 and Birc5) (Figure S6A). Interestingly, at a number of peaks, binding intensity was strongly increased upon stimulation in both Tconv and Treg similarly, confirming that NF-κB activation and binding to chromatin is induced upon TCR+CD28 signaling (Figure 7A, B). We defined a set of shared peaks in both Treg and Tconv stimulated cells that were also differentially enriched in stimulated versus unstimulated Treg cells, but not significantly differentially occupied between Treg and Tconv (see Methods). We named this subset “Treg-Tcon shared” and found that this gene set corresponding to 3552 genes was enriched for genes such as cytokines and cytokine receptors, e.g Tnf, as well as known NF-κB-induced genes such as Nfkbia (IκBα) (Figure S6B). We also detected a number of peaks that were significantly higher in stimulated Treg compared to Tconv. This final set of p65 peaks specific to stimulated Treg cells was defined by identifying stimulated Treg binding sites differentially enriched over both unstimulated Treg and stimulated Tconv cells. Of the 432 genes putatively associated with these 859 peaks, several fall within the well-described set of Treg signature genes such as Foxp3, Ikzf2 and Lrrc32 (Figure 7F). GSEA of this “Treg-stim enriched” population was enriched in genes highly expressed in Treg versus Tconv (Figure S6C). Only a few peaks were specifically detected in stimulated Tconv but not Treg cells, suggesting that p65 binding could be influenced by a Treg-specific environment that does not exist in a Tconv-specific landscape. The distribution of peak distances from the nearest transcription start site (TSS) and the distribution of genomic annotations associated with those peaks showed significant variation between different samples (Figure S6D). In contrast with constitutive peaks, Treg stim-enriched peaks showed marked increase in the fraction of peaks located in distal intergenic regions, suggesting that they might be binding to putative enhancers. These peaks contributed to the overall increase in Treg stim peaks (Figure 7A). Our data demonstrated that TCR+CD28-induced NF-κB binding is different in Treg and Tconv, suggesting that the Treg-lineage environment influences NF-κB binding and function.

We next looked for p65 binding motif enrichment in each gene subset using SeqGL (Setty and Leslie, 2015), as described in Methods. We found an enrichment for a Rel Homology domain-like motif nearby both Treg stim-enriched and Treg-Tcon shared summits (Figure 7C). The peaks shared by Tconv and Treg were enriched with motifs for various transcription factors such as CREB, ETS and RUNX1 in addition to NF-κB REL-binding motifs (Figure 7C). On the other hand, Treg-enriched peaks were predominantly enriched only with REL-binding motifs. This suggests that preferential binding of p65 to those sites in Treg over Tconv may not be due to unique unconventional binding motifs.

We then analyzed whether Treg-specific p65-binding could simply rely on a differential chromatin accessibility in Treg and Tconv cells. DNase hypersensitivity (HS) experiments have previously demonstrated a discrete chromatin accessibility in Treg and Tconv (Samstein et al., 2012). To assess a possible role for accessibility in p65-binding, we aligned their DNase HS sequencing data in Treg and Tconv cells to our p65 ChIPseq dataset. Interestingly, in Treg-enriched peaks, chromatin accessibility was significantly higher in Treg versus Tconv, whereas the opposite trend was observed in Treg-Tconv shared peaks (Figure 7D). This was illustrated by higher DNase HS signals in Treg at numerous Treg-associated genes, such as Foxp3, Ikzf2 or Lrrc32 (Figure 7G). Similar results were obtained when using a recently reported (Kitagawa et al., 2017) ATAC-Seq dataset (Figure 7E). This correlation suggests that chromatin accessibility yields increased p65 binding at Treg-enriched loci, resulting NF-κB-dependent regulation of the Treg specific transcriptional program.

To further elucidate the direct contribution of p65 binding to the differential gene expression of Tconv and Treg cells, we next focused our analysis on Treg and Tconv specific genes. We first defined a set of 587 “Treg signature” genes whose expression was significantly higher in WT Treg than WT Tconv, based on public datasets (Ye et al., 2014) (Figure 7F). As a counterpart, we used a “Tconv signature” set composed of 211 genes, as well as the total coding genome. We assessed the potential association between p65 binding at Treg-stim enriched, Treg-Tcon shared and constitutive peaks, FoxP3 binding peaks (Samstein et al., 2012), Tconv—Treg and Treg Rela^−/−Rel^−/− —WT differentially expressed genes, and Tconv—Treg signature genes. We found a strong enrichment of Treg-specific p65 binding in Treg-signature genes. Expression of many of these genes was dependent on NF-κB, as their expression was down-regulated in Rela^−/−Rel^−/− Tregs (Figure 7F). Interestingly, the fraction of genes bound by both FoxP3 and p65 was higher in Treg-signature genes than in the rest of the genome (Figure 7F and S6E), further underlining the importance of these transcription factors in Treg-specific gene expression. This was illustrated by FoxP3 binding in the vicinity of p65 binding at Foxp3, Ikzf2 and Lrrc32 loci (Figure 7G). Taken together, our results showed that NF-κB directly directed the expression of Treg-essential genes by accessing specific, open loci in Treg cells.

CONTACT FOR REAGENT AND RESCOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Sankar Ghosh (sg2715@cumc.columbia.edu).

EXPERIMENTAL MODELS

METHOD DETAILS

QUNATIFICATION AND STATISTICAL ANALYSIS

We used the unpaired Student t-test for two comparison groups. N represents the number of animals. Detailed description of statistics for each figure can be found in the figure legend. 5-way ANOVA was used to analyze the replicates of RNAseq.

To compare DNase I hypersensitivity signals in Treg and Tconv, we calculated the mean, quantile normalized, library-size normalized coverage signals using the ENCSR000COC and ENCSR000COA, respectively, from data made available by the mouse ENCODE Project. We calculated Mann-Whitney U test p-values comparing the Treg to Tconv signal over +/−150bp around peak summits for each peak set. We evaluated statistically associated features within the context of the previously described 587 Treg and 211 Tconv signature genes. One-sided Fisher exact tests were performed comparing the count of genes with affirmative binary status with regards to presence of an associated peak (within upstream 10kb to downstream 5kb), significant over-expression in Treg compared Tconv or WT compared to c-Rel/p65 DKO, or a member of the Treg signature instead of the Tconv signature.

KEY RESOURCES TABLE