TGF-β signaling controls Foxp3 methylation and T reg cell differentiation by modulating Uhrf1 activity

Regulatory T (T reg) cells maintain immune system homeostasis by keeping a delicate balance between activation and suppression (Johanns et al., 2010; Dias et al., 2017). T reg cells can differentiate in the thymus and migrate to peripheral tissues as natural T reg (nT reg) cells (Curotto de Lafaille and Lafaille, 2009). In addition, induced T reg (iT reg) cells differentiate from naive CD4⁺ T cells in the periphery under a variety of conditions (Workman et al., 2009; Bilate and Lafaille, 2012). Although they have different gene expression and epigenetic patterns (Haribhai et al., 2011), both types of T reg cell express the master transcription factor Foxp3 (Chen et al., 2011). Continuous expression of Foxp3 maintains full T reg cell suppressive capacity (Williams and Rudensky, 2007), and Foxp3 is sufficient to confer T reg cell–like suppressive function in conventional T cells by ectopic expression (Fontenot et al., 2003; Hori et al., 2003). Thus, Foxp3 is crucial for T reg cell differentiation and inhibitory function (Williams and Rudensky, 2007; Lee and Lee, 2018).

Epigenetic modifications regulate gene expression in a heritable manner without affecting genomic sequences (Trerotola et al., 2015). DNA methylation is the most well-established form of epigenetic regulation in T reg cell differentiation and controls gene expression, which is mediated by DNA methyltransferases, such as Dnmt1 and Dnmt3a/b (Liang et al., 2002; Razin and Kantor, 2005; Chen et al., 2011). It has been demonstrated that regulation of DNA methylation is indispensable for Foxp3 expression, lineage determination, and the maintenance of T reg cells (Kim and Leonard, 2007; Ohkura et al., 2013; Huehn and Beyer, 2015). Within the Foxp3 locus, a proximal promoter and three intronic enhancers that are designated as conserved noncoding sequences (CNSs) are the primary targets of epigenetic regulation and are essential for modulating its expression (Kim and Leonard, 2007; Huehn and Beyer, 2015). The Foxp3 promoter CpG motifs are almost completely demethylated in both nT reg cells and iT reg cells, while conventional CD4⁺ T cells have much heavier methylation in the promoter region (Zheng et al., 2010; Huehn and Beyer, 2015). CNS2 contains CpG islands that are highly demethylated only in committed T reg cells, and demethylation of this region is required for stable expression of Foxp3 (Floess et al., 2007; Wieczorek et al., 2009; Yue et al., 2016; Someya et al., 2017).

TGF-β signaling has essential roles in the transcriptional regulation of Foxp3 expression and T reg cell suppressive function (Chen et al., 2003; Marie et al., 2005; Tone et al., 2008). Smad2 and Smad3 are phosphorylated and activated by TGF-β and can subsequently form a heterotrimer with Smad4 (Souchelnytskyi et al., 1997; Derynck and Zhang, 2003). This heterotrimer binds to CNS1 in the Foxp3 locus, which is essential for initiation of Foxp3 expression, and thus promotes T reg cell phenotype acquisition in peripheral tissues and in vitro (Schlenner et al., 2012; Kanamori et al., 2016). Continuous exposure to TGF-β can prevent the loss of Foxp3 expression in iT reg cells (Selvaraj and Geiger, 2007). However, how TGF-β signaling and epigenetic modifications are coordinated to induce Foxp3 expression has not been fully clarified.

Uhrf1 is an important epigenetic regulator that contains multiple domains, which enables it to participate in various molecular processes (Bostick et al., 2007; Liang et al., 2015; Tian et al., 2015; Kent et al., 2016; Zhao et al., 2017). Among these processes, Uhrf1 is considered to play an important role in the maintenance of DNA methylation and repression of gene expression (Sharif et al., 2007; Chen et al., 2018). During DNA replication, Uhrf1 binds methylated and hemi-methylated DNA via the SET- and RING-associated domain and recruits Dnmt1 to ensure accurate transmission of DNA methylation patterns (Liu et al., 2013). Recent studies have revealed that aberrant Uhrf1 expression is relevant to many different human malignancies and promotes cancer progression (Mudbhary et al., 2014; Sidhu and Capalash, 2017). T cell development and function involve multiple rounds of proliferation (Au-Yeung et al., 2014; Chen et al., 2015), and Uhrf1 is essential for the maintenance of DNA methylation during DNA replication, but the role of Uhrf1 in T cell development and function remains to be further investigated.

Here, we found that Uhrf1 is significantly up-regulated upon TCR stimulation. T cell–specific ablation of Uhrf1 leads to T reg cell–biased differentiation in naive CD4⁺ T cells, with DNA hypomethylation upon TCR stimulation. Uhrf1 maintains Foxp3 DNA methylation by recruiting Dnmt1 during cell division induced by TCR stimulation. Further analysis in WT iT reg cells revealed that Uhrf1 is sequestered in the cytoplasm through phosphorylation and undergoes subsequent proteasome-dependent degradation in response to TGF-β signaling, hence increasing Foxp3 passive demethylation and expression. Altogether, our study demonstrates a critical role of Uhrf1 in orchestrating DNA methylation, TGF-β signaling, and Foxp3 induction.

To understand the function of Uhrf1 in T cells, we generated Cd4-Cre–mediated Uhrf1 conditional knockout mice (Uhrf1^−/− mice), and Uhrf1 was efficiently ablated from T cells (Fig. S1 A). Our recent findings suggested that Uhrf1 is required for invariant natural killer T cell development (Cui et al., 2016). Therefore, in this study, we investigated how Uhrf1 functions in conventional T cells. Unexpectedly, the overall composition of conventional T cell subsets and T cell activation showed no evident differences after Uhrf1 deletion (Fig. S1, B–D).

As Uhrf1 functions during cell proliferation and TCR stimulation drives the activation and proliferation of naive T cells, we measured Uhrf1 expression in naive T cells before and after proliferation upon TCR stimulation. Both Uhrf1 mRNA and protein levels were significantly increased after TCR stimulation (Fig. 1, A and B). RNA-sequencing (RNA-seq) analysis was performed to evaluate the roles of Uhrf1 in TCR-stimulated naive T cells. Genes involved in the immune response and immune system processes were significantly enriched in Uhrf1^−/− naive T cells according to gene ontology (GO) enrichment analysis (Fig. S1 E). Interestingly, genes encoding molecules associated with T reg cells (e.g., Il2ra, Foxp3, Gzmb, and Tnfrsf4) were most significantly up-regulated among all 826 differentially expressed genes (Fig. 1 C). We then selected signature genes specific for T helper (Th) cells or T reg cells and found that Uhrf1^−/− naive T cells seemed to attain a T reg cell phenotype, with higher expression of T reg cell signature genes, after TCR stimulation, which was further confirmed by real-time PCR analysis (Fig. 1, D and E). To investigate the DNA methylome changes of naive T cells after Uhrf1 deficiency, we performed whole-genome bisulfite sequencing (WGBS). Consistent with the RNA-seq data, genes involved in T reg cell differentiation were moderately enriched in Uhrf1^−/− naive T cells according to the GO enrichment analysis in the regulation of T cell differentiation category (Fig. S1 F). Gene methylation profiles showed that T reg cell–specific genes were hypomethylated in Uhrf1^−/− naive T cells upon TCR stimulation (Fig. 1 F). These data suggested that the DNA methylation pattern was altered, and naive T cells showed a T reg cell–biased transcriptome in Uhrf1^−/− naive T cells upon TCR stimulation.

To assess the role of Uhrf1 in T reg cell development, we first analyzed T reg cell differentiation in Uhrf1^−/− mice. The proportion of T reg cell was not enhanced in the thymus, spleen, or lymph nodes of Uhrf1^−/− mice compared with WT littermates (Fig. S1 G). Next, we cultured naive T cells from WT and Uhrf1^−/− mice under iT reg cell skewing conditions, and no significant increase in iT reg cell differentiation was observed in Uhrf1^−/− mice (Fig. 2 A). As both Uhrf1^−/− and WT iT reg cells were induced up to 98%, we speculated that the difference might be obscured by TGF-β saturation. Therefore, a TGF-β concentration titration assay was performed, and as predicted, more iT reg cells were induced from Uhrf1^−/− naive T cells with decreasing doses of TGF-β (Fig. 2 B). An explanation could be that Uhrf1^−/− naive T cells have enhanced sensitivity to TGF-β stimulation at lower doses of TGF-β. To test this possibility, we measured the expression of TGF-β receptors (TβR) I and II and the level of phosphorylated Smad2/3 protein, and there was no substantial difference (Fig. 2 C and Fig. S2, A–C). Based on this observation and the aforementioned T reg cell–biased transcriptome in Uhrf1^−/− naive T cells, we hypothesized that Uhrf1 deletion might induce T reg cell differentiation in the absence of TGF-β signaling. To test this hypothesis, we cultured naive T cells with TCR stimulation alone and found significantly enhanced induction of Foxp3 expression in Uhrf1^−/− naive T cells compared with WT naive T cells (Fig. 2, D and E). In addition, cytokine production showed no significant difference after TCR stimulation, which was consistent with the RNA-seq data and confirmed the T reg cell–biased differentiation and special regulation of Foxp3 in Uhrf1^−/− T cells upon TCR stimulation (Fig. S2 D). To better clarify the role of TGF-β signaling on Foxp3 induction in TCR-stimulated Uhrf1^−/− T cells, we blocked TGF-β signaling by using either a neutralizing TGF-β antibody or an inhibitor of TGF-β receptor (SB431542), and the induction of Foxp3 in Uhrf1^−/− naive T cells was still notably elevated in the absence of TGF-β signaling compared with corresponding WT controls, although slightly decreased compared with vehicle-treated Uhrf1^−/− T cells (Fig. 2 F). Thus, Uhrf1^−/− naive T cells could initiate Foxp3 expression without TGF-β signaling.

To functionally characterize Foxp3⁺ T cells induced from Uhrf1^−/− naive T cells, we crossed Uhrf1^−/− or WT mice with Foxp3^EGFP mice to mark Foxp3⁺ T cells with an enhanced green fluorescent protein (EGFP) reporter, isolated CD4⁺EGFP⁺ T cells, and performed an in vitro suppression assay. Uhrf1^−/− Foxp3⁺ T cells suppressed conventional T cell proliferation as effectively as WT iT reg cells, though less effectively than WT nT reg cells (Fig. 2 G). To further investigate the function of Uhrf1^−/− Foxp3⁺ T cells in vivo, we induced colitis by adoptive transfer of naive T cells from CD45.1 mice together with TCR-stimulated CD4⁺EGFP⁺ cells from Foxp3^EGFP Uhrf1^−/− mice, Foxp3^EGFP WT iT reg cells, or Foxp3^EGFP WT nT reg cells into Rag1^−/− mice, and the suppressive capacity was similar to that observed in the in vitro suppression assay (Fig. 2 H). Taken together, these findings demonstrated that Uhrf1^−/− naive T cells could differentiate into functional T reg cells in the absence of TGF-β signaling.

To further evaluate the importance of Uhrf1-mediated regulation of CD4⁺ T cell differentiation under pathophysiological conditions in vivo, we examined the Uhrf1^−/− naive T cell phenotype in a T cell transfer model of colitis and excluded the influence of Uhrf1^−/− invariant natural killer T cells. I.v. injection of Rag1^−/− mice with Uhrf1^−/− naive T cells failed to induce colitis, with no obvious body weight loss in the mice (Fig. 3 A). Gross examination of the colon did not show enlargement or thickening as observed in WT naive T cell–transferred mice, and H&E staining of colon tissues indicated a disruption of epithelial damage, with massive infiltration of leukocytes in WT but not Uhrf1^−/− naive T cell–transferred mice (Fig. 3, B and C). The differentiation of the transferred naive T cells was analyzed, and a significant increase in iT reg cell generation from Uhrf1^−/− naive T cells was observed, which was consistent with the in vitro results (Fig. 3 D). As previously reported, the proinflammatory cytokines IFN-γ and IL-17A produced from Th1 and Th17 cells, respectively, are major contributors to inflammatory bowel disease. Flow-cytometric analysis of cytokines recovered from colon lamina propria lymphocytes (cLPs), mesenteric lymph nodes (MLNs), and spleen of Rag1^−/− mice revealed that Uhrf1^−/− naive T cells produced less IFN-γ and IL-17A than WT naive T cells (Fig. 3 E).

To investigate whether Uhrf1 deletion might affect Th1 and Th17 cell differentiation and lead to the ameliorated colitis, in vitro effector T cell differentiation experiments were performed, and no significant difference was observed between WT and Uhrf1^−/− T cells (Fig. S2 E). Given that Uhrf1^−/− T cells could induce T reg cells after TCR stimulation in the absence of TGF-β, the other possibility was that the T reg cell–biased differentiation in Uhrf1^−/− naive T cells might inhibit Th1 and Th17 cell differentiation potential. Indeed, after TCR stimulation of WT and Uhrf1^−/− naive T cells for 2 d, further Th1 and Th17 cell differentiation showed significantly less IFN-γ and IL-17A production from Uhrf1^−/− effector T cells compared with WT effector T cells (Fig. S2 F). Collectively, these data suggested that Uhrf1^−/− naive T cells were more prone to induce iT reg cells under pathophysiological conditions, which suppressed proinflammatory cytokine production and ameliorated colitis.

Given that Uhrf1 is induced by TCR stimulation and that DNA methylation patterns are established during cell division, we wondered whether the maintenance of Foxp3 methylation relies on Uhrf1 and cell division. To this end, naive T cells from WT and Uhrf1^−/− mice were stimulated via TCR for different times. As the culture time increased, DNA methylation levels of CpG islands in the promoter and CNS2 region of the Foxp3 gene decreased, and more T reg cells were induced in Uhrf1^−/− T cells (Fig. S3, A and B). In addition, blockade of TGF-β signaling did not alter the Foxp3 DNA methylation pattern when Uhrf1 was deleted (Fig. S3 C), which was consistent with Foxp3 expression shown in Fig. 2 F and confirmed the induction of Uhrf1^−/− T reg cells without TGF-β signaling. These results demonstrated that Uhrf1 maintained Foxp3 DNA methylation to restrict its expression during TCR stimulation.

To further clarify the relationship between the maintenance of Foxp3 DNA methylation and cell proliferation, we analyzed the pattern of Foxp3 expression versus cell division at designated time points. The proportion of CD4⁺Foxp3⁺ T cells in Uhrf1^−/− T cells continued to increase in each successive generation, while the proportion of WT T cells was stable at low levels, with even slightly lower levels after cell division (Fig. 4, A and B). Consistent with a Foxp3 expression pattern, the DNA methylation level of Uhrf1^−/− T cells was reduced in each successive generation, suggesting that Uhrf1-mediated Foxp3 DNA methylation depended on cell division (Fig. 4 C). We next investigated whether entry into the cell cycle was required to induce Foxp3 expression in Uhrf1^−/− naive T cells upon TCR stimulation. After cell cycle arrest, Uhrf1^−/− naive T cells were barely able to induce Foxp3⁺ cells, and the Foxp3 locus showed similar DNA methylation patterns to those of WT controls (Fig. 4, D and E). In addition, Foxp3 mRNA levels were comparable upon TCR stimulation between Uhrf1^−/− and WT naive T cells before T cells were divided (Fig. S3 D). Furthermore, we found that WT iT reg cells were harder to induce after cell cycle arrest (Fig. S3 E). Therefore, cell division was necessary for Uhrf1 to maintain Foxp3 DNA methylation and subsequently restrict Foxp3 expression.

Next, we wanted to test whether Uhrf1 maintains Foxp3 DNA methylation by recruiting Dnmt1 during T cell stimulation. As expected, TCR stimulation–induced interaction between Uhrf1 and Dnmt1 was observed in WT CD4⁺ T cells (Fig. S3 F). Chromatin immunoprecipitation (ChIP) experiments with anti-Dnmt1 antibody revealed that Dnmt1 was enriched at the promoter and CNS2 region of the Foxp3 locus and that Uhrf1 deficiency suppressed this enrichment (Fig. 4 F). Furthermore, we used the H346G Uhrf1 mutant to interrupt the targeting of Dnmt1 to replication foci (Qin et al., 2015), and analyzed Foxp3 expression. Retroviral expression of H346G Uhrf1 mutant did not “rescue” the reduced Foxp3 induction and increased DNA methylation level observed in Uhrf1 retrovirus-transduced cells (Fig. S3, G and H). These results indicated the involvement of Uhrf1 in regulating Foxp3 DNA methylation not only after TCR stimulation but also in iT reg cell differentiation during cell proliferation.

Many definitive studies have uncovered the essential role of TGF-β signaling in the initiation and transcriptional regulation of Foxp3 expression. Our aforementioned data demonstrated that Uhrf1^−/− naive T cells can induce Foxp3 expression in the absence of TGF-β signaling. These findings prompted us to hypothesize that TGF-β signaling might initiate Foxp3 expression by reversing the Uhrf1-mediated maintenance of DNA methylation. As most functions ascribed to Uhrf1 occur in the nucleus, we first examined whether TGF-β induced Foxp3 expression by modulating the localization of Uhrf1. Immunofluorescence analysis confirmed the specificity of our homemade Uhrf1 antibody (Fig. S4 A). Notably, the retention of Uhrf1 in the cytoplasm was significantly increased after TGF-β treatment, which was further confirmed by immunoblot analysis (Fig. 5, A and B; and Fig. S4 B). Accordingly, the DNA methylation level of the Foxp3 locus was decreased (Fig. 5 C). These results indicated that TGF-β signaling might sequester Uhrf1 in the cytoplasm, which resulted in decreased DNA methylation of Foxp3 locus.

It was reported that protein phosphorylation could have either an enhancing or an inhibitory effect on nuclear translocation (Nardozzi et al., 2010). To determine whether phosphorylation is required for the change in Uhrf1 localization during iT reg cell differentiation, we first measured Uhrf1 phosphorylation in WT naive T cells after TGF-β treatment. Total serine phosphorylation of Uhrf1 was noticeably up-regulated after culture with TGF-β for 2 d (Fig. 5 D). Next, we treated cells with roscovitine, a known inhibitor of cyclin-dependent kinases that has been reported to inhibit Uhrf1 phosphorylation (Ma et al., 2012), and analyzed Uhrf1 localization. As expected, roscovitine treatment led to prominent nuclear localization of Uhrf1, and the corresponding DNA methylation level and Foxp3 expression were increased and down-regulated, respectively (Fig. 5, E–G). A previous study suggested that phosphorylation of Uhrf1 (p-Ser-661) could change its localization in human cell lines (Chu et al., 2012). To directly examine the effect of Uhrf1 phosphorylation on its localization and Foxp3 expression, Uhrf1^−/− naive T cells were transiently transfected with an empty vector, Uhrf1, or Uhrf1^S656E and cultured under iT reg skewing conditions. After retroviral expression of S656E Uhrf1 mutant, Uhrf1 localization was exclusively cytoplasmic compared with that of Uhrf1-transfected T cells (Fig. 5 H). The Foxp3 methylation status and expression results were consistent with the observed changes in Uhrf1 localization (Fig. 5, I and J). Collectively, these results suggested that TGF-β signaling induced Uhrf1 phosphorylation and sequestered Uhrf1 in the cytoplasm, which reversed DNA methylation-mediated repression and consequently initiated Foxp3 expression.

We next investigated whether the alteration of Uhrf1 localization affected its protein expression after TGF-β stimulation. Indeed, the abundance of the Uhrf1 protein was observed to decrease in a dose-dependent manner after TGF-β treatment for 4 d, whereas IL-2 plus TCR or TCR stimulation alone showed no effect (Fig. 6 A and Fig. S5 A). In addition, Uhrf1 mRNA levels had no change in T cells after TGF-β treatment (Fig. S5 B). The abundance of the Dnmt1 protein was barely altered upon TGF-β stimulation (Fig. S5 C). There are two major mechanisms by which animal cells degrade proteins: the lysosome and the proteasome (Ciechanover, 2005). Proteasome and lysosome inhibitors were used to clarify the Uhrf1 degradation mechanism. Only MG132 (a proteasome inhibitor) restored Uhrf1 expression after TGF-β treatment, and this effect was dose dependent (Fig. 6 B and Fig. S5 D). Immunoblot analysis showed that ubiquitination of Uhrf1 was enhanced in T cells after TGF-β treatment compared with that in TCR-stimulated T cells (Fig. 6 C and Fig. S5 E). These results suggested that TGF-β specifically induced proteasome-dependent degradation of Uhrf1.

Phosphorylation can serve as a marker that promotes subsequent ubiquitination, in particular when ubiquitination leads to degradation (Hunter, 2007; Nguyen et al., 2013). So we investigated the relationship between phosphorylation and ubiquitination of Uhrf1 induced by TGF-β and found that the Uhrf1 ubiquitination level was elevated, accompanied by increasing phosphorylation levels (Fig. 6 D). To evaluate whether Uhrf1 phosphorylation affected its degradation and protein abundance, we inhibited Uhrf1 phosphorylation by roscovitine and found that blocking phosphorylation significantly suppressed Uhrf1 degradation after TGF-β stimulation (Fig. 6 E). Therefore, Uhrf1 phosphorylation was required for its degradation.

To investigate which lysine residue of Uhrf1 was responsible for its ubiquitination, we analyzed endogenous ubiquitin modifications of Uhrf1 by mass spectrometry. Lys24, Lys50, Lys84, Lys172, and Lys196 of Uhrf1 were identified as sites modified by ubiquitin (Fig. S5 F). Protein half-life assays were used to confirm that Lys84, Lys172, and Lys196 were responsible for the ubiquitination of Uhrf1, based on the observation that replacing those lysine residues with arginine (K84R, K172R, and K196R) completely rescued Uhrf1 degradation (Fig. S5 G). Then, Uhrf1^−/− naive T cells were transiently transfected with empty vector, Uhrf1, or Uhrf1^K84/172/196R, and iT reg cells were induced. Retroviral expression of the K84/172/196R Uhrf1 mutant restored Uhrf1 expression after TGF-β stimulation compared with that of Uhrf1-transfected T cells (Fig. 6 F).

To further clarify the mechanism of Uhrf1 degradation after TGF-β stimulation, mass spectrometry was performed, and the deubiquitylase Usp7 was identified, which was consistent with previous reports (Qin et al., 2011; Zhang et al., 2015; Fig. S5 H). Immunoprecipitation assays confirmed the interaction between endogenous Uhrf1 and Usp7 in WT CD4⁺ T cells, and TGF-β attenuated this interaction (Fig. 6 G). As Usp7 is a deubiquitinase and prevents protein degradation, Uhrf1 protein abundance decreased in T cells after treatment with the Usp7 inhibitor (Fig. 6 H). We next examined whether blocking Uhrf1 degradation could inhibit iT reg cell differentiation. Intriguingly, the Uhrf1^K84/172/196R mutant moderately inhibited Foxp3 induction to an extent similar to that of Uhrf1 (Fig. 6 I). In summary, these results indicated that TGF-β induced Uhrf1 phosphorylation and sequestered it outside the nucleus, which led to decreased Foxp3 DNA methylation and initiation of Foxp3 expression, and cytoplasmic Uhrf1 underwent proteasome-dependent degradation through inhibition of Usp7-mediated deubiquitination as a consequence.

DNA methylation can be erased directly by either an active demethylation mechanism involving ten-eleven translocation (TET) proteins or a passive demethylation process through replication-coupled dilution of 5-methylcytosine (5-mC). Previous studies have shown that TET proteins mediate the loss of 5-mC in the Foxp3 T reg cell–specific demethylated region in T reg cells (Yue et al., 2016), while our data suggested that Uhrf1 directly participated in the maintenance of Foxp3 DNA methylation during cell division and that TGF-β signaling reversed Foxp3 DNA methylation by modulating Uhrf1 activity. To understand the relationship between TET proteins and Uhrf1 in the modulation of Foxp3 DNA methylation, we first measured whether Uhrf1 deletion would affect the abundance of TET proteins and found that there was no difference in TET protein levels after Uhrf1 ablation (Fig. 7 A). Vitamin C and S-2-hydroxyglutarate (S-2HG) are TET protein agonist and inhibitor, respectively (Losman et al., 2013; Yue et al., 2016). Treatment with vitamin C or S-2HG failed to alter Foxp3 methylation pattern or expression level of Uhrf1^−/− naive T cells upon TCR stimulation (Fig. 7, B–E). Thus, TET-mediated active demethylation was not involved in the induction of Foxp3 expression in Uhrf1^−/− naive T cells upon TCR stimulation.

To further clarify the relationship between active and passive demethylation during Foxp3 induction, we used the bisulfite sequencing and oxidative bisulfite methods to measure the enrichment of 5-mC and 5-hydroxymethylcytosine (5-hmC) within the Foxp3 promoter and CNS2 region during iT reg cell differentiation. The 5-mC and 5-hmC conversion efficiencies were confirmed using 5-mC and 5-hmC standard DNA. We calculated the percentage of 5-mC, 5-hmC, and C/5fC/5caC at CpGs in the promoter and CNS2 regions, respectively. In the CNS2 region, all CpGs showed progressive increase of 5-hmC following the loss of 5-mC after 48 h, which was consistent with the previously reported role of TET proteins in improving Foxp3 stability (Yue et al., 2016; Fig. 7 F). However, in the promoter region, there was moderate 5-hmC generation, but accompanied by rapidly reduced 5-mC after 24 h, C/5fC/5caC level was noticeably increased (Fig. 7 G). These results suggested that passive demethylation triggered by inhibition of Uhrf1 activity mainly functioned in the promoter region of Foxp3 locus, while active demethylation mediated by TET proteins acted primarily in the CNS2 region during iT reg cell differentiation.

Based on the findings described above, we assumed that overexpression of Uhrf1 would increase the methylation level of the Foxp3 promoter region. Retroviral expression of Uhrf1 in WT T cells resulted in attenuated passive demethylation, followed by higher methylation levels of the Foxp3 promoter region and eventually less Foxp3 induction (Fig. 7, H and I). Altogether, these data supported the notion that Uhrf1 was involved in passive demethylation of the Foxp3 promoter region and contributed to Foxp3 induction, while active demethylation of the Foxp3 CNS2 region via TET proteins determined Foxp3 stability.

In T reg cells, DNA demethylation has been found to be essential for Foxp3 transcription (Lal et al., 2009), and TGF-β signaling plays a crucial role in Foxp3 transcriptional regulation (Tone et al., 2008). However, how DNA methylation-mediated repression is reversed before Foxp3 gene transcription is less clear. Here, we identified Uhrf1 as a DNA-methylation adaptor regulated by TGF-β signaling that controlled Foxp3 methylation and iT reg cell differentiation. Uhrf1^−/− naive T cells underwent T reg cell–biased differentiation upon TCR stimulation, and these Foxp3⁺ T cells had suppressive functions both in vitro and in vivo. Uhrf1 maintained Foxp3 DNA methylation by recruiting Dnmt1 during cell division induced by TCR stimulation. In iT reg cells, Uhrf1 was involved in the passive demethylation of the Foxp3 promoter region upon TGF-β stimulation, with sequential initiation of Foxp3 expression. Mechanistically, Uhrf1 was phosphorylated upon TGF-β treatment and largely excluded from the nucleus. Phosphorylated Uhrf1 underwent proteasomal degradation, and DNA methylation-mediated repression of Foxp3 promoter region was reversed as an outcome. Taken together, our findings revealed a unique mechanism of TGF-β signaling-mediated epigenetic regulation of Foxp3 methylation and iT reg cell differentiation, which was accomplished through modulating Uhrf1 activity.

Naive T cells are able to both proliferate and differentiate (Jelley-Gibbs et al., 2000). TCR stimulation in the absence or presence of environmental cytokines induces a network of downstream signaling pathways, which eventually lead to naive T cell proliferation or differentiation into various inflammatory effector subsets (e.g., Th1, Th2, and Th17 cells) or T reg cells (Zhou et al., 2009). However, it remains unclear how individual T cells respond to these signaling pathways to make critical cell fate decisions. Here we reported that Uhrf1 was up-regulated and Foxp3 methylation was increased upon TCR stimulation in naive T cells. Uhrf1^−/− naive T cells underwent T reg cell–biased differentiation after TCR stimulation, which was correlated with enhanced expression and DNA hypomethylation of T reg cell–specific genes. Moreover, adoptive transfer of these Uhrf1^−/− naive T cells could significantly suppress colitis. These results revealed an important role for Uhrf1: conventional T cells up-regulated Uhrf1 expression upon TCR stimulation, which maintained conventional T cell identity during activation and proliferation, thus maintaining immune homeostasis.

Many studies have revealed the critical role of TGF-β in inducing Foxp3 expression, and most of those studies focused on transcriptional regulation by the activation of Smad2/3, which occurs almost immediately after TGF-β treatment (Poncelet et al., 1999; Zi et al., 2011). However, the expression of Foxp3 mRNA reached a high level after 48 h upon TGF-β stimulation (Ichiyama et al., 2008; Lu et al., 2010); this time lag between Foxp3 expression and Smad2/3 activation suggested that DNA demethylation may exist to initiate Foxp3 expression after TGF-β stimulation. In the current study, we found that Uhrf1^−/− naive T cells could differentiate into phenotypically similar WT iT reg cells with TCR stimulation alone in the absence TGF-β, which suggested that TGF-β might induce iT reg cell differentiation by inhibiting Uhrf1 function. Indeed, Uhrf1 was observed to be phosphorylated upon TGF-β treatment and was mainly sequestered in the cytoplasm instead of the nucleus, where it performed the DNA methylation function and ultimately underwent proteasome-dependent degradation. As a consequence, the Foxp3 promoter region underwent replication-coupled dilution of 5-mC during cell division, which made the gene loci accessible for transcription factors to initiate its expression. Notably, after TGF-β signaling blockage, Uhrf1^−/− T cells still showed increased Foxp3 expression upon TCR stimulation compared with WT T cells, although the expression was moderately lower than vehicle-treated Uhrf1^−/− T cells. From these data, we thought TGF-β signaling might initiate Foxp3 induction in two processes. First, TGF-β could decrease Foxp3 methylation via blocking Uhrf1 function. Second, when Foxp3 locus was demethylated, TCR- and TGF-β–induced downstream factors might cooperatively promote Foxp3 transcription, which was consistent with the previous reports that Smad protein and transcription factors could cooperate to induce Foxp3 expression through its enhancer (Chen et al., 2003; Tone et al., 2008; Ruan et al., 2009). These findings supported a unique epigenetic regulatory model to explain how TGF-β signaling promotes Foxp3 expression by modulating Uhrf1 activity.

The Uhrf1–Dnmt1 complex is well established to be indispensable for maintaining global DNA methylation following DNA replication (Qin et al., 2015). However, whether these two components are of equal importance and which component determines target specificity of DNA methylation during T cell differentiation are largely unknown. Here, our study found that Foxp3 expression was increased in Uhrf1^−/− naive T cells with TCR stimulation alone, which was consistent with the phenotype of T cells deficient in Dnmt1 (Josefowicz et al., 2009). In addition, we also demonstrated that TGF-β regulated Uhrf1 translocation and induced its degradation to induce iT reg cell differentiation, whereas the expression of Dnmt1 was barely altered upon TGF-β stimulation. These findings revealed that although both Uhrf1 and Dnmt1 repressed Foxp3 expression by maintaining its methylation, only upstream Uhrf1 could respond to TGF-β signaling and determine the target specificity of DNA methylation during iT reg cell differentiation. In addition, we observed a distinct T reg cell–biased differentiation after Uhrf1 deletion, whereas the expression of cytokines was promiscuous in Dnmt1-deficient T cells (Lee et al., 2001; Makar and Wilson, 2004). This phenotypic difference suggested that DNA demethylation was sufficient to initiate Foxp3 transcription, while expression of cytokines might include other functions of Uhrf1, such as transcriptional regulation via chromatin modifications and ubiquitin ligase activity (Jenkins et al., 2005; Kim et al., 2009). Collectively, these findings suggested that despite the cooperation of Uhrf1 and Dnmt1 in DNA methylation maintenance, there was subtle difference between these two molecules, and Uhrf1 determined the target specificity of DNA methylation during iT reg cell differentiation, which made Uhrf1 a better therapeutic drug candidate of inflammatory diseases for generating stable iT reg cells.

In T reg cells, DNA demethylation of the Foxp3 promoter and CNS2 region is important for inducing and stabilizing Foxp3 expression, and maintaining T reg cell identity (Kim and Leonard, 2007; Ohkura et al., 2012; Feng et al., 2014; Li et al., 2014). Moreover, disturbance of this fine-tuned epigenetic process can result in overt pathological consequences. It has been reported that TET-mediated active DNA demethylation maintains the demethylated status of CNS1 and CNS2 in the Foxp3 locus and regulates the stability of Foxp3 expression (Yue et al., 2016). In the current study, we revealed that TGF-β regulated passive demethylation of the Foxp3 promoter region and further Foxp3 expression during iT reg cell differentiation through suppressing Uhrf1 function. In addition, this passive DNA demethylation exerted its function during cell division upon TCR stimulation. Altogether, passive DNA demethylation involving Uhrf1 cooperated with TET-mediated active DNA demethylation to induce Foxp3 expression and control Foxp3 stability, respectively.

Uhrf1 has been reported to specifically facilitate the proliferation and maturation of colonic regulatory T cells but not the maintenance of extracolonic T reg cells, and Uhrf1^−/− mice developed spontaneous colitis before 10 wk of age under specific pathogen–free (SPF) conditions, while they did not display any inflammation under germ-free conditions (Obata et al., 2014). Here we did not observe spontaneous colitis of Uhrf1^−/− mice raised under SPF conditions. As differences in animal housing facilities and diet might affect study outcome (Hirayama et al., 1990; Reliene and Schiestl, 2006), we speculated the different environment between their and our animal facilities might determine whether or not spontaneous colitis occurred. In addition, Obata et al. (2014) found that Uhrf1 facilitated colonic T reg cell proliferation to restrain colitis, while our data demonstrated that Uhrf1 inhibited iT reg cell differentiation to keep the identity of conventional T cells. Notably, increasing evidence suggested that the same factors might play distinct roles in different T reg cell subsets (Pabbisetty et al., 2014; Kitagawa et al., 2017; Wu et al., 2017; Ghosh et al., 2018). Colonic T reg cells and iT reg cells were generated in different locations and in response to different environmental signals (Zhou et al., 2009; Josefowicz et al., 2012; Arpaia et al., 2013; Furusawa et al., 2013; Luu et al., 2017), and a single-cell transcriptomics study showed that T reg cells generated from lymphoid organs and colon were quite heterogeneous, which might be due to their adaptation to the local environments (Miragaia et al., 2019). In this context, Uhrf1 might exert distinct functions in different T reg cell subsets.

In summary, our study revealed a unique regulation of Uhrf1 in restraining T reg cell differentiation during conventional T cell activation and proliferation and demonstrated that TGF-β functioned to unleash this regulation via sequestering Uhrf1 in the cytoplasm. In addition, our study could broaden the underlying mechanisms by which TGF-β mediates T reg cell generation through regulating DNA methylation and may provide a novel potential therapeutic target in inflammatory diseases.

All animal experiments were conducted according to the National Institutes of Health guidelines and were approved by the Institutional Animal Care and Use Committee of the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. The Uhrf1^fl/fl mice were obtained as described (Cui et al., 2016) and were backcrossed for ≥10 generations with C57BL/6J. Cd4-Cre transgenic mice were obtained as described previously (Sun et al., 2014). CD45.1 mice (strain: B6.SJL-Ptprc^aPepc^b/BoyJ) and Rag1^−/− mice (strain: B6.129S7-Rag1^tm1Mom/J) were purchased from The Jackson Laboratory. Foxp3-EGFP mice (strain: B6.Cg-Foxp3^tm2Tch/J) were obtained from H.L. Wang (Shanghai Jiao Tong University School of Medicine, Shanghai, China). 6–10-wk-old sex-matched mice were used unless otherwise indicated. We used Uhrf1^fl/fl littermates as WT mice. All the mice were bred in the SPF facility and genotyped by PCR before experimentation (Uhrf1-forward primer: 5′-CAG​CCA​GTG​TTC​TTA​ACC​TCC​AA-3′; Uhrf1-reverse primer: 5′-CTC​CCA​CGT​AAC​CCT​AGA​TCC​TT-3′).

The following antibodies from BD Biosciences were used: anti-CD4 (RM4-5), anti-CD8 (RPA-T8), anti-CD44 (IM7), anti-CD62L (MEL-14), anti-CD69 (H1.2F3), anti-CD45.2 (104), anti-CD45.1 (A20), anti–IL-4 (11B11), and anti-Smad2 (pS465/pS467)/Smad3 (pS423/pS425) (072–670). The following antibodies from BioLegend were used: anti-CD25 (PC61), anti-CD45RB (C363-16A), anti-IFN-γ (XMG1.2), and anti–IL-17A (TC11-18H10.1). The following antibodies were from eBioscience: anti-Foxp3 (FJK-16s), anti-Ki-67 (SolA15), and neutralized TGF-β antibody (16–9243-85). The following antibodies were from R&D Systems: anti-TβRI (FAB5871P), and anti-TβRII (FAB532P). Antibody against Dnmt1 (60B1220.1) was purchased from Novus Biologicals. Antibody against β-actin (ab40009) was purchased from MultiSciences. Antibodies against lamin B1 (sc-374015), α-tubulin (sc-23948), Ub (sc-8017), goat anti-rabbit IgG-HRP (sc-2004), and m-IgGκ BP-HRP (sc-516102) were purchased from Santa Cruz Biotechnology. Antibodies against phosphoserine (ab6639), Usp7 (ab157132), and Tet1 (ab191698) were purchased from Abcam. Antibody against Tet2 (21207–1-AP) was purchased from Proteintech. A homemade rabbit antibody directed against a glutathione-S-transferase fusion protein containing amino acids 594–701 of mouse Uhrf1 was derived using conventional methods (Leenaars and Hendriksen, 2005). Mimosine (M0253), hydroxyurea (H8627), collagenase D (11088866001), and DNase I (11284932001) were purchased from Sigma-Aldrich. Roscovitine and SB431542 were purchased from Selleck. Taq Master Mix was purchased from Vazyme Biotech.

Cells were isolated from spleens and lymph nodes by grinding with cell strainers. For cLP isolation, colons were excised and placed in cold PBS. After being minced into ∼0.5-cm pieces, the colons were incubated in RPMI 1640 medium (Gibco) containing 10 mM Hepes and 5 mM EDTA at 37°C for 20 min. The remaining tissues were further minced into ∼1-mm pieces, and digested with 5% FBS, 0.2 mg/ml collagenase D, and 0.1 mg/ml DNase I in RPMI 1640 medium for 1 h at 37°C with constant agitation. The resulting cell suspensions were filtered through a cell strainer (40 µM; BD Biosciences), and lymphocytes were purified on a 40%/70% Percoll (GE Healthcare) gradient by centrifugation.

Surface markers were stained in HBSS with 0.1% BSA for 30 min at 4°C (antibodies are listed above). For cytokine staining, cells were stimulated with 50 ng/ml PMA (Sigma-Aldrich) and 1 mg/ml ionomycin (Merck) in the presence of 1,000× brefeldin A (eBioscience) for 4 h at 37°C. Cells were first stained with antibodies against indicated cell surface markers followed by 10 min of fixation with 2% formaldehyde solution in PBS at room temperature and 5 min of permeabilization in Perm/Wash Buffer (BD Biosciences) at 4°C. For Foxp3 staining, cells were fixed and permeabilized with the Foxp3 Staining Buffer Set according to the manufacturer’s instructions (eBioscience). To track cell division, T reg cells were labeled in vitro with CFSE (Sigma-Aldrich) or CellTrace Violet (Life Technologies) according to the manufacturer’s protocol. Cell fluorescence was acquired on a four-laser BD LSR Fortessa II or a two-laser BD FACS Calibur and was analyzed with FlowJo software (TreeStar). Cell sorting was carried out by a BD FACSAria II after surface staining. Sorted cell purity was over 95%.

Naive CD4⁺ T cells (CD4⁺CD25⁻CD44^lowCD62L^high) were isolated from mouse total spleen and lymph node cells using the EasySep Mouse Naive CD4⁺ T Cell Isolation Kit (STEMCELL Technologies), following the manufacturer’s instructions. For analysis of the in vitro T reg cell suppression assay, naive CD4⁺ T cells were sorted from CD45.1⁺ mice using a BD FACS Aria II Flow Cytometer. Isolated naive CD4⁺ T cells were stimulated for 4 d in 48-well flat-bottomed plates with plate-bound anti-CD3 and anti-CD28 (1 µg/ml each; BD Biosciences), β-mercaptoethanol (50 µM; Shanghai Genebase Gene-Tech), and 10% FBS in RPMI 1640 medium unless otherwise indicated. For iT reg skewing conditions, the cells were cultured in RPMI 1640 medium supplemented with recombinant murine IL-2 (10 ng/ml; PeproTech) and recombinant human TGF-β1 (10 ng/ml unless otherwise indicated; PeproTech). For Th1 cell condition, the cells were cultured in RPMI 1640 medium supplemented with recombinant murine IL-2 (10 ng/ml), recombinant murine IL-12 (10 ng/ml; PeproTech) and anti–IL-4 (10 µg/ml; BD Biosciences). For Th17 cell condition, the cells were cultured in RPMI 1640 medium supplemented with recombinant murine IL-6 (20 ng/ml; PeproTech), recombinant human TGF-β1 (5 ng/ml), recombinant murine IL-1β (10 ng/ml; PeproTech), recombinant murine IL-21 (50 ng/ml; PeproTech), recombinant human IL-23 (20 ng/ml; PeproTech), anti–IFN-γ (10 µg/ml; BD Biosciences), and anti–IL-4 (10 µg/ml; BD Biosciences).

Sorted naive CD4⁺CD25⁻CD45RB^high T cells (6 × 10⁴ cells) were labeled with 5 µM CFSE for 10 min at 37°C and then cultured together with sorted WT nT reg cells (CD4⁺GFP⁺), WT iT reg cells (CD4⁺GFP⁺), or stimulated (anti-CD3 and anti-CD28) Uhrf1^−/− T cells (CD4⁺GFP⁺) at the indicated ratio at 37°C in 200 µl of RPMI medium plus 10% FBS, supplemented with 50 µM β-mercaptoethanol in the presence of Dynabeads Mouse T-Activator CD3/CD28 (Thermo Fisher Scientific) in round-bottomed 96-well dishes. After 72 h, the cells were harvested, and CFSE dilution was measured by flow cytometry.

Sorted naive CD4⁺CD25⁻CD45RB^high T cells (5 × 10⁵ per mouse) from CD45.1⁺ mice in combination with WT or Uhrf1^−/− T reg cells (5 × 10⁵ per mouse) were i.v. injected into Rag1^−/− recipient mice. Naive T cells were also transferred without T reg cells as a control group in the suppression assay. In the transfer model of colitis, naive CD4⁺CD25⁻CD45RB^high T cells (5 × 10⁵ per mouse) from WT or Uhrf1^−/− mice were transferred to Rag1^−/− mice by i.v. injection. The transferred mice were monitored daily for weight loss. The mice were euthanized when some mice reached a 20% reduction of initial weight, and colons and other lymphoid organs were collected for flow-cytometric analyses and H&E staining.

Mouse tissues were fixed in 10% formalin and stored in 70% ethanol until staining. Paraffin-embedded sections were cut and stained with H&E. Slides were then examined on an Olympus BX51 microscope.

Confocal microscopy for Uhrf1 localization was performed as previously described (Zhao et al., 2017). Briefly, isolated naive T cells were stimulated with anti-CD3 and anti-CD28 in the presence or absence of TGF-β for indicated times. Cells were then harvested, fixed with 4% paraformaldehyde in PBS, and settled on coverslips coated with poly-L-lysine (Sigma-Aldrich) at 37°C for 1 h. Cells were permeabilized with 0.5% Triton X-100 and blocked with 1% BSA in Tris-buffered saline and Tween 20. Then cells were stained with Uhrf1 antibody diluted in blocking buffer (1:100), followed by incubation with secondary antibody conjugated to Cy3 (111–165-144; Jackson ImmunoResearch) together with DAPI (1 µg/ml; Cell Signaling Technology). Slides were analyzed on a Leica confocal microscope TCS SP8. ImageJ software was used to determine fluorescence intensity of pixels.

The HEK293T cell line was purchased from the Cell Bank (Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences). The PlatE cell line was obtained from G. Pei (Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences, Shanghai, China) and was originally purchased from Cell Biolabs. The cells were cultured in a humidified incubator at 37°C with 5% CO₂ in DMEM (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco).

cDNAs of mouse genes were cloned into the pMX-IRES-GFP retroviral vector. The virus was packaged in the PlatE cell line, and after transfection for 48 h, viral culture supernatants were harvested, supplemented with 8 µg/ml polybrene (Sigma-Aldrich), and added to previously stimulated T cells (1 × 10⁶ per well, plate-bound anti-CD3 and anti-CD28 for 24 h). Then, the cells were centrifuged at 1,500 g for 2 h at 32°C. Retroviral supernatants were replaced with fresh RPMI culture medium with cytokines after transduction.

Cells were lysed with 1× SDS sample buffer (50 mM Tris-HCl, pH 6.8, 100 mM dithiothreitol, 2% SDS, and 10% glycerol) and incubated at 95°C for 10 min. Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes (GE Healthcare). Membranes were incubated with 5% BSA in Tris-buffered saline and Tween 20 (0.5 M NaCl, Tris-HCl, pH 7.5, and 0.1% [vol/vol] Tween-20) for 60 min at room temperature. Proteins were detected with antibodies as described above. Membranes were visualized by immunoblot analysis with the enhanced chemiluminescence detection system (Thermo Fisher Scientific).

For immunoprecipitation, cells were lysed in lysis buffer (150 mM NaCl, 1% Triton X-100, 1 mM EDTA, and 50 mM Tris-HCl), supplemented with protease inhibitor cocktail, 1 mM NaF, and 1 mM Na₃VO₄. Cell lysates were incubated with the primary antibody-conjugated protein A/G beads (Santa Cruz Biotechnology) for 4–6 h at 4°C. Beads were washed extensively, and eluted with 1× SDS sample buffer. The eluates were subjected to immunoblot analysis and mass spectrometry.

Purified naive T cells were stimulated with precoated anti-CD3 and anti-CD28 for 4 d, and ChIP assay was performed based on the manufacturer’s instructions (Millipore). Briefly, stimulated WT and Uhrf1^−/− naive T cells were cross-linked for 30 min on ice with 1% formaldehyde and lysed for 10 min on ice. The lysates were then sonicated to obtain DNA fragments of an average length of 500 bp. The fragmented lysates were subjected to immunoprecipitation with the indicated antibodies. The recovered DNA was used as templates for real-time PCR quantification. Data were normalized by input DNA for each sample. The following primers were used for quantitative PCR: Foxp3 promoter forward primer: 5′-TTT​CAG​ATG​ACT​TGT​AAA​GGG​CAA​AG-3′; Foxp3 promoter reverse primer: 5′-GAG​TGT​GTG​TGC​TGA​TAA​TTG​CAG​G-3′; Foxp3 CNS2.2 forward primer: 5′-CCC​TCT​GGC​ATC​CAA​GAA​AG-3′; Foxp3 CNS2.2 reverse primer: 5′-GGG​TGC​TAG​CGG​ATG​TGG​TA-3′; Foxp3 CNS2.4 forward primer: 5′-CCC​AAC​AGA​CAG​TGC​AGG​AA-3′; and Foxp3 CNS2.4 reverse primer: 5′-AAA​GAG​GAC​CTG​AAT​TGG​ATA​TGG-3′.

Total RNA was extracted from isolated or cultured T cells with TRIzol (Invitrogen), and reverse-transcribed with the SuperScript III First-Strand Synthesis System (Invitrogen). Real-time PCR was performed on the Rotor Gene 6000 machine (Corbett Life Sciences) with SYBR Green QPCR Master Mix (Toyobo). GAPDH was used as internal control. The sequences of quantative PCR primers for the genes examined are listed below:Uhrf1 forward primer: 5′-CTG​GCT​ATG​GTG​TGG​GTC​ACA​G-3′; Uhrf1 reverse primer: 5′-CTG​GGC​CTC​AAA​CCA​TGC​AC-3′; Ifng forward primer: 5′-GCG​TCA​TTG​AAT​CAC​ACC​TG-3′; Ifng reverse primer: 5′-ACC​TGT​GGG​TTG​TTG​ACC​TC-3′; Tbx21 forward primer: 5′-CAA​CAA​CCC​CTT​TGC​CAA​AG-3′; Tbx21 reverse primer: 5′-TCC​CCC​AAG​CAG​TTG​ACA​GT-3′; IL4 forward primer: 5′-AGA​TGG​ATG​TGC​CAA​ACG​TCC​TCA-3′; IL4 reverse primer: 5′-AAT​ATG​CGA​AGC​ACC​TTG​GAA​GCC-3′; Gata3 forward primer: 5′-AGA​ACC​GGC​CCC​TTA​TGA​A-3′; Gata3 reverse primer: 5′-AGT​TCG​CGC​AGG​ATG​TCC-3′; Il17A forward primer: 5′-GCT​CCA​GAA​GGC​CCT​CAG​ACT​A-3′; Il17A reverse primer: 5′-CAG​GAT​CTC​TTG​CTG​GAT​GAG​AAC​AG-3′; Rorc forward primer: 5′-TTC​ACC​CCA​CCT​CCA​CTG-3′; Rorc reverse primer: 5′-GTG​CAG​GAG​TAG​GCC​ACA​TT-3′; Foxp3 forward primer: 5′-GGC​CCT​TCT​CCA​GGA​CAG​A-3′; Foxp3 reverse primer: 5′-GCT​GAT​CAT​GGC​TGG​GTT​GT-3′; Gapdh forward primer: 5′-ACT​CCA​CTC​ACG​GCA​AAT​TCA-3′; and Gapdh reverse primer: 5′-GCC​TCA​CCC​CAT​TTG​ATG​TT-3′.

WGBS was performed as previously described (Kulis et al., 2015; Zhang et al., 2017). Briefly, ∼5 µg genomic DNA was spiked with unmethylated 26 ng λ DNA. DNA was fragmented into 200–300 bp with Covaris S220 (Covaris), followed by end repair and adenylation. Then these DNA fragments were treated twice with bisulfite using the EZ DNA Methylation Gold Kit (Zymo Research). The resulting single-strand DNA fragments were PCR-amplificated using the KAPA HiFi HotStart Uracil + ReadyMix (2X; Kapa Biosystems). The library concentration was quantified by Qubit 2.0 Flurometer (Life Technologies) and quantitative PCR (iCycler, BioRad Laboratories), and the insert size was assayed on the Agilent Bioanalyzer 2100 system. The library preparations were sequenced on an Illumina Hiseq X Ten platform (Illumina) to generate 150 bp paired-end reads by the Novogene Bioinformatics Institute.

Bisulfite sequencing was started with 1 × 10⁶ cells and converted with the EZ DNA Methylation-Direct Kit (Zymo Research). Selected genomic regions were PCR-amplified by the Taq HS enzyme (Takara). PCR products amplified from a DNA sample were recovered using the Gel Extraction Kit (Qiagen) and ligated into the pMD 18-T vector (Takara) for sequencing. Sequencing data were analyzed with BISMA (http://services.ibc.uni-stuttgart.de/BDPC/BISMA/).

The sequences of the Foxp3 primer sets were as follows: promoter forward, 5′-GTG​GTG​AGG​GGA​AGA​AAT​TAT​ATT​T-3′ and reverse, 5′-TAA​CAC​CCA​CCC​TCA​ATA​CCT​CT-3′; CNS2 forward, 5′-TTG​GGT​TTT​TTT​GGT​ATT​TAA​GAA​AGA-3′ and reverse, 5′-CCC​TAT​TAT​CAC​AAC​CTA​AAC​TTA​ACC​AA-3′.

RNA was extracted, purified, and checked for integrity using a RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies). Libraries were generated for sequencing using NEBNext Ultra RNA Library Prep Kit for Illumina (NEB) following the manufacturer’s recommendations. The library preparations were sequenced on an Illumina Hiseq platform, and 125 bp/150 bp paired-end reads were generated. The sequenced reads were aligned to the mouse genome (mm10) using HISAT2 (v2.0.5), and featureCounts (v1.5.0-p3) was used to count the reads numbers mapped to each gene. Fragments per kilobase per million mapped reads of each gene was calculated based on the length of the gene and reads count mapped to this gene. Differential expression analysis was performed using the DESeq2 R package (1.16.1). The P values were adjusted using the Benjamini and Hochberg method, and genes with an adjusted P value <0.05 found by DESeq2 were assigned as differentially expressed.

For GO enrichment analysis, selected differentially regulated genes between WT and Uhrf1^−/− T cells with an adjusted P value <0.05 were analyzed by the GOseq R package.

Statistical analyses were performed with GraphPad Prism 6.0 (GraphPad). Unless indicated, all experiments shown were performed at least three times. All data are shown as means ± SD, and to compare the statistical significance of two independent groups, Student’s t test was used to calculate P values unless otherwise indicated. For all experiments, *, P < 0.05; **, P < 0.001; ***, P < 0.0001; and N.S., no significance.

All sequencing data that support the findings of this study have been deposited in the Gene Expression Omnibus of the National Center for Biotechnology Information under accession no. GSE128480.

Fig. S1 shows T cell development and GO enrichment analysis after Uhrf1 deletion. Fig. S2 shows effects of Uhrf1 deficiency on the expression of TGF-β receptors and effector T cell differentiation. Fig. S3 shows DNA methylation patterns and expression of Foxp3 in the absence of Uhrf1 and that maintenance of Foxp3 methylation depends on the Uhrf1–Dnmt1 interaction. Fig. S4 shows that TGF-β signaling sequesters Uhrf1 in the cytoplasm. Fig. S5 shows regulation of Uhrf1 ubiquitination after TGF-β treatment.

We thank Baojin Wu and Guoyuan Chen for the animal husbandry and Wei Bian for the support of cell sorting.

This work was financially supported by the National Natural Science Foundation of China (31621003, 31530021, 91542122, and 31500717), and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB19000000).

The authors declare no competing financial interests.

Author contributions: X. Sun designed and performed most experiments, analyzed data, and prepared the manuscript. Y. Cui designed and performed experiments, and analyzed data. H. Feng performed the genotyping and plasmids construction experiments. H. Liu helped with mice construction and breeding. X. Liu conceptualized the research, directed the study, and prepared the manuscript.