Wnt/β-catenin activation epigenetically reprograms Regulatory T cells in IBD and progression to dysplasia.

The molecular and functional diversity of regulatory T-cells (T reg s) in health and in disease remains unclear. We previously described in colorectal cancer (CRC) patients a subpopulation of RORγt + T reg s with elevated expression of β-catenin and pro-inammatory properties. Here we observed progressive expansion of RORγt + T reg s in inammatory bowel disease (IBD) patients during inammation and early dysplasia. Activating Wnt/β-catenin signaling in human and murine T reg s was sucient to recapitulate the disease-associated increase in frequencies of RORγt + T reg s expressing IL-17, IFN-γ, and TNFa. We found that binding of the β-catenin interacting partner, TCF-1, to DNA overlapped with Foxp3 binding at enhancer sites of pro-inammatory pathway genes. Sustained Wnt/β-catenin activation induced newly accessible chromatin sites in these genes and upregulated their expression. These ndings indicate that TCF-1 and Foxp3 together limit the expression of pro-inammatory genes in T reg s. Activation of (cid:0) -catenin signaling interferes with this function and promotes the disease-associated RORγt + T reg phenotype. enrichment of DNA methylation marks


Introduction
Regulatory T cells (T reg s) are essential to sustain peripheral tolerance and prevent autoimmune pathology. A growing body of evidence suggests that T reg s adapt phenotypically and functionally to their anatomic location and in response to the in ammatory millieu 1,2 . Multiple T reg sub-populations have been described that express additional T helper (T H ) cell lineage transcription factors 3 including T-bet 4 , GATA-3 5 , and RORγt 6, 7 . Adopting these programs equips T reg s with the ability to speci cally regulate different T H responses. The intestine comprises a highly diverse T reg compartment. Speci cally, RORγtexpressing T reg s are found in the colon, small intestine (SI), and at lower frequencies in the peripheral lymphoid organs of healthy mice 6,7,8 . These cells are necessary to maintain the balance of intestinal in ammatory responses. In humans, however, it remains unclear whether RORγt + /Foxp3 + T reg s have physiological regulatory functions.
T reg s are de ned by a core regulatory program established during late T reg differentiation by Foxp3 binding to gene enhancers made accessible by Foxo1 9 . Moreover, Foxp3 acts together with other factors forming multi-molecular complexes with different modes of action 10,11 ; functioning as an activator in complex with RelA, Ikzf2 (Helios), Ep300, and Kat5, and a repressor with EZH2, YY1, and Ikzf3 12 . This seminal work 9, 10, 11, 12 , however, suggests that Foxp3 has a complex interactome, which requires further study. Recently, it was suggested that overlap of Foxp3 binding with the transcription factor TCF-1 may affect the maintenance of T reg function 13 . We further showed that TCF-1 binding also overlaps with multiple other factors including Ikaros, Runx, Ets, and HEB in double positive thymocytes. In particular, it cooperates with HEB to initiate T cell-speci c programs through active enhancer (AE) binding and regulation of chromatin accessibility 14 . Furthermore, TCF-1 plays critical roles in the differentiation of T H lineages 15 , memory formation 16 , and counteracting exhaustion 17 . In T cells, TCF-1 is the bona de DNA-binding partner of β-catenin, the central effector of the Wnt-signaling cascade. The classical Wntsignaling dogma de nes TCF-1 as a DNA-bound repressor until Wnt signals stabilize β-catenin, which translocates to the nucleus and binds TCF-1 to initiate transcription 18 . β-catenin is also stabilized upon T cell receptor (TCR) engagement in T cells 19 . Excessive Wnt/β-catenin activation in T reg s may impact T reg function 20 and is linked to autoimmunity in the context of multiple sclerosis 21 . Foxp3 binding to DNA has been suggested to overlap with β-catenin in human T reg s, which parallels the overlap between TCF-1 and Foxp3 DNA occupancy in mice 13 .
Immune suppression through recruitment of T reg s is a major immune escape mechanism in cancer 22 and increased tumor-in ltrating and/or circulating T reg s are correlated with detrimental outcomes 23 . Their role in colorectal cancer (CRC) remains unclear 24,25,26,27 , as T reg s also suppress chronic in ammation, which is a cancer-promoting factor in in ammatory bowel disease (IBD)-associated CRC 28,29 . We described a sub-population of RORγt + T reg s that expands in the tumor and peripheral blood (PB) of CRC patients throughout disease progression 30 . While these cells still suppress T cell proliferation, they are also proin ammatory, as they express IL-17 and promote mast cell degranulation. In murine polyposis models, we causatively linked the pro-in ammatory skewing of T cells to activation of Wnt/β-catenin-signaling during thymic development 31 .
Here, we show that RORγt + T reg s expressing the pro-in ammatory cytokines IL-17, IFNg, and TNFa expand progressivelly in both IBD and IBD/CRC patients and in mouse models of colitis and colitis/dysplasia. We establish that enhanced Wnt/β-catenin signaling is responsible for the induction of pro-in ammatory RORγt + T reg s. Moreover, we provide a molecular mechanism that explains this disease-associated phenomenon. We found that TCF-1 and Foxp3 co-bind regulatory elements, most importantly active enhancers, of genes that are involved in pro-in ammatory processes and limit their expression. Upon Wnt/β-catenin activation, these genes gain newly accessible chromatin sites and become transcriptionally upregulated. Hence, the upregulation of β-catenin imposes a pro-in ammatory phenotype onto T reg s by interfering with TCF-1/Foxp3-mediated gene suppression.

Results
Frequencies of β-catenin high RORγt + T reg s increase in the PB and colonic mucosa of IBD patients during in ammation and malignant disease progression. We previously established that RORγt + T reg s expand in sporadic CRC patients 30 . To understand the etiology of this expansion, we investigated IBD, which represents a unique platform to study the mechanisms underlying the emergence and distribution of RORγt + T reg s during in ammation and progression to CRC. We analyzed colonic tissue and/or PB samples from 65 IBD patients, with or without dysplasia (IBD/Dys), and compared them to a group of 20 healthy donors (HD ; Table S1).
We rst determined the frequencies of circulating fractions (Fr.) of naïve T reg s (Fr.I=CD45RA + Foxp3 int ), activated T reg s (Fr.II=CD45RA -Foxp3 high ), and activated conventional T cells (Fr.III= CD45RA -Foxp3 int ) within CD4 + T cells as suggested by   32,27 . Previously, we reported that Fr.II of activated T reg s selectively increased in sporadic CRC compared to HDs 30 . Likewise, here we show a progressive enrichment of Fr.II T reg frequencies from HDs to IBD and IBD/Dys (Fig. 1c). We also observed a progressive increase of non-T reg Fr.III T cells, which were suggested to be prognostic for CRC clinical outcomes (Fig. 1c) 27 . We previously de ned proin ammatory T reg s in patients by their expression of RORγt. We therefore assessed the proportions of RORγt-expressing cells within the Sakaguchi T reg fractions ( Fig. 1b-d) as well as in the classically de ned CD25 + Foxp3 + CD127 -T reg population (Fig. S1a-c).
In accordance with our previous work, RORγt + cells were enriched within Fr.II of activated T reg s in IBD/Dys patients (Fig. 1d). Similarly, frequencies of RORγt + within CD25 hi Foxp3 + CD127 -T reg s also increased in IBD and IBD/Dys patients compared to HDs (Fig. S1b, c). Conclusively, PB RORγt + T reg frequencies were elevated in IBD and IBD/Dys patients.
Previously, we causatively linked the skewing of CD4 + conventional T cells towards a Th17/proin ammatory phenotype to the activation of the Wnt/β-catenin pathway 31 . Hence, we assessed the expression of β-catenin in RORγt + and RORγt -Fr.II T reg s and in CD25 hi Foxp3 + CD127 -T reg cells. Indeed, RORγt expression in T reg s uniformly correlated with enhanced β-catenin protein levels (Fig. 1b,. Speci cally, β-catenin expression was signi cantly higher in RORγt + compared to RORγt -Fr.II T reg s in patients and HDs (Fig. 1e). Further validating previous ndings by our group and others 21, 30 , the RORγt + but not the RORγt -CD25 + Foxp3 + CD127 -PB T reg s produced proin ammatory cytokines upon stimulation with PMA/Ionomycin . More precisely, RORγt + PB T reg s intracellularly accumulated IL-17, IFNg, and TNFa ( Fig. 1 f-h) and a fraction of them were even found to be double producers for TNFa and IL-17 (Fig. 1i). Importantly, a signi cantly higher percentage of RORgt + PB T reg s produced IFNg in IBD and IBD/Dys patients compared to HDs . Furthermore, compared to RORgt -, co-expression of IL-17 and TNFa was signi cantly more frequent in RORgt + PB T reg s of IBD and IBD/Dys patients but not of HDs (Fig. 1i). These ndings show that chronic in ammation in IBD and IBD/Dys conincides with the increase in frequencies of the previously described RORgt + T reg population in the PB which produces multiple pro-in ammatory cytokines.
Next, we determined the frequencies of tissue-resident RORγt + T reg s in IBD patients by purifying mononuclear cells (MNCs) from in amed (INF) and less in amed ('margin', M) colonic mucosa samples.
We found substantially higher frequencies of total Foxp3 + T cells (~30%) within tissue-resident CD4 + T cells compared to the PB. Moreover, tissue-resident total Foxp3 + T cells were signi cantly increased in IBD/Dys compared to IBD patients (Fig. 1j) for both M and INF areas. We also found that frequencies of RORγt + T reg s increased from the margin to the in amed mucosa (Fig. 1k-l). Like their circulating counterparts, tissue-resident RORγt + T reg s expressed signi cantly higher levels of β-catenin compared to RORγt -T reg s ( Fig. 1m-n) and produced IL-17 upon stimulation (Fig. S1h, Fig. 1o). Thus, circulating RORγt + T reg s in IBD/Dys patients share major similarities with tissue-resident RORγt + T reg s and are likely tissue/tumor derived.
To explore possible RORγt + T reg tumor in ltration we analyzed datasets from The Cancer Genome Atlas (TCGA, Research Network: https://www.cancer.gov/tcga) and determined how expression changes of Wnt/β-catenin, T reg , and Th17 pathway genes were connected. To do this, we calculated the average zscores over all genes in the 'human WNT signature' (KEGG_human_WNT) 33, 34 and signatures containing genes that are transcriptionally up-regulated in Th17 cells (TH17_UP) and T reg s (T reg _UP) that were kindly provided to us by Dr. Benoist and colleagues 10 . The TH17_UP positively correlated with the KEGG_human_WNT signature (p<0.001, Spearman-score r=0.5604, red dots Fig. 1p) and mirrored the equally strong positive correlation between the T reg _UP and the KEGG_human_WNT (p<0.001, Spearmanscore r=0.6161, blue dots Fig. 1p) signatures. Also, the TH17_UP and T reg _UP signatures showed a strong positive correlation (p<0.001, Spearman-score r=0.7836, Fig. S1i). This suggests that the T reg in ltrate in CRC tumors with an activated Wnt signature may possess Th17-like traits.
We further assessed whether the enhanced expression of T reg /Th17 signature genes could be correlated with adverse survival in the TCGA cohort, as the expression of Th17 associated genes was previously linked to detrimental outcome in CRC 35 . The effect on survival was tested with a machine learning approach. Coe cient values were derived from the machine learning approach for each gene in the TH17_UP, Treg_UP, and the combined TH17_UP/Treg_UP signatures via Cox proportionalhazards regression (example for the TH17_UP signature is shown in Fig. S1k,l) ( Table S2). The genes whose enhanced expression predicted decreased survival included, amongst others, LEF-1 and MAF. We then divided patients, based on the median score of the TH17_UP, Treg_UP and combined TH17_UP/Treg_UP signature (TH17_UP signature shown as example in Fig. S1i). We interogated the survival outcomes of groups with above versus below the median score of the weighted signatures. These analyses indicated that above-median signature scores for the TH17_UP (Fig. 1q, p<0.0001, logrank test, n=188 -below and above median) T reg _UP (Fig. 1r, p<0.0001) and most importantly for the combined TH17_UP/T reg _UP genes (Fig. 1s, p<0.0001) correlated with reduced overall survival. The signi cant detriment in survival for the combined weighted signature (TH17_UP/Treg_UP) further supports the suggestion from the unweighted Spearman correlations that a T reg tumor in ltrate with Th17-traits might results in adverse survival.
Ex vivo stabilization of β-catenin in human T reg s is su cient to induce the pro-in ammatory phenotype.
Given our nding that RORγt expression in T reg s correlates with increased β-catenin levels in IBD(/Dys) patients, we investigated whether β-catenin stabilization was su cient to induce RORγt and proin ammatory cytokine expression in HD T reg s. Therefore, we treated HD PBMCs ex vivo with the GSK-3β inhibitor Chiron (CHIR99021) to stabilize β-catenin protein. Levels of β-catenin (Fig. 2a) and RORγt (Fig. 2c) were assessed in CD4 + CD25 + Foxp3 + T reg s after 4 (d4) and 7 (d7) days of culture with Chiron. Intracellular β-catenin (Fig. 2b) as well as RORγt (Fig. 2d) levels were signi cantly elevated in T reg s after Chiron treatment at both time points. While β-catenin expression plateaued on d4, RORγt expression in Chiron-treated T reg s increased signi cantly between d4 and d7 (Fig. 2d).
Alternatively, in human primary T cell cultures 19 and in murine thymocytes 42 it was shown that TCR engagement stabilizes β-catenin. To determine the physiological effect of β-catenin activation, we cultured HD PBMCs for 4d in the presence of its natural activator, Wnt3a, in combination with CD3/CD28beads, CD3/CD28-beads alone, or vehicle control (Fig. 2e). CD3/CD28-beads were removed from cultures on d4 and Fr.II T reg cells (Fig. 2f) were analyzed for β-catenin and RORγt expression (Fig. 2g,h) as well as cytokine production (Fig. 2i,j) on d6. Intracellular levels of β-catenin and RORgt directly correlated (Fig 2g) and increased signi cantly in Wnt3a/CD3/CD28 and CD3/CD28 compared to vehicle control treated Fr.II T reg s (Fig. 2h). Similarly, a signi cantly higher frequency of Wnt3a/CD3/CD28 and CD3/CD28 treated Fr.II Conclusively, TCR stimulation in combination with activating Wnt-signals induced a pro-in ammatory phenotype in Fr.II T reg s that is indentical with that observed in IBD(/Dys) patients ( Fig. 1). Hence, these observations imply that β-catenin stabilization in primary human PBMCs is su cient to drive the proin ammatory phenotype of T reg s observed in IBD and CRC patients.
Activated RORγt + sub-populations of T reg s expressing gut homing receptors peripherally expand during disease progression in a murine IBD/CRC model. To recapitulate our ndings in patients and further analyze IBD/CRC-associated RORγt + T reg s, we used our previously established murine APC Δ468 (APC Δ ) polyposis model 43 . In this model, translation of the adenomatous polyposis coli (APC) protein prematurely terminates due to ablation of exons 11 and 12, resulting in a truncated, nonfunctional 468 aa protein. The pathology of APC Δ resembles APC Min/+ mice. They develop intestinal polyposis that spreads to the colon over time, and succumb to disease between 6-8 month of age 44,45 . In previous work we used this model to demonstrate that T reg -speci c ablation of Rorc reduced polyp burden in APC Δ mice 30 .
Treatment of APC Min/+ mice with dextran sodium sulfate (DSS), which causes colitis in mice 46 , leads to the development of invasive CRC 47, 48 . We established invasive colonic lesions (Fig. 3b), by treating 3-4month-old APC Δ mice with 3 rounds of 2% DSS in the drinking water for 7 days followed by 2 weeks recovery (Fig. 3a). This regimen led to colon shortening and a highly signi cant increase in colonic adenomas compared to untreated APC Δ mice (Fig. 3c).
Next, we assessed the frequencies of total colonic T reg s (Fig. 3d) and RORγt + T reg s (Fig. 3e). Each of the three pathologic conditions colitis (WT+DSS), polyposis (APC Δ -DSS), and IBD/CRC (APC Δ +DSS) displayed increased colonic CD25 + Foxp3 + T reg (Fig. 3d) and RORγt + T reg (Fig. 3e) frequencies compared to naïve mice (WT-DSS). As observed in IBD patients, β-catenin expression was signi cantly higher in RORγt + T reg s than in RORγt -T reg s for all treatment groups, (Fig. 3f). However, total T reg s (Fig. 3d) and RORγt + T reg s (Fig. 3e) trended towards even higher frequencies in IBD/CRC compared to colitis or polyposis alone. We speculated that speci c sub-populations of RORγt + T reg s increased under IBD/CRC conditions, but the effect was masked by the overall increase of T reg s. Therefore, we designed two ow cytometric panels comprising CD4, Foxp3, CD25, RORγt, and β-catenin to distinguish distinct RORγt + and RORγt -T reg subsets. The rst panel assessed tissue and in ammation homing markers; gut-homing receptor CCR9, in ammation-homing receptor CCR6, and tissue-residency marker CD103. The second panel focused on proliferation and activation markers Ki67, CD44, CD69, and CD62L (Fig. 3g). We gated on CD4 + CD25 + Foxp3 + T reg s from spleen (SPL) and colon MNCs, concatenated the populations from all experimental groups, and performed tSNE analysis. RORγt + T reg gates were then superimposed onto the respective tSNE landscape revealing unique populations ( Fig. S2a,b). These landscapes show that the complexity of RORγt + T reg populations in the colon is greater than in the spleen (tissue and in ammation homing panel: colon=8, SPL=2 populations, activation panel: colon=8, SPL=3 populations, Fig 3g). To trace the origin of these cells, we focused on populations that expressed the same markers between the spleen and colon (red arrows, Fig. 3g). Analysis of expression pro les ( Fig. S2a,b) revealed that RORγt + T reg populations shared between the colon and spleen expressed CCR9/CD103, and Ki67/CD44 (Fig. 3g).
Cumulative analysis con rmed that CCR9 + CD103 + (Fig. 3h) and CD44 + Ki67 + (Fig. 3i) RORγt + T reg frequencies in the IBD/CRC group (APC Δ +DSS) increased compared to the untreated WT, colitis (WT+DSS), and polyposis (APC Δ -DSS) groups for colonic and splenic samples. Moreover, polyposis (APC Δ -DSS) or colitis (WT+DSS) alone also showed elevated frequencies of the shared T reg populations in the colon, compared to the untreated group. This suggested that during IBD/CRC carcinogenesis a subpopulation of highly activated and proliferative RORγt + T reg s with gut-homing properties expanded in the colon and the periphery (SPL). It further indicated diversity within the RORγt + T reg subset, particularly in the colon.
T reg -speci c β-catenin stabilization in mice results in a pro-in ammatory T reg phenotype in mice. Ex vivo Wnt/β-catenin pathway activation in primary human HD T reg s recapitulated the phenotype of RORγt + T reg s observed in patients (Fig. 2). Accordingly, we examined in mice whether T reg -speci c activation of the Wnt/β-catenin pathway was su cient to induce RORγt + T reg s. We introduced the ctnnb1 (ex3) allele 49 into the Foxp3 YFP-Cre mice established by Rudensky and colleagues 50 . Cre-excision of exon 3 encoding the β-catenin degradation domain leads to intracellular accumulation of β-catenin protein.
These observations mirrored those of a recent study using a different Foxp3Cre 53 with the same oxed βcatenin allele 21 .
The observed pathology in CAT mice implied a lymphoproliferative disorder. Hence, we next assessed the activation status of conventional T cells in the peripheral lymphoid organs and thymi of 21-day-old mice.
Compared to Cre-only, CAT mice had increased frequencies of total CD3 + cells in the spleen and pLNs; with a decrease of CD4 + T helper (T H ) cells in the pLNs and mLNs and an increase of CD8 + cytotoxic T cells (CTLs) in the spleen (Fig. S3e), indicating imbalanced T cell proliferation. This was further supported by the increase of Ki67 + T H cells and CTLs in CAT compared to WT mice (Fig S4a-b). Moreover,

CD25 -Foxp3 -CD4 + T H cells and CTLs were strongly activated in CAT mice. The frequencies of T H cells and
CTLs expressing the activation markers CD44, CD69, and CD25 were signi cantly increased in all peripheral lymphoid organs whereas the fractions of T cells expressing the naïve T cell marker CD62L was reduced. (Fig. S4a-b).
The highly activated effector T cell compartment in CAT mice suggested that β-catenin stabilization altered T reg function. T reg frequencies strongly decrease within mLNs and spleens of CAT mice, however thymic T reg generation was unaltered (Fig. 4a, Fig. S4c). In peripheral T reg s, Foxp3 expression mildly increased (Fig. 4b), whereas expression of the T reg marker Neuropilin (Fig. S4d) did not change. Similar to the human ex vivo cultures, β-catenin stabilization (Fig. 4c) was su cient to uniformely upregulate RORγt ( Fig. 4d) levels in CAT (β-catenin high ) compared to Cre T reg s. Moreover, β-catenin high T reg s were highly activated, as evidenced by increased frequencies of CD44 + (Fig. 4e) and CD69 + (Fig. 4f) T reg s and decreased CD62L + T reg s (Fig. 4g) in CAT mice. Likewise, T reg s in CAT mice were more proliferative, demonstrated by Ki67 staining (Fig. 4h). Lastly, we tested the ability of β-catenin high T reg s to suppress proliferation of polyclonally activated, CFSE-labeled CD4 + CD25 -T effector cells (T eff ) in vitro. Compared to control Cre T reg s, β-catenin high T reg s had signi cantly reduced suppressive function for all T reg :T eff ratios tested (Fig. 4i).
Conclusively, β-catenin stabilization in murine T reg s was su cient to up-regulate RORγt and induce an activated, pro-in ammatory phenotype that mirrored the systemically expanding RORγt + T reg subpopulation observed in the APC Δ /DSS model (Fig. 3i). The lymphoproliferative disease in CAT mice could thus be attributed to a combination of the activated phenotype and the reduced suppressive function of β-catenin high T reg s. β-catenin high T reg s are competitively disadvantaged in a chimeric setting and spontaneously express proin ammatory cytokines. Foxp3 YFP-Cre heterozygous females represent natural chimeras of βcatenin high /YFP + versus WT/YFP -T reg s due to random X-chromosome inactivation. These mice provide the opportunity to test the pathogenic potential of β-catenin high /RORγt high T reg s in a competitive chimeric setting. Interestingly, compared to hemizygous males, heterozygous Foxp3 Cre(+/-) females that carry the Ctnnb1 (ex3) allele are healthy.
When comparing Cre (Foxp3 YFP-Cre(+/-) WT) and CAT (Foxp3 YFP-Cre(+/-) Ctnnb1 (ex3) ) female mice, total T reg numbers in the spleen and thymus were identical between genotypes (Fig. 5a). The frequency of YFP + T reg s, however, was drastically reduced in CAT compared to Cre females (Fig. 5b). Cumulative analysis of YFP + and YFP -T reg fractions showed that YFP + T reg s only accounted for 1-3% of total T reg s in all peripheral lymphoid organs of female CAT mice (Fig. 5c). The thymic output of YFP + T reg s was also reduced to 20 % in CAT females. In Cre females the ratio of YFP + /YFP -T reg s was also slightly reduced to 40%/60% (Fig. 5b,c). This could be attributed to the YFP-Cre transgene rendering Foxp3 expression in YFP + T reg s mildly hypomorphic compared to T reg s with an unaltered Foxp3 locus 54 . The persisting YFP + / β-catenin high T reg s had elevated RORγt expression (Fig. 5d) and an activated phenotype, as evidenced by increased CD44 expression (Fig. 5e) compared to CAT YFP -T reg s but also compared to Cre YFP + /YFP -T reg populations. In accordance to our ndings in patients' RORgt + T reg s, CAT YFP + /β-catenin high compared to YFPand to Cre YFP + /YFP -T reg populations spontaneously expressed proin ammatory cytokines IL-17, IFNg, and TNFa. Further mirroring our nding in patients (Fig. 1), a signi cant proportion of these YFP + /β-catenin high T reg s co-expressed IL-17 and IFNg or IL-17 and TNFa ( Fig. 5f-i). In accordance to ndings in human samples, the gut-homing receptor CCR9 was upregulated (Fig. S5a) as shown for wnt3a/anti-CD3/28-treated human T reg s (Fig. 2l). Conclusively, in a non-in ammed setting, βcatenin high /RORγt high T reg s have a competitive disadvantage in chimeric mice. Although they do not spontaneously cause disease, they are poly-functional in producing proin ammatory cytokines.
The DNA-binding partner of β-catenin, TCF-1, and Foxp3 co-bind accessible chromatin at gene loci that are crucial for T reg function. Previous work indicated that β-catenin 20 /TCF-1 13 occupy DNA together with Foxp3. Thus, we anticipated that precise mapping of TCF-1 and Foxp3 co-binding in T reg s, could provide a molecular explanation for the change in the phenotype of T reg s that overexpress β-catenin. To address this, we analyzed TCF-1 DNA binding in T reg s through chromatin immunoprecipitation and deep sequencing (ChIPseq) in Foxp3 YFP-Cre WT T reg s. Furthermore, we assessed regions of accessible chromatin in T reg s via transposase-accessible chromatin approach and deep sequencing (ATAC-seq). We also utilized available data 55 for Foxp3-binding (Foxp3-ChIPseq), histone marks including; monomethylated histone 3 at Lys4 (H3K4me1), tri-methylated H3 Lys4 (H3K4me3), acetylated H3 at Lys27 (K3K27Ac), trimethylated H3 at Lys27 (H3K27me3), and Methyl-CpG binding domain-based capture and sequencing (MBD-seq).
We assigned TCF-1 (Fig. S6a-c) and Foxp3 (Fig. S7a-c) ChIPseq-peaks to speci c gene regulatory regions by performing K-means clustering guided by the surrounding chromatin modi cations. 14.6% of TCF-1 (Fig. S6a) and 24.0% of Foxp3 (Fig. S7a) binding sites, were assigned to active enhancers (AE, enriched for H3K4me1, H3K27Ac), 41.6% and 39.3% of transcription factor binding sites to poised enhancers (PE, enriched for H3K4me1, lacking H3K27Ac), and 44.0% and 36.7% to promoters (Pr, enriched for H3K4me3, H3K27ac, and accessible chromatin (ATAC-seq signal)). Both TCF-1 and Foxp3 preferentially bound open and poised chromatin as their binding sites rarely overlapped with the repressive histone mark H3K27me3 (Fig. S6b-c, Fig. S7b-c). We next performed motif analysis for TCF-1-and Foxp3-bound sites. TCF-1-bound enhancer sites were enriched for the TCF-1 consensus motif, while both promoter and enhancer Foxp3-bound sites were enriched for the Foxp3 motif (Fig S6d, Fig. S7d). The high enrichment for motifs of other factors like Ets, RUNX and YY family members agreed with previous evidence that TCF-1 14 and Foxp3 12 act in multi-molecular complexes. Pathway enrichment analysis (http://www.metascape.org) revealed that TCF-1 (Fig. S6e) and Foxp3 (Fig. S7e) binding to promoter and enhancer sites marked genes involved in T cell activation and other T cell processes, as well as in general DNA and RNA metabolism processes.
We further analyzed directly overlapping TCF-1 and Foxp3 co-bound sites (Fig. 6), which included 504 AE, 389 PE, and 1101 Pr (Fig. 6a). Examples for TCF-1 and Foxp3 co-bound gene loci at AE (Tcf7), PE (Ccr7), and Pr (Stat1) sites are visualized as IGB tracks (Fig. 6f). The TCF-1 consensus motif, but not the Foxp3 binding motif, was highly enriched in the overlapping AE and Pr sites (Fig. 6d). Pathway enrichment analysis revealed that co-binding of TCF-1 and Foxp3, particularly at AE sites, marked genes involved in Th17 differentiation, T cell activation, and cytokine production pathways (Fig. 6e). Thus, in T reg s TCF-1 and Foxp3 together occupied genes whose upregulation could explain the phenotype of RORγt + T reg s observed in IBD/CRC patients and mouse models. β-catenin stabilization drives the proin ammatory phenotype of RORγt + T reg s through epigenetic and transcriptional regulation of critical Foxp3 and TCF-1 co-bound genes. The classical Wnt signaling model posits that TCF-1 acts as a transcriptional repressor that becomes an activator once β-catenin binds and mediates epigenetic changes. Hence, one could speculate that, in a Wnt-off state, TCF-1 is part of Foxp3 repressor complexes 12 , which act together to limit the expression of genes that should not be permanently expressed in bona de T reg s. Thus, we postulated that β-catenin stabilization speci cally affects the transcription and epigenetic states of these TCF-1/Foxp3 co-bound genes.
To explore this possibility, we performed ATAC-seq analysis of YFP + CD25 + β-catenin high T reg s FACS sorted from CAT (Foxp3 YFP-Cre Ctnnb1 (ex3) ) and WT (Foxp3 YFP-Cre ) mice (Fig. 7a). β-catenin high T reg s (23,084 peaks) had fewer accessible sites than Cre T reg s (29,566 peaks). However, the sites that lost accessibility from Cre to β-catenin high (Cre unique sites, 9831 peaks), already had low accessibility in Cre T reg s and yet were not fully closed in β-catenin high T reg s. The majority of sites remained consistently open for both genotypes (common, 19735 peaks) and 3349 sites gained de novo accessibility in β-catenin high T reg s. We speci cally analyzed the newly accessible sites for transcription factor binding motifs (Fig. 7b).
One of the most strongly enriched motifs was that of TCF-1 (Tcf7), indicating that β-catenin together with TCF-1 may directly alter the accessibility of corresponding genes. Additionally, motifs of transcription factors that are involved in Th17 differentiation (Maf, RARg) and T cell activation (JunB) were highly enriched. The 3349 newly accessible sites corresponded to 2582 genes (Fig. 7a) of which 370 were cobound by TCF-1 and Foxp3 (Fig. 7c). Pathway enrichment analysis of these 370 genes revealed that T cell activation, cytokine production, and Th17 differentiation pathways were most signi cantly enriched (Fig. 7d).
We next performed RNA-seq expression pro ling to investigate how the changes in chromatin accessibility impacted transcription in β-catenin high T reg s. Differential expression testing yielded 3190 upand 2795 down-regulated (q-value≤0.05) genes in β-catenin high compared to Cre T reg s. The upregulated list comprised genes that are essential for T reg function, including CTLA-4 and Il2ra (CD25, Fig. 7e).
Pathway enrichment analysis for signi cantly up-regulated genes that were co-bound by TCF-1 and Foxp3 ( Fig. 7f-g) revealed an enrichment for the very same pathways identi ed by the newly accessible chromatin sites, namely, T cell activation and Th17 differentiation (Fig. S8).
We next assessed the dynamic chromatin accessibility changes following β-catenin stabilization of genes co-bound by TCF-1/Foxp3 in different regulatory regions. Therefore, we compared the log 2 -fold accessibility changes in β-catenin high T reg s over Cre T reg s for promoter, PE, and AE co-bound to that of all differentially accessible genes (Fig. 8a). Although, genes co-bound by TCF-1/Foxp3 always gained accessibility, the AE-bound genes showed the strongest increase in accessibility in β-catenin high T reg s ( Fig. 8a). As AE-bound genes are most affected by β-catenin stabilization, we speci cally searched for those that were up-regulated, gained accessibility, and were TCF-1/Foxp3 co-bound at AE sites in T reg s ( Fig. 8b). This yielded in 49 genes, that were identi ed to belong to T cell activation and Th17 differentiation pathways (Fig. 8b). We further validated these ndings via an inverse, non-supervised approach. We retrieved gene lists for the Th17 differentiation (102 genes) pathway (GSEA/MSigDB database), and used the same T reg_ UP signature (195 genes) used for TCGA screening. The leukocyte migration (242 genes) pathway (GSEA/MSigDB database) was also assessed due to the observed increase in CCR9 expression in RORγt + T reg s in humans (Fig. 2l) and mice (Fig. 3, Fig. S5d) suggesting their migration from the colon to the periphery. Comparing accessibility of genes in these pathways showed strong de novo accessible sites that arise after β-catenin activation in T reg s (red lines) compared to common (black lines) and WT sites (green lines, Fig. 8c). Moreover, gene set enrichment analysis (GSEA 56 ) showed a signi cant positive upregulation for Th17 differentiation and leukocyte migration pathways (Fig. 8d, Fig. S8). The T reg_ UP signature, however, was not consistently changed between WT and β-catenin high T reg s (Fig. 8d). This indicated that the core T reg program was not drastically altered, but non-T reg /effector T cell functions were acquired upon β-catenin activation. For instance, activation of βcatenin led to the acquisition of a newly accessible site in the IFNγ locus (Fig. 5e, black arrow), which may account for the ability of β-catenin high T reg s to express IFNγ. Similarly, the IL-17 locus gained accessibility (Fig. 8e), which may account for spontaneous IL-17 production in β-catenin high T reg s (Fig   5h,i).
Overall, our ndings indicated that β-catenin activation caused epigenetic and transcriptional changes of crucial Foxp3/TCF-1 co-regulated genes. This resulted in the induction of transcriptional programs responsible for Th17 differentiation/T cell activation that were super-imposed onto the T reg program and potentially conferred the observed pro-in ammatory phenotype to β-catenin high T reg s.

Discussion
Diverse T reg populations that express transcriptional programs in addition to the core T reg signature as functional adaptation to the tissue microenvironment and in ammatory milieu have been described 1,2,3,4,5,6,7 . Here we address mechanisms that potentially underlie T reg functional changes in chronic in ammation and cancer, focusing on a population of RORγt + T reg s in the context of IBD and CRC. Our results show that during IBD and progression to dysplasia, RORγt-expressing T reg s that produce proin ammatory cytokines selectively expand in the colon and PB of patients. We linked this RORγt + T reg phenotype to enhanced Wnt/β-catenin signaling in both humans and animal models. Our molecular analyses provide evidence that TCF-1, the DNA-binding partner of β-catenin, together with Foxp3, controls the expression of genes involved in T cell activation and Th17-proin ammatory processes. The stabilization of β-catenin upregulates the expression of these genes by promoting chromatin accessibility and thereby superimposes pro-in ammatory traits onto the core T reg program.
In healthy mice, a speci c consortium of bacteria induces RORγt + T reg s mainly in the colon, where they regulate local in ammatory responses 6, 7 . During IBD and progression to cancer, antigens from bacteria and from the chronically-in amed, damaged mucosal tissue mediate persistent TCR-stimulation 57 .
Furthermore, CRC tumors and the in amed mucosa were shown to secrete enhanced levels of activating Wnt proteins 58, 59 . This persistent TCR-stimulation 19, 42 in a Wnt rich environment has the potential to strongly activate Wnt/β-catenin in T cells. Furthermore, mucosal barrier dysfunction in IBD and IBD/Dys likely enables the PB dissemination of the RORγt + T reg s in an antigen-driven manner leading to their systemic occurance. Indeed, our ex vivo experiments in human PBMCs indicate that TCR-stimulation and activating Wnt signals stabilize β-catenin in T reg s and increase the frequencies of RORγt + T reg s, which express the pro-in ammatory cytokines IL-17, IFNγ, and TNFa. A certain proportion of these RORγt + T reg s showed a poly-functional cytokine pro le by co-expressing IL-17 and IFNg or IL-17 and TNFa simultaneously. Furthermore, genetic β-catenin stabilization in our Foxp3 Cre -driven mouse model induces T reg s that are homogenously RORγt high and spontaneously produce IL-17, IFNγ, and TNFa even in the absence of overt in ammation in a competitive chimeric setting. These ndings suggest that constitutive Wnt/β-catenin signaling in human as well as murine T reg s is su cient to promote the pro-in ammatory RORγt + T reg -phenotype.
We previously showed increased frequencies of circulating RORγt + T reg s in patients with sporadic CRC 30 .
Here we found that in IBD RORγt + T reg s also expand systemically and have proin ammatory properties, promoting the notion that colonic and circulating RORγt + T reg s may be related. This is supported by several ndings. First the observation that β-catenin stabilization both in human PBMCs and in the chimeric mouse model upregulates gut-homing receptor CCR9. Second, the nding that the expanding RORγt + T reg s in the IBD/CRC mouse model express CCR9 as well as tissue-residency marker CD103 undeline their colonic/tissue origin. Finally, the observation that β-catenin stabilization transcriptionally upregulates leukocyte migration genes including CCR7, which promotes cell entry to lymphoid tissues 60 , suggests altered migration. As mentioned above, mucosal barrier dysfunction 61, 62 can cause systemic microbial and tissue-speci c antigen-spread during IBD/CRC and contribute to systemic immune dysregulation 57 . It was previously reported that T reg migration in breast cancer can be Ag-speci c 63, 64 .
Similarly, the above conditions may facilitate migration of gut-tissue-resident RORγt + T reg s and/or support their maintenance in the periphery, thereby increasing their frequencies in IBD and CRC.
The epigenetic and transcriptional studies presented here focus on determining the mechanisms by which T reg -speci c β-catenin-activation promotes the disease-associated RORγt + T reg phenotype. In line with previous evidence that TCF-1 interacted with the signature T reg transcription factor Foxp3 20 , here we show that the genome-wide TCF-1-DNA binding in T reg s substantially overlaps with that of Foxp3 13 . Importantly, we nd that TCF-1 and Foxp3 co-bind active enhancer regions of genes responsible for Th17 differentiation, T cell activation, and leukocyte migration. Recent studies have shown that Foxp3 binding to active enhancers 9 can activate or suppress expression of the associated genes depending on local interactions with other speci c cofactors. In particular, Foxp3 likely acts as a repressor in complexes with YY1, EZH2, and Ikzf3 12 . Based on current evidence 20, 13 and our own results, TCF-1 possibly participates in repressive regulatory complexes with Foxp3. The nding that the conserved YY1 binding motif is strongly enriched in the Foxp3/TCF-1 co-bound DNA-sites further supports this suggestion. Moreover, our ndings that β-catenin stabilization enhances chromatin accessibility and the expression of certain TCF-1/Foxp3 co-bound genes indicates that the TCF-1/Foxp3-mediated regulation of these genes depends on β-catenin expression levels. Enhanced β-catenin binding to TCF-1 may alter the TCF-1/Foxp3 regulation of these gene loci by displacing co-repressors to promote transcription. Indeed, the regions that gain de novo accessibility with β-catenin stabilization are enriched for the TCF-1 binding motif, suggesting that the β-catenin-mediated epigenetic changes involve TCF-1. These molecular changes mediated by βcatenin promote a duality in the phenotype of RORγt + T reg s by enhancing pro-in ammatory Th17 differentiation and leukocyte migration pathways without abrogating the regulatory core T reg signature.
In conclusion, our study establishes the progressive and systemic expansion of RORγt + T reg s during IBD and progression to dysplasia and highlights activation of β-catenin as a major underlying molecular process. We show that TCF-1 and Foxp3 function together to control the expression of a proin ammatory program through binding to active enhancer regions of the associated genes. We further de ne that this regulation is driven by β-catenin stabilization, which promotes chromatin accessibility and enhances transcription of the TCF-1/Foxp3 co-regulated genes. β-catenin mediates upregulation of T cell activation and Th17 signature genes and superimposes this activated proin ammatory phenotype onto the core T reg program. Future studies should establish the ontology and TCR speci cities of the diseaseassociated RORγt + T reg s, as well as their roles in perpetuating in ammation. It is curcial to understand the exact contributions of RORγt + T reg s to chronic intestinal in ammation and malignancies, which rst and foremost should address the effects of pro-in ammatory cytokine production by these cells on disease progression.
The present data connects to our previous report that frequencies of pro-in ammatory T reg s increase in the blood of CRC patients 30 and expands it to a pre-malignant setting of chronic in ammation. The step wise increase in the frequencies of RORγt + T reg s expressing IL-17, IFN-γ, and TNFa from IBD to early dysplasia and CRC emphasizes the involvement of these cells in pro-in ammatory immune dysregulation fostering chronic in ammation and therewith acting cancer-promoting. Early intervention in cancer treatment is greatly needed. In this context frequencies of circulating RORγt + T reg s could serve as a biomarker in IBD and CRC. Furthermore, our detailed molecular analysis of genes and processes affected by -catenin stabilization in T reg s could help identify targets to limit the systemic expansion of potentially disease-promoting pro-in ammatory T reg s during CRC.

Materials And Methods
Patients and Healthy Donors. 43 patients undergoing major surgery for Crohn's disease (CD)/ulcerative colitis (UC) treatment (proctocolectomy, colectomy, partial ileectomy, or combinations of the aforementioned procedures) were consented to donate 45ml of peripheral blood either directly preop or directly postop. Up to 4g of fresh intestinal mucosal tissue was collected from in amed, margin (in most cases less in amed), and dysplastic areas from the surgical specimens (if made available for research purposes by pathology) from the same patients. Additionally, 22 IBD patients undergoing routine colonoscopy check-up visits and 20 healthy individuals participated to the study by donating up to 45ml of peripheral blood. Informed consent was obtained from all patients and healthy donors. The protocol was approved by the University of Chicago Division of the Biological Sciences Institutional Review Board (BSD IRB) and was performed in accordance to Federal Law.
Isolation and ow cytometric analysis of PBMCs and MNCs from patient and healthy donor samples.
PBMCs were isolated from peripheral blood samples using LymphoprepTM (Serumwerk Bernburg AG for Alere Technologies AS, via Stemcell 7851) density gradient centrifugation. MNCs were isolated from mucosal tissue samples using the Tumor Dissociation Kit, human (Miltenyi 130-059-929) and gentleMACS C Tubes (Miltenyi 130-093-237) following the instructions from the Lamina Propria Dissociation Kit, mouse (Miltenyi 130-097-410). MNCs isolated from blood and intestinal mucosa were either rst re-stimulated with Cell Stimulation Cocktail plus protein transport inhibitors 500x (eBioscience 00-4975-93) for three hours in X-Vivo-20 media (LONZA 04-448-Q) or used directly for ow cytometric staining. Cells were stained for viability using the LIVE/DEAD Aqua uorescent reactive dye (Molecular Probes -life technologies L34963), followed by surface staining with varying combinations of the following antibodies: CD3 (clone OKT3, BioLegend; UCHT1, ThermoFisher), CD4 (clone RPA-T4, BD), CD8 (RPA-T8, eBioscience/ThermoFisher Scienti c/BD), CD25 (clone M-A251, BD), CD45RA (clone HI100, BD/BioLegend), CD127 (clone HIL-7R-M21, BD), CD279/PD-1 (clone EH12.1.H7, BioLegend), CD152/CTLA-4 (clone 14D3, eBioscience/ThermoFisher Scienti c), CCR9 (L053E8, BioLegend (V27-580, BD) and respective manufacturer isotype controls. Following surface staining, cells were xed and permeabilized using Foxp3/Transcription Factor Fixation/Permeabilization Concentrate and Diluent (Affymetrix/eBioscience/ThermoFisher Scienti c, 00-5521) and intracellular staining was performed for Average z-scores are calculated as relative mRNA expression value of an individual gene and tumor, normalized to the gene's expression distribution in a reference population (de ned as all samples that are haploid for the respective gene). Then, the z-scores of all genes in each individual signature were averaged over the number of genes in the respective signature, yielding average z-scores for the Th17_UP, Treg_UP, and KEGG_human_WNT signatures for each patient. Regularized Regression of TCGA Gene Expression; In order to determine the ability of the TH17 gene signature to inform patient-level survival, we deployed Cox proportional-hazards regression on colorectal cancer clinical data extracted from TCGA. We used Ridge regression regularization 65 and chose an optimal penalty coe cient λ using 10-fold cross validation optimized for partial likelihood deviance. The TH17 gene-based score for each patient was generated by multiplying the coe cient vector by the patient-level gene expression. Patient scores were dichotomized at the median and Kaplan-Meier curves were plotted. Statistical coding was performed using R version 3.1.2, and packages survival 66 , glmnet 67 , and survminer 68 .
Mice. Foxp3 YFP-Cre Ctnnb1 (ex3) 49 , Foxp3 YFP-Cre 50 , APC Δ468 43 , CD45.1 and WT mice were used in all experiments described. All mice were maintained on the C57BL/6 background. Mice were housed under speci c pathogen-free (SPF) conditions in the animal facilities at the University of Chicago in accordance with protocol #71880, approved by the University of Chicago Institutional Animal Care and Use Committee.
Mouse model for IBD-associated colon cancer. 3-4 months old APC Δ468 (male and female) and agedmatched WT mice were treated with 3 rounds of 2% dextran sulfate sodium salt (colitis grade, 160110, MP) in the drinking water ad libitum over 7 days with a 2-week recovery period on normal drinking water between each round (see experimental treatment scheme Fig. 3a). Weight loss was monitored every other day.
Histological assessment of in amed tissues. Peripheral lymphoid organs and vital organs were resected from mice and directly ex vivo formalin xed (HT5014, Sigma-Aldrich). Colons were harvested from mice, ushed free of feces with ice-cold HBBS, formalin xed, and Swiss rolled. All xed organs were para n embedded and 4mm sections were used for H&E staining. Embedding, sectioning and H&E staining service was provided by the University of Chicago Human Tissue Resource Center (HTRC).
Isolation of lymphocytes from thymi, peripheral lymphoid organs, and the intestinal mucosa of mice. Thymi, peripheral lymph nodes, mesenteric lymph nodes, and spleens were resected. Single After 48h, CD4 + CD25 + T reg s from Foxp3 YFP-Cre Ctnnb1 (ex3) and Foxp3 YFP-Cre mice were added, respectively. T reg s were puri ed from peripheral lymph nodes and spleens using the CD4+CD25+ Regulatory T cell Isolation Kit II (130-091-041, Miltenyi). The assay was performed in triplicates with T reg to T effector ratios of 1:1, 1:2, and 1:4. Samples were incubated for another 48h and then collected for ow cytometric read out. Cells were stained for viability and then surface stained with CD45.1 (clone A20,  Genome mapping and data analysis. Sequenced ChIP and ATAC datasets were mapped with the Galaxy (https://usegalaxy.org/) suite of tools. Data were groomed and aligned to the mouse mm10 genome with Bowtie, allowing up to one mismatch and retaining only uniquely mapped reads, and unmapped reads were ltered. For transcription factors (TCF-1, Foxp3) peak calling was performed with MACS via HOMER 71 . Transcription factor peak calling was performed relative to input controls with the requirement that peaks be at a minimum 5-fold enriched over input and meet a p-value cutoff of 1 x 10 -5 . Openchromatin (ATAC-seq) peaks were called with Macs2 72 with --nomodel set and no background provided.
Differential accessibility testing of ATACseq data was performed using the edgeR Bioconductor package 73 . Sequenced RNA datasets were aligned to the mouse mm10 genome like the ChIP-seq datasets. Differential gene expression analysis was performed with Cuffdiff2 74 . Genes with transcript abundance differences below q<0.05 were considered to be signi cantly differentially expressed. Heat maps of normalized reads for gene subsets were generated with using Genepattern 75 . Motif analysis was performed with the HOMER motif-discovery algorithm, and transcription-factor overlap analysis was conducted with the HOMER mergePeaks command, considering only peaks that directly overlapped. Peaks were annotated to the mm10 genome with annotatePeaks.pl in HOMER. Histograms for transcription factors and histone modi cations were generated with ngs.plot software 76 . K-means clustering of ChIP-seq datasets and heat maps were also generated with ngs.plot. Transcription factor binding was visualized with the Integrated Genome Browser software 77 . Pathway enrichment analysis for genes identi ed by ChIP-seq and RNA-seq analysis was performed via Metascape (http://metascape.org) 78 .
Statistical analysis. Results from biologically distinct experiments were combined and analyzed with the indicated statistical tests in Prism 7 (GraphPad). The statistical signi cance of RNA-seq data was determined with Cuffdiff and for ATACseq data with edgeR. ChIP-seq (factor enrichment) and ATAC-seq (chromatin accessibility) p-value cutoffs were determined with MACS2. Gene pathway enrichment pvalues were determined with Metascape. Data are presented as mean ± SEM. Ex vivo stabilization of β-catenin in human Tregs is su cient to induce the pro-in ammatory phenotype.   β-catenin DNA-binding partner TCF-1 co-binds accessible chromatin together with Foxp3 at gene loci crucial for bona de Treg integrity and function. a, Venn diagrams of overlapping TCF-1 (blue) and Foxp3