TCF1 controls T reg functions that regulate inflammation, CD8 T-cell cytotoxicity, and severity of colon cancer.

TCF1 is essential for the development and function of T regulatory cells (T reg s). However, how TCF1 regulates T reg function in homeostasis or under pathogenic conditions is poorly understood. Here, we ablated TCF1 in T reg s to elucidate its role in T reg specification in healthy mice and mice with colon cancer. RNAseq revealed that TCF1-deficient T reg s maintain their core transcriptional signature, but promote T-cell receptor, Tgfβ receptor, T H 17, and Wnt/β-catenin signaling pathways. Single-cell RNAseq identified central-memory-T reg s with low Klf2 or high Mif expression, which upon downregulation of TCF1 gained T H 17 characteristics and matured into Maf expressing effector T reg s. Tcf1-deficient T reg s exhibited enhanced suppression of T-cell proliferation and cytotoxicity but were compromised in controlling CD4 + T-cell polarization and inflammation. In mice with polyposis, Tcf1-deficient T reg s promoted inflammation and tumor growth. Thus, TCF1 differentially regulates T reg control of T H 17 inflammation and T-cell cytotoxicity, and its action in T reg s determines colorectal cancer outcome. and in mice with polyposis. We report that T reg s lacking TCF1 become activated and gain T H 17 properties while maintaining their core T reg gene expression signature, molecular heterogeneity, and robust T reg characteristics. Functionally, TCF1 deficiency compromises the ability of T reg s to suppress T H 1 or T H 17 polarization of CD4 T-cells, but enhances their suppression of T-cell proliferation and antigen-specific cytotoxicity. T reg specific deficiency of TCF1 increases tumor load and aggression in mice predisposed to polyposis. We conclude that TCF1 differentially controls independent T reg suppressive functions, and that deregulation of this mechanism renders them pathogenic and tumor promoting in genetically susceptible hosts. demonstrate that TCF1 coordinates activation and independent Treg suppressive activities, in a manner that is relevant to the role of T reg s in It is noteworthy that T reg s with pro-inflammatory properties are not restricted to the cancer pathology, and are also in individuals in fungal and bacterial infections, reviewed in 126, Thus, the has physiological relevance.


Introduction
Regulatory T-cells (Tregs) are a heterogenous population of cells of thymic and extrathymic origins with diverse immune suppressive functions. Expression of the lineage-determining transcription factor Foxp3 is essential for maintaining Treg identity 1, 2, 3 , but is not sufficient to account for their substantial functional diversity 4 . In addition to FOXP3, Tregs can express other transcription factors that are normally associated with T-helper cell functions, namely RORγt 5, 6, 7 , GATA3 8,9 , or TBET 10,11,12 . More than half of gut-infiltrating Tregs in healthy mice express RORγT 6,7 and cMAF 13,14,15 , and these Tregs critically maintain immunotolerance and host microbe homeostasis.
Recent single-cell RNA sequencing (scRNAseq) studies have identified two transcriptionally distinct subpopulations of activated/effector-Tregs (eTregs) in addition to central-memory-Tregs (cTregs) 7,24,25 . One expresses elevated levels of RORC and its regulator c-Maf and the other expresses IKZF2/HELIOS, GATA3, and downstream ST2/IL33-receptor. The mouse RORγT + Tregs are generated from naïve conventional CD4 + Tcells (Tconv) after stimulation by bacteria antigens and also through interaction with the enteric nervous system 7,24,25 . The IKZF2/HELIOS Tregs are mainly of thymic origin, although a subset of these that potentially originate from Tconv cells can convert to RORγt + Tregs 25 . Analysis of Tregs in humans also revealed several subpopulations of eTregs and cTregs 26 . Tregs expand in response to inflammation and tumor growth. Treg adapt to environmental challenges to re-establish homeostasis or drive pathogenic conditions, but the molecular underpinning of these responses at the single cell level is still poorly understood.
Tregs expressing RORγT and IL17 have been described in fungal infections of mice 27 and were attributed with proinflammatory functions in humans 23,28,29 . In colorectal cancer (CRC) a notable subset of tumor-infiltrating Tregs express IL17, RORγT 23,29 , and high levels of β-catenin 30 . This subset is readily detectable in circulating blood of CRC patients but rapidly decline with surgical removal of the primary tumor 29,30 . Expansion of RORγT + Tregs in CRC is tumor dependent and coincides with suppression of T-cells, but overall loss of Treg expression of IL10 and compromise of their anti-inflammatory properties 29,30 . In mouse models of hereditary colon cancer Tregs over express β-catenin, RORγT, and IL17, and promote tumor growth 23,29,31 . Treg-specific ablation of RORγt attenuates inflammation and polyposis 23 . Using mouse models, we demonstrated that β-catenin increases chromatin accessibility at the Rorc locus and other genes associated with TH17 inflammation to render Tregs proinflammatory 30 (Quandt et al, Nature Immunology in press). These findings were corroborated by an independent report of Tregs in multiple sclerosis, which express elevated expression of β-catenin, secret Ifnγ , and promote inflammation, but maintain their ability to suppress T-cells 32 . In total, these obserations indicate a pathogenic βcatenin and RORγT axis in Tregs that alters Treg functions, rendering them pro-inflammatory but highly T-cell suppressive in diseases such as cancer and autoimmunity. The contrast between the properties of RORγT + Tregs in health and disease is strinking and urges better understand of the molecular underpinning of proinflammatory RORγT + Tregs.
TCF1 is the T-cell specific DNA binding partner of β-catenin 33,34 . TCF1 interacts with FOXP3 35,36 to repress the MAF-RORγ axis 37 and preserve Treg suppressive functions. However, germline TCF1 deficiency induces premature expression of FOXP3 in double-positive thymocytes 38 and expands thymic Tregs 39 , suggesting a role in Treg specification rather than maintaining Treg identity. Here, we address how the loss of TCF1 post thymicselection alters Treg properties in healthy mice and in mice with polyposis. We report that Tregs lacking TCF1 become activated and gain TH17 properties while maintaining their core Treg gene expression signature, molecular heterogeneity, and robust Treg characteristics. Functionally, TCF1 deficiency compromises the ability of Tregs to suppress TH1 or TH17 polarization of CD4 T-cells, but enhances their suppression of T-cell proliferation and antigen-specific cytotoxicity. Treg specific deficiency of TCF1 increases tumor load and aggression in mice predisposed to polyposis. We conclude that TCF1 differentially controls independent Treg suppressive functions, and that deregulation of this mechanism renders them pathogenic and tumor promoting in genetically susceptible hosts.

TCF1 deficiency programs Tregs for activation, TH17 characteristics, and TGFβ signaling.
To understand how Tcf1 (encoded by Tcf7) alters Treg properties, we generated mice double homozygous for the conditional Tcf7 fl/fl (European Mouse Mutant Archive) 40 and Foxp3 Cre alleles 41 . FACS analysis of mononuclear cells from the mesenteric lymph nodes (MLNs) of the Foxp3 Cre Tcf7 fl/fl mice confirmed that TCF1 expression was lost in Tregs ( Fig. S1a-b) and not in conventional CD4 + T-cells (Tconvs) (Fig. S1c). We first performed bulk RNAseq analysis comparing purified Tregs from pooled lymph nodes of 12-week-old Foxp3 Cre Tcf7 fl/fl and control Foxp3 Cre mice (n=3 independent biological replicates). We noted 1,090 upregulated genes (fold change >1. 5  and 422 downregulated genes including Ctla4, Ikzf2/Helios, and Gzmb (Fig. 1a). To identify pathways that are affected by TCF1, we performed gene set enrichment analysis (GSEA) on all Kegg pathways comparing transcriptomes of knockout and wildtype Tregs (FDR<0.25; Fig. 1b, Table S1). TCF1-affected genes were not differentially enriched in core Treg signature genes (Fig. S1d), suggesting that the Treg core program is not compromised in Tcf1-deficient cells. Genes more highly expressed in TCF1-deficient Tregs than in the control were strongly enriched in WNT signaling, MAPK signaling, TH17 differentiation, IL17 signaling, TGFβ signaling, and T-cell receptor (TCR) signaling (Fig. 1b). The enhanced WNT signature could be the result of loss of TCF1mediated transcription inhibition 34 (Fig. 1c). Collectively, our data show that TCF1 deficiency affects large numbers of genes and modulates several pathways related to the activation, function, and polarization of Tregs.
To confirm these data at the protein level, we next performed FACS analysis (Fig. 2). TCF1-deficient Tregs had elevated ratios of multiple cell surface activation markers 56,57 including CD69, ICOS, PD1, and CD44 + CD62L - (Fig. 2a). The Foxp3 Cre Tcf7 fl/fl mice had higher Treg to CD4 T-cell ratios, and higher frequency as well as absolute numbers of Tregs in secondary lymphoid organs than the Foxp3 Cre mice (Fig. 2b). However, fewer TCF1deficient Tregs expressed CD25 than TCF1-sufficient Tregs (Fig. 2b). Tconv cells in the lymph nodes and spleens of Foxp3 Cre Tcf7 fl/fl mice were more activated than in control Foxp3 Cre mice, in agreement with an earlier study 36 (Fig. 2c). We also observed elevated frequencies of RORγt-expressing Tregs in the MLN, spleen, small bowel, and colon, regardless of their expression of HELIOS (Fig. 2d). TCF1-deficient Tregs expressed higher levels of TGFβR1, TGFR2, and pSMAD2/3 (Fig. 2e). In addition, pSTAT5 and pS6, a downstream target of mTORC1 activity, were more highly expressed in TCF1-deficient Tregs (Fig. 2e). This is notable because the high basal activity of mTORC1 is an in vivo feature of Tregs 58, 59 (reviewed by Chapman and Chi 60 ). In conclusion, our protein level data show that TCF1 deficiency results in the activation of core Treg and TH17 signatures, consistent with the RNA data.

Single-cell transcriptomics delineate distinct Treg subpopulations in the mesenteric lymph nodes.
To better understand how TCF1 regulates Treg properties and possible relevance to colon cancer, we performed scRNAseq of Tregs from MLNs of 5.5-month-old mice with four different genotypes (n=2 replicates per genotype): Foxp3 Cre Tcf7 fl/fl and its control Foxp3 Cre mice, the polyposis prone Apc D468 mice and control WT C57BL/6J (B6) mice. We enriched Tregs to over 90% purity, using the untouched magnetic bead purification of CD4 + T-cells followed by positive selection for CD25 (Miltenyi) (Fig. S2), and subjected them to scRNAseq using the 10X genomics platform. An unbiased integrative analysis across all four genotypes after regression for potential artifacts using the seurat platform 61 resulted in 14,487 cells grouped into 10 major subpopulations on UMAP projection (Fig. 3a, Table S2; see Methods). These subpopulations were annotated according to the most salient identified cell marker (Fig. 3b). We identified three clusters of Tregs with activated/effector characteristics (eTregs), and annotated them as Maf, Ikzf2, and Mif based on their high expression of the corresponding genes.
The Maf cluster had the highest expression of Maf as well as S100a4, Rorc, Hif1a, Icos, and Rora (Fig. 3b,c; S3a). The Ikzf2 cluster had the highest expression of Ikzf2/Helios as well as Rora , and Gata3, IL7r, and Klrg1, and the second highest expression of Maf and Icos (Fig. 3b,c; Fig. S3a). HELIOS is a member of the IKAROS transcription factor family that regulates several Treg suppressive functions 62 and is preferentially but not exclusively expressed by thymus-derived naïve/cTregs 63 . The Ikzf2 cluster was enriched for Gata3-expressing Tregs, which are largely thymus derived and have strong IL10-dependent anti-inflammatory properties important for maintaining gut homeostasis 16,64 . The first two of these clusters have been described earlier as the RORγT + and the HELIOS + subsets of Tregs 25 in mice, or as nonlymphoid T-cell like (nLT) Tregs in mice and eTregs in humans 26 . The Mif (macrophage migration inhibitory factor) cluster has not been previously described, and could represent an early stage of pTreg differentiation or T-follicular suppressor Tregs 65,66 . MIF is the receptor for CXCR2 and CXCR4 67, 68 and regulates TLR4 expression and TCR signaling 69 . Of note, CXCR4 is a key chemokine for attracting extrafollicular T-lymphocytes 70  and Hif1a relative to the other clusters (Fig. 3b). It also expressed Maf, Rorc, and Gata3 but less than the other two eTreg clusters. To obtain insight into the function of each subpopulation, we performed gene ontology pathway analysis on the genes upregulated in each cluster when compared to others using Metascape 78 (Fig. 3d). Genes upregulated in Maf and Ikzf2 clusters represented pathways associated with lymphocyte activation, immune response, negative regulation of immune system process, positive regulation of cytokine production, and high apoptotic signaling; by comparison, genes upregulated in the Mif cluster represented fewer pathways, the most outstanding one being regulation of response to cytokine stimulus (Fig. 3c).
Five other clusters (Klf2 -, Klf2 + , Klf2 ++ , Ncoa3) expressed naïve/central-memory markers (cTregs), comparable to the earlier descriptions of human central Treg (cTreg) 26 . Three of these cTreg clusters were named by their relative expression of Kruppel-like Factor 2 (Klf2) as Klf2 -, Klf2 + , and Klf2 ++ (Fig. 3a-c). Klf2 is highly expressed by early thymic emigrants (ETE) 79,80 . The Klf2 ++ cluser expressed other ETE markers including S1pr1 81,82,83 , Sell (Lselectin/CD62L) 84 , and Igfbp4 85 . KLF2 86 and S1PR1 87 are involved in Treg migration to and establishment of immunological tolerance by naïve Tregs in secondary lymphoid organs 86 , and IGFBP4 blocks signaling by insulinlike growth factor and inhibits extrathymic induction of Tregs 88 . The Klf2cluster stood out from the rest by its low expression of Klf2 and ETE markers (Fig. 3b). Since Klf2and Klf2 ++ were the two largest Treg subpopulations ( Fig. 3a), we directly compared their upregulated genes and identified the 20 most enriched pathways using Metascape 78 . The most significantly enriched pathways of the Klf2 ++ cluster included T-cell migration and leukocyte cell-cell adhesion (Fig. 3e). By contrast, Tregs of the Klf2cluster were enriched for TH17 cell differentiation, IgA production, and cytokine production (Fig. 3e), indicating response to microbial antigens. Both clusters were enriched for leukocyte activation and immune response pathways, indicative of Treg function within the secondary lymphnodes (Fig. 3e). The Ncoa3 cluster expressed low to intermediate levels of Klf2, but was unique in its high expression of Ncoa3, a nuclear co-activator partner of arylhydrocarbone receptor 89,90 and estrogen receptor alpha 91 , as well as relatively high expression of Notch2 (Fig. 3b). The potential distinct function of this subpopulation is currently unknown. Although small, the Ifn subpopulation was conspicuous by its expression of multiple interferon response genes including Stat1, Ifit1, Ifit3, Ifit1bl1, and Ifit3b (Fig. 3b,c).
Functionally, genes upregulated in this cluster were enriched in response to type-1 interferon, response to Ifnγ, and regulation of innate immune response (Fig. 3d). The Vsp8 cluster shared cTreg expressed genes, except for the strong expression of Vps8, a subunit of the CORVET complex that coordinates lysosome fusion with endosomes 92, 93 (Fig. 3a,b,d). The remaining cluster, namely Cd63, was spatially isolated from the other Treg clusters, and is likely not to be Tregs because of its expression of myeloid cell makers and low to undetectable expression of the core Treg transcript 36,42,94 Izumo1r (FOLR4) (Fig. 3a,b).
These analyses corroborate and earlier findings that indicate the existence of two eTreg and multiple cTreg clusters in mice, and further define a new Mif cluster with T-follicular suppressor classificataions and a novel Ifn cluster with narrow expression of interferon response genes. We classify the remaining cTreg clusters with respect to their relative expression Klf2, the significance of which becomes more obvious with our velocity analysis of the pseudotime trajectory of the clusters relative to the eTreg clusters, as described below.

TCF1-deficient and sufficient Tregs show distinct effector functions.
To better understand the function of Tcf1 in Tregs, we focused on scRNAseq data from Foxp3 Cre Tcf7 fl/fl Tregs and control Foxp3 Cre Tregs. Side-by-side comparison of UMAP projections did not reveal any noticable change in the numbers or cellular contents of the Treg clusters with ablation of Tcf7 (Fig. 4a). The quantitative analysis of the Tregs in each cluster further coroborated similar cellularities across the two genotypes (Fig. 4a, right panel). To identify genes that are regulated by Tcf1, we directly compared the expression profiles of Tcf1-deficient to Tcf1sufficient cells in each cluster ( Fig. 4b; see Table S3 for full data). Several genes were broadly elevated across all TCF1-deficient Treg clusters including Dnaja1 that ecodes a heatshock protein cochaperone, and Erdr1 a bacteria-sensitive secreted apoptic factor 95, 96 ( Fig. 4b and Fig. S4). Others were uniformly downregulated, such as Igfbp4, an inhibitor of insulin-like growth factor receptor signalling 88 ( Fig. 4b and Fig. S4). Cells lacking TCF1 had elevated expression of Maf across all Treg clusters with the exception of Klf2 + and Ifn clusters, and elevated Hsph1 in all but the Ncoa3 cluster ( Fig. 4b-c). cMAF is essential for the generation of RORγT + Tregs 13,14,15 , and is negatively regulated by TCF1 37 . Hsph1 encodes a heat shock protein that marks Treg activation and is important for suppressive functions 97,98 . Other genes with strongly elevated expression in the Maf and Ikzf2 clusters, and their potential precursors the Klf2and Ncoa3 clusters, included CCR9, that encodes a gut tropic chemokine receptor which is essential for Treg regulation of TH17 inflammation 99,100 , and fibrinogen-like-protein-2 (Fgl2), a downstream target of TIGIT and an indispensable molecule for Treg suppression of autoimmunity 101 ( Fig. 4b-c). Enhanced expression of the gut-associated intergrin Itgae/CD103/aE-integrin was limited to the Klf1and Maf clusters (Fig. 4b).
The Ifn cluster was exceptional in showing the least change with the ablation of TCF7, except for Dnaja1 (Fig. 4b) and Erdr1 (Fig. S4). The intercluster relations were highly stable across all mouse genotypes analysed ( Fig. S3b) and did change with the ablation of TCF1 in Tregs.
To gain further insight into the function of TCF1, we performed GSEA (Stubbington) comparing TCF1-deficient and sufficient cells in each Treg cluster. Lack of TCF1 had minimal effects on the core Treg program ( Fig. 4d with representative GSEA plot shown in Fig. 4e) consistent with our observation from bulk RNAseq (Fig. S1d). TCF1 deficiency broadly enhanced the expression of TH17 (Fig. 4d) and/or IL17 ( Fig. 4f) program genesets across Treg clusters, using independent Stubbington and Kegg pathway module scores. The Vps8 cluster was exeptional in having highly-induced TH1 signaling along with TH17 signature, raising speculation that this cluster may be an intermediate to pathogenic TH17 cells that co-express TH1 and TH17 cytokines 102, 103 (Fig. 4d). Collectively, these findings are compatible with loss of TCF1 resulting in the broad activation and gain of TH17 properties by Tregs, with the Maf cluster exhibiting the strongest change in gut homing potential and the Vsp8 cluster potentially differentiating into Teff cells.
To determine the intercluster relations and how these may change with the loss of Tcf1, we overlaid RNA velocity vectors on the UMAP projection of Tregs (Fig. 4g). RNA velocity predicts the future state of the cells using unspliced and spliced mRNAs from scRNAseq data 104 . The velocity vectors suggested that the Maf Treg cluster mainly derives from the Mif cluster, but to some extent also from the Ikzf2 and therefore indirectly from two cTreg clusters, the Klf2and Ncoa3. The lack of relation of the Mif cluster to cTreg clusters indicated a potential extrathymic origin (Fig. 4g). This finding is compatible with the notion that Rorγt + Tregs, which are enriched in the Maf cluster, are extrathymically generated by bacterial stimulation of CD4 + Tconv cells 7,24,25,42,105 . The Ikzf2 cluster derived from the Klf2and Ncoa3 cTreg clusters (Fig. 4g). These findings are compatible with earlier reports that the Ikzf2/Helios + Tregs are largely of thymic origin, and capable of converting to Rorγt + Tregs but with relatively lower efficiency than CD4 + Tconv cells 25 . The Vsp8 cluster vectors pointed away from all other Treg clusters, in agreement with their potential differentiation into Teff cells. Other inter-relations between the cTreg clusters such as Ncoa3 with Klf2and Klf2 ++ with Ifn suggest plasticity and/or sequential stages of maturation of the cTregs.
The above findings were supported by scRNA analysis of Tregs from WT and polyposis ridden Apc D468 mice . Treg clusters and relative frequencies of cells within each cluster did not change with polyposis in Apc D468 mice ( Fig.   S5a; Table S4). The two activated Maf and Ikzf2 eTreg clusters had lower expression of Tcf7 as compared with the cTreg clusters, suggesting that Treg activation requires downregulation of TCF1 (Fig. S5b). However, there were differences in gene expression between Tregs from polyposis and WT mice. Expression of Maf was enhanced in the Maf cluster, Jund and Tgfb1 were stronger in the KLf2cluster, and Jund and Soc3 were elevated in the KLF2 ++ Treg clusters of polyposis mice as compared to WT mice (Fig. S5c&d). JunD, encodes an AP1 transcription factor that is activated downstream of the TCR 106, 107 , Socs3, encodes a major regulator of IL-23mediated STAT3 phosphorylation and TH17 generation 108 , and Lag3, encodes a ligand for major histocompatibility complex class II (MHC-II) 109, 110 that is a critical mediator of immune suppression by Tregs 111,112 . Therefore, these transcriptional changes are compatible with activation of select cTreg clusters and enhanced immune suppressive activity of the Maf eTreg cluster during polyposis. Velocity analysis revealed similarities conserved intercluster relations among Tregs of polyp-ridden and healthy mice. One possible exception was the enhanced differentiation of Tregs from the Ikzf2 to the Maf cluster in polyp-bearing mice Apc D468 mice relative to healthy WT control mice (Fig. S5e). Collectively, these findings are consistent with the activation and enhanced differentiation of Maf cluster Tregs in polyposis relative to WT mice. The changes introduced in Tregs by polyposis are similar to those caused by the loss of TCF1 in Tregs of healthy mice.

TCF1-deficient Tregs suppress viral antigen-specific CD8 + T-cell cytotoxicity and T-cell proliferation.
Treg suppression of CD8 + T-cell cytotoxicity is TGFβ dependent 51,52,113 . Given their activated expression profile, preservation of the core Treg signature, and the enhanced TGFb signature, we predicted that TCF1-deficient Tregs would efficiently suppress CD8 + T-cell cytotoxic activity. To test this, we compared in vivo responses of  Fig. 5b,c). Treatment of mice with the LY3200882 inhibitor abrogated this difference, and increased the killing of VP2121 pulsed splenocytes in Foxp3 Cre Tcf7 fl/fl mice to over 55% (Fig. 5b,c). Our data suggest that the Treg-specific loss of TCF1 in Foxp3 Cre Tcf7 fl/fl mice augments the suppression of anti-viral CD8 + T-cell response. Mechanistically, the enhanced suppressive activity of Foxp3 Cre Tcf7 fl/fl Tregs can be related to TGFβR1 signaling signaling as it was abrogated with LY3200882 a small molecule inhibitor of TGFβR1 signaling.
These differences can be accounted for by enhanced TGFb-dependent inhibition of CD8 + T-cell cytotoxicity in mice.
To further assess the mechanisms of CD8 + T-cell suppression, we monitored the fate of viral antigen-specific CD8 + T-cells in the same mice using VP2121-130 tetramers. TMEV infection of Foxp3 Cre mice triggered the expansion of VP2121-130-specific CD8 + T-cells nearly 14 fold from the steady-state frequency of 0.07% to almost 1% (p=0.004) of total CD8 + T-cells at the peak of response to TMEV. This expansion was less than half in the infected Foxp3 Cre Tcf7 fl/fl mice, showing significant difference with baseline (p= 0.004) and with the expansion of the specific CD8 + T-cells in mice with TCF1-deficient Tregs (p=0.016). Thus, the proliferation of VP2121-130-specific CD8 + T-cells was more strongly suppressed in Foxp3 Cre Tcf7 fl/fl mice than in the Foxp3 Cre mice (Fig. 5d). To independently validate the inhibition of T-cell proliferation by TCF1-deficient mice, we performed in vitro proliferation inhibition assays (Fig. 5e). FACS purified CD4 + CD25 + YFP + Foxp3 Cre Tregs were cocultured with an equal number of purified naïve CD4 + CD25 -T-cells and then stimulated with allogeneic splenocytes and aCD3.

TCF1-deficient Tregs fail to suppress TH1 or TH17 polarization of CD4 + Tconv cells.
Inflammation requires CD4 + T-cell help and we therefore compared Tcf1-deficient and sufficient Tregs for their ability to suppress polarization of naïve CD4 + T-cells to TH1 or TH17 type cells. For the in vitro assays, we stimulated splenocytes from Foxp3 Cre Tcf7 fl/fl mice and Foxp3 Cre mice with aCD3 under TH1 116,117,118 (Fig. 6a) or TH17 119, 120 (Fig. 6d) polarization conditions. We evaluated polarization by intracellular staining for Ifnγ or IL17.
Next we assessed our findings in an in vivo setting using well established conditions that elicit TH1 or TH17 immunity in mice. Mice infected with TMEV mount a robust TH1 response that promotes viral-specific CD8 + Tcell cytolytic activity. We therefore infected mice with TMEV and after seven days isolated and re-stimulated mononuclear cells from spleen or MLNs with PMA/ionomycin/Golgistop (Fig. 6g). Expression of Ifnγ by CD4 + and CD8 + T-cells was significantly higher in Foxp3 Cre Tcf7 fl/fl mice than in control Foxp3 Cre mice (Fig. 6h,i)  activates TH17 polarization in the small bowel 121 . We injected mice i.p. with aCD3 and three days later quantified the expression of IL17 by CD4 T-cells in the small bowel by FACS (Fig. 6j). Injection of antibody produced significantly more IL17-expressing CD4 Teff cells in Foxp3 Cre Tcf7 fl/fl mice than in control Foxp3 Cre mice (p< 0.01), indicating poor control of TH17 polarization in mice with Tcf1-deficient Tregs (Fig. 6k). Collectively, our data show that TCF1-deficient Tregs are compromised in their ability to suppress TH1 and TH17 polarization in vitro and in vivo.

TCF1-deficient Tregs promote inflammation and tumor growth in polyposis-prone Apc D468 mice.
To assess the tumor-promoting properties of TCF1-deficient Tregs, we crossed the Foxp3 Cre Tcf7 fl/fl and control Foxp3 Cre mice to the polyposis-prone Apc D468 mice 122 , and generated two new compound mutant mouse strains: the TCF1 defective Apc D468 Foxp3 Cre Tcf fl/fl mice and control Apc D468 Foxp3 Cre mice. At 5.5 months of age, the TCF1-defective mice had significantly more colon polyps than control mice (12% versus 4% p<0.0001) (Fig. 7a), but similar numbers of polyps in the small bowel (Fig. 7b). Nuclear β-catenin staining revealed that Apc D468 Foxp3 Cre Tcf fl/fl mice had high incidence of pre-invasive colon tumors (Fig. 7c) and small bowel tumors (Fig. 7d) while these were rare in control Apc D468 Foxp3 Cre mice. The pre-invasive lesions were defined by the accumulation of abberant epithelial cells at the submucoal boundary (Fig. 7e), in contrast to their accumulation at the luminal boundary of benign polyps in Tcf1-sufficient mice (Fig. 7f).
The tumor-distant healthy tissues also had higher densities of Gr1 + cells in the Apc D468 Foxp3 Cre Tcf fl/fl mice than the Apc D468 Foxp3 Cre mice (colon: 1.2 versus 0.4 per FOV; p<0.009 and small bowel: 2 versus 1 per FOV; p<0.01) ( Fig. 7g-j). Together, these findings establish that loss of Tcf1 in Tregs leads to gain of pro-inflammatory and tumor-promoting properties in the colon of mice with genetic predisposition to polyposis.

Discussion
Our findings highlight the role of TCF1 in regulating multiple independent mechanisms by which Tregs fine tune immunity. TCF1 deficient Treg gained a "split personality" similar to that observed with Tregs in CRC 123  Since expression of Rorc by Tregs is bacterial dependent 7 , the cMaf cluster is likely to have originated from naïve conventional CD4 + T-cells (Tconv) through interaction with the gut microbiota. Consistently, our velocity analysis indicated that this cluster largely derives from the Mif cluster, a less activated Treg cluster which does not relate to the other cTreg clusters. Ikzf2 encodes HELIOS, which serves to enhance Treg fitness 62 , and when expressed by naïve Tregs can be an indication for their thymic origin 63 . It is reported that the Ikzf2 cluster of Tregs can convert to RORγT + Tregs albeit less efiiciently than conventional CD4 + T-cells 25  Therefore, tumors exploit the b-catenin/TCF1/RORgt axis to its' own advantage, and this pathway is a potential target for immune therapy of CRC.
Our detailed analyses of genes and processes affected by the loss of TCF1 in Tregs elucidate the molecular basis of regulation of Treg functional diversity, and have translational implications. The identified genes and pathways can be targeted to modulate Treg functions and limit disease. We posit that signaling pathways which overlap between down regulation of TCF1 and upregulation of b-catenin may be most relevant to the gain of proinflammatory and tumor properties by Tregs. Therefore, we think that our findings could help to both understand fundamental mechanisms of regulation of Treg functions and provide the basis for future translational studies aimed at cancer detection and therapy.

Mice
Mouse strains described below were housed and bred at the Mayo Clinic animal facility. and Foxp3 Cre Il10.Thy1.1 mice. Animal experiments were approved by the Animal Ethics Committee of the institutes responsible for housing the mice. Unless otherwise specified, all experimental procedures were performed on 5.5-6-month-old laboratory mice.

In vivo cytotoxicity assay
In vivo CTL assays followed established protocols 52,133  infection. Then the cytotoxicity was measured as described above.

Dissociation of mesenteric lymph nodes (MLNs) and spleen
A single cell suspension was obtained from MLNs and splenocytes after physical dissociation with a 40 μm mesh (Falcon). Red blood cell lysis on splenocytes was performed using 1 ml of ACK lysis buffer (Lonza) for 1 min on ice and washed in PBS-2% FBS (F8067; Sigma) buffer.

Enzymatic dissociation of small bowel and colon
Tissue was dissociated using the following steps. Fat layers were removed, washed, and opened longitudinally. All flow cytometry data were acquired on LSRII or LSR Fortessa (BD Biosciences) and analyzed with Flowjo software (Tree Star).

In vivo TH17 polarization and intracellular IL17 staining
Mice were injected intraperitoneally three times with CD3-specific antibody (20 µg per mouse; 2C11; BioLegend) or PBS at 0, 48 and 96 h 121 . 100 h after the first injection, the small bowel was enzymatically dispersed, intraepithelial cells (ILC) and lamina propria (LP) cells were isolated, and re-stimulated with PMA/Ionomycin and 5 h later were stained for intracellular IL-17A (TC11-18H10; BD Biosciences).

In vitro T cell polarization assay
Total

Histology and immune staining
Gut tissues were harvested, opened longitudinally and fixed using 10% formalin for 12-18 h, and routinely paraffin embedded and processed. For immune staining, 5-micron-thick tissue sections were deparaffinized in xylene and rehydrated in ethanol. Following rehydration, slides were immersed in target retrieval solution (S1699; Dako), and heat-induced epitope retrieval was performed in a Decloaking Chamber (Biocare Medical).

Single cell RNA-Seq data analysis
The 10X Genomics Cellranger (v2.0.2) mkfastq was applied to demultiplex the Illumina BCL output into FASTQ files. Cellranger count was then applied to each FASTQ file to align reads to mm10 reference genome and generate barcode and UMI counts. We followed the Seurat (v3.2.2) integrated analysis and comparative analysis workflows to do all scRNA-Seq analyses 61  To calculate the RNA velocity, the loom files were generated from the bam files by Velocyto 104 ; The RNA velocity was then calculated using the RunVelocity function in Velocyto.R package. The velocity for each sample was shown by show.velocity.on.embedding.cor function in Velocyto.R package.

Quantification and statistical analysis
Except for deep-sequencing data, statistical significance was calculated with GraphPad Prism software. Error bars in graphs indicate standard error of the mean (SEM) and statistical comparisons were done by unpaired Student's t-test. p values of ≤ 0.05 were considered statistically significant.  * p < 0.05; ** p < 0.01; *** p < 0.001.