A temporally dynamic Foxp3 autoregulatory transcriptional circuit controls the effector Treg programme

Abstract Regulatory T cells (Treg) are negative regulators of the immune response; however, it is poorly understood whether and how Foxp3 transcription is induced and regulated in the periphery during T‐cell responses. Using Foxp3‐Timer of cell kinetics and activity (Tocky) mice, which report real‐time Foxp3 expression, we show that the flux of new Foxp3 expressors and the rate of Foxp3 transcription are increased during inflammation. These persistent dynamics of Foxp3 transcription determine the effector Treg programme and are dependent on a Foxp3 autoregulatory transcriptional circuit. Persistent Foxp3 transcriptional activity controls the expression of coinhibitory molecules, including CTLA‐4 and effector Treg signature genes. Using RNA‐seq, we identify two groups of surface proteins based on their relationship to the temporal dynamics of Foxp3 transcription, and we show proof of principle for the manipulation of Foxp3 dynamics by immunotherapy: new Foxp3 flux is promoted by anti‐TNFRII antibody, and high‐frequency Foxp3 expressors are targeted by anti‐OX40 antibody. Collectively, our study dissects time‐dependent mechanisms behind Foxp3‐driven T‐cell regulation and establishes the Foxp3‐Tocky system as a tool to investigate the mechanisms behind T‐cell immunotherapies.

Foxp3 expression is a major determinant of Treg phenotype and function, and Foxp3 interacts with transcription factor complexes, such as those involving NFAT and Runx1, to repress IL-2 transcription and convert the effector mechanisms in T cells into a suppressive one (5)(6)(7). Treg have activated phenotypes, and upon TCR signals, Treg suppress the activities of conventional T cells (8). TCR signaling is the major regulator of Treg differentiation in the thymus, as T cells that have received strong TCR signals preferentially express CD25 and Foxp3 and differentiate into Treg (9).
Accumulating evidence indicates that Foxp3 expression is dynamically controlled in Treg and non-Treg. TCR stimulation induces Foxp3 expression in human (13) and mouse T cells (14) in vitro. Studies using T-cell receptor (TCR) transgenic systems have shown that Foxp3 expression is induced in non-Treg in some inflammatory conditions in vivo (15). Although such induced Foxp3 expression is often dismissed as "transient expression", the dynamic induction of Foxp3 expression may have functional roles during T cell responses if this reactive Foxp3 expression occurs in activated polyclonal T cells during inflammation in vivo (16). In addition, Foxp3 expression can be dynamically downregulated in Treg. Fate-mapping experiments showed that, while most of thymus-derived Foxp3+ T cells stably express Foxp3, some Foxp3+ cells downregulate Foxp3 to become ex-Foxp3 cells in the periphery, joining the memory-phenotype T cell pool (14). PD-1 KO mice with a partial Foxp3 insufficiency leads to generation of ex-Foxp3 effector T cells (17), indicating that the mechanism of T cell activation is involved in the dynamic regulation of Foxp3 transcription. These findings lead to the hypothesis that Foxp3 acts as a cell-intrinsic and transcellular negative feedback regulator for T cell activation among self-reactive T cell repertoires (16), challenging the thymus-central view of Treg-mediated immune regulation.
The key question is whether and how frequently activation of new Foxp3 transcription is induced in non-Treg cells in physiological conditions, and how expression is, the more suppressive Treg are (18,19). Thus, it is fundamental to investigate the temporal dynamics of Foxp3 transcription over time in individual T cells in vivo, but this has been technically difficult to do to date.
Here we use our novel Timer of Cell Kinetics and Activity (Tocky) system (20) to reveal the time and frequency of Foxp3 transcription during peripheral immune responses. In the Foxp3-Tocky system, the transcriptional activity of the Foxp3 gene is reported by Fluorescent Timer protein, which emission spectrum spontaneously changes from Blue to Red fluorescence after translation (21), and the in vivo dynamics of Foxp3 transcription is analysed.
We show that during inflammation, the flux of Foxp3to Foxp3+ T cells dramatically increases within the periphery. In addition, we demonstrate that the real-time frequency of Foxp3 transcription determines the effector Treg differentiation. Thus, we provide experimental evidence that Foxp3 expression is dynamically regulated in Treg and non-Treg during inflammation in vivo, providing fresh insight into Foxp3-driven T cell regulation.

Timer-Blue fluorescence reports real-time Foxp3 transcription
Fluorescent Timer protein (Timer) is an mCherry mutant, and when translated, the chromophore of Timer is an unstable blue-form, which spontaneously and irreversibly matures to become a stable red-form (21). We hypothesised that  1A). This indicates that Timer transcripts are well correlated to Foxp3 ones in Foxp3-Tocky T cells, in which Timer transcripts report the transcriptional activity of the Foxp3 gene (20). Next, we analysed the half-life of Timer-Blue fluorescence in activated T cells in vitro using a short-term treatment with cycloheximide (CHX) to inhibit new protein synthesis. Whilst a previous study estimated the maturation half-life of Timer-Blue to be 7.1 h, using purified Timer proteins and by fitting data to a pharmacological kinetic model (21), here we aimed to experimentally determine the half-life of Timer-Blue fluorescence by flow cytometric analysis of Foxp3-Tocky T cells. Upon CHX treatment, flow cytometric plots showed that Foxp3-Timer Blue fluorescence rapidly decayed, while Timer-Red fluorescence was more stable (Fig. 1B).
The half-life of Timer-Blue fluorescence was estimated to be 4.1 h (Fig. 1C) whilst Timer-Red fluorescence gradually increased (Fig. 1D)

transcription compared to splenic CD4+ T cells in neonatal mice
In neonatal mice, Foxp3+ T cells are actively produced in the thymus (22). In Foxp3-Tocky mice, thymic CD4-single positive cells showed a higher percentage of Blue+ cells in Timer+ cells than splenic CD4+ cells (i.e. 95% and 63% of all Timer+ cells were Blue+, Fig. 2A). Timer-Blue expression levels in Timer+ cells were much higher in the thymus than in the spleen, while Timer-Red expression levels were higher in the spleen (Fig. 2B). These data indicate that the rate of Foxp3 transcription is higher in the thymus than the spleen, while splenic Foxp3+ cells have transcribed the Foxp3 gene for a longer time on average than thymic Foxp3+ cells. These results thus further confirm that Foxp3-Tocky captures real-time Foxp3 transcription by Timer-Blue fluorescence and its history and cumulative activity by Timer-Red in vivo.

The flux of new Foxp3 expressors and the rate of Foxp3 transcription are increased in tissue-infiltrating T cells during the immune response
Next, we tested if Foxp3 transcriptional activity is dynamically regulated during the immune response. We used a skin contact hypersensitivity (CHS) model to analyse the dynamics of Foxp3 transcription in draining lymph nodes (dLN) and skin-infiltrating T cells after sensitising mice with the hapten oxazolone and challenging them on their ears (Fig. 3A). T cells were not present in sufficient numbers to be recovered from Vehicle-treated control skins. Most Timer+ cells in dLN were Red+, and these expressed various levels of Blue fluorescence (Fig. 3B), which indicates that Foxp3 transcription levels vary between individual cells. While immunisation with oxazolone resulted in minimal effects in dLN, skin-infiltrating T cells showed uniformly high Timer-Blue fluorescence, indicating that these cells had high-frequency persistent Foxp3 transcription ( Fig. 3B and 3C). On the other hand, Timer-Red fluorescence was lower in skin-infiltrating T cells than dLN cells, indicating that these skin T cells had less accumulated Foxp3 transcripts, and thus that these cells are short-lived or rapidly dividing.
Blue-high Red-high expression is the hallmark of high-frequency persistent transcription, which can be the most effectively identified by trigonometric data transformation to calculate the angle from the Blue axis (Timer-Angle) and the distance from the origin of individual cells (Timer-Intensity) in the Blue-Red plane (20) (Fig. 3D). Timer expression starts from Blue+Red-(the New locus), these New cells mature and acquire Red Fluorescence within 4 hours ( Cells in the Arrested locus can however immediately acquire Blue fluorescence again when they re-initiate Foxp3 transcription (Fig. 3D), indicating that the Timer-Angle between Persistent -Arrested loci represents the recent frequency of Foxp3 transcriptional activity (20).
The distribution (precisely, kernel-density estimation, i.e. a smoothed version of histogram) of Timer-Angle values was markedly right-skewed and high values in dLN, whether Vehicle or Oxazolone treated, while the Timer-Angle distribution in the skin showed peaks around 0˚ and 60˚ (Fig. 3E), with a mean Timer-Angle value of 45˚ (Fig. 3F). We classified cells according to Timer-Angle: New (Angle = 0), Persistent (Angle 30˚ -60˚), and Arrested (Angle = 90˚). The areas between New and Persistent and between Persistent and Arrested are designated as NP-t and PA-t, respectively (20).
This Timer Locus analysis clearly showed that skin-infiltrating T cells were enriched with New, NP-t and Persistent cells, compared to dLN T cells (Fig.   3G). Collectively these results indicate that flux from Foxp3-to Foxp3+ T cells is increased at sites of inflammation, and that skin-infiltrating Foxp3+ T cells have more active Foxp3 transcription and are short-lived. These cells are enriched with newly born cells, and may have rapid cell divisions and/or accelerated cell death.

Foxp3 transcription
The dynamic regulation of Foxp3 transcription during the immune response suggests that Foxp3 protein itself is involved in the regulation of Foxp3 transcription. In order to understand the mechanism regulation of temporallydynamic Foxp3 transcription, we addressed this possibility by investigating the dynamics of Foxp3 transcription in the absence of Foxp3 protein. To achieve this, we crossed Foxp3-Tocky mice with Foxp3 Egfp/Scurfy mice. Treg from females from this line will all express the Foxp3-driven Timer protein, but half will express Foxp3/EGFP and the other half (EGFP-negative) express the non-functional Scurfy Foxp3 mutant protein (Fig. 4A) due to random Xinactivation (23,24). This triple transgenic system thus permits the analysis of Scurfy cells (Fig. 4B). Timer-Blue levels were remarkably decreased, and Timer-Red levels were also moderately decreased, in Scurfy cells (Fig. 4C).
Timer-Angle was significantly higher and approached 90˚ in Scurfy cells (Fig.   4D). Scurfy Timer+ cells had very few cells displaying Persistent or PA-t transcriptional dynamics compared to GFP+ Treg, with nearly all cells located in the Arrested locus (Fig. 4E). Furthermore, in the absence of Foxp3 protein, Timer-expressing Persistent/PA-t cells actively transcribed Il2 (Fig. 4F), a characteristic of antigen-stimulated effector T cells. These findings reveal that which made another cluster (Fig. 5E). Differential expression analysis identified unique clusters of genes (Fig. 5F). Persistent Foxp3 expressors highly expressed the co-inhibitory/costimulatory molecules Ctla4, Icos, and Tigit, and the serine protease Gzmb, both of which are all involved in cancer immunity (25). In addition, persistent Foxp3-expressors transcribed other effector-Treg signature genes such as Nkg7, Fgl2, and Irf4, while they had low expression of naïve T cell-specific genes such as Ccr7, Bach2, and Il7r.
These features are fully compatible with the previously reported effector-Treg phenotype (12,26). The specific increase in the activation gene Mki67 (Ki67) in persistent Foxp3 expressors also supports the idea that these cells are highly activated. Intriguingly, Pearson correlation analysis showed that Ctla4 and effector-Treg markers (Icos, Lag3, Tigit, Ccr4) had relatively low correlation to Foxp3 levels (R 2 was between 0.30 -0.65) ( Supplementary   Fig. 1A), while these markers showed very high correlation to the activation gene, Mki67 (R 2 was between 0.90 -0.98) (Supplementary Fig. 1B). This suggests that the T cell activation process predominantly regulates these markers and that persistent Foxp3 expressors express these markers because they are highly activated. In contrast, Tnfrsf18 (GITR) and Tnfrsf4 (OX40) showed high correlations to Foxp3 (R 2 = 0.91 and 0.82, respectively), but only moderate correlations to Mki67 (R 2 = 0.49 and 0.74, respectively), suggesting that Foxp3 may more closely regulate GITR and OX40. Il2ra is ubiquitously highly expressed by all Foxp3+ cells, and not by Foxp3-cells, and showed only low correlations to both Foxp3 and Mki67 levels (R 2 = 0.26 and 0.43), suggesting that CD25 expression is stabilized in Foxp3+ cells but regulated primarily by other factors, such as IL-2 signalling once Foxp3 is expressed (Supplementary Fig. 1A and 1B). Persistent Foxp3 expressorspecific genes were enriched with cell cycle-related pathways. In contrast, the PA-t and Arrested-specific genes were enriched with cytokine-signalling pathways but not cell-cycle-related pathways (Fig. 5G). Collectively, RNA-seq analyses showed that temporally-persistent Foxp3 transcription is an underlying mechanism for effector Treg differentiation.

Identification of distinct groupings of cell surface markers for targeting T cells with specific Foxp3 transcriptional dynamics
Next, we analysed protein expression of candidate surface proteins identified by RNA-seq for targeting T cells with specific Foxp3 transcriptional dynamics in vivo. T cells from Foxp3-Tocky mice were immunized with Oxazolone, and analysed by flow cytometry for the expression levels of candidate markers from Fig. 5 in each Timer Locus. Hierarchical clustering of MFI levels showed that these candidate genes were grouped into two main groups, in addition to Neuropilin 1 (Nrp1) which had its own separate cluster (Fig. 6A). Group I markers, showed high expression throughout New -Persistent Foxp3 expressors, but fell drastically as Blue fluorescence decreases in PA-t and Arrested Foxp3 expressors (Fig. 6B), suggesting that these markers are associated with the initiation of new Foxp3 transcription in activated T cells and with high-frequency Foxp3 transcription. Group II markers are less expressed in new Foxp3 expressors, and increased in persistent and PA-t Foxp3 expressors (Fig. 6C), suggesting that these markers can be sustained by moderate frequencies of Foxp3 transcription once the Foxp3 autoregulatory loop is established. Nrp1 is characteristically high in PA-t Foxp3 expressors, which may explain why this marker has been used to identify thymus-derived Treg (Fig. 6D). Importantly, the expression levels of both Group I and II markers markedly declined in Arrested Foxp3 expressors.
Together these results further support the notion that the temporal frequency (rate) of Foxp3 transcription regulates the Treg phenotype and function.

Foxp3-Tocky allows visualistion of the manipulation of T cells with specific Foxp3 transcriptional dynamics upon immunotherapy
We asked if surface proteins from Group I and Group II in Fig. 6 can be used to target T cells with specific Foxp3 transcriptional dynamics. If successful, this approach may lead to future precision immunotherapy for controlling Foxp3-mediated immune regulation by manipulating its temporal dynamics.
Both TNFRII (Group I) and OX40 (Group II) are targets for immunotherapy and may be used to manipulate T cell responses, and currently clinical trials are ongoing for melanoma and other solid cancers (27). However, it is unknown whether these immunotherapy antibodies affect Foxp3 transcriptional dynamics.
We used the CHS model, and antibodies were administered during the challenge phase (Fig. 7A). Treatment with anti-TNFRII antibody increased the frequency of Blue+Red-T cells, and Blue+Red+ T cells at the inflamed site ( Fig. 7B). Although average Blue and Red levels were overall not affected by the antibody (Fig. 7C and 7D), Timer-Angle tended to be lower within the anti-TNFRII treated group (p=0.055), likely reflecting the increased proportion of pure blue T cells at the inflamed site (Fig. 7E). Analysis of the Timer-Angle distribution revealed that Isotype control treatment showed one dominant peak around the 45˚ mark, whilst anti-TNFRII treatment produced a more flattened, peak from 25-50 degrees (Fig. 7F). Analysis of Timer locus revealed that anti-TNFRII increased both New and NPt stage Foxp3 expressors, indicating that anti-TNFRII immunotherapy enhances the flux of Foxp3-into Foxp3+ T cells (Fig. 7G). There was no significant change in the high-frequency/persistent Foxp3 transcribers, and anti-TNFRII treatment did not significantly alter the course of inflammation (Fig. 7H). In contrast, anti-OX40 treatment appeared to dramatically reduce the frequency of Blue+Red+ Foxp3+ T cells in inflamed skin (Fig. 7I). Whilst Blue levels were unaffected (Fig. 7J), both Timer-Red and Timer-Angle were significantly lower in anti-OX40 treated mice (Fig. 7K and 7L). Interestingly, the peak of Timer-Angle distribution in anti-OX40 treated mice was shifted towards the NPt (Fig. 7M).
Timer locus analysis revealed that anti-OX40 reduced the frequency of more mature Foxp3 expressors, as both Persistent and PA-t loci were significantly reduced in anti-OX40 treated mice (Fig. 7N) These data indicate that, in the presence of anti-OX40 antibody, Treg are very short lived at the inflamed site, and that high-frequency Foxp3 expressors are either depleted or undergo activation-induced death. Interestingly, this reduction in high-frequency Foxp3 expressors led to delayed resolution of inflammation (Fig. 7O), demonstrating the importance of sustained Foxp3 transcription for regulating the T-cell response. In summary, both of the effects of the immunotherapies tested were in keeping with the Foxp3 transcriptional dynamics identified in Figure 6, showing how Foxp3-Tocky may be used to identify new therapeutic targets and mechanism of drug action in pre-clinical models.

Discussion
Sustained Foxp3 transcription is a cardinal feature of the functional maturation of Treg, or effector-Treg (Fig. 5). Importantly, sustained Foxp3 transcription requires functional Foxp3 protein itself. This type of transcriptional regulation (i.e. transcription factor regulates its own gene) is defined as positive autoregulation, which allows a slow and sustained transcriptional response to external signalling cues (28). This may constitute the mechanism behind effector Treg differentiation in inflammatory environments. It is known that, among the conserved non-coding DNA sequence (CNS) elements of the Foxp3 gene, the CNS2, which contains the TSDR, is required for maintenance of Foxp3 expression after cell division (29). The CNS2 regions is bound by Foxp3 (29) cytokine-deprivation death) (34). If some of the reactive Foxp3 expressors survive beyond the resolution of inflammation, such cells will be memory T cells by definition. Some cells may sustain Foxp3 expression and survive as "memory Treg", and some others may lose Foxp3 expression after the resolution of inflammation and survive as memory-phenotype (memory-like) T cells (16). Intriguingly, ex-Treg are enriched in the memory T cell pool and prone to express Foxp3 more readily than naïve T cells (14,35). These hypotheses will be best addressed in future studies of T cell memory, and will require new approaches to combine the Tocky technology and the analysis of cellular dynamics several weeks after the onset of inflammation.
In the hapten model, ~10% of Timer+ cells are identified at the New locus in the skin (Fig. 3G). Given that these new cells quickly move to NP-t/Persistent locus within 4 h (Fig. 1B), a significant proportion of cells at the Persistent locus may have been supplied from new cells within 72 h before the analysis.
It is also possible that cells at the PA-t and Arrested loci move to the Persistent locus by increasing Foxp3 transcription. These two possibilities are not mutually exclusive and are compatible with evidence that the effector Treg population arises from both thymic and peripherally-induced Treg (12), while the ratio of the contributions may vary between antigens (16). Future studies using TCR repertoire sequencing will fully answer this question.
The current study has shown that Foxp3-Tocky is effective in designing immunotherapy strategies to target a specific phase of Foxp3 transcriptional dynamics, and also, for visualising alterations of the dynamics upon immunotherapy. In fact, we successfully manipulated in vivo Foxp3 dynamics by targeting Timer-stage specific surface receptors. By targeting Group I (TNFRII) or Group II (OX40) marker of Foxp3 dynamics (Fig. 6) we either increased the flux of new Foxp3 transcription (anti-TNFRII) or decreased persistent Foxp3 expressors (anti-OX40). While the precise cellular mechanism is to be determined, anti-OX40 treatment resulted in either the depletion or death of persistent Foxp3 expressors within the skin, which are enriched with effector Treg. OX40 is an emerging target for cancer immunotherapy, and its agonistic antibodies for augmenting the effector function of T cells are currently tested in clinical trials (36). It is currently thought that OX40 is expressed mainly by activated effector T cells, and is also constitutively expressed on all Treg (37). Such OX40 high expression is found relatively more frequently on persistent Foxp3 expressors compared to Foxp3-negative effector T cells.
In summary, we have dissected the previously hidden dynamics of Foxp3 transcription that regulate the functions of Treg through time. In addition, we have shed light on how Foxp3-mediated molecular mechanisms coordinate T cell responses in existing Treg and non-Treg, introducing a new dimension into studies on Foxp3-driven T cell regulation.

Study Design
Sample size for anti-OX40 CHS experiments was estimated based on a power calculation with alpha= 0.05 and power 80%, which gave a sample size of 8. For analysis of Foxp3-Tocky Foxp3 EGFP/Scurfy mice, 4 mice were analysed over two independent experiments (age 6-10 weeks). For anti-OX40 experiments, a pilot experiment was performed to obtain estimates for effect size and standard deviation, which yielded an experimental group size of 9.
Two independent experiments were then performed to test the null hypothesis that anti-OX40 treatment had no effect on ear swelling at 96 h. The null hypothesis was rejected in both experiments, with no significant interexperiment variation. Therefore, data from both experiments were pooled for presentation of results. For all other experiments a minimum of 3 animals were used. Where male and female mice were used in the same study, these were randomised into experimental groups according to sex and age.
Similarly, wild type littermates from transgenic mice were randomised into experimental groups for the anti-OX40 experiments. The individual making the ear measurement was blinded to the treatment group of the mouse until after the measurement was taken.

Timer data analysis
Sample data including a negative control were batch gated for T cell populations and exported by FlowJo (FlowJo LLC, OR) into csv files, including all compensated fluorescence data in the fcs file. The code developed in this study imports csv files into R, preprocesses and normalises data, and automatically identifies Timer + cells and performs Trigonometric data transformation, producing Timer-Angle and -Intensity data for individual cells in each sample, as previously described (20). Basic procedures for flow cytometric data analysis have been previously described elsewhere (19).

Statistical analysis and data visualisation
Statistical analysis was performed on R or Prism 6 (Graphpad) software.