An engineered cellular Timer for analysing the time domain of cellular differentiation in vivo

Understanding in vivo mechanisms of cellular differentiation is challenging because the temporal regulation of cellular signalling and differentiation in vivo is heterogeneous and highly dynamic. Here we establish a novel cellular technology, Timer of cell kinetics and activity (Tocky [toki], the Japanese word for time). Tocky uses Fluorescent Timer protein, which spontaneously shifts its emission spectrum from blue-to-red, in combination with computer algorithms to reveal the timing of cellular signalling and differentiation events in vivo. Using a transcriptional target of T cell receptor (TCR)-signalling, we establish Nr4a3-Tocky to analyse the temporal dynamics of in vivo TCR signalling activity. Nr4a3-Tocky determines the temporal order of events driving regulatory T cell (Treg) differentiation and identifies their immediate developmental precursors. Furthermore, by generating Foxp3-Tocky, we for the first time reveal in vivo dynamics of DNA demethylation, showing that Foxp3 transcription precedes demethylation of its enhancer region. Thus, Tocky empowers cellular biologists to ask previously inaccessible questions, revealing time-dependent differentiation mechanisms and developmental states.


Abstract
Understanding in vivo mechanisms of cellular differentiation is challenging because the temporal regulation of cellular signalling and differentiation in vivo is heterogeneous and highly dynamic. Here we establish a novel cellular technology, Timer of cell kinetics and activity (Tocky [toki], the Japanese word for time). Tocky uses Fluorescent Timer protein, which spontaneously shifts its emission spectrum from blue-to-red, in combination with computer algorithms to reveal the timing of cellular signalling and differentiation events in vivo. Using a transcriptional target of T cell receptor (TCR)-signalling, we establish Nr4a3-Tocky to analyse the temporal dynamics of in vivo TCR signalling activity. Nr4a3-Tocky determines the temporal order of events driving regulatory T cell (Treg) differentiation and identifies their immediate developmental precursors. Furthermore, by generating Foxp3-Tocky, we for the first time reveal in vivo dynamics of DNA demethylation, showing that Foxp3 transcription precedes demethylation of its enhancer region. Thus, Tocky empowers cellular biologists to ask previously inaccessible questions, revealing time-dependent differentiation mechanisms and developmental states.

Main Text
It is a central question in cell biology how cellular differentiation progressively occurs through the activities of temporally-coordinated molecular mechanisms (1,2). It is, however, challenging to investigate in vivo mechanisms at the T cells migrate through the body (6), and their activation and differentiation statuses are almost exclusively determined by flow cytometric analysis (7). In T cells, T cell receptor (TCR) signaling triggers their activation and differentiation (8), and is the central determinant of thymic T cell selection (1) (including negative selection (9) and Treg selection (10)) and antigen recognition in the periphery (8). While the temporal dynamics of proximal TCR signaling, which are in the time scale of seconds, have been comprehensively and quantitatively analysed (9,11), it is still unclear how transcriptional mechanisms for activation and differentiation respond to TCR signals over time in vivo.
TCR signalling activates NFAT, AP-1 and NF-B, which initiate the transcription of immediate early genes within a few hours (12), but their effects on T cell differentiation over the time scale of hours and days are obscure. In order to analyse TCR signal strength, currently, Nur77 (Nr4a1)-EGFP reporter mouse is commonly used (13), but the long half-life of the reporter gene, EGFP (~56 h (14)) prevents its application for the analysis of the temporal dynamics of the events downstream of TCR signaling in vivo.
In this study, we have established a new Fluorescent Timer technology, Timer of cell kinetics and activity (Tocky), which uniquely reveals the time and frequency domains of cellular differentiation and function in vivo. Fluorescent Timer proteins have been used to analyse in vivo protein dynamics and receptor turnover (15,16), as well as identify progenitor cells, i.e. those cells expressing only immature fluorescence during embryogenesis and pancreatic beta cell development (17)(18)(19)(20). However, those studies were qualitative and did not recognise the quantitative power of Fluorescent Timer. Here we develop a new Fluorescent Timer approach to quantitatively analyse the time and frequency domains of gene transcription within individual cells in vivo. By identifying a downstream gene of TCR signalling (Nr4a3) and developing a Fluorescent Timer reporter mice for the gene, we experimentally establish and validate the Tocky system for TCR signalling. Furthermore, we apply the Tocky approach to the lineage-specific transcription factor of regulatory T cells (Foxp3), revealing in vivo dynamics of Treg differentiation. Thus, Tocky technology reveals time-dependent mechanisms of in vivo cellular differentiation and developmental states following key signalling pathway or lineage commitment, which cannot be analysed by existing technologies.

Design of Tocky system for analysing the time and frequency domains of signal-triggered activation and differentiation events
Given the long half-life of stable fluorescent proteins (FP), like GFP (56 h (14)), the dynamics of gene transcription cannot be effectively captured using conventional FPs expression as a reporter. We therefore chose to use Fluorescent Timer protein (Timer), which forms a short-lived chromophore that emits blue fluorescence (Blue), the half-life of which is ~ 7 h, before producing the mature chromophore that emits red fluorescence (Red), which has a decay rate longer than 20 h (20) (Fig. 1A).
We investigated in silico how two different transcriptional activities would influence the production of blue and red fluorescent form of the protein through time, and how the light emitted would be detected by flow cytometry.
We compared the reporting of transient, pulse-like transcription with that of persistent transcription, both of which are relevant to TCR-mediated transcription (21) (Fig. 1B). By plotting fluorescence against time, we observed that Blue was a better readout for the real-time level of transcription than Red and GFP, which rather reported the cumulative activity of transcription. By plotting blue fluorescence against red fluorescence, there are thus 3 key loci in the Blue-Red plane (Fig. 1C). Firstly, when new transcription occurs in Timer-negative cells, Timer + cells acquire pure Blue, and are identified in the New locus. Secondly, if transcriptional activity is persistent and/or sufficiently frequent, cells accumulate in and around the steady-state diagonal line (Persistent locus). Lastly, when transcription is diminished, cells lose Blue and stay in the Arrested locus, until the red protein decays (Fig.   1C). Importantly, Timer decay is unidirectional and irreversible as the chromophore matures from Blue to Red in a clockwise direction. In the case of individual Timer + cells, movement from New to Persistent loci is also unidirectional and irreversible, because the half-life of red protein is longer than the half-life of the blue form to mature into red. Therefore transition from the New to the persistent loci captures the time domain of cellular differentiation. In contrast, cells in the Arrested locus could re-initiate transcription to express new Blue protein and move anti-clockwise back into the Persistent locus (Fig. 1C). Thus movement between Persistent and Arrested loci more specifically captures how frequently transcriptional activities occur in mature cells. These three loci can be identified and quantified more effectively by analysis of the angle of individual cells from the blue axis, and the 2-dimensional blue vs. red Timer fluorescence data can be transformed by trigonometric data transformation to provide new variables: the angle from the blue axis is defined as Timer-Angle, while the distance from the origin is defined as Timer-Intensity (Fig. 1D). We therefore coined the system Timer of cell Kinetics and activity (Tocky; とき [toki] means "Time" in Japanese). Timer-Angle provides time-related composite information: the first phase (New  Persistent) is for analysing how differentiation mechanisms are regulated over time (i.e. time domain analysis); and the latter half (Persistent -Arrested) is for analysing the relative frequency of transcriptional activation (i.e. frequency domain analysis). Thus, we designate this composite time-axis as Tocky-Time (Fig. 1E).

Nr4a3-Tocky reveals the time domain of TCR signal-triggered activation and differentiation events
T cell differentiation is triggered by TCR signals. In order to track the in vivo dynamics of transcription downstream of TCR signal transduction, we firstly identified genes immediately downstream of TCR signalling using a dataoriented multidimensional analysis, Canonical Correspondence Analysis (CCA) (22) (Supplementary Fig. 1A). Nr4a3 (Nor1) was identified as the gene with the highest correlation with anti-CD3 mediated T cell activation (which mimics TCR activation) and in vivo TCR signals in the thymus, while Nr4a1 (Nur77) and Rel also scored highly (Supplementary Fig. 1B and C).
In agreement, upon anti-CD3 stimulation, Nr4a3/Nor1 was rapidly induced in T cells and peaked within 2 hours of stimulation (Supplementary Fig. 1D).
Having established in silico that Tocky can temporally report transcription, we generated BAC transgenic reporter Nr4a3-Tocky mice (Supplementary Fig.   2 Fig. 2A), which captured the reduction in frequency of TCR signalling.
In order to analyse effectively the continuous progression of Timer fluorescence in Fig. 2a, Timer fluorescence data were transformed into Angle and Intensity data (Fig. 2B). Importantly, the progression of Timer Angle becomes slower and cells accumulate as cells approach ~45˚ by density plots, which visualise the distribution of angle values, reflecting sustained or high frequency TCR signalling (Fig. 2C). In order to quantify the maturation of the Timer chromophore since first onset of its transcription, we grouped cells into 5 populations (i.e. data categorisation) according to their Angle values: from the 'New' population (Angle = 0˚) representing cells within the first 7 hours of initiation of TCR-mediated transcription to the 'Arrested' population (Angle = 90˚), which represents cells in which all Timer protein has matured to red, and transcriptional activity has decreased below cytometer detection thresholds (Fig. 2D). The area between 30˚ -60˚ was defined as the Persistent locus. The areas between New and Persistent (NP-t) or between the Persistent and Arrested (PA-t) contained cells from the surrounding loci that are in the process of changing their frequency of Nr4a3 transcription (Fig.   2E). The analysis of the percentage of cells in these loci (designated as Timer locus analysis) neatly captured the change in TCR-mediated transcription through time, as cells shifted from New to Persistent, which represent the time domain. Furthermore, it showed that removal of TCR signals lead to a complete loss of New, NP-t and Persistent signalling within 8 hours, and cells migrated to Arrested transcriptional dynamics, which represents sparse or no signalling activity (Fig. 2E). Thus, Nr4a3-Tocky recaptured the predicted kinetics of Timer-expressing cells, validating the Tocky model (Fig. 1E).
We hypothesised that Timer-Intensity reflects both the signal strength and the duration/ frequency of TCR signalling, as Timer proteins accumulate in individual cells in response to strong and/or repeated TCR signals. Timer Intensity in antigen-stimulated Nr4a3-Tocky T cells increased over time, and fell after removal of the TCR signal (Fig. 3A). Furthermore, Timer Intensity was increased in a dose-dependent manner by cognate antigen (Fig. 3B).
CD28 signals did not have statistically significant effects on the progression of Timer Angle and Intensity upon TCR engagement using plate-bound anti-CD3 and anti-CD28 antibodies (Supplementary Fig. 3). In addition, Timer Intensity showed a high correlation to cell surface CD25 expression, which is a marker of activated T cells (R 2 = 0.93) (Fig. 3C). Thus, using Nr4a3-Tocky, Timer Intensity reflects the cumulative transcriptional outcome of signals in a given cell, as the reporter protein accumulates in response to sustained or highly frequent signals.

Tocky mice identify Treg developmental precursors
TCR signaling is the major determinant of regulatory T-cell (Treg) differentiation in the thymus, as T cells that have received strong TCR signals preferentially express CD25 and Foxp3 and differentiate into Treg (23,24).
In order to demonstrate the quantitative power of the Tocky technology, we investigated the temporal sequence of thymic CD4 T cell differentiation following TCR signals by analysing the time domain of ex vivo T cells from Nr4a3-Tocky thymus. Timer expression occurred in the CD4 + CD8 + double positive (DP), CD4 + single positive (CD4SP), and CD8SP populations, with CD4SP displaying the highest frequency (8.9  3.7%, Fig. 4A). The cell surface expression of CD69, which is highly expressed on immature CD4SP, was high in CD4SP cells in the New locus, and progressively declined as the Timer protein matures. In contrast, the majority of CD4SP cells in the Persistent locus expressed Foxp3 and CD25 (Fig. 4B). With the assumption that in neonatal mice, nearly all cells in the Persistent locus would be derived from the New locus, these results led to the hypothesis that Treg differentiation requires persistent TCR stimulation.
The sequence of thymic Treg development is controversial. Some studies suggest that CD25 + Foxp3 -Treg are the major Treg precursors (25,26) while other studies argue that CD25 -Foxp3 + cells are Treg precursors (27) (Fig.   4C). Analysis of Timer expression in the different populations enabled us to distinguish between these 2 possibilities. The majority of Treg (CD25 + Foxp3 + ) were Timer + (89.0  6.1%), and both of the proposed Treg precursor populations (CD25 + Foxp3and CD25 -Foxp3 + ) also had high proportions of Timer + cells (63.3  18.6% and 57.5  15.1%, respectively, Fig. 4D), indicating that these three populations have all received either strong or frequent TCR signals. To place the three populations in time following TCR signal transduction, we analysed thymi from various ages of Nr4a3-Tocky neonates, and measured Foxp3 and cell-surface CD25 expression, in addition to Nr4a3 Timer fluorescence, and applied our Timer locus analysis. Timer expressing CD25 + Foxp3cells were mostly Blue+Red-, while those of CD25 + Foxp3 + Treg and CD25 -Foxp3 + cells were mostly Blue+Red+ (Fig. 4E).
Density plot showed that CD25 + Foxp3cells only had a remarkable peak of Angle 0, while CD25 + Foxp3 + Treg showed a clear peak around 30-50˚ and CD25 + Foxp3cells had a higher peak at 90˚ (Fig. 4F). The mean values of Timer-Angle in the CD25 + Foxp3cells were significantly lower than the values in CD25 -Foxp3 + cells and CD25 + Foxp3 + Treg (Fig. 4G), indicating that the CD25 + Foxp3population are the cells that have most recently signalled through the TCR, and therefore the earliest of these three subsets. The mean values of Timer-Intensity were lowest in CD25 -Foxp3 + , and highest in CD25 + Foxp3 + Treg (Fig. 4H), indicating that the CD25 -Foxp3 + subset had received the weakest and/or least frequent TCR signals among these subsets. Next, in order to define the temporal relationships between the three populations (CD25 + Foxp3 -, CD25 -Foxp3 + and CD25 + Foxp3 + ), we quantified the proportion of Timer + cells in each maturation stage of Timer fluorescence (Fig. 4I). This showed that most of the CD4SP cells in the New locus were CD25 + Foxp3cells, while most of CD4SP cells in the Persistent and PA-t loci were CD25 + Foxp3 + Treg. The relative contribution of CD25 -Foxp3 + was greatest in the Arrested locus, suggesting that these cells are enriched with those with aborted TCR signalling (Fig. 4I).
Collectively, these analyses demonstrate that the major pathway for Treg differentiation is from CD25 + Foxp3 -Treg precursors, in which persistent TCR signals induce Foxp3 expression. The CD25 -Foxp3 + subset is enriched with Foxp3 + cells that have received relatively weaker and/or less sustained TCR signals. Following TCR signalling, cell surface CD69 expression peaks within 4 to 8 hours. CD25 expression is induced in this early phase in the Timer Blue (New) population and steadily accumulates as T cells receive TCR signals.
Foxp3 expression is the most delayed and occurs most efficiently after T cells have persistently interacted with antigen (Fig. 5). Importantly, the Tocky system has for the first time directly shown these in vivo dynamics of Treg differentiation. Thus, our investigations show that Nr4a3-Tocky and Timer locus analysis effectively unravel the temporal dynamics of T cell differentiation following TCR signals.

Foxp3-Tocky successfully identifies newly generated Treg
Next, in order to address whether the Tocky system can be applied to another gene, and to validate the system, we developed Foxp3-Tocky mice using the same approach as Nr4a3-Tocky (Fig. 6A). By investigating Foxp3 protein staining (Fig. 6B) and Foxp3 IRES-GFP Foxp3-Tocky double transgenic mice (Fig. 6C), Timer expression showed high correlation with GFP and with Foxp3 protein. As expected, when naïve CD4 + T cells were stimulated in the presence of IL-2 and TGF- (i.e. induced-Treg [iTreg] conditions), new Foxp3 expression (Blue + Red -) was induced at 22 h, which gained Red proteins by 50 h as the Timer chromophore matured (Fig. 6D). Trigonometric Timer data analysis showed ex-vivo splenic Foxp3 + T cells from adult mice had high Timer-Angle values throughout the culture. In contrast, iTreg showed very low Angle values at 22 h, which slowly increased overtime (Fig. 6E-6F).

Foxp3-Tocky reveals that the initiation of Foxp3 transcription precedes the demethylation of the Foxp3 gene in vivo
Next, taking advantage of the Tocky system, we investigated the transcription and methylation of the Foxp3 gene during Treg generation. Whilst it is known that demethylation can occur within a few hours in cultured cells (28), there is no available method to investigate the in vivo dynamics of demethylation in the mouse body.
It is well established that DNA demethylation of the Treg-Specific Demethylated Region (TSDR) is a hallmark of committed Treg (29). It is, however, unknown whether demethylation occurs before or after the initiation of Foxp3 transcription. To address this question, differentiating thymic T cells were isolated by flow-sorting thymic Blue high T-cells into Timer-Angle Low, Medium, and High (Fig. 7A). Analysis showed that the isolated populations exhibited distinct Timer Angles from a mean of 10 to 80 degrees (Fig. 7B &   C). DNA was isolated from the population and their TSDR demethylation was analysed (Fig. 7D). Importantly, the newest Foxp3 + cells (Low) were still mostly methylated and not significantly different from Timer(-) cells in the degree of TSDR demethylation, while rapid demethylation occurred as cells moved to the Persistent locus (Fig. 7D). Timer-Angles showed a strong correlation with TSDR demethylation rates (Spearman's correlation coefficient Fig. 7E)

Discussion
Here we have established two uses of the Tocky system. First, Tocky allows investigators to determine the relative temporal order of the molecular events following the activation of key signalling pathways, and thereby provides a new way to identify immediate precursors and relatively mature cells using the time domain analysis. Previous studies using Fluorescent Timer proteins identified newly generated cells, as they expressed solely immature fluorescence without mature fluorescence (17)(18)(19). In this study, in addition to the rigorous determination of New cells (Blue + Red -), we established a quantitative method to measure the relative age of cells which are maturing in vivo. This has allowed us to simultaneously investigate the temporal dynamics of the demethylation process and the transcriptional activities in vivo.
Surprisingly, Foxp3 transcription preceded the DNA demethylation of its enhancer region, which is considered to regulate the transcriptional activity (30). This means that the Tocky system provides an unprecedented mean to investigate the in vivo regulation of molecular mechanisms. Thus, we determined the temporal sequences of molecular events during thymic Treg differentiation after receiving TCR signals using Nr4a3-Tocky, and identified the major precursor of these cells as the earliest cells among the presumptive precursor populations. In addition, using Foxp3-Tocky, we identified and characterised the earliest stages of Foxp3 + Treg differentiation.
Second, the Tocky system allowed identification of cells with persistent or high-frequency transcription of the reporter gene through frequency domain analysis. The persistent dynamics can be maintained if the interval of recent transcriptional activities is sufficiently smaller than the half-life of Blue.
Importantly, when Timer proteins are produced in a persistent manner, the progression of Timer Angle becomes slower as cells approach the Persistent locus (i.e red-to-blue ratio gradually increases; Fig. 1D), as the speed of cells decreases in the Blue-Red plane, resulting in a ritardando towards a pause.
Consequently, the accumulation of cells within a defined red-blue space is indicative of cells displaying a similar average frequency of transcription/ similar signalling patterns over time, as reflected by tissue-infiltrating antigenspecific T-cells. Therefore, we use the term 'frequency domain' to define not individual cells, but groups of cells with similar transcriptional patterns that can be related to biological functions. It is highly likely that at the individual cell basis, transcriptional frequencies are highly stochastic. We anticipate that the development of future fluorescent timers with a shorter maturation time will enable a more precise estimation of true transcriptional frequencies, which will be important for analysing transcriptional circuits occurring in the time frames of minutes.
In fact, using Nr4a3-Tocky, we showed that thymic T cells that have received sustained TCR signals differentiate into Treg. Consistent with these findings, thymic Foxp3 + cells show slower and more confined migration, compared to other thymic T cells by two-photon microscopy (31). These differentiating  Fig. 4), which is in a non-linear relationship with 'real' time, and represents a relative chronological readout for events occurring following a cellular differentiation cue. The advantages of the Tocky system over other gene reporters are summarised in Supplementary Table 1. In the future Tocky mice for key transcription factors and genes will be promising tools to reveal the in vivo dynamics of gene transcription and cellular differentiation (e.g. Bcl6, for T-follicular helper cells; Rag2, for TCR recombination; and, Oct4, for stem cell-ness); these cannot be investigated otherwise.
In summary, we have established the Tocky system as a pioneering tool to investigate cellular differentiation and gene dynamics in vivo in immunology, developmental biology and stem cell biology.

Study Design
Two independent lines were established for both of the Nr4a3-Tocky and Foxp3-Tocky transgenic reporter strains. Both lines exhibited highly similar phenotypes and frequencies of Timer + cells.

Transgenesis and mice
The BAC clones RP23-122N18 and RP23-330D17 were obtained from the BACPAC Resources Center at Children's Hospital Oakland Research Institute (CHORI), and were used for generating Nr4a3-Tocky and Foxp3-Tocky, respectively. BAC DNA was modified by the BAC recombineering approach using the SW106 strain bacteria (35).
We used a Timer knock-in knock-out approach for BAC transgenic reporter constructs. Precisely, for Nr4a3-Tocky, the first coding exon of the Nr4a3 gene in RP23-122N18 was targeted and replaced with the transgene cassette containing the Timer gene (20), a poly-A tail and a floxed neomycin resistance gene (neo, Supplementary Fig. 2). For Foxp3-Tocky, the first coding exon of the Foxp3 gene in RP23-330D17 was targeted and replaced by the same transgene cassette. Subsequently, neo was excluded by arabinose-inducible BioScience. Sequence traces and CpG methylation rates were analysed and determined by ESME software (Epigenomics) (36).

Canonical Correspondence Analysis
The gene signature and sample scores in Supplementary Fig. 1 were calculated by a cross-dataset analysis using CCA (22). Briefly, the expression data of GSE15907 (37) was regressed onto the log2 fold change of activated CD4 + T cells (2 h after activation) and naïve T cells from GSE48210 (38) as the explanatory variable, and Correspondence Analysis was performed for the regressed data and the correlation analysis was done between the new axis and the explanatory variable. CCA was performed by the CRAN package vegan as previously described (22). The analysis was undertaken using only transcription factor genes, which were selected by the Gene Ontology database by including the genes that are tagged with GO:0003677 (DNA binding) and GO:0005634 (nucleus) and not with GO:0016020 (membrane) using the Bioconductor package GO.db.

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. Data were preprocessed and scaled, the negative cloud was collapsed (i.e. set a threshold for each fluorescence), and Timer + cells were automatically identified. Trigonometric data transformation was performed to produce Timer-Angle and -Intensity data for individual cells in each sample. Basic procedures for flow cytometric data analysis have been previously described elsewhere (7).

Statistical analysis and data visualisation
Statistical analysis was performed on R or Prism 6 (Graphpad) software.
Percentage data for Timer + and Timer locus analysis was analysed by Mann- Scatterplots were produced by the CRAN packages ggplot2 and graphics.
Dose response data in Fig. 2g were fitted to a dose-response curve using the three-parameter log-logistic function of the CRAN package drc as previously described (39). All computations were performed on Mac (version 10.11.6).

Data and code availability
All R codes are available upon request. Data will be made available upon reasonable requests to the corresponding author.