Multiplexed live visualization of cell fate dynamics in hPSCs at single-cell resolution

Live imaging can provide powerful insights into developmental and cellular processes but availability of multiplexable reporters has been limiting. Here we describe ORACLE, a cell fate reporter class in which fluorescent proteins fused with the nucleoporin POM121 are driven by promoters of transcription factors of interest. ORACLE’s nuclear rim localisation therefore enables multiplexing with conventional nuclear reporters. We applied ORACLE to investigate the dynamics of pluripotency exit at single-cell level, using human pluripotent stem cells (hPSCs) imaged by multi-day time-lapse high-content microscopy. Using an ORACLE-OCT4 pluripotency marker we reveal that G1 phase length and OCT4 level are strongly coupled and that spatial location in a colony impacts the timing of pluripotency exit. Combining ORACLE-OCT4 and an ORACLE-SOX1 early neuronal differentiation marker, we visualize in real-time the dynamics of cell fate transition between pluripotency and early neural fate, and show that pluripotency exit and differentiation onset are likely not tightly coupled in single-cells. Thus ORACLE is a powerful tool to enable quantitative studies of spatiotemporal cell fate control.


INTRODUCTION
Human pluripotent stem cell (hPSC) technologies stand to enable the regenerative medicine of the future 1 . Although hPSCs can in principle be differentiated into almost all cell types in our bodies [2][3][4][5] , targeted differentiation often also leads to generation of undesired cell types and cellular heterogeneity incompatible with safe and effective cell therapies 6 .
Understanding the origin of such heterogeneity is critical to learn how to efficiently, specifically, and safely produce hPSC-derived cells and tissues for both basic and therapeutic applications.
Single-cell RNA sequencing (scRNAseq) studies of hPSCs differentiating into many different cell types -such as retina, cardiac or endodermal cells -have revealed how complex and poorly understood this heterogeneity really is 7-10 . Although lineage trajectories can be inferred computationally, scRNAseq approaches are terminal and therefore unable to access some fundamental aspects of the spatiotemporal dynamics of cell fate control. These include the dynamics of transcription factor (TF) heterogeneity 11 in cells, and the real-time sequences, durations, and spatial changes in TF levels through tissues, as well as key cell biological aspects of fate dynamics like cell movement, cell death and progression through the cell cycle and how they are coordinated and linked with fate. Such dynamical and cell biological layers are only accessible by 'live' microscopy 12,13 . Hence, refined live highcontent microscopy tools and approaches are needed to extract this critical information in order to clarify the mechanisms controlling cell fate establishment and how to harness those mechanisms to optimize tissue derivation.
An example illustrating this need is the cell cycle, which is known to regulate fate control in hPSCs. It has been shown that CyclinD/CDK4-6 controls hPSCs differentiation capacity during the G1 phase 14, 15 , while, CyclinB1 promotes pluripotency maintenance in G2, at least for human embryonic stem cells (hESCs) 16 . Reciprocally, pluripotency factors have also been shown to provide feedback to cell cycle control. For instance, the pluripotency determinants OCT4 and SOX2 control miR-302 expression -a CyclinD1 translational repressor -in G1 17 , and OCT4 down-regulation blocks cell cycle progression in G0/G1 via p21 18 . These findings derived from cell population studies provide evidence that hPSCs coordinate cell cycle entry with early cell fate changes, yet thus far it has not been possible to investigate this coordination directly. Such studies require tools allowing visualization of both cell cycle and cell fate status at a single-cell level in real-time.
In the case of cell fate dynamics, it has been shown that TF heterogeneity can greatly impact early cell fate decisions during development 19 as well as clinical tissue design 20,21 .
Recently pioneering studies have begun to explore temporal TF dynamics and cell-to-cell TF heterogeneity in live single cells using transgenic reporter cell lines and quantitative timelapse microscopy 22 . For instance, continuous long-term cell lineage tracking of mouse embryonic stem cells (mESCs) expressing VENUS-tagged pluripotency TFs Oct4 or Nanog allowed quantification and comparison of their protein expression changes. This analysis revealed a strong heterogeneity in Nanog expression between cells and mESCs subsets with differing Nanog levels, with both factors differently impacting TF correlation networks and differentiation propensities 23 . Similarly, quantitative multi-generational analysis of an OCT4-mCherry fusion in hESCs revealed that cells' pre-existing OCT4 pattern can predict their extraembryonic mesoderm differentiation propensity 24 . Additionally, in a more recent study, quantitative live monitoring of mESCs co-expressing the two pluripotency TF fusion proteins Sox2-SNAP and Oct4-Halo showed that small endogenous fluctuations of Sox2 and Oct4 with respect to each other can bias early neuroectodermal versus mesendodermal fate decisions 25 . This illustrates the power of integrating live dynamical information from multiple TFs in clarifying otherwise inaccessible layers of cell fate control.
At a general level, analyses that resolve multiple spatiotemporally integrated mechanisms live remain rare, in part because live multiplexable cell reporters compatible with simultaneous visualization are scarce. This is particularly limiting for TF analysis, as TF reporters are all localised to the nucleus, which means that typically few different factors (labelled by fluorophores that emit in different colours) can be co-visualized and distinguished within cells. In addition, functionality/physiology can be impacted when TFs are tagged 22 . Furthermore, frequent time-lapse, multi-day imaging -as required to track singlecells and quantitatively monitor their cell state and fate changes across generations -can be particularly phototoxic and damaging to delicate cells such as stem cells 13,26,27 . These issues combined have made it particularly difficult to apply such approaches to hPSCs, and to our knowledge, no study to date has visualized live the spatiotemporal dynamics of multiple TF in human stem cells at single-cell resolution.
Here, we develop a novel class of 'live' reporters of TF status named ORACLE, which are localised to the nuclear envelope, making them multiplexable and compatible with conventional nuclearly-localised TF or cell proliferation reporters. We demonstrate their power and broad applicability using classical experimental paradigms in stem cell biology: exit from the pluripotent state and onset of early neuroectodermal differentiation in hPSCs.
By establishing hPSC lines expressing ORACLE reporters alone or in combination with conventional TF or cell proliferation reporters, we are able to reveal the spatiotemporal dynamics of cell fate transitions in hPSCs over multiple days and at single-cell level, by optimised multi-day/multi-colour time-lapse high-content microscopy. Using an ORACLE-OCT4 reporter, we show that there is significant cell-to-cell heterogeneity in the temporal loss of pluripotency among hPSCs differentiating into early neural lineage, and that spatial location likely determines the timing of pluripotency exit. By co-visualizing ORACLE-OCT4 and the nuclear cell cycle reporter FUCCI, we demonstrate that cell cycle duration is a poor predictor of pluripotency exit and that, instead, G1 phase length and OCT4 levels are coupled during pluripotency exit, suggesting that they are mechanistically linked. Finally, by combining ORACLE-OCT4 and an ORACLE-SOX1 reporter, we visualize in real-time the transition between pluripotency and early neural differentiation onset and reveal that pluripotency exit and differentiation onset are not tightly coupled in single-cells.

RESULTS
ORACLE cell fate reporter design. In order to expand the available toolkit for cell fate monitoring, we set out to establish a new class of fluorescent cell fate reporters. We wanted these to be amenable to multiplexing and visually compatible with commonly used existing reporters of spatiotemporal cell dynamics. All existing reporters -such as fluorescentlylabelled histone H2B chromatin 28 , the FUCCI cell cycle reporter 29 and fluorescently-labelled transcription factors (TFs) -are restricted to the nucleus. We therefore reasoned that fate reporters localised to other subcellular localisations would be valuable, complementary tools to enable cell biological discoveries. We considered designing fate reporters localised to cellular compartments such as microtubules, the Golgi apparatus, or the plasma membrane, but reasoned that robustly assigning such reporter signals to specific cells growing in compact colonies, such as hPSCs, would be technically challenging. We instead focused on the nuclear rim, as signals from reporters localised there would likely be clearly distinguishable from nuclear reporter signals and easy to assign to the cell of origin. We trialled several fluorescent protein fusions including KASH domains and LEM proteins 30,31 , nuclear lamina proteins (Emerin, Lamin A and B 32,33 ) and nuclear pore proteins 34 , and eventually selected the nuclear pore protein POM121. Relative to the other candidates, POM121 has ideal nuclear-rim protein properties: (a) fluorescently tagged expression in cells does not lead to any detectable morphological defects 34 , (b) it has well characterized and quality controlled published fusions 35 , (c) it has little to no diffuse cytoplasmic or nuclear signal, allowing for a distinct nuclear rim localisation, and (d) it has a high enough protein turnover 34,36,37 that it could be used as a proxy to monitor transcription factor turnover 24 .
Based on these properties, we assessed the suitability of fluorescently-labelled POM121 as a reporter of transcription factor level in hPSCs ( Fig. 1a and b, left, and Supplementary Fig.   1a and b). In our first strategy (Fig. 1a, left), we generated "exogenous" expression constructs in which fluorescently-labelled POM121 expression can be driven from the TF promoter of choice, by integrating a cassette comprising the TF promoter sequence and the labelled POM121 sequence at genomic safe harbours such as AAVS1 and ROSA26, using CRISPR knock-in (see Methods for detailed descriptions from here on). This approach allows the original TF to function with minimal disruption, provided a promoter sequence is known and well defined for the TF of interest. In the second strategy (Fig. 1b, left), we generated "endogenous" constructs allowing fluorescently-labelled POM121 expression to be controlled directly by the endogenous TF's promoter, by integrating the labelled POM121 sequence directly at the TF locus via translational coupling using the 2A system 38,39 . This ensures that expression of the endogenous TF is not affected. We call these types of constructs ORACLE for their ability to reveal fate.
ORACLE-OCT4 reports OCT4 expression with high fidelity at single-cell level. We used both strategies to generate live exogenous ORACLE ('exoORACLE') and endogenous ORACLE ('endoORACLE') reporters of OCT4. OCT4 is a pluripotency TF widely used as a marker of pluripotency, whose expression is both necessary to establish and maintain the pluripotent state and lost when cells exit pluripotency 40 . We tagged POM121 with fluorescent tdTomato and stably knocked-in exo and endo constructs in hESCs (see Methods). exoORACLE-OCT4 and endoORACLE-OCT4 reporters' expression led to bright nuclear rim-localised signal in all cells, which after triggering neural differentiation was lost after 5 days, as would be expected for OCT4 ( Fig. 1a and b (right) and Supplementary Fig.   1c and d).
To assess how faithfully ORACLE-OCT4 expression reports endogenous OCT4 expression, we induced neural differentiation and fixed cells at days 0 (untreated), 1, 2, 3, 4 and 5 following neural induction, and then analysed samples for ORACLE-OCT4 level and OCT4 immunofluorescence. We performed comparative quantitative analysis both at a cell population level and on a cell-by-cell basis for thousands of cells. As shown in Fig. 1c-h, the signals of both exoORACLE-OCT4 and endoORACLE-OCT4 gradually decreased throughout the time course, most visibly starting on day 3 of differentiation, similarly to OCT4 protein levels. This data indicate that ORACLE-OCT4 loss mirrors OCT4 loss at the population level ( Fig. 1c-d, f-g). When we compared ORACLE-OCT4 signal and OCT4 protein level at a single-cell level, we found a high degree of correlation for both constructs ( Fig. 1c and f). This correlation was particularly high for endoORACLE-OCT4, where it was typically >0.86 and up to 0.93. Thus, ORACLE levels quantitatively mirror transcription factor levels and changes at a single-cell level, indicating that they can be used to indirectly monitor TF levels with high fidelity.
Visualizing spatiotemporal cell-to-cell heterogeneity during pluripotency exit in hPSCs. We then sought to investigate OCT4 spatiotemporal dynamics at the single-cell level by multi-day time-lapse microscopy. To this end, we generated an endoORACLE- Pluripotent cells have been shown to express heterogeneous levels of OCT4 before on-set of differentiation 24 . Interestingly, through manual tracking of individual cells and their progeny throughout the 5 days of differentiation, we observed broad cell-to-cell heterogeneity also in the temporal pattern of OCT4 loss. Some differentiating hPSCs lost most of their OCT4 signal already after 2.5 days, while others only did so only after 4 days, i.e. more than 1 cell cycle later ( Fig. 2b and c). Importantly, such heterogeneity could also be observed between progeny from the same lineage ( Fig. 2d and e). This indicates that the cell of origin 24 alone cannot account wholly for the dynamics of pluripotency exit in hPSCs.
Notably, we observed that as pluripotency exit proceeds, cells at the periphery of colonies appeared to retain higher OCT4 levels for longer than cells away from the periphery (see Thus, spatial location is likely a key determinant of the timing, but not the kinetics, of pluripotency exit.

Co-visualizing cell cycle changes and pluripotency exit dynamics at single-cell level.
We then sought to exploit the ORACLE system to investigate how hPSCs coordinate cell cycle dynamics with pluripotency exit. To this end, we generated a triple CRISPR knock-in cell line co-expressing endoORACLE-OCT4, H2B-miRFP670 and the two-colour FUCCI cell cycle reporter 29  Similar to what we had observed with OCT4 variation, we found broad heterogeneity across individual cells in their cell cycle pattern following on-set of differentiation, with some cells shortening and others lengthening their cell cycle (Fig. 3f). In agreement with this data, cell cycle length across many cells correlated poorly with both time into differentiation ( Fig. 3g; r s = -0,25) and OCT4 level ( Fig. 3h; r s = 0,24). By contrast, G1 length was highly correlated with both time of differentiation and OCT4 levels (r s = 0,65 and r s = -0,64, respectively), whereby cells with highest OCT4 levels tended to have shortest G1 phase durations, in some cases as short as 1.5h or 2h. The correlation between OCT4 level and G1 length could be observed also at the single cell level (Fig. 3i). By contrast, neither cell cycle length nor S/G2 length correlated well with OCT4 level (Fig. 3i). Thus cell cycle length is a poor predictor of pluripotency exit. Conversely, G1 phase length is a good predictor of OCT4 level and therefore of pluripotency status during pluripotency exit, suggesting that these features are mechanistically linked.

Multiplexed visualization of TF levels in live cells.
Next, we set out to exemplify how ORACLE can be leveraged to enhance the number of different TF reporters that can be simultaneously visualized in cells. To this end, we generated a triple CRISPR knock-in cell line expressing endoORACLE-OCT4, H2B-miRFP670 and a live nuclear reporter of the pluripotency gene SOX2 42 . In the SOX2 construct, a mCherry-NLS protein is driven by the endogenous SOX2 promoter (pSOX2) by placing mCherry-NLS after the SOX2 coding sequence via the 2A system. As both the OCT4 (nuclear rim-localised) and SOX2 (nuclear localised) reporters are red fluorescent, this should allow testing whether two TF levels can be monitored simultaneously within a single fluorescence channel, effectively doubling the number of TF reporters that can be visualized in cells. Indeed, co-expression of pSOX2driven mCherry-NLS with ORACLE-OCT4 confirmed that distinct OCT4 and SOX2 signals could be visualized within individual cells, despite being in the same fluorescence channel (Fig. 4a). Furthermore, quantitation of the SOX2 nuclear signal showed that it could be unequivocally distinguished from the nuclear background signal of cells expressing ORACLE-OCT4 alone (Fig. 4b). Given this specificity, we next examined the relative abundance of OCT4 and SOX2 with respect to each other on a cell-by-cell basis. By quantifying ORACLE-OCT4 and pSOX2-driven mCherry-NLS signals across a number of cells, we found that OCT4 and SOX2 levels are highly correlated (Fig. 4c), consistent with the known cooperative role and interaction between OCT4 and SOX2 in pluripotency control and in agreement with previous findings in mESCs 23,43 . Taken together, these findings demonstrate that ORACLE allows for simultaneous observation and quantitation of dynamic behaviour of multiple TFs in live cells even within the same fluorescence channel, thereby multiplexing TF observation.
Visualizing live transcription factor 'handover' at single-cell level. Finally, we sought to exploit the ORACLE system to reveal the dynamics of transition between different cell states 'live' at single-cell level. To this end, we selected SOX1, a neuroectoderm TF and fate determinant 44 , and designed an endoORACLE-SOX1 reporter. Using this reporter we generated a triple CRISPR knock-in cell line co-expressing endoORACLE-OCT4, endoORACLE-SOX1 and H2B-miRFP670 (Fig. 4d, and Supplementary Video 3). We then induced neural differentiation in these cells to visualize the transition from pluripotency to early neural stem cell fate in individual cells, i.e. to observe the dynamics of OCT4 loss and SOX1 gain, hereafter referred to as OCT4-to-SOX1 'handover'. Importantly, use of H2B-miRFP670 allowed us to track without interruption single-cells and their progeny even in instances where they had neither OCT4 nor SOX1 signal, allowing us e.g. to capture dividing cells even if they had lost OCT4 and compare the SOX1 expression patterns of their daughter cells. To minimize phototoxicity, fluorescent signals were again sampled at different intervals (H2B-miRFP670 imaged every 10 minutes, ORACLE-OCT4 and ORACLE-SOX1 imaged every 3 hours). Imaged colonies carrying this combination of reporters gradually lost OCT4 and gained SOX1 on differentiation, as expected (Fig. 4d).
Uninterrupted tracking of cells and their progeny throughout the 5 days of differentiation revealed that, just like for OCT4, cells display an ample range in the dynamics of SOX1 expression. Some cells switched on SOX1 already at day 3, while others did so at day 4 or even later (Fig. 4d). Unexpectedly, we found that the timing of OCT4 loss in cells occurred independently of the onset of SOX1 expression. Specifically, we observed instances where two cells had a similar timing of SOX1 activation but a different timing of OCT4 loss (Figs. 4e, f), and conversely, instances where cells with different timing of SOX1 activation had a similar pattern for OCT4 loss (Figs. 4g, h). This indicates that hPSCs display broad differences in how they coordinate the transition between pluripotency exit and onset of early differentiation. It also suggests that OCT4 and SOX1 are not mechanistically coupled in this fate transition. Thus, the ORACLE system is equally capable of detecting and can help distinguish between TF-driven mechanisms that are and that are not coupled.

DISCUSSION
In this study, we expand the number of available 'live' fate and proliferation reporters that can be used to simultaneously and quantitatively monitor cellular behaviour. Our choice to localise ORACLE at the nuclear rim makes it particularly well suited for hPSCs, which grow as dense cell colonies with difficult to visualize compact cytoplasm. However, the ORACLE system can be easily modified in other cellular contexts to alternative (e.g. cytoplasmic, cell membrane) localisations. To exemplify the breadth of ORACLE's multiplexing capacity, we combined it with H2B, FUCCI and nuclear TF probes. This shows that ORACLE can easily be combined with almost any type of reporter, including a wide variety of nuclear (e.g. signalling like ERK-KTR 45 or YAP/TAZ 46 ) and cytoplasmic (e.g. cytoskeletal, polarity and metabolic) cell biological reporters and biosensors 47 . Because ORACLE exploits TF promoter-driven gene expression but does not involve tagging or mutation of TFs per se, it is not detrimental to cells 22 and does not modify pluripotency or differentiation propensity, and because it relies on targeted CRISPR knock-in it allows for control both of knock-in locus and copy number. This makes it an excellent non-invasive alternative to fluorescent TF fusion reporters. Furthermore, ORACLE's modular design is adaptable to any TF or gene of interest. The sum of all these features makes ORACLE a flexible tool well suited both for basic discovery as well as tissue engineering and quality control applications 48,49 .
The enhanced multiplexing capacity of ORACLE will enable studies of previously inaccessible regulatory layers, such as coordination of cell cycle/proliferation and cell fate transitions, TF handover and TF stoichiometry at single-cell level. To illustrate this we provide four examples that all warrant future investigation. In the first, we quantitated OCT4 levels during pluripotency exit and early NE differentiation by combining ORACLE-OCT4 with a H2B-miRFP670 reporter, allowing unequivocal tracking of cells before, during and after loss of the TF signal. We revealed broad cell-to-cell heterogeneity in OCT4 dynamics during early NE differentiation, in agreement with previously reported heterogeneity during early extraembryonic mesoderm differentiation 24 . In addition, we showed that OCT4 loss does not happen uniformly within the differentiating tissue, and in particular happens later in cells in the periphery of cell colonies. This finding is in contrast with previous reports 50 and suggests that colony size and shape may have an unappreciated impact on differentiation efficiency and patterning 51,52 . In the second example, we found that G1 phase length, and not cell cycle length, correlates with OCT4 level in hPSCs undergoing early NE differentiation. This finding indicates that cell cycle length per se is not informative about pluripotency exit status in hPSCs and reveals a tight connection between OCT4 and G1 length control that can now be investigated with the tools described here. This close relationship might be mediated via the CDK inhibitor (CKI) p21, as OCT4 has been shown to negatively regulate p21 in mouse embryonic stem cells 18,53 , and/or by other mechanisms coupling OCT4 and the CDK4-6/CyclinD1-3 gearbox directly. Although OCT4 inheritance in daughter cells in hPSCs has been proposed to be a determinant for readiness to initiate differentiation (prior to presentation of any lineage cues) 24 , our results suggest that this readiness may in fact stem from the underlying cell cycle properties and G1 length in cells.
In a third example, we used an ORACLE-OCT4 pSOX2-driven mCherry-NLS system and confirmed that OCT4 and SOX2 abundances correlate at single-cell level, consistent with the known OCT4-SOX2 cooperativity in pluripotent stem cells. Finally, in a fourth example we used a dual ORACLE-OCT4/ORACLE-SOX1 system to probe directly TF handover at the single-cell level, an event that until now has never been visualized in real-time. With this system, we discovered that there is variable lag time between OCT4 loss and SOX1 gain, sometimes as long as 24h, which is within one typical cell cycle duration (equivalent to 1 generation) difference in time. These observations suggest that OCT4/SOX1-controlled early cell fate decisions are possibly more loosely coupled than previously appreciated, suggesting that cells exiting pluripotency may have a window of time before integrating or executing instructive fate signals. Taken together, these examples illustrate the power of this novel class of reporters in a wider range of applications.
Understanding how cells make complex and integrated cell fate decisions in space and time within tissues remains a quintessential topic in developmental biology and an important goal in regenerative medicine. Single-cell approaches, particularly scRNAseq and molecular barcoding technologies, have brought important advances in both areas, but as these are endpoint analysis-based, they are not able to reveal how cellular cause-effect decisions take place in real time in tissues. The tools and approach described here lend themselves naturally to large-scale 'live' microscopy phenotyping approaches aimed at generating comprehensive spatiotemporal maps of cell fate and proliferation dynamics in real time and space. We anticipate that such approaches will enable orthogonal avenues of research in stem cell biology and predictive synthetic tissue design.

Maintenance and differentiation of hESCs
WA09 (H9) hESC line was purchased from WiCell (wicell.org) and maintained in Essential 8 medium (A151700, Thermo Fisher Scientific) on hESC-qualified growth factor reduced Geltrex-coated (A1413302, Thermo Fisher Scientific) 6 well plates. Cells were split into 6 well plates at 1:10 ratio when cells become confluent using 0.5 mM EDTA. Medium was changed everyday for maintenance of hESCs. To differentiate into neural stem cells (NSCs), essential 8 medium changed into PSC neural induction medium (A1647801, Thermo Fisher Scientific) on the next day after passage of hESCs. Cells were grown in the induction medium for 7 days to complete differentiation into NSCs then, the medium was changed every other day.
Then, hOCT4 promoter was removed and POM121 fused tdTomato was placed before the stop codon of endogenous OCT4 locus, followed by T2A to allow expression of target transcription factor genes together with nuclear rim expression. For the ORACLE-SOX1 construct, the fragments of 5' and 3' homology arms to target endogenous SOX1 locus were synthesized as gene fragments (Eurofins genomics). Then, POM121 fused to 3 copies of GFP (p30474, Euroscarf) was placed before the stop codon of endogenous SOX1 locus, followed by T2A sequence. SOX2-mCherry construct was modified from SOX2-t2a-mCherry (Plasmid #127538, Addgene) by replacing fragments of the 3' homology arm to target the UTR region of SOX2 after stop codon using gene synthesis (Eurofins genomics) and inserting additional NLS sequences at the end of mCherry. For H2B-miRFP670 and FUCCI constructs, the fragments of 5' and 3' homology arms to target ROSA26 locus were subcloned into H2B-670 (modified from pmiRFP670-N1, #79987, Addgene) or FUCCI (kind gift from Ludovic Vallier's lab, U. of Cambridge). All of the cloning procedures were performed using In-fusion HD Cloning kit (639650, Takara) for seamless DNA cloning.     Figure 2 were carried out using a custom-made image processing pipeline using MATLAB (MathWorks). A median background value was calculated individually for each image and images were background subtracted, nuclei were segmented using the reference nuclear DNA DAPI signal, and the median fluorescence per