Abstract
A fundamental challenge in cell biology is to understand how an individual cell commits to a particular fate1–3. A classic example is the fate decision between self-renewal and differentiation, which plays an essential role in the biology of human embryonic stem cells (hESCs)4. Despite significant advances in our understanding of development of the human embryo5,6, it is still unclear how, and when, an individual stem cell makes the decision to differentiate. Here, we used time-lapse fluorescence microscopy to capture differentiation of hESCs to trophoblast—the first cell fate decision during mammalian development. By tracing the histories of both self-renewing and differentiating cells, we found that each population displayed distinct levels of the pluripotency factor OCT4 long before they were exposed to a differentiation stimulus. The levels of OCT4 were lineage dependent; however, each mother cell distributed unequal levels of OCT4 to its daughter cells randomly during cell division. The resulting ratio of OCT4 between daughter cells— established immediately after division—determined downstream fates: cells receiving a greater ratio of maternal OCT4 showed sustained increases in OCT4 and a reduced capacity to differentiate. We propose a simple formula, pdaughter = (pmother)r, that successfully predicts the probability that a daughter cell will differentiate based on its mother cell’s differentiation probability and its inherited ratio of OCT4. Our study reveals that the balance between self-renewal and differentiation is altered by the ratiometric distribution of OCT4 during cell division. These findings imply that a stem cell’s fate is already largely determined by the time the cell is born.
In the early human embryo, hESCs give rise to either pluripotent cells of the inner cell mass that will form the embryo proper, or extraembryonic trophoblast cells that will become placental tissue5,7. Differentiation of hESCs to trophoblasts leads to reduced expression of the core pluripotency factor OCT4 and accumulation of the caudal type homeobox 2 (CDX2) transcription factor (Fig. 1a)5,8. This cellular fate decision can be recapitulated in vitro by treating hESCs with bone morphogenetic protein 4 (BMP4)9. After 24 h of BMP4 treatment, quantitative immunofluorescence (IF) reveals two emerging populations of cells: a pluripotent population with low CDX2 expression that retains the ability to differentiate into other cell types (Extended Data Fig. 1); and a differentiating population of cells with reduced OCT4 expression, increased CDX2 expression, and enlarged morphology (Figs. 1b-c). BMP4-treated hESC colonies adopted a radially symmetric pattern of differentiation that resembles the human gastrula10 (Extended Data Fig. 2). This spatial configuration—with pluripotent cells located at the interior of the colony and differentiated cells near the periphery—was recently shown to arise from a gradient of receptor polarization and diffusible factors6. However, the self-organization of a seemingly uniform starting population still raises the fundamental question: how does a single stem cell choose between self-renewal and differentiation?
To address this question, we developed a fluorescent reporter system to monitor expression of human OCT4, a canonical marker of the pluripotent state11, in live hESCs. We used CRISPR-mediated genome editing to fuse a monomeric red fluorescent protein (mCherry) to the endogenous OCT4 protein in WA09 (H9) hESCs and isolated a clonal population of single-allele knock-in reporter cells (Fig. 1d and Extended Data). The OCT4-mCherry fusion protein showed accurate co-localization with the endogenous OCT4 protein; similar degradation kinetics; and the same chromatin binding pattern near the promoters of OCT4 target genes (Extended Data Fig. 3). Moreover, cells bearing the OCT4-mCherry reporter were competent to differentiate into multiple differentiated cell types (Extended Data Fig. 4), and time-lapse imaging did not alter their proliferation characteristics (Extended Data Fig. 5). For each cell, we calculated a single OCT4 expression level by averaging OCT4-mCherry intensity over its cell cycle duration (Fig. 1e-f). In addition, we examined the time-series profile of OCT4 dynamics for individual cells and found that the majority of hESCs (68%) displayed sporadic bursts of OCT4 expression that lasted ~1.5 h, with some cells showing as many as 7 bursts (Fig. 1g). Finally, we calculated individual cell cycle durations, which ranged from 10-24 h with a mean duration of 14.6 h (Fig. 1h), consistent with the reported population doubling time of ~16 h12. Thus, our reporter system enabled the reliable analysis of single-cell OCT4 dynamics in hESCs and revealed considerable heterogeneity in untreated stem cells.
With this system in place, we set out to capture the fate decisions of hESCs in real time. First, we performed time-lapse fluorescence imaging of H9 OCT4-mCherry hESCs for 42 h under basal conditions (Fig. 2a). We then treated these cells with 100 ng/mL BMP4 to induce differentiation while continuing to monitor their responses. Within 12 h of treatment, each cell began to follow one of two distinct fate paths: sustained accumulation of OCT4; or a precipitous decrease in OCT4. After 24 h, cells were fixed and stained for expression of CDX2 to determine their final differentiation status (Fig. 2b). We imposed a strict cutoff to classify each cell as either pluripotent or differentiated based on its OCT4 and CDX2 expression levels. By fitting the data in Fig. 2b to a 2-component Gaussian distribution (Extended Data Fig. 6), we selected only those cells that belonged exclusively to either the pluripotent distribution (pdiff < 0.01) or the differentiated distribution (pdiff > 0.99), where pdiff represents the probability that a given cell has differentiated. We then traced both populations back through time—spanning multiple cell division events—and labeled each earlier cell according to its “pro-fate”—the fate to which it (or its progeny) would ultimately give rise. The majority of cells were either pro-pluripotent (71%, red traces in Fig. 2a), giving rise to only self-renewing cells; or pro-differentiated (24%, green traces in Fig. 2a), giving rise to only differentiated cells. Approximately 5% of cells were “pro-mixed” and gave rise to both fates (yellow traces in Fig. 2a). Overall, 89% of sister cells chose the same fate, suggesting a large degree of heritability in cell fate and the absence of classically described “asymmetric” cell divisions13. Thus, time-lapse imaging allowed us to group hESCs by their eventual fate categories before they had received a differentiation signal or had made a clear fate decision. Although the majority of progenitor cells gave rise exclusively to a single fate, a small but significant group of cells gave rise to two different fates.
We next asked whether there were preexisting differences between pro-pluripotent, pro-mixed, and pro-differentiated cell populations that might influence their fate decisions. Indeed, pro-pluripotent cells showed significantly higher OCT4 expression levels than either pro-differentiated or pro-mixed populations (Fig. 2c). This result echoes the observation that repression of Oct4 in mouse ESCs induces loss of pluripotency and differentiation to trophectoderm14. Pro-pluripotent cells also showed greater burst frequency (number of OCT4 bursts per hour) than pro-mixed cells (Fig. 2d) and had shorter cell cycle durations than both pro-mixed and pro-differentiated populations (Fig. 2e). The latter finding is consistent with reports that hESC self-renewal is linked with a shortened G1 cell cycle phase15. Furthermore, we observed that both burst frequency and cell cycle duration were strongly correlated with mean OCT4 levels (Fig. 2f-g). Taken together, these results show that undifferentiated hESCs display heterogeneous OCT4 levels, burst dynamics, and cell cycle durations. These single-cell features, which were evident as early as 2 days before the differentiation stimulus was presented, were associated with alternate cell fate decisions. Although the preexisting differences between pro-fate populations were statistically significant, these populations of cells still showed considerable heterogeneity and overlapping measurements, suggesting that there are additional factors or events that influence final cell fate.
Because OCT4 levels were the strongest predictors of cell fate (Fig. 2c), we next asked how heterogeneity in OCT4 levels arises in a population of hESCs. To identify the source of cell-to-cell heterogeneity, we monitored OCT4 expression continuously in proliferating, undifferentiated hESCs for 72 h and generated lineage trees of single-cell relationships (Fig. 3a). Visual inspection of the lineages revealed that OCT4 levels were most similar among closely related cells (i.e., cells emerging from a common cell division event), providing further support that OCT4 levels are heritable from mother to daughter cell. To quantify this heritability pattern, we calculated the differences in OCT4 levels between pairs of cells as a function of their shared history. Sister cells showed the most similarity in OCT4 levels, followed by “cousin” and “second cousin” cells (Fig. 3b). Both sister and cousin cells, but not second cousins, were more similar than randomly paired cells, indicating that similarity in OCT4 levels can persist for at least two cell cycle generations16. Suspecting that each cell division event introduced variability in OCT4 levels, we detected a strong correlation between the number of cell divisions and the difference in OCT4 levels between all pairs of cells (Extended Data Fig. 7). Thus, OCT4 levels are heritable from mother to daughter cell, but each division event introduces incremental variability in OCT4 expression levels.
Close examination of cell division events at high temporal resolution revealed the precise time during which variability in OCT4 levels arises during cell division. As cells entered mitosis, OCT4 became strongly associated with the condensed chromosomes (Fig. 3c, left panel). This compacted state persisted throughout anaphase until the two daughter chromatids could be visibly distinguished. We used this first time point—before cytokinesis was complete—to quantify the levels of OCT4 in both newly born daughter cells (Fig. 3c, center panel). Comparison of OCT4-mCherry intensity between daughter cells revealed that the distribution of OCT4 was not perfectly symmetric but instead adopted a bell-shaped distribution that was centered around a mean ratio of 1 (r = 1/1) (Fig. 3d). Approximately 38% of divisions produced daughter cells with r = 5/6 or a more extreme ratio; 12% of divisions resulted in r = 3/4 or a more extreme ratio; and 3% of division events resulted in r = 1/2 or a more extreme ratio. These differences in OCT4 ratios were not due to measurement error because the distribution of r between sisters remained consistent for several hours after cytokinesis both before and after BMP4 treatment (see below). In addition, we determined the half-life of OCT4 to be ~8 h (Extended Data Fig. 3), making it unlikely that asymmetric ratios were due to stochastic differences in protein degradation during the first 5 minutes of daughter cell lifetime. Moreover, OCT4 ratios were not correlated with nuclear area or radial position within the colony (Extended Data Fig. 8). Thus, significant differences in OCT4 protein levels between sister-cell pairs were established at the moment of cell division.
We next tested whether the ratio of OCT4 inherited by a particular daughter cell influenced its downstream behavior. By comparing OCT4-mCherry intensities between sister chromatids at the moment of cell division, we found that the inherited ratio of OCT4 established within the first 5 minutes of daughter cell separation was predictive of the final OCT4 level in each cell (Fig. 3e). Daughter cells receiving the larger proportion of OCT4 (r > 1) showed increased levels of OCT4 relative to the mother cell, whereas daughters receiving the smaller proportion of OCT4 (r < 1) showed permanent decreases in OCT4. This trend was also observed after differentiation (Fig. 3e, right panel) and became stronger as more time elapsed after cell division (Extended Data Fig. 9). Furthermore, the ratio of OCT4 immediately after division predicted the difference in final OCT4 expression between sister cell pairs (Extended Data Fig. 10). Thus, the ratio of OCT4 established immediately after cell division—before the nuclear envelope is formed—determined the amount of OCT4 that was maintained throughout the lifetime of a cell.
To summarize thus far, differences in OCT4 expression levels arise through asymmetric distribution of OCT4 to daughter cells (Fig. 3c-e). Precise levels of OCT4 are transmitted from mother to daughter cells as reflected by both the similarity among cells that share a common lineage (Fig. 3a-b) as well as the observation that most progenitor cells (89%) give rise to a group of cells with the same fate (Fig. 2a). Moreover, OCT4 levels are strongly predictive of cell fate decisions (Fig. 2c). Taken together, these results suggest that a cell’s probability of differentiation has both a heritable component—transmitted through the mother cell—and a random component that depends on the inherited ratio of OCT4 at the moment of cell division. This behavior can be expressed in a simple formula (Fig. 4a) in which the probability that a daughter cell will differentiate, pdaughter, is equal to the probability that its mother cell will give rise to differentiated cells, pmother, raised to the power of r, the inherited ratio of OCT4:
Here, pdaughter can represent either the probability that a daughter cell has differentiated (as shown in Fig. 2b) or the probability that the daughter cell will give rise to differentiated daughter cells (i.e., pro-differentiated, green circles in Fig. 2c). As such, Equation 1 is a recursive formula that can be used to assign a probability to every cell in a lineage tree. By fitting Equation 1 to our measured values of pdiff and r, we assigned a differentiation probability to each cell in a proliferating population including those cells that were observed before the differentiation stimulus was given (Fig. 4b). As expected, pro-pluripotent cells had the lowest differentiation probabilities (p < 0.5), pro-differentiated cells the highest differentiation probabilities (p > 0.5), and pro-mixed cells had intermediate differentiation probabilities (0.2 < p < 0.8).
Because all probabilities lie between 0 and 1 (pdaughter, pmother ∈ [0,1]) and because the OCT4 ratio is always a positive rational number (r ∈ ℚ+), Equation 1 has the following desirable properties: (i) As less OCT4 is inherited by a daughter cell (r → 0), it becomes more likely to differentiate (pdaughter → 1). (ii) As more OCT4 is inherited by the daughter cell (r → ∞), it becomes less likely to differentiate (pdaughter → 0). (iii) If OCT4 is split evenly between daughter cells (r = 1), each daughter cell has the same differentiation potential as the mother cell (pdaughter = pmother). Theoretical predictions of Equation 1 are illustrated in Fig. 4c. Each line shows how the differentiation probability of a daughter cell, pdaughter, changes as a function of the mother cell’s differentiation probability, pmother, and the inherited ratio of OCT4, r. For example, when the mother cell is likely to differentiate (p ≈ 0.9), a 1/2 ratio of OCT4 inherited by a pair of daughter cells should alter their differentiation probabilities by ~10%. Given the experimentally measured distribution of OCT4 ratios (blue shading in Fig. 4c), in which only 3% of cells inherit ratios in this extreme range (Fig. 3d), Equation 1 provides an explanation for why only 5% of cells gave rise to both pluripotent and self-renewing cell fates (pro-mixed group in Fig. 2a). In other words, the fairly narrow distribution of OCT4 ratios observed experimentally accounts for why most daughter cells choose the same fate.
However, Equation 1 also makes an unexpected prediction. Fig. 4c indicates that cells with intermediate differentiation probabilities, such as the pro-mixed group, should be especially sensitive to the inherited ratio of OCT4 (note that the slopes of the curves in Fig. 4c are steepest at r = 1 for p = 0.25 and p = 0.5) whereas cells that are already likely to differentiate (p near 1) should be relatively insensitive to r. We tested this prediction by comparing the change in final OCT4 levels versus inherited ratios in the pro-pluripotent, pro-mixed, and pro-differentiated populations. In fact, the pro-mixed population showed the most sensitivity to OCT4 ratio while the pro-differentiated population showed the least sensitivity (Fig. 4d). Mechanistically, this increase in sensitivity to OCT4 at intermediate levels is consistent with the finding that the OCT4 and CDX2 transcription factors are reciprocally inhibitory8. Such a “double-negative feedback loop” would lead to a bistable system17 in which a small change in OCT4 levels drives cells strongly toward either pluripotency or differentiation (Fig. 4e).
In conclusion, we developed an endogenous fluorescent reporter for the canonical pluripotency factor OCT4 to capture differentiation of human embryonic stem cells in real time. We found that the decision to differentiate to trophoblast is largely determined before the differentiation stimulus is presented to cells and can be predicted by a cell’s preexisting OCT4 levels, bursting frequency, and cell cycle duration. These results in human cells harmonize with studies of mouse ESCs in which differences in OCT4 expression14,18–20 and degradation kinetics21,22 are associated with different developmental fate decisions. However, we identify a precise window of time during which these differences arise by showing that the strongest predictor of trophoblast fate—OCT4 levels—is established during cell division through ratiometric distribution of OCT4 to daughter cells. Thus, a cell’s probability to differentiate is a mixture of both heritable and random components. These observations challenge a prevailing view of cell fate decisions by suggesting that the fate of a cell is already largely determined by the time it emerges from its mother cell. As such, the concept of pluripotency—currently defined as the capacity of a given cell to undergo differentiation—may be more properly comprehended as a heritable trait that applies to an entire lineage of proliferating stem cells.
METHODS
Culture and treatment of hESCs
WA09 (H9) hES cell line was purchased from WiCell (Wisconsin) and maintained in mTeSR1 (05850, StemCell Technologies) on growth factor reduced Matrigel (354230, BD). Cells were passaged every three days using 0.5% EDTA in PBS.
Genome editing
H9 cells were cultured on 10 cm dishes and when 80% confluent, were dissociated using 0.5mM EDTA. 10 × 106 cells were resuspended in 800ul ice-cold PBS-/-containing 25ug of the OCT4-mCherry donor vector and 25ug of the guideRNA/Cas9 vector. Cells were electroporated in 100ul tips (Neon, ThermoFisher Scientific) using program 19 of the optimization protocol (1050V, 30ms and 2 pulses) and resuspended in mTeSR1 supplemented with Rock inhibitor (S1049, Selleck Chemicals) at 10uM final concentration. When the colonies that expressed mCherry reached the size of a nickel, they were marked and picked in Matrigel coated 24-well plates.
Endogenous OCT4 levels
Endogenous OCT4 levels in H9 wild-type cells and H9 OCT4-mCherry clone 8-2 were determined by antibody staining using a rabbit anti-OCT4 antibody (ab19857, Abcam). Immunostaining was performed using standard protocols. Briefly, cells were fixed for 15 min in 4% paraformaldehyde and permeabilized and blocked for 30 minutes in 5% goat serum with 0.3% Triton X-100 in TBS. Incubation with primary antibody was performed overnight and the incubation with the secondary antibody (Molecular Probes) was done at room temperature for 45 minutes. Nuclei were visualized using NucBlue Fixed Cell Stain ready Probes reagent (R37606, Molecular Probes).
Live-cell imaging analysis
Asynchronous H9 OCT4-mCherry cells were plated on 12-well glass bottom plates (Cellvis) in phenol-red free or clear DMEM/F-12 (Gibco) supplemented with mTeSR1 supplement (05850, STEMCELL Technologies) approximately 24 hours before being imaged. Cells were imaged using a Nikon Ti Eclipse microscope operated by NIS Elements software V4.30.02 with an Andor ZYLA 4.2 cMOS camera and a custom stage enclosure (Okolabs) to ensure constant temperature, humidity, and CO2 levels. Fresh media with or without BMP4 was added every 24 h. Images were flat-field corrected using NIS Elements.
Image analysis
A custom ImageJ plugin (available upon request) was used to perform automated segmentation and manually tracking of hESCs. Fluorescence intensity was quantified using an adapted threshold followed by watershed segmentation of the OCT4-mCherry channel. The program tracked the cell ID, parent ID, frame number, mean intensity and exported this information to MATLAB for analysis.
ACKNOWLEDGEMENTS
We thank Paul Lerou, Galit Lahav, Allon Klein, Jean Cook, Paul Maddox, Bill Marzluff, Scott Bultmann, Peijie Sun, and members of the Purvis Lab for helpful discussions and technical suggestions. This work was supported by NIH grant DP2-HD091800-01, the W.M. Keck Foundation, and the Loken Stem Cell Fund.
AUTHOR CONTRIBUTIONS
S.C.W., R.D., and J.C. constructed the OCT4-mCherry reporter cell line. S.C.W. and R.D. performed validation studies. S.C.W. and C.D. performed live-cell imaging. C.D., R.A.H., and K. K. conducted image analysis and cell tracking. C.D., R.A.H., and J.E.P. performed computational analysis. S.C.W. and A.S.B. carried out lineage differentiation experiments. J.E.P wrote the manuscript with contributions from all authors.