Exit from naive pluripotency proceeds with variable latency but without asymmetric division to generate population heterogeneity

Summary

Mouse embryonic stem (ES) cells are pluripotent cell lines established from the preimplantation epiblast. ES cells undergo transition from naive to formative pluripotency in defined conditions. Here, we used the Rex1-GFPd2 (RGd2) fluorescent reporter to track ES cells traversing the cell state boundary by long-term single-cell imaging (LTSCI). We saw no reporter re-expression, indicating irreversible differentiation. We observed only symmetric divisions, consistent with a simple population dynamics model. Genealogical analysis showed that transitions were highly correlated between sister cells. The collapse of naive identity was invariably abrupt but was preceded by a variable lag period. Across the population, however, exit times varied by more than 15 h. The delay in entering transition extended up to three generations. Thus, ES cell differentiation timing is not directly determined by cell cycle. We speculate that asynchronous departure from the naive state may safeguard pluripotency progression in the embryo founder population.

Highlights

• •Live tracking of Rex1-GFPd2 ES cells reveals unidirectional, asynchronous transition
• •Sister cells exit naive pluripotency synchronously through symmetric divisions
• •Exit follows an abrupt transition after a variable lag, up to three generations
• •Exit timing is independent of cell cycle and population density

In this article, Martello and colleagues use long-term live tracking of Rex1-GFPd2-labeled mouse ES cells to dissect exit from naive pluripotency at single-cell resolution. They show that exit is unidirectional and asynchronous, proceeds through symmetric divisions, and follows an abrupt transition after a variable lag, independently of cell cycle and population density.

Introduction

Embryonic stem (ES) cells are derived from the preimplantation mouse epiblast (Brook and Gardner, 1997; Evans and Kaufman, 1981; Martin, 1981) and closely resemble their in vivo counterpart (Boroviak et al., 2014). Their defining features are self-renewal and naive pluripotency (Jaenisch and Young, 2008; Martello and Smith, 2014; Nichols and Smith, 2012; Smith, 2001). Self-renewal is defined as the repeated generation of daughter cells with unchanged proliferative capacity and differentiation potency. Self-renewal divisions can be symmetric (both daughter cells inherit the potency of the mother cell) or asymmetric (one daughter cell remains pluripotent and the other differentiates) (Morrison and Kimble, 2006). Naive pluripotency is the potential to give rise to all tissues in the adult body, including germ cells (Nichols and Smith, 2012; Smith, 2001), and is governed by a recursive transcription factor network (Dunn et al., 2014; Jaenisch and Young, 2008; Yamane et al., 2018; Young, 2011). The naive state of mouse ES cells is maintained by the cytokine leukemia inhibitory factor (LIF) (Smith et al., 1988; Williams et al., 1988) plus serum, or by two molecule inhibitors (2i): CHIR99021 (glycogen synthase kinase-3 inhibitor) and PD0325901 (MEK inhibitor that blocks Erk1/2 signaling) (Wray et al., 2011; Ying et al., 2008).

ES cells transition from naive to formative pluripotency upon withdrawal of 2i or LIF. Loss of self-renewal in 2i/LIF marks functional extinction of the ES cell state (Betschinger et al., 2013; Kalkan and Smith, 2014). Although in 2i conditions ES cells are substantially homogeneous, exit occurs asynchronously over ∼48 h (Betschinger et al., 2013; Kalkan et al., 2017; Mulas et al., 2017). Fluctuations in transcription factor expression may contribute to variability and delay in differentiation timing. Recent single-cell studies described ES cell cycle and division during early differentiation (Chaigne et al., 2020; Waisman et al., 2019) but did not address asynchrony.

REX1 (ZFP42) is a transcription factor tightly coupled to the naive pluripotency network. RGd2 reporters are widely employed to monitor ES cell status (Guo and Smith, 2010; Nakai-Futatsugi and Niwa, 2013; Toyooka et al., 2008; Wray et al., 2010). In RGd2 ES cells (Kalkan et al., 2017), one allele of the Zfp42 genes is replaced by a destabilized GFP protein. GFP fluorescence reflects Zfp42 transcription, and its loss coincides with self-renewal extinction (Betschinger et al., 2013; Kalkan et al., 2017; Leeb et al., 2014). Here, we used long-term single-cell imaging (LTSCI) of RGd2 cells to track the naive to formative transition at single-cell resolution.

Results

Exit from naive pluripotency remains asynchronous at clonal cell density

To monitor exit from naive pluripotency, we measured loss of self-renewal ability upon restoration to 2i (Figure 1A) (Betschinger et al., 2013). We used 2i rather than 2i/LIF to facilitate clonal tracking, as LIF delays the exit by ∼24 h (Dunn et al., 2014). To determine whether cell-cell interactions affected the transition, we used germline-competent chimera RGd2 cells (Kalkan et al., 2017) and plated them in 2i on laminin at either 500 cells/10 cm² (“clonal density”) or at 5,000 cells/10 cm² (“population density”) (Figure 1B). After 12 h, 2i was withdrawn to trigger differentiation and subsequently restored at different time points. For the clonal density assay, individual colonies were imaged for nuclei (Hoechst), GFP, and alkaline phosphatase (AP) staining, while flow cytometry was applied to quantify GFP-expressing cells for the higher density plates (Figure 1B).

Figure 1.

Exit from naive pluripotency remains asynchronous at clonal cell density

(A) ES cells are maintained in the naive pluripotent state by 2i. Upon 2i withdrawal, ES cells dwell in a reversible phase, where 2i can be reapplied, and cells return to the naive state. Prolonged withdrawal results in irreversible exit and reapplication of 2i have no effect. RGd2 allows monitoring of the exit dynamics in near real time.

(B) Experimental strategy used to measure the dynamics of exit from naive pluripotency in cells cultured in 2i.

(C) Representative images of recovery kinetics assays at clonal density after the indicated hours of 2i withdrawal. Colonies were scored by both alkaline phosphatase staining (AP, top row) and RGd2 fluorescence (bottom row). Nuclei staining by Hoechst33342 (middle row) allowed identification of GFP-negative colonies.

(D) The fraction of each colony type is shown for alkaline phosphatase (squares in top) and RGd2 (circles in bottom). Markers show mean and error bars show SEM of 8 biological replicates, from 4 independent experiments, where filled markers show positive colonies, empty markers show negative colonies, and half-filled markers show mixed colonies. The Weibull distribution (CDF) was fit to both the positive fraction (solid line) and negative fraction (dashed line) of colonies. The dotted line connects the mixed fraction of colonies.

(E) Flow cytometry profiles from the population recovery kinetics. Two biological replicates from a representative experiment are shown. x axis: log10GFP. y axis: count. Each profile consists of at least 10,000 cells.

(F) Quantification of the flow cytometry profiles from population recovery kinetics showing the fraction of RGd2-positive (filled markers) and negative (empty markers) cells. Markers show mean and error bars show SEM of 8 biological replicates from 4 independent experiments. The Weibull distribution was fit to both the positive fraction (solid line) and negative fraction (dashed line) of cells.

(G) Experimental strategy to perform LTSCI of ES cells exiting the naive state.

(H) Representative time course showing BF (top row) and RGd2/Gap43-mCherry (bottom row) with arrows indicating cells from a single branch of the lineage tree (I).

(I) Quantified metrics from the lineage in (H) with color corresponding to arrows and vertical black lines corresponding to time points.

See also Figure S1 and S2.

Colony types were scored as positive, mixed, or negative based on AP staining and GFP fluorescence (Figures 1C and S1A). A small fraction of negative colonies were apparent after 2i withdrawal for 16 h, and this number increased dramatically at 24 and 36 h (Figures 1C and 1D). By 48 h, the majority of colonies (79% ± 3%) were negative for both AP and GFP. We observed a high level of correlation between AP staining and RGd2 expression across all colony types (all Pearson’s ρ > 0.992, all p < 1.01 × 10⁻⁴). For the higher density cultures (Figure 1E), we resolved GFP-positive and negative populations by flow cytometry (Figure S1B) and dynamics similar to those of ES cells at clonal density (Figure 1F).

As exit from naive pluripotency can be thought of as collapse, or failure, of a transcription factor network, we fitted exit timings to a 2-parameter Weibull cumulative distribution function (CDF) (Figures 1D and 1F; Tables S1 and S2), widely used to model failure of processes (Weibull, 1951). The shape parameter (γ) indicates how the process evolves in time (Tables S1 and S2). It was >1 in all conditions, implying that collapse of the pluripotency network increases over time.

These analyses showed that exit from pluripotency occurs asynchronously over 48 h, as reported (Betschinger et al., 2013; Kalkan et al., 2017; Leeb et al., 2014; Mulas et al., 2017). Importantly, these dynamics were not affected by density, suggesting intrinsic control.

Single-cell tracking of ES cell exit from naive pluripotency

Preservation of transition dynamics at clonal density allowed us to apply LTSCI and tracking. To facilitate cell segmentation, we introduced into RGd2 cells a plasma membrane fluorescent reporter (Gap43-mCherry, methods), which did not affect exit kinetics (p = 0.6892, two-way ANOVA) (Figures S2A–S2C).

Cells were plated at clonal density in 2i and imaged by confocal microscopy for 72 h (Figures 1G and 1H). By semi-automated tracking and analysis (methods), we determined position, area, fluorescence, genealogy, and fate of each cell (Figures 1I and S2D). In 2i, cells were viable and maintained GFP signal, thus ruling out photobleaching (Figures S2E–S2I). Under exit conditions (Figures 1G–1I), GFP signal decreased from 16 h onward, as reported (Kalkan et al., 2017). We tracked 2,624 cells from two experiments up to the fifth generation (Table 1).

Table 1.

Summary of sample sizes for conditions and experiments

		Experiment 1			Experiment 2
Cell line	Condition	Cells	Lineages	Data points	Cells	Lineages	Data points
Rex1-GFPd2	N2B27	1,384	64	19,138	573	43	8,691
Rex1-GFPd2	2i	405	17	6,141	154	18	2,438
E14 IVC	2i	56	6	664	52	6	1,258

Exit from naive pluripotency is associated with a shorter cell cycle, cell spreading, and increased motility

From LTSCI data, we measured cell-cycle length as the time between two cell divisions. Cell-cycle length in 2i (10.20 ± 2.14 h) was longer than in N2B27 (9.26 ± 2.03 h) (p = 2.771 × 10⁻⁹; rank sum) (Figure 2A). The cell-cycle length in 2i remained constant over generations (p = 0.2269; Kruskal-Wallis), while in N2B27 it progressively shortened (p < 2.2 × 10−16) (Figure 2B). Analysis of static images of cells exiting from naive pluripotency at clonal density confirmed this shortening (Figures S3A–S3E).

Figure 2.

Exit from naive pluripotency is associated with a shorter cell cycle, transient cell spreading, and increased motility

Boxplots superimposed with swarm plots, each point represents a single cell. All analyses shown are from experiment 1; experiment 2 yielded similar results.

(A) Cell-cycle length in 2i and after 2i withdrawal. ES cells in 2i show a mean cell-cycle length of 10.20 ± 2.14 h (N_cells = 196), which is longer than in N2B27 (9.26 ± 2.03 h, N_cells = 602; p = 2.77 × 10−9, two-tailed Wilcoxon rank-sum test).

(B) Cell-cycle length is unchanged across generations in 2i (p = 0.2269, Kruskal-Wallis test with Bonferroni correction) but decreases over generations in N2B27 (p < 2.2 × 10−16, Kruskal-Wallis test).

(C) RGd2-positive cells have a longer cell cycle (9.68 ± 1.87 h, N_cells = 275) than RGd2-negative cells (8.91 ± 2.10 h, N_cells = 327, p = 1.108 × 10⁻⁸, two-tailed unpaired Wilcoxon rank sum).

(D) Cell-cycle length decreases over generations for both RGd2-positive cells (p = 5.209 × 10⁻¹¹, Kruskal-Wallis test with Bonferroni post hoc correction) and RGd2-negative cells (N2B27, p < 2.2 × 10−16, Kruskal-Wallis test with Bonferroni post hoc correction).

(E) Representative lineages for 2i and N2B27; each line represents a cell.

(F) ES area is smaller in 2i (596 ± 257 μm², N_cells = 207) than in N2B27 (768 ± 303 μm², N_cells = 725, p = 4.071 × 10⁻¹³, two-tailed unpaired Wilcoxon rank sum).

(G) Cell area decreases over generations in both 2i (p = 9.600 × 10⁻⁸, Kruskal-Wallis test with Bonferroni post hoc correction) and N2B27 (p = 5.693 × 10⁻¹³, Kruskal-Wallis test).

(H) RGd2-positive cells have a larger cell area (873 ± 318 μm², N_cells = 304) than RGd2-negative cells, with an area (693 ± 268 μm², N_cells = 421) (p = 2.609 × 10⁻¹⁴, two-tailed unpaired Wilcoxon rank sum).

(I) Cell area decreases over generations in RGd2-positive cells (p = 9.600 × 10⁻⁸, Kruskal-Wallis test with Bonferroni post hoc correction), whereas in RGd2-negative cells in N2B27 a significant difference is observed only between the first and second generation (generation 1 vs. 2: p = 0.0436; all other comparisons: p > 0.05, Kruskal-Wallis test with Bonferroni post hoc correction).

(J) Representative lineages for 2i and N2B27; each line represents a cell.

(K) Average cell velocity is lower in 2i (0.14 ± 0.07 μm/min, N_cells = 196) than in N2B27 (0.16 ± 0.08 μm/min, N_cells = 602, p = 1.391 × 10⁻⁴, two-tailed unpaired Wilcoxon rank sum).

(L) Average velocity decreases over generations in 2i (p = 1.049 × 10⁻⁶, Kruskal-Wallis test with Bonferroni post hoc correction) and increases in N2B27 (p = 2.379 × 10⁻¹⁵, Kruskal-Wallis test).

(M) RGd2-positive cells are less motile (0.13 ± 0.06 μm/min, N_cells = 275) than the RGd2-negative cells (0.18 ± 0.08 μm/min, N_cells = 327, p < 2.2 × 10−16, two-tailed unpaired Wilcoxon rank sum).

(N) Average velocity is unchanged across generations in RGd2-positive cells (p = 0.0559, Kruskal-Wallis test with Bonferroni post hoc correction), whereas RGd2-negative cells become more motile (p = 7.082 × 10⁻⁹, Kruskal-Wallis test with Bonferroni post hoc correction).

All boxes indicatethe interquartile range (25th - 75th percentile), the central line indicates the median and the wiskers extend to 1.5x IQR. Individual dots represent single cells as agglomerate swarm plots.

See also Figure S3.

We examined if cell-cycle shortening begins immediately after 2i withdrawal or later. GFP-positive cells in N2B27 had a shorter cell cycle (9.68 ± 1.87 h) than cells in 2i, but longer than GFP-negative cells (8.91 ± 2.10 h) (Figures 2C and S3F) (p = 1.108 × 10⁻⁸). Therefore, shortening of the cell cycle begins prior to exit from naive pluripotency, as recently proposed (Waisman et al., 2019), and continues over successive generations (Figure 2D).

Previous studies in human pluripotent stem cells (PSCs) showed that distinct cell-cycle phases differ in their responsiveness to differentiation cues (Pauklin and Vallier, 2014). In murine naive PSCs, studies based on fluorescence-activated cell sorting (FACS) reached opposing conclusions about the G1 phase being more receptive to differentiation stimuli (Chaigne et al., 2020; Coronado et al., 2013; Waisman et al., 2017). Analysis by single-cell RNA sequencing also failed to detect any differentiation propensity for a cell-cycle phase (Jayaram et al., 2023). We looked for any propensity in the time of exit across the cell cycle of each cell in our dataset. In both biological replicates, we found no deviation from a uniform distribution (one-sample Kolmogorov-Smirnov test; replicate 1: D = 0.096, p = 0.316, n = 96; replicate 2: D = 0.105, p = 0.404, n = 69), indicating that exit does not preferentially occur in any specific cell-cycle phase (Figure S3G).

We evaluated changes in cell size during transition, measured as cell area (Figure 2E). ES cells in 2i were smaller (596 ± 257 μm²) than exiting cells (768 ± 303 μm², p = 4.071 × 10⁻¹³) (Figure 2F) as previously reported (Neagu et al., 2020; Waisman et al., 2017). After this initial cell spreading, in either condition, cell area decreased over generations as the number of cells in a colony increased (Figure 2G). However, GFP-positive cells exhibit a greater surface area (873 ± 318 μm²) than negative cells (693 ± 268 μm²) (p = 2.609 × 10⁻¹⁴) or 2i ES cells (Figures 2H and 2I). Thus, cells flatten and spread out prior to exit. The decrease in cell area occurred between the first and second generation (p = 0.0436) after GFP loss and then stabilized. Statistical analysis revealed no preference for exit at any particular colony size (binomial tests, Bonferroni-adjusted p > 0.05) (Figure S3H), indicating that exit timing is not biased by the number of cells in the lineage.

We observed increased motility from around GFP downregulation time (Figure 2J). ES cells in 2i moved slower (0.14 ± 0.07 μm/min) than in N2B27 (0.16 ± 0.08 μm/min) (p = 1.391 × 10⁻⁴) (Figure 2K), in agreement with previous observations (Waisman et al., 2019). Over generations, average velocity decreased in 2i (p = 1.049 × 10⁻⁶), whereas increased in N2B27 (p = 2.379 × 10⁻¹⁵) (Figure 2L). In N2B27, GFP-positive cells were less motile (0.13 ± 0.06 μm/min) than GFP-negative cells (0.18 ± 0.08 μm/min) (p < 2.2 × 10⁻¹⁶) (Figure 2M). Moreover, motility increased over generations in the GFP-negative cells (p = 7.082 × 10⁻⁹) (Figure 2N).

Our observations confirm that exit from naive pluripotency is preceded by modest cell-cycle acceleration, an increase in cell surface area, but unchanged low motility (Waisman et al., 2019). After exit, the cell cycle shortens further, cell area reduces and then remains constant, and motility increases.

Related cells exit naive pluripotency at similar times

To explore the ES cell transition asynchronicity, we asked whether exit dynamics were more similar between related than non-related cells. From each GFP trajectory, we extracted four parameters: initial, final, mean value, and integrated area under the curve (AUC) (Figure S4A). We then performed genealogical comparisons within generations (Figure S4B): sister cells (ss) and first cousins (cc), and across generations (Figure S4C): mothers-daughters (md) and grandmothers-granddaughters (gd) (Figures 3A and 3B).

Figure 3.

Related cells exit naive pluripotency at similar times

(A) Sister-cell (ss) comparison for final RGd2 level in 2i (n = 82). Wilcoxon rank-sum test, p = 0.2193, Spearman ρ = 0.4587 (permutation test).

(B) Genealogical comparisons for RGd2 level in N2B27. Sister (ss): rank-sum p = 0.7416, Spearman ρ = 0.9006, n = 271. First cousin (cc): rank-sum p = 0.3388, Spearman ρ = 0.8411, n = 386. Mother-daughter (md): rank-sum p < 2.2 × 10−16, Spearman ρ = 0.6988, n = 600. Grandmother-granddaughter (gd): rank-sum p < 2.2 × 10−16, Spearman ρ = 0.0692, n = 474. All p values from Wilcoxon rank-sum tests; correlations assessed by Spearman’s correlation with permutation testing.

(C) Paired rank-sum differences for 5 parameters across genealogical relations, calculated for two independent experiments.Color indicates the Wilcoxon rank-sum test p value. The colormap distinguishes significant (blue) from non-significant (red) comparisons.

(D) Spearman correlation for 5 parameters across genealogical relations, calculated for two independent experiments.Correlations not reaching significance (p > 0.05, permutation test) are shown as zero. The colormap represents effect size following Cohen (2013): blue for significant difference, red for no significance. Black for no effect (ρ < 0). Dark purple for small effect (0.1 ≤ ρ < 0.5), light purple for medium effect (0.5 ≤ρ < 0.8), and bright grays for large effect (0.8 ≤ ρ).

(E) Schematic of the population model. Cells divide at rate λ and die at rate δ, yielding a net growth rate γ. Undifferentiated cells are shown as green ovals, differentiated cells as gray pentagons.

(F) Division outcomes for RGd2-positive cells: symmetric self-renewal χ_R, symmetric exit χ_S, and asymmetric χ_A.

(G) Time-course of fraction of RGd2-positive cells from the single-cell snapshot data used to infer model parameters. The “x” shows the mean while error bars indicate SEM. Superimposed is the leading-order polynomial model simulated using inferred parameters (solid line).

(H and I) Time course of (H) symmetric self-renewal divisions and (I) exit divisions from the single-cell tracking data. The “x” shows the mean, error bars indicate SEM and the solid line indicates model simulation.

(J) SSE of the number of divisions between the empirical data and the simulations over the constraints of minimum division time and symmetry bias. Colors indicate SSE.

See also Figure S4.

Each comparison combined rank-sum p values to test for differences and Spearman correlations to quantify similarity. Agreement between the two measures supports the robustness of the observed relationships.

The rank-sum p value is higher for more similar pairs and lower for dissimilar pairs. In 2i, where all cells remain GFP positive (Wray et al., 2010), sister cells showed no significant difference (rank-sum p = 0.2193) and only a modest correlation (ρ = 0.4587) (Figures 3C and 3D), consistent with a small effect size (Cohen, 2013). In N2B27, ss and cc showed little difference between pairs of cells (rank-sum p = 0.7416 and 0.3388, respectively) but high correlation (0.9006 and 0.8411, large effect size). Across generations, both md and gd pairs differed significantly (rank-sum p < 2.2 × 10⁻¹⁶). However, the correlation was moderate for md (ρ = 0.6988) and negligible for gd (ρ = 0.0692).

Similar analyses for all GFP metrics and cell-cycle length (Figures 3C and 3D) confirmed that ss and cc pairs in N2B27 showed no significant differences and remained highly correlated. The high correlation between ss and cc we observed could be due to the fact that cells belonging to the same generation display similar levels of GFP. We thus repeated the analyses within each generation and again found strong positive correlations between ss and cc (Table S3).

We conclude that sister cells are born with comparable GFP levels, maintain similar levels over their lives, as indicated by the mean and AUC fluorescence, and divide after nearly identical cell-cycle lengths, yielding equivalent final intensities. These observations show that sister ES cells adopt the same fate and follow similar dynamics during exit from naive pluripotency. To independently assess division symmetry in sister cells, we examined cases where GFP-negative daughters were generated from positive mothers. Divisions were classified based on daughters’ final fluorescence values and then validated by visual inspection. Out of 141 divisions analyzed, 139 (98.5%) were symmetric, with both daughters GFP negative (Figure S4D). Thus, sister ES cells divide symmetrically during exit from naive pluripotency, in the absence of exogenous patterning cues.

Population dynamics modeling substantiates a symmetric division mode during exit from naive pluripotency

The preceding observations show that exit from naive pluripotency occurs via symmetric divisions. Building on theoretical frameworks used for tissue stem cell populations (Clayton et al., 2007; Lopez-Garcia et al., 2010; Simons and Clevers, 2011), we modeled how the probability of exit evolves over time and whether divisions are symmetric or asymmetric. We modeled exit in a population of GFP-positive (P) and negative (N) cells. Cells divide at rate λ and die at a logistic rate that increases with total cell number, determined by parameters δ and K (Figure 3E). The carrying capacity is K=((λ−δ) · K₀)/δ, with K₀ defining cell death dependency on cell density, and the net growth rate is γ = λ−δ. We assert that exit is an irreversible process coupled to the divisions of P cells (Figure 3F). Having these data, we modeled the probability of exit choosing from a set of phenomenological candidate functions f(P,N,t). Exit divisions could be symmetric with probability ρ or asymmetric with probability 1−ρ. These dynamics yield coupled first-order nonlinear ordinary differential equations,

$$\frac{dP}{dt}=\left(\gamma \left(1-\frac{N+P}{K}\right)-\left(1+\rho \right)f·\lambda \right)P\text{,}$$ (Equation 1)

$$\frac{dN}{dt}=\gamma \left(1-\frac{N+P}{K}\right)N+\left(1+\rho \right)f·\lambda P\text{.}$$ (Equation 2)

Parameters were inferred from cell counts in individual clones measured every 8 h, with cells classified as P or N, irrespective of lineage. These data (“x” in Figures 3G and S4E–S4H) provided mean colony size and composition over time. Model parameters were inferred by constrained optimization, minimizing the sum of squared errors (SSEs) between simulations and empirical data.

Using the inferred parameters, we test the model’s predictive power by simulating the total number of divisions (Figure S4I) and the self-renewal and exit divisions (Figures 3H and 3I) of P cells and comparing against single-cell tracking data. To identify the best-fitting parameter set and test function, we computed the SSE across combinations of symmetry bias and cell-cycle length. The optimal fit corresponded to a ∼10-h cell cycle and ρ ≈ 1 (Figures 3J and S4J), consistent with symmetric division and the experimental observation of 98.5% symmetric events.

Analytically solving the model with a time-explicit test function shows the solution follows a Weibull distribution (supplemental methods, Equation 26). Weibull parameters derived from the inferred leading-order polynomial (LOP) values closely matched those obtained experimentally (Figure 1). This analytical result supports the view that exit from naive pluripotency can be viewed as a time-dependent failure process, with symmetric exit divisions and an increasing probability of exit over time.

ES cells exit at different time points but with similar and rapid half-life decay kinetics

Our model assumes that the probability of exit increases over time, independent of cell-cycle phase or generation. Indeed, unlike cells in 2i, which remain GFP positive across generations (Figure 4A), cells in N2B27 downregulated GFP at variable times during the cell’s life and at different times post 2i withdrawal (Figure 4B). We next asked whether exit frequency depends on colony size, as ES cells could affect the behaviors of sisters and cousins via pro-differentiation cytokines (Kunath et al., 2007; Mulas et al., 2017). However, no significant bias was detected for any colony size (binomial tests, Bonferroni-corrected p > 0.05; Figure S3H), indicating that exit occurs with similar probability in small and large colonies and is thus intrinsically regulated. We next tested whether intrinsic factors, such as initial RGd2 levels in the first cell of each lineage (matriarch), influence exit timing, as initially higher GFP could delay transition. We compared multiple metrics of matriarchal GFP intensity (initial, maximum, and final values) with the timing of the first exit event in each lineage, assessed in real time and by generation. No significant correlation was found between exit timing and either initial or maximum GFP intensity (for all Spearman ρ: p > 0.05). Only one replicate showed a weak and significant correlation with final GFP intensity (Figure S4K). Thus, RGd2 levels in the matriarch did not show consistent association with the timing of lineage exit.

Figure 4.

ES cells exit at different time points but with similar and rapid half-life decay kinetics

(A) RGd2 trajectory for a representative ES cell lineage self-renewing in 2i. Each line is a cell. The dashed gray line marks the RGd2 threshold, calculated as the 95% confidence interval of the positive control distribution (Figures S2H and S3F), and used to classify cells as RGd2 positive or negative.

(B) RGd2 trajectory for 4 representative lineages exiting in N2B27, downregulating the reporter at disparate times and in different generations. The dashed gray line shows the same RGd2 threshold as in (A).

(C) Schematic of time-translation method for trajectory alignment.

(D) Real-time trajectories for 1,384 cells from 64 lineages in the N2B27 condition superimposed.

(E) Heatmap of the real-time trajectories for all lineages in the N2B27 condition with the time average shown as a white line. Colors indicate population density, ranging from 0 to 1.

(F) Real-time percent of cells that have crossed the signal threshold for all lineages.

(G) Translated-time trajectories for all lineages in the N2B27 condition superimposed.

(H) Heatmap of the translated-time trajectories for all lineages in the N2B27 condition with the time average shown as a white line. Colors indicate population density, ranging from 0 to 1.

(I) Translated-time percent of cells that have crossed the signal threshold for all lineages.

We then analyzed GFP signal kinetics. None of the 1,957 tracked cells reactivated the reporter after threshold crossing up to the fifth generation, confirming the unidirectional nature of exit in defined conditions (Kalkan et al., 2017; Smith, 2017). The GFP signal also decayed immediately before exit, consistent with the short half-life of the destabilized reporter and its transcriptional repression before exit. To assess whether GFP decay was consistent across lineages, we aligned all trajectories by time translation, setting time zero to the point when the first cell in each lineage crossed the GFP-loss threshold (black arrows in Figure 4C). We then compared real-time and translated kinetics (Figures 4D–4I) by superimposing all tracked lineages, computing the mean GFP trajectory for the population, and examining exit dynamics on a per-cell basis. Time translation reduced GFP signal heterogeneity (Figures 4D and 4G), revealing an exponential decay at the population level (Figures 4E and 4H). Despite being applied to entire lineages, the alignment also decreased variability among individual cells. All cells reduced GFP intensity by 50% within 3 h of the first exit, further indicating that cells of the same lineage exit naive pluripotency synchronously. Finally, comparing real-time and translated data revealed that exit kinetics shifted from a sigmoidal (Figure 4F) to a hockey-stick curve (Figure 4I). Thus, GFP downregulation occurs acutely and exit proceeds with fast and simultaneous kinetics.

Discussion

In this study, we applied continuous LTSCI to examine the exit of mouse ES cells from naive pluripotency in defined conditions, a system recapitulating aspects of peri-implantation epiblast progression (Carbognin et al., 2023; Kalkan et al., 2017; Lackner et al., 2021; Smith, 2017). We tracked 2,624 cells in two replicate experiments.

Our analyses revealed that, prior to exit, cells exhibit a modest shortening of the cell cycle and transient spreading. After exit, the cell cycle continues to shorten, but cell size decreases. Motility increases only post transition. These features parallel the “rosette pluripotency” state recently described as an intermediate between naive and formative phases (Neagu et al., 2020). Indeed, independent studies performed under our experimental conditions observed cell spreading and acquisition of polarity (i.e., apical actin accumulation) (De Belly et al., 2021; Hamouda et al., 2025). RGd2 re-expression was never observed after downregulation, consistent with an irreversible transition process. As in mass culture, exit occurred asynchronously over ∼48 h, yet synchrony was maintained among lineally related cells. Our population dynamics model accounts for this behavior: asynchrony arises due to a variable initial lag phase of 15.04 ± 10.56 (SD) h that precedes an acute exponential transition process.

A previous time-lapse study of ES cells withdrawn from 2i reported cell spreading and changes in motility (Waisman et al., 2019) but lacked a marker distinguishing naive and formative states. Use of the RGd2 reporter allowed us to discriminate cells before and after transition, thereby revealing that cell spreading increases prior to exit, while cell motility increases after exit. Direct measurements of generation time confirmed that the cell cycle starts shortening upon 2i withdrawal (Waisman et al., 2019). Cell cycle shortening may signal impending differentiation. However, ES cells maintained in serum/LIF display a faster cycle with a shortened G1 phase due to Erk signaling (ter Huurne et al., 2017). Thus, the initial cell cycle change may be a direct response to Erk activation rather than differentiation itself. Continued shortening after exit may instead mirror the rapid proliferation of the post-implantation epiblast (Snow, 1977). While this study focuses on the 2i condition, future work could examine whether alternative innervation of the naive pluripotency network through different culture conditions, such as 2i/LIF, CHIR/LIF, PD03/LIF (Dunn et al., 2014; Jaenisch and Young, 2008; Yamane et al., 2018; Young, 2011), and a2i (Shimizu et al., 2012), affects the dynamics of exit or introduces differences in intercellular coordination.

Specific cell cycle phases have been linked to different cell fate propensities in human conventional PSCs (Pauklin and Vallier, 2014), but evidence in murine naive ES cells remains contradictory. Two studies would suggest that G1 phase is more receptive to differentiation stimuli (Chaigne et al., 2020; Waisman et al., 2017), whereas another reported this effect only under retinoic acid (RA) exposure (Coronado et al., 2013). The 3 studies are based on FACS of cells in G1. In stark contrast, a recent study performed FACS for all different cell cycle phases and failed to detect any difference (Jayaram et al., 2023). Methodological differences (e.g., RA presence, sorting strategy) may explain such inconsistencies. The Buecker lab, with single-cell RNA sequencing, failed to detect any differentiation propensity for a cell-cycle phase.

Using our orthogonal approach, we found that exit events were uniformly distributed across the cell cycle, indicating that exit does not preferentially occur at any specific phase.

In contrast to ES cells cultured in serum (Toyooka et al., 2008), reversibility of Rex1-GFP reporter has been tested under serum-containing media, reaching inconclusive results (Marks et al., 2012; Toyooka et al., 2008), possibly because of confounding effects of serum or differences in the cell lines or reporter systems used. Under chemically defined conditions, like 2i in N2B27, no reactivation was observed by flow cytometry and cell sorting analyses (Kalkan et al., 2017). We did not detect any reactivation of the Rex1-GFP reporter after downregulation, thus indicating that the transition is irreversible.

Notably, Rex1 signal starts to decrease 1 or 2 generations before exit occurs (Figures 1I and 4B). Although qualitative, this observation suggests a transition from the naive state, in which Rex1 promoter is fully active, to a partially active state, in which cells are still responsive to signals supporting naive pluripotency, and then to a precipitate full inactivation of Rex1 transcription, coinciding with exit from naive pluripotency. Future quantitative and modeling analyses could be based on these observations, potentially informing us on the variable duration of the lag period observed before exit.

Asymmetric fates are observed after cell division in various stem and progenitor cell systems (Simons and Clevers, 2011). Live imaging has suggested that such divisions may occur during differentiation of heterogeneous populations of ES cells (Nakamura et al., 2018) or in response to spatially localized Wnt signals (Habib et al., 2013). However, we observed no asymmetry in Rex1 expression after 2i withdrawal. RGd2 reporter followed very similar dynamics between sister cells, indicating that exit predominantly proceeds through symmetric divisions. This inference is supported by population dynamics modeling, which describes exit as a time-dependent symmetric process, and by high correlation in sister cell division times, consistent with previous studies (Cannon et al., 2015; Nakamura et al., 2018; Waisman et al., 2019).

We also directly measured and observed the symmetry of division of GFP-positive mothers generating at least one GFP-negative daughter. In 139 cases of 141, both daughters were GFP negative, indicating that sister ES cells divide symmetrically during exit from naive pluripotency.

After cytokinesis, ES cells may remain connected by cytoplasmic bridges (Chaigne et al., 2020), potentially allowing exchange of signals between sisters and synchronizing their developmental progression, which may contribute to symmetric divisions.

In contrast, exit times between unrelated cells were uncorrelated, occurring at different times after division and within all generations. Time-translated alignment of lineages revealed a common sudden exponential decay in reporter expression. Thus, exit occurs abruptly but after a variable lag period. Importantly, transition in the peri-implantation epiblast in vivo is also asynchronous (Acampora et al., 2016; Malaguti et al., 2019; Neagu et al., 2020). Staggered exit may prevent premature exhaustion of naive epiblast (10–12 cells in the mouse blastocyst) or help establish heterogeneity in the formative epiblast. It is therefore of interest to determine the cause of variability in transition latency.

Variability in Erk signaling (Albeck et al., 2013; Aoki et al., 2013), a major driver of transition, may contribute to different lag periods between unconnected cells. Indeed, reducing negative feedback in the Erk pathway accelerates transition (Nett et al., 2018). Moreover, comparing exit rates at two plating densities or colony sizes revealed no significant differences, suggesting that exit is controlled primarily by intrinsic factors, such as phosphorylation of Erk. Quantitative variation in the protein levels of naive factors in the starting population may modulate network stability. Genetic studies have already identified multiple regulators of the transition (Betschinger et al., 2013; Leeb et al., 2014), while LIF is known to delay the transition by ∼24 h, most likely by strengthening the naive pluripotency transcription factor network (Dunn et al., 2014). It should be instructive to delineate which components affect duration of the lag phase as opposed to the exponential phase.

Methods

Cell culture

Cells were maintained in N2B27 medium supplemented with 2i (Mulas et al., 2019). Low oxygen improved cell viability during differentiation (Mazumdar et al., 2010). Full details are in the supplemental methods.

Factor withdrawal and recovery assays

RGd2 ES cells underwent 2i withdrawal and reapplication to assess clonal and population-level recovery and proliferation dynamics. The fraction of RGd2- or AP-positive cells was quantified over time to model exit kinetics. More details are in supplemental methods.

Long-term single-cell imaging

Cells were imaged at clonal density for 72 h under 2i or differentiation conditions using confocal microscopy. RGd2 fluorescence, cell morphology, and genealogy were tracked over time. Details of imaging setup and acquisition parameters are in supplemental methods.

Semiautomated lineage tracking

Manual cell tracking and segmentation were performed using Tracer (England et al., 2006), and quantitative analyses of fluorescence, cell shape, and genealogy were carried out with the oTracks software suite (Blanchard et al., 2009) (see supplemental methods).

Formal analysis

Formal analysis of the single-cell tracking data output from oTracks was performed using a custom CLASSDEF written in MATLAB 2019a. This includes comparison of generational distributions, genealogical correlations, and multi-trajectory time-translations (see supplemental methods).

Mathematical modeling and parameter inference

Model simulations and parameter inference were performed in MATLAB 2019a. Parameters were inferred by constrained optimization using FMINCON over the SSE between the number of RGd2-positive and negative cells (see supplemental methods).

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Graziano Martello (graziano.martello@unipd.it).

Materials availability

The RGd2 cell line generated in this study is available from the lead contact, Prof. Graziano Martello, upon request.

Data and code availability

All code and process data from this study are available at https://github.com/stanleystrawbridge/strawbridge_and_corridori_et_al_2025_exit_dynamics (https://doi.org/10.5281/zenodo.17523884).

Raw imagining data are available as Figshare Data: https://doi.org/10.6084/m9.figshare.30776795.

Acknowledgments

We would like to thank Wellcome-MRC Cambridge Stem Cell Institute Core Facilities especially support from Peter Humphreys with imaging and Andy Riddell with flow cytometry, Richard Adams for adapting the Tracer tool used in the semi-automated lineage tracing, Tüzer Kalkan, Amy Li, Carla Mulas, Masaki Kinoshita, and Paul Bertone for discussions and technical input, Mariam Hergnyan for help with code optimization and analysis of division symmetry, Alexander G. Fletcher for assistance with the parameter inference and, together with Sara-Jane Dunn, for helpful feedback on the mathematical modeling content.

S.E.S. was supported by a Microsoft Research PhD studentship (MRL 2013-028), a Sir Henry Wellcome Postdoctoral Fellowship (224070/Z/21/Z), and a University of Sheffield Strategic Research Fellowship in the Physics of Life and Quantitative Biology. C.C. was supported by a Microsoft Research PhD studentship (MRL 2020-035). A.S. is a Medical Research Council Professor. G.M. was supported by the Giovanni Armenise-Harvard Foundation, the Telethon Foundation (TCP13013), and an ERC Starting Grant (MetEpiStem). H.K. was supported by the Horizon 2020 Research and Innovation Program through the Bio4Comp project (732482) and the Israel Science Foundation (190/19).

Author contributions

Conceptualization, S.E.S., A.S., H.K., and G.M.; methodology, S.E.S. and G.M.; software, S.E.S., C.C., and G.B.B.; validation, S.E.S., C.C., and G.B.B.; formal analysis, S.E.S. and C.C.; investigation, S.E.S.; resources, A.S.; data curation, S.E.S. and C.C.; writing – original draft, S.E.S., A.S., H.K., and G.M.; writing – review and editing, S.E.S., C.C., A.S., H.K., and G.M.; visualization, S.E.S. and C.C.; supervision, A.S., H.K., and G.M.; project administration, A.S., H.K., and G.M.; funding acquisition, A.S., H.K., and G.M.

Declaration of interests

The authors declare no competing interests.

Published: February 12, 2026

Supplemental information

Video S1. Positive control

Example time-lapse imaging of RGd2/Gap43-mCherry embryonic stem cells cultured in 2i. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merge of the RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right.

Video S2. Exit 1

Example time-lapse imaging of RGd2/Gap43-mCherry embryonic stem cells cultured in N2B27. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merge of the RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right

Video S3. Exit 2

Example time-lapse imaging of RGd2/Gap43-mCherry embryonic stem cells cultured in N2B27. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merge of the RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right

Video S4. Negative control

Example time-lapse imaging of Gap43-mCherry embryonic stem cells cultured in 2i. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merge of the RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right

Video S5. Asymmetric 1

Time-lapse imaging of one of two asymmetric divisions observed in a RGd2/Gap43-mCherry embryonic stem cell cultured in N2B27. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merged RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right. Manual segmentation is overlaid in yellow, with cell IDs labeled in cyan. Mother cell 3taab divides to produce Rex1-negative daughter 3taaba and Rex1-positive daughter 3taabb. The progeny of 3taaba remain Rex1-negative, while the progeny of 3taabb remain Rex1-positive

Video S6. Asymmetric 2

Time-lapse imaging of one of two asymmetric divisions observed in a RGd2/Gap43-mCherry embryonic stem cell cultured in N2B27. Images were acquired every 30 min over 48 h. The brightfield image is shown on the left, and the merged RGd2 signal (green) and Gap43-mCherry signal (red) is shown on the right. Manual segmentation is overlaid in yellow, with cell IDs labeled in cyan. Mother cell 2taa divides to produce Rex1-negative daughter 2taaa and Rex1-positive daughter 2taab. The progeny of 2taaa remain Rex1-negative, while the progeny of 2taab remain Rex1-positive