Live imaging endogenous transcription factor dynamics reveals mechanisms of epiblast and primitive endoderm fate segregation

SUMMARY

The segregation of the epiblast (EPI) and primitive endoderm (PE) cell types in the preimplantation mouse embryo is not only a crucial decision that sets aside the precursors of the embryo proper from extraembryonic cells, respectively, but also has served as a central model to study a key concept in mammalian development: how much of developmental patterning is predetermined vs. stochastically emergent. Here, we address this question by quantitative live imaging of multiple endogenously tagged transcription factors key to this fate decision and trace their dynamics at a single-cell resolution through the formation of EPI and PE cell fates. Strikingly, we reveal an initial symmetry breaking event, the formation of a primary EPI cell lineage, and show that this is linked to the dynamics of the prior inner cell mass/trophectoderm fate decision through the expression of SOX2. This primary EPI lineage, through fibroblast growth factor (FGF) signaling, induces an increase in the transcription factor GATA6 in other inner cell mass cells, setting them on the course toward PE differentiation. Interestingly, this trajectory can switch during a defined developmental window, leading to the emergence of secondary EPI cells. Finally, we show that early expression levels of NANOG, which are seemingly stochastic, can bias whether a cell’s trajectory switches to secondary EPI or continues as PE. Our data give unique insight into how fate patterning is initiated and propagated during unperturbed embryonic development through the interplay of lineage-history-biased and stochastic cell-intrinsic molecular features, unifying previous models of EPI/PE segregation.

Graphical Abstract

In brief

Kim-Yip, Denberg, et al. visualize epiblast and primitive endoderm specification by live imaging key transcription factors. They reveal systematic and stochastic features by identifying a primary epiblast lineage biased by early SOX2 expression and a response to FGF in neighboring cells influenced by stochastic variability in NANOG levels.

INTRODUCTION

The early mammalian embryo is famed for its regulative nature, which hinges on cells gauging their surroundings through cell-cell communication and consequently making crucial fate decisions. Although this regulative nature is undisputed, there is disagreement about how initial symmetry breaking events arise during development, specifically whether these first differences are rooted in emergent stochastic variabilities of molecular regulators or whether they are biased by some feature of the cell’s history.¹ For example, the initiation of the second cell fate decision in the preimplantation embryo, when the cells of the inner cell mass (ICM) segregate into epiblast (EPI) (precursors of the fetus) and primitive endoderm (PE) cells (precursors of the yolk sac endoderm), has been the subject of such inqury.^2,3

ICM cells initially co-express two transcription factors (TFs), NANOG and GATA6, which become mutually exclusive as fates segregate, with NANOG expressed in EPI and GATA6 in PE cells (Figure 1A).^4–7 NANOG and GATA6 constitute a mutual inhibition circuit that interacts with the fibroblast growth factor (FGF)/extracellular signal-regulated kinase (ERK) signaling pathway.^2,8–14 Specifically, the ligand FGF4 originates from EPI cells and activates FGF/ERK signaling in a paracrine manner to induce PE fate, with ERK activity promoting GATA6 and inhibiting NANOG expression.^{2,7–9,11–13,15–22} FGF signaling differences are required for EPI/PE segregation; however, how initial differences arise, either in the sourcing of the ligand or in the receptivity to it, remains unknown (Figure 1A).^{8,9,18,19,23,24}

Figure 1. Live imaging and quantification of NANOG-mCherry and GATA6-eGFP dynamics from morula to blastocyst stages.

(A) Cartoon schematic of preimplantation development with gray box highlighting ICM specification into EPI/PE.

(B) Example images from time-lapse imaging of an embryo expressing NANOG-mCherry; GATA6-eGFP and H2B-miRFP720 (NG-1) from the ~13- to the 71-cell stage. A z stack was acquired every 15 min for H2B and every 1 h for each TF. Maximum intensity projections are shown. Scale bar, 10 μm. Time (t) shown in hours (h).

(C) Image processing pipeline illustrated for two sample time points from the time-lapse imaging data, segmented via Stardist-3D using the H2B-miRFP720 channel, registered, and tracked between frames.

(D) Lineage tree of NG-1. Gray lineages contribute only to the TE, black lineages have final ICM contribution. One TE lineage has been highlighted with a red dotted line, one ICM lineage has been highlighted with a blue dotted line.

(E) Example traces of min-max normalized NANOG-mCherry (magenta) and GATA6-eGFP (teal) fluorescence intensities for corresponding highlighted TE and ICM lineages. Dashed lines indicate cell divisions.

Related to Figure S1. See also Video S1.

One model, based on gene expression heterogeneity found in ICM cells, proposes that purely stochastic fluctuations in gene expression, either in FGF signaling components or other cell fate determinants that feed into FGF signaling, initiate symmetry breaking among ICM cells that are then amplified into stable fate decisions.^{18,19,21,22,25,26}

An alternative model suggests molecular asymmetries based on the positional history of cells.²⁷ Specifically, it is well appreciated that ICM cells are segregated from the outer trophectoderm (TE) and positioned to the inner compartment at different times: some are internalized as a result of asymmetric divisions at the 8-to-16-cell stage, while others are only pushed inside during the 16-to-32-cell divisions.^28–30 Two studies examined whether the duration of surface exposure would bias ICM cells toward EPI or PE fates and arrived at opposite conclusions: one study did report such bias, with shorter surface exposure favoring EPI fate and longer PE fate, while the other measured no bias.^2,3 Two subsequent studies refined the former conclusion by showing that the strength of bias depended on the number of ICM cells internalized early, which can be variable from embryo to embryo.^31,32

Proponents of the early bias model argue that longer surface exposure of cells results in prolonged engagement in the TE differentiation program and that this underlies the EPI/PE fate bias.^27,31–33 Specifically, TE-induced expression of FGF receptor 2 (Fgfr2) was suggested to link longer surface exposure to PE fate,^31,33 albeit later studies found that loss of Fgfr2 resulted in only minor changes to cell fate composition.^34,35 Alternatively, early internalization—and thus early isolation from TE fate—was shown to result in higher expression of Fgf4.³² Supporters of the stochastic mechanism on the other hand point to variable expression of several cell fate determinants^18,19,25,26 and the role of NANOG in eventually coordinating gene expression and initiating EPI fate.²⁶ In summary, it remains to be determined whether the dynamics of the first cell fate decision, and thus a somewhat systematic feature, influence EPI and PE fates or whether initial differences arise stochastically.

Critically, addressing this question requires not only tracking individual cells through development to assess their lineage history but also simultaneous visualization of key molecular determinants to follow cell fate choices in real time. To achieve this, we establish endogenous fusion reporter mouse lines for TFs NANOG and GATA6, as well as use our previous SOX2 reporter.³⁶ We simultaneously live image and track the dynamics of up to three cell fate determinants at a single-cell resolution in unperturbed embryos, which allows us to read out how cellular states arise. We discover that an EPI-biased precursor emerges first at the 16-cell stage that is marked by early expression of SOX2, a TF characteristic of the ICM. Therefore, symmetry breaking initiating EPI/PE segregation is linked to the dynamics of the ICM/TE decision. In agreement with FGF4 being sourced from EPI cells and signaling in a paracrine manner, we show that other ICM cells respond to FGF signaling by increasing their expression of GATA6 and initiating differentiation toward the PE fate. Strikingly, we also find that some lineages switch trajectory and decrease GATA6, leading to the emergence of a second wave of EPI cells. Interestingly, these switching points occur during a defined developmental window, during the time the embryo divides from 32 to 64 cells. Finally, we demonstrate that stochastic NANOG levels during the 32-cell stage can bias whether a lineage’s trajectory switches or is maintained.

In summary, we uncover the origin of symmetry breaking, which is linked to the lineage history of cells as well as a stochastic component that influences a cell’s eventual fate.

RESULTS

Live imaging endogenous NANOG and GATA6 dynamics

To simultaneously visualize NANOG and GATA6 dynamics in individual cells over the course of EPI and PE segregation, we generated a Nanog-mCherry and a Gata6-eGFP knock-in reporter mouse line. Briefly, both lines were established using 2C-HR-CRISPR engineering, targeting the coding sequence of mCherry or eGFP to the C terminus of Nanog and Gata6, respectively, creating endogenous fusion reporters (Figure S1).³⁶ Homozygous animals were bred individually for both lines, indicating no adverse effects of tagging on viability. To examine whether mCherry and eGFP faithfully report on endogenous NANOG and GATA6 expression dynamics, respectively, we performed immunofluorescence with antibodies against NANOG or GATA6 protein and the respective tags on embryos heterozygous for each reporter allele (Figure S1). We found significant positive correlation between nuclear fluorescent signal intensities between the endogenous protein and their respective tags from the morula to the late blastocyst stages. Additionally, we analyzed the lineage composition of embryos homozygous for either Nanog-mCherry or Gata6-eGFP at the late blastocyst stage and found no significant difference in the proportions of different lineages compared with matched control embryos (Figure S1). This indicates that both reporters are suitable to monitor NANOG and GATA6 expression dynamics in preimplantation embryos.

Next, we crossed mice with Nanog-mCherry; Gata6-eGFP and H2B-miRFP720³⁷ alleles to obtain preimplantation embryos that were heterozygous for all three reporters. Embryos were isolated at the 8- or 8/16-cell stages (embryonic day [E]2.5) and live imaged for ~40–48 h on a light sheet microscope until the mid-blastocyst stage (64- to ~80-cell stage), acquiring a z stack every 15 min for H2B-miRFP720 and every 30 or 60 min for each TF (Figure 1B; Video S1). We verified that embryos did not show excessive apoptosis, reached expected final cell numbers in the given time frame, and developed cell fate proportions comparable with a previous report (see cell fate classification later) (Figure S1).⁶ Time-lapse datasets of the H2B fluorescent signal were subjected to 3D nuclear instance segmentation using our previously established machine-learning-based approach (Figure 1C).³⁷ Individual cell lineages were then tracked over time with our tracking pipeline to reconstruct lineage trees (Figures 1C, 1D, and S2).³⁷ We used the nuclear instance segmentation masks to extract fluorescence intensities for NANOG-mCherry and GATA6-eGFP at each time point in the time-lapse dataset, allowing us to plot TF intensity traces for each cell lineage in the embryo (Figure 1E). Using these tools, we obtained complete lineage trees of seven embryos (embryos NANOG-mCherry/GATA6-eGFP [NG]-1 through −7, totaling 155 ICM lineages) developing from the 8/16- to the 64/80-cell stages, with simultaneous recording of two key TFs involved in EPI and PE lineage segregation.

Classification of ICM cell fates at the blastocyst stage allows backtracking of lineages to query fate segregation dynamics

To be able to query the dynamics of cell fate choices, cell fates must be first assigned at the final time point of each time-lapse dataset. TE fate was assigned based on the outer position of cells and low NANOG-mCherry and GATA6-eGFP fluorescence intensities. At this stage of development ICM cells have previously been categorized into EPI (NANOG high, GATA6 low), PE (NANOG low, GATA6 high), and double positive (DP) (expressing both NANOG and GATA6) groups.⁶

Plotting globally min-max normalized NANOG-mCherry and GATA6-eGFP intensities of individual ICM cells at the final time point of the movies failed to show obvious EPI, PE, and DP cell fate clusters (Figure 2A). Therefore, we sought to leverage additional information, specifically analyzing the relative dynamics of both NANOG-mCherry and GATA6-eGFP over time. To represent this, we designed a measure θ defined as an arctangent of the ratio of normalized NANOG-mCherry and GATA6-eGFP intensities. This measure can be intuitively understood as the angle at which a cell lies on a scatterplot of the two intensities at a given time point (Figure 2A). Graphing θ throughout a cell’s history by tracing a final cell back in the lineage tree to an initial cell in the movie yielded an ‘‘angle trace’’ (Figure 2B). Angle traces effectively track the relationship of NANOG-mCherry to GATA6-eGFP over time and have a benefit of being insensitive to optical aberrations caused by the cell’s position in 3D.

Figure 2. Live imaging NANOG-mCherry and GATA6-eGFP dynamics from morula to blastocyst stages and classification of inner cell mass fates.

(A) Scatterplot showing min-max normalized NANOG-mCherry and GATA6-eGFP fluorescence intensity levels at the mid-blastocyst final time point for NG-2. Illustration for measuring θ, the angle at which a cell lies on a scatterplot is shown.

(B) All angle traces (θ in rad) over time in hours (h) of all ICM cells in NG-2. Dashed line indicates mean ICM division time of 8/16, 16/32, or 32/64 cell divisions, with gray area indicating the range of ICM division times.

(C) For NG-2, average angle traces (θ in rad) plotted for EPI (red), DP (purple), and PE (blue) cell fates as determined by k-means clustering over time (h), corresponding shaded area indicates 1 standard deviation. Dashed line indicates mean ICM division time of 8/16, 16/32, or 32/64 cell divisions, with gray area indicating the range of ICM division times.

(D) From NG-2, example traces of min-max normalized NANOG-mCherry (magenta) and GATA6-eGFP (teal) fluorescence intensities for cells classified as EPI (top), DP (middle), and PE (bottom). Dashed lines indicate cell divisions.

(E) The accuracy of fate calls evaluated at different end points post the average 32/64 division time. The accuracy was evaluated by comparing unsupervised k-means classifications to manual ground truth classifications at the full cell-fate-sorted time point. NG-1 and NG-2 datasets were used for analysis.

(F) Lineage tree of NG-2. Gray lineages contribute only to the TE, black lineages have final ICM contribution. Final k-means clustering-assigned ICM cell fate indicated: solid red, EPI; purple outline, DP; and solid blue, PE. * indicates EPI-1 lineages.

Related to Figures S1, S2, S3, and S4.

Angle traces revealed a variety of ICM trajectories that tended to diverge at the late 32-/early 64-cell stages (Figure 2B). We chose to assign EPI, DP, or PE cell fates in an unbiased manner using k-means clustering on portions of the movies where the variance increased using the same rules across movies (detailed description of the clustering method in STAR Methods). The resulting three characteristic trajectories were as follows: angle traces that trended upward toward the end of the movie and/or ended on a high θ were classified as EPI cells, angle traces that trended downward and/or resulted in a low θ were classified as PE cells, and angle traces that hovered in between these two extremes were classified as DP cells (Figures 2C and S2). We note that we did not observe any double-negative cells until the mid-blastocyst stage (Figure S4). NANOG-mCherry and GATA6-eGFP intensities for representative cells within each cluster are shown in Figure 2D. To verify the accuracy of these fate assignments we selected two embryos (NG-1 and NG-2) that we were able to image and track beyond the mid-blastocyst stage to a time point where all but one ICM cell could be unambiguously categorized as EPI, PE, or, in a few cases, double-negative (due to late downregulation of NANOG) using a conventional thresholding method (see Figure S4 and STAR Methods for details).⁶ All ICM lineages in NG-1 and NG-2 embryos are shown in Figure S4. Although we observed moderate cell death starting at the late 64-cell stage (4 and 9 complete ICM lineages in NG-1 and NG-2 embryos, respectively), we found that these events did not alter the trajectories of the surviving ICM cells (Figures S3 and S4). Therefore, we reasoned that we could use the final fates of surviving lineages as ground truth to assess the accuracy of k-means-based classification at earlier time points. We found that k-means classification performed with 87% accuracy already at 6 h post the average 32/64 cell division and increased to 100% accuracy by 9 h (Figures 2E and S3). Based on these data, we chose to classify cell fates of all our movies at a minimum of 6 h post the average 32/64 cell division or, if possible, >9 h (given the movie was sufficiently long and no substantial cell death occurred). Such fate assignments were mapped onto the lineage trees (Figures 2F, S2, and S4), giving us access to the history of each cell from the 8- or 16-cell stage, well into the segregation of EPI and PE fates.

The EPI fate emerges earlier than other fates

Previous studies, relying either on the analysis of fixed embryos or on mathematical modeling, proposed that FGF4-producing EPI cells^13,18,19 likely emerge first and initiate differentiation of other ICM cells.^{8,9,12,14,20–22,38–40} Our live imaging data offer means to directly test this hypothesis, as well as to investigate at what developmental stage different fates emerge.

To address these questions, we calculated the percent of cell fate contributions for each lineage at different heights of the lineage tree (e.g., at the 16-, 32-, and 64-cell stages), similar to other lineage analyses performed previously.^41,42 Briefly, as shown in a toy example lineage (Figure 3A), the end-point fate assignments are back-propagated as the fraction of progeny of a cell that adopt a given cell fate. This showed that at the 16-cell stage we more often observed cells with high EPI contribution compared with DP or PE contributions (Figure 3B). Specifically, in most embryos (5/7), we could identify one (N = 3 embryos) or two cells (N = 2 embryos) at the 16-cell stage that gave rise to 75%–100% EPI progeny, which we termed primary EPI (EPI-1) lineages (Figures 2F and S2). We noted that the two other embryos (NG-5 and NG-7) had only a single 16-cell-stage lineage that resulted in only ICM cells, and, in both embryos, this lineage gave rise to 50% EPI and 50% DP fates (Figure S2). Most other EPI lineages emerged at the 32-cell and 64-cell stages (Figure 3B). Together, we termed these secondary EPI (EPI-2) lineages. By contrast, we found that DP and PE cells emerged mostly only later, at the 32- and 64-cell stages.

Figure 3. Epiblast fate emerges earlier than primitive endoderm via a primary epiblast lineage.

(A) Toy example of lineages giving rise to ICM cells. Color of the endpoint indicates sample fate assignment (solid red for EPI, purple outline for DP, and solid blue for PE). Thinner black lines indicate branches giving rise to TE cells. Branch color indicates the fraction of progeny cells that adopt a given fate (top, EPI; middle, DP; and bottom, PE). Gray dashed lines indicate the approximate sampling times. An example EPI-1 lineage can be seen in yellow at the top.

(B) Fraction of lineages at indicated sampling times (16-, 32-, and 64-cell stages) that have at least 75% bias to have progeny adopt a given fate (top, EPI; middle, DP; and bottom, PE). Data combined from N = 7 embryos (NG-1, NG-2, NG-3, NG-4, NG-5, NG-6, and NG-7) totaling n = 146 ICM lineages at the 64-cell stage.

(C) Fraction of lineages at indicated sampling times (16-, 32-, and 64-cell stages) with at least 75% bias for a given fate (top, EPI; middle, DP; and bottom, PE) from 10,000 sets of shuffled fate classifications. Empirical p-values calculated as the number of simulations where the fraction of highly biased (≥75%) lineages matched or exceeded empirical values and are indicated above a given cell stage. Boxplots show median values (middle bars) and first to third interquartile ranges (boxes), whiskers indicate 1.5× the interquartile ranges, and dots indicate outliers.

Related to Figures S2 and S4.

To estimate the statistical significance of these observations, we generated 10,000 simulations where the terminal fate assignments were shuffled between the ICM cells, keeping lineages and fate proportions conserved, and the proportion of simulations producing the same or a higher number of high-bias lineages as the empirical measurements was used as the p-value (Figure 3C). We found that at 16- and 32-cell stages, EPI deviated from randomized scenarios significantly and outperformed both DP and PE fates. Based on this, we conclude that EPI cells indeed emerge earlier, typically with one or two cells of a 16-cell-stage embryo already showing significant bias toward the EPI fate.

Early ICM identity marked by SOX2 expression biases toward primary EPI fate

Having identified the first EPI-biased precursors at the 16-cell stage, we next explored which features set these cells apart. Specifically, we investigated whether a cell-intrinsic property, such as TF dynamics, could differentiate EPI-1 cells from other ICM. We first plotted average NANOG-mCherry and GATA6-GFP fluorescence intensity traces over time for EPI-1 and other ICM lineages but observed no consistent differences at the 16-cell stage (Figures 4A, 4B, and S5).

Figure 4. Early Halo-SOX2 expression biases toward primary epiblast fate.

(A) Average normalized NANOG-mCherry intensity in EPI-1 cells (dark red) vs. other ICM cells (gray) in NG-1 (left) and NG-2 (right). Corresponding shaded area indicates 1 standard deviation. Dashed line indicates mean ICM division time of 8/16, 16/32, or 32/64 cell divisions, with gray area indicating the range of ICM division times.

(B) Average normalized GATA6-eGFP intensity in EPI-1 cells (dark red) vs. other ICM cells (gray) in NG-1 (left) and NG-2 (right). Corresponding shaded area indicates 1 standard deviation. Dashed line indicates mean ICM division time of 8/16, 16/32, or 32/64 cell divisions, with gray area indicating the range of ICM division times.

(C) Example images from time-lapse imaging of an embryo (NGS-1) expressing NANOG-mCherry, GATA6-eGFP, Halo-SOX2, and H2B-miRFP720 from the 15- to the 78-cell stage. A z stack was acquired every 15 min for H2B and every 1 h for each TF. Maximum intensity projections are shown. Scale bar, 10 um. Time (t) shown in hours (h).

(D) Lineage tree of NGS-1. Lineages are colored by min-max normalized Halo-SOX2 fluorescence intensity levels. Final k-means clustering-assigned ICM cell fates indicated: solid red, EPI; purple outline, DP; and solid blue, PE. * indicates EPI-1 lineage. Black triangles mark lineages that initiate Halo-SOX2 early, at the 16-cell stage.

(E) For embryos NGS-1, 2, 3, 4, and 5, ICM cells were manually classified into early or late Halo-SOX2 based on the time Halo-SOX2 expression was initiated (early: at the 16-cell stage; late: after the 16-cell stage). The lineage composition of cells that express Halo-SOX2 early (left) or late (center), as well as all ICM cells (right), are depicted as pie charts. Binomial tests for enrichment or depletion of EPI, DP, and PE fates relative to frequency observed in all ICM lineages yielded p = 2.37 × 10⁻³, 0.13, 4.56 × 10⁻⁵ for early Halo-SOX2 group and p = 0.074, 0.66, 2.11 × 10⁻² for late Halo-SOX2 group. Asterisks denote levels of significance.

Related to Figure S5. See also Video S2.

We then asked whether the timing of ICM fate acquisition might set these lineages apart. ICM fate is characterized by active Hippo signaling, cytoplasmic YAP localization, and the expression of SOX2.^43–45 We therefore used a previously established Halo-Sox2 reporter line³⁶ to investigate SOX2 expression dynamics in the embryo. First, we crossed mice to generate embryos with Halo-Sox2, Nanog-mCherry, and Gata6-eGFP alleles and confirmed that they showed expected lineage proportions at the late blastocyst stage (Figure S5). We then live imaged embryos with four reporters simultaneously (embryos NANOG-mCherry, GATA6-eGFP, and Halo-SOX2 (NGS)-1 through NGS-5, totaling 137 ICM traces) (Figure 4C; Video S2). As before, we constructed lineage trees and classified cell fates using k-means clustering of NANOG-mCherry/GATA6-eGFP angle traces (Figures 4D and S5). Similar to previous studies, we found that SOX2 is expressed in a few ICM cells starting at the late 16-cell stage (1 cell in 3 embryos, 2 cells in 1 embryo, and 3 cells in 1 embryo) (Figures 4D and S5).⁴³ Additionally, in agreement with our previous study, we found that not every cell internalized at the 16-cell stage initiates SOX2 expression (out of 24 internalized cells at the 16 cell stage in 5 embryos, 8 turn on SOX2).⁴⁶ To address whether the timing of SOX2 expression would bias toward EPI fate, we plotted the proportions of cell fate contributions of early SOX2 (SOX2 expression initiated at the late 16-cell stage), late SOX2 (SOX2 expression initiated at 32- or 64-cell stages) lineages, and cell fate proportions of all ICM lineages (Figure 4E). Strikingly, we found that early SOX2-expressing lineages showed a significant bias toward contributing to EPI, while at the same time PE fate contributions were significantly underrepresented. Moreover, when specifically examining EPI-1 lineages (as defined previously, by having at least 75% of progeny at the 64-cell stage contributing to EPI), we found that in three out of four embryos this lineage resulted from an early SOX2-expressing cell (NGS-1, −2, and −4) (Figures 4D and S5). We noted that in embryo NGS-3. the EPI-1 lineage resulted from the second ICM lineage to initiate SOX2 expression (at the 32-cell stage), while its early SOX2 lineage resulted in 100% DP cells (Figure S5). NGS-5 did not have a clearly identifiable EPI-1 lineage, and its early SOX2 lineage produced 75% DP and 25% EPI cells.

Together, these data demonstrate that the origins of the first EPI lineage, emerging at the 16-cell stage, are correlated with the timing of ICM identity acquisition, as marked by SOX2 expression. Therefore, we reveal a systematic symmetry breaking event that initiates EPI/PE segregation.

The initial response to FGF signaling is an increase in GATA6

Although it is unknown whether early SOX2 alone can confer EPI fate, it has been shown to be directly responsible for one hallmark of EPI identity: the expression of FGF4, as knockout or overexpression of Sox2 results in reduced or increased Fgf4 expression in the embryo, respectively and direct SOX2 binding increases Fgf4 expression in vitro.^43,47,48 We therefore set out to investigate the first effects of FGF signaling activity.

The requirement of FGF/ERK signaling for EPI/PE segregation was previously shown by modulating FGF/ERK activity during embryo development coupled to end point fate analysis.^{2,4,6,8–10,23} However, due to the limited temporal resolution afforded by fix and stain experiments, the exact timing of signaling activity and whether NANOG, GATA6, or both simultaneously respond to it remains unknown. We therefore leveraged our live imaging setup to address these questions. We visualized NANOG-mCherry and GATA6-eGFP dynamics in individual cell traces in the absence of FGF/ERK signaling by live imaging embryos from the 8/16-cell stages in the presence of a mitogen-activated protein kinase kinase (MEK) inhibitor (which inhibits ERK activity^2,6) (Figure 5A; Video S3). By the blastocyst stage, all ICM cells showed NANOG-mCherry expression and the absence of GATA6-eGFP, similar to the previously reported behavior of endogenous NANOG and GATA6, which further verified the utility of our reporters.^{2,4,6,8–10,23}

Figure 5. Live imaging NANOG-mCherry and GATA6-eGFP dynamics in the absence of FGF/ERK signaling reveals a lack of increase in GATA6-eGFP at the 32-cell stage.

(A) Example images from time-lapse imaging of an embryo expressing NANOG-mCherry; GATA6-GFP and H2B-miRFP720 from the 16- to the 64-cell stage in the presence of MEK inhibitor. A z stack was acquired every 15 min for H2B and every 1 h for each TF. Maximum intensity projections are shown. Scale bar, 10 μm. Time (t) shown in hours (h).

(B) Min-max normalized NANOG-mCherry (magenta, top) and GATA6-eGFP (teal, bottom) fluorescence intensity traces for all ICM cells in a control embryo. Dashed line indicates mean ICM division time of 8/16, 16/32, or 32/64 cell divisions, gray area indicates range of ICM division times. Duration of imaging on x axis in hours (h).

(C) Min-max normalized NANOG-mCherry (magenta, top) and GATA6-eGFP (teal, bottom) fluorescence intensity traces for all ICM cells in a MEK-inhibited embryo. Dashed line indicates mean ICM division time of 16/32, 32/64, or 64/128 cell divisions, gray area indicates range of ICM division times. Duration of imaging on x axis in hours (h).

(D) Slope sign of best-fit lines of NANOG-mCherry (top) and GATA6-eGFP (bottom) fluorescence intensity traces at the 32-cell stage for control (N = 4 embryos, n = 47 ICM cells at 32-cell stage) and MEK-inhibited embryos (N = 3 embryos, n = 37 ICM cells at 32-cell stage). Distribution of slope signs represented as stacked bar plots. Statistical significance was tested using a two sample binomial test. P = 0.10 for NANOG-mCherry and p = 2.97 × 10⁻⁵ for GATA6-eGFP.

Related to Figure S6. See also Video S3.

To address when and how FGF activity impacts the dynamics of these TFs, we contrasted NANOG-mCherry and GATA6-eGFP traces in control and MEK-inhibited embryos (Figure 5B, 5C, and S6). This revealed that GATA6-eGFP levels did not increase when FGF signaling was inhibited, as it did in the control, starting around the late 32-cell stage. To quantitatively analyze which TF responded earlier to MEK inhibition, we plotted the sign of slopes of individual NANOG-mCherry and GATA6-eGFP traces during the 32-cell stage as a proxy for the rate of change during this window. We found that although the sign of slopes of NANOG-mCherry traces was not different, the sign of slopes of GATA6-eGFP traces significantly decreased in the presence of MEK inhibition (Figures 5D and S6). In conclusion, these experiments show that an increase in GATA6 during the 32-cell stage is the first response in the embryo due to FGF signaling activity. Moreover, the increase in GATA6 is evident in nearly all ICM lineages during the 32-cell stage, except for EPI-1 lineages, where GATA6 dynamics are variable among embryos (with some embryos showing an increase, whereas others do not) (Figures 4A, 4B, and S5). Therefore, these data outline a model where the first effects of FGF signaling set non-EPI-1 ICM cells on the course of PE differentiation by inducing GATA6 expression.

Divergence in GATA6 dynamics at the 32/64-cell transition is biased by NANOG levels at the 32-cell stage

Non-EPI-1 ICM cells eventually give rise to either PE or EPI-2 fates (Figure S4). We therefore next asked when these fates diverge and how this divergence is manifested in TF dynamics. To gain better resolution of PE and EPI-2 fates, we focused on embryos NG-1 and −2, which have a total of 12 PE and 9 EPI-2 lineages at the 64-cell stage. Examining GATA6-eGFP dynamics between these groups revealed a striking pattern: while in PE lineages GATA6 either continued to increase or remained high during the 64-cell stage, in EPI-2 lineages the initial GATA6 rise switched and showed a decrease in fluorescence intensity (Figures 6A and S4). This switching point typically occurred during the window when ICM cells are dividing from the 32- to the 64-cell stage (Figure 6B). These data show that lineages can switch their GATA6 trajectories within a limited developmental window, leading to the emergence of EPI-2 fates. Next, we asked whether any feature would predict which lineages maintain or switch their GATA6 trajectories during the 32/64-cell divisions. We plotted average NANOG-mCherry and GATA6-eGFP fluorescence intensity traces over time for PE and EPI-2 lineages (Figures 6A and 6C). Although GATA6 levels did not show any appreciable difference until the switching point (Figure 6A), we found that PE lineages displayed on average lower NANOG peaks during the 32-cell stage compared with EPI-2 lineages (Figure 6C). This observation was corroborated by examining average NANOG-mCherry dynamics in five additional embryos (NG-3 to −7), where, despite the presence of a DP population representing mixed PE and EPI-2 fates, the PE-classified population also showed slightly, yet consistently, lower NANOG levels at the 32-cell stage (Figure S7). In conclusion, we find that lineages displaying lower NANOG levels during the 32-cell stage are biased toward the PE fate.

Figure 6. Segregation of PE and EPI-2 fates.

(A) Average normalized GATA6-eGFP intensity in EPI-2 cells (yellow) vs. PE cells (dark blue) in NG-1 (left) and NG-2 (right). Corresponding shaded area indicates 1 standard deviation. Dashed line indicates mean ICM division time of 8/16, 16/32, 32/64, or 64/128 cell divisions, with gray area indicating the range of ICM division times.

(B) Distribution of GATA6 switching times for EPI-2 cells in NG-1 (left) and NG-2 (right). For each EPI-2 cell, the time point at which GATA6-eGFP switches from an increasing to a decreasing trend was manually curated (see Figure S5, time points of switches marked with triangles). Histogram of these time points is plotted with average ICM 32–64 cell divisions in each respective embryo marked with dotted line and the range of ICM 32–64 cell divisions shaded in gray.

(C) Average normalized NANOG-mCherry intensity in EPI-2 cells (yellow) vs. PE cells (dark blue) in NG-1 (left) and NG-2 (right). Corresponding shaded area indicates 1 standard deviation. Dashed line indicates mean ICM division time of 8/16, 16/32, 32/64, or 64/128 cell divisions, with gray area indicating the range of ICM division times.

Related to Figures S4 and S7.

Stochastic origins of NANOG variability

Given that NANOG levels at the 32-cell stage show a cell fate bias, we next investigated the origins of NANOG-mCherry variability among ICM cells. Our imaging showed that NANOG-mCherry displayed fluctuations (Figures 4A, 6C, and S4) reminiscent of previously reported dynamics for NANOG in embryonic stem cells (ESCs), which are derivatives of EPI cells.⁴⁹ Interestingly, all ICM cells showed NANOG-mCherry fluctuations; however, PE progenitors displayed on average lower amplitudes of fluctuations at the 32-cell stage (Figures 4A, 4B, 6C, S5, and S7). To probe this further, we examined three factors that might contribute to these differences in NANOG-mCherry levels: (1) the time when NANOG-mCherry first turns on in cell lineages; (2) the length of cell cycles, as NANOG has been shown to be regulated by a cell cycle kinase in other systems⁵⁰; and (3) a positive input from another pluripotency factor, SOX2, which has also been shown in ESCs to increase NANOG levels.^51–53

NANOG-mCherry expression is initiated at the late 16-cell stage. The time in each cell’s trace when NANOG-mCherry surpasses a threshold of 20% of the maximum fluorescence intensity revealed no significant correlation to NANOG-mCherry peak height or average intensity during the 32-cell stage (Figure S7). Similarly, the length of the 32-cell-stage cell cycle only showed a weak positive correlation with peak height (R² = 0.2082) or average intensity (R² = 0.1657) (Figure S7). Finally, to assay whether SOX2 might boost NANOG expression, we generated embryos with Halo-Sox2, Nanog-mCherry, and H2B-miRFP720 reporters. Live imaging these reporters simultaneously revealed no coordination between the levels of these two TFs, suggesting that in the embryo SOX2 does not directly increase NANOG expression (Figure S7).

In summary, we did not detect any significant contribution to NANOG-mCherry variability from the timing of initial NANOG expression, cell cycle lengths, or SOX2 dynamics. These observations are compatible with a model of stochastic NANOG behavior at the 32-cell stage.

DISCUSSION

The segregation of EPI and PE fates has been an extensively studied problem. Mostly through the analysis of fixed samples, we have gained understanding of the population-level dynamics of this event, and through genetic and pharmacological perturbations, we know the key molecular players and their interactions with one another. Here, we directly visualized how cells undertake these fate decisions in unperturbed embryos by tracking and quantifying the temporal dynamics of multiple endogenously tagged transcription factors.

We show that the initial molecular symmetry breaking event is the expression of SOX2 in a few ICM cells at the late 16-cell stage, which then show a bias toward giving rise to EPI fate (Figure 7). Our data showing that each embryo typically produces one or two such cells at the 16-cell stage (which we term EPI-1) support previous observations that cells internalizing early show a strong EPI bias only when few ICM cells form during the first round of cell internalizations.^31,32

Figure 7. Schematic of proposed model for epiblast and primitive endoderm segregation.

We have recently investigated the dynamics of SOX2 induction during the segregation of ICM and TE fates in the embryo and showed that removal of YAP (which acts as a repressor of Sox2) from the nucleus is required, but not sufficient, for SOX2 expression.⁴⁶ This indicates that although the dynamics of the first cell fate decision, through YAP subcellular localization dynamics, influence the initiation of EPI/PE segregation they also do not fully define it and that additional factors influencing SOX2 expression need to be identified.

Our observation that early SOX2-expressing cells tend to give rise to EPI fates raises the question of whether SOX2 itself promotes EPI identity or whether early ICM fate is needed, of which SOX2 is a marker. The two hallmarks of EPI cells are the production of FGF4 ligand^13,18,19 and the ability to protect themselves from autocrine FGF/ERK signaling activity, which would otherwise induce PE fate.⁵⁴ SOX2 has been shown to regulate Fgf4, as Sox2 knockout embryos expressed lower levels of Fgf4⁴³ and SOX2 overexpression induced Fgf4 expression in morula-stage embryos⁴⁷ as well as in vitro.⁴⁸ Therefore, SOX2 may contribute to this aspect of EPI identity. Whether it is also sufficient to protect against autocrine FGF/ERK activity remains to be determined.

We hypothesize that early expression of FGF4 by the EPI-1 cell (s) may suppress the progression of EPI identity in other ICM cells through paracrine FGF/ERK signaling. Interestingly, in two embryos out of seven, we found two EPI-1 lineages. This suggests that there may be cases where two cells tie in the race to produce FGF4 early on.

The temporal resolution afforded by live imaging has allowed us to identify that the first response to FGF signaling is an increase in GATA6 expression starting at the 32-cell stage, with a response in NANOG trailing behind (Figure 7). This is in agreement with previous work that showed that FGF signaling indirectly downregulates NANOG, in a GATA6-dependent manner, in the embryo¹¹ but is in contrast to the mechanism reported in ESCs, which represent the mature EPI state, when exit from naive pluripotency is induced by FGF/ERK.⁵⁵ Here, ERK-mediated phosphorylation of NANOG was shown to destabilize it and target it for degradation. Therefore, the mode of ERK action during PE fate induction and at the exit of the naive pluripotent state are likely distinct.

The EPI/PE fate decision is a textbook example for producing cell types in reproducible ratios: ~40% EPI and 60% PE cells.^6,39,56 However, the mechanisms that ensure such robustness of cell type proportions are not well understood. EPI and PE precursors have been shown to exhibit varying degrees of plasticity in experimental settings that challenge fate commitment^{2,6,23,24,39,57,58}; therefore, it is possible that cell fate could fluctuate until the appropriate EPI/PE ratio is reached. However, our understanding of whether such fluctuations occur in the context of unperturbed embryonic development is limited.²⁴ Our data offer unique insight into this question by tracking the emergence of both EPI and PE fates simultaneously. Although we find that one or two EPI-1 lineages are set aside early and largely maintain their fate, we also demonstrate that other ICM cells increase their expression of GATA6, which sets the course toward PE differentiation. Strikingly, we also show that this trajectory can switch, marked by a transition to decreasing GATA6 levels, which leads to the emergence of EPI-2 lineages. Moreover, we reveal that the majority of these inflection points take place within a defined developmental window, when the ICM is dividing from the 32- to the 64-cell stage. Thus, our data show a unique lineage segregation dynamic: in the first wave, EPI-1 cells break symmetry and likely initiate FGF4 production, leading to FGF/ERK signaling activity and GATA6 increase in other ICM cells at the 32-cell stage. In a second wave during the 32/64-cell divisions, EPI-2 lineages split from PE lineages by switching their GATA6 dynamics.

We hypothesize that the switching point is likely crucial for adjusting the correct final ratios of EPI and PE cells. This is also supported by the observation that despite cell death occurring shortly after (during the late 64-cell stage), we do not observe a change in trajectories of surviving lineages, which implies that fates are rapidly committed after the window of EPI-2/PE segregation. This rapid commitment is also in line with a recent study that found no cell fate switching between E3.5 and E4.5, when EPI and PE cells physically sort into separate compartments.⁵⁶

A key question that remains is what determines whether a non-EPI-1 lineage switches or not? Our measurements show that PE cells display on average lower NANOG levels during the 32-cell stage compared with EPI-2 lineages.

Heterogenous NANOG levels are a feature of both ESCs and the early embryo.^17,25,59,60 This variability was previously shown to be independent of cell position in the embryo.²⁵ We now show that it is also not significantly correlated with time of initial expression, duration of cell cycles, and levels of SOX2, in agreement with a model for stochastic NANOG expression. Nanog heterogeneity was also observed at the transcriptional level, and studies in ESCs point to the randomness of transcriptional noise as the most likely source.^19,49,61,62

It is important to note, however, that although NANOG levels show a cell fate bias, they are not deterministic (Figures 6C and S7), suggesting additional inputs to the PE/EPI-2 choice. The observation that switching mostly occurs during a wave of cell divisions raises the intriguing possibility that access to FGF4 may be perturbed as cell numbers increase and cell positions are shuffled in the ICM. Indeed, previous work showed brief ERK activity differences among ICM cells following the 32/64-cell division that correlated with later expression of NANOG and GATA6.⁶³ Distance dependent FGF signaling from the ligand source was shown using a stem cell-based model of EPI/PE segregation, where FGF acted up to only 1–1.5 cell diameters away from the source.²⁰ It will be important to investigate the spatial properties of FGF signaling in the embryo using similar long-term live imaging, as presented in this study, and by directly visualizing the source of the ligand as well as cell-cell contact dynamics during the 32/64-cell divisions.

RESOURCE AVAILABILITY

Lead contact

Requests for further information or resources should be directed to and will be fulfilled by the lead contact, Eszter Posfai (eposfai@princeton.edu).

Materials availability

Unique materials used in this study are available from the lead contact.

Data and code availability

Codes required to recapitulate the analysis can be accessed at https://github.com/ddenberg/Cell-Fate-Specification.

All data reported in this paper will be shared by the lead contact upon request, and any additional information required to reanalyze the reported data in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mouse lines

Mouse lines used in this study are the following: CD-1(ICR) (Charles River), Gata6-GFP (generated in this study), Nanog-mCherry (generated in this study), Halo-Sox2,³⁶ and H2B-miRFP720.³⁷ For generation of Gata6-GFP and Nanog–mCherry lines, animal work was carried out at the Centre for Phenogenomics, Toronto by following the Canadian Council on Animal Care Guidelines for Use of Animals in Research and Laboratory Animal Care under protocols approved by the Centre for Phenogenomics Animal Care Committee (20–0026H). For performing experiments, mice were housed at Princeton University in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited facility following the Guide for the Care and Use of Laboratory Animals. Animal maintenance and husbandry followed the laboratory Animal Welfare Act. Princeton University’s Institutional Animal Care and Use Committee approved all animal procedures (IACUC protocol number 2133). Mice were housed in a facility with ambient temperature of 21°C, average ambient humidity of 48% and a daily light cycle of 14h.

METHOD DETAILS

2C-HR-CRISPR reagents

Cas9 mRNA and sgRNAs were synthesized as previously described.^36,65 Briefly, pCS2-Cas9 plasmid (Addgene 122948) was linearized with NotI restriction digestion (New England Biolabs, R3189L) and used as a template for in vitro transcription using a mMESSAGE mMACHINE SP6 Transcription Kit (Thermo Fisher Scientific, AM1340). sgRNA target sequences were selected around the stop codon for C-terminal targeting using the CRISPOR design tool (http://crispr.mit.edu/). Nanog C-term sgRNA: TATGAGACTTAC GCAACATCTGG (TGG is the PAM), Gata6 C-term sgRNA: GGTAGCACCAGCTCAGGCCAGGG (GGG is the PAM). sgRNA sequences were cloned into the pX330 vector (Addgene plasmid 42230) as previously described.³⁶ To synthesize sgRNAs, sgRNA coding sequences were PCR-amplified from the pX330 plasmids with primers containing the T7 promoter and were used as templates for in vitro transcription using MEGAshortscript T7 Transcription Kit (Thermo Fisher Scientific, AM1354). Cas9 mRNA and sgRNAs were purified with the RNeasy Mini Kit (Qiagen, 74104) using the cleanup protocol according to manufacturer’s instructions. Targeting vectors were cloned by PCR amplifying −2–3kb homology arms from genomic DNA, immediately up- and downstream of the STOP codon for both Nanog and Gata6. Arms were cloned into the pBluescript (Nanog) or pUC57 (Gata6) plasmid backbones, along with the coding sequences for mCherry and eGFP, respectively. Targeting plasmids were prepared using an endotoxin-free Maxi prep kit (Macherey-Nagel, NucleoBond Xtra Maxi EF, 740424.50). For sgRNA sequences, details of targeting vectors and targeting strategy used see Figure S1.

Generation of Gata6-GFP and Nanog-mCherry mouse lines

Mice were generated via 2-cell cytoplasmic injection as described previously.³⁶ Briefly, female CD-1 (ICR) (Charles River) mice between 5–8 weeks of age were each injected with 5 IU pregnant mare serum gonadotropin (PMSG, Sigma, G4527) and 5 IU human chorionic gonadotropin (hCG, Sigma, 9002–61-3), 48 hours apart. The females were then mated with males between 8–10 weeks of age. Vaginal plugs were checked the following morning. Plugs were counted as embryonic day (E) 0.5. 2-cell embryos were collected from superovulated CD-1 females on E1.5 in M2 media (Zenith Biotech). Microinjection mixes were prepared in 15 uL total nuclease-free injection buffer (10 mM Tris-HCl, pH 7.4 and 0.25 mM EDTA) with 30 ng/uL targeting plasmid (Figure S1), 100 ng/uL Cas9 mRNA and 50 ng/uL sgRNA (Figure S1). Microinjection was performed under negative capacitance using a Leica microscope and micromanipulators (Leica Microsystems). Injection pressure was provided by a FemtoJet (Eppendorf) and negative capacitance was generated with a Cyto721 intracellular amplifier (World Precision Instruments). Microinjections were performed in M2 media (Zenith Biotech) in an open glass chamber. After microinjections, embryos were immediately transferred into pseudopregnant females and gestated until birth. Founder mice were identified using genotyping PCR with primers spanning a homology arm. Primers used: Gata6-GFP gt F: GAATGTTGCCAGAACCAGGAGGAT; Gata6-GFP gt R: CTGGACTGCTGGACAATATCAGACAC; Nanog-mCherry gt F: CTATCTGGTGAACGCATCTGGAAG; Nanog-mCherry gt R: GAAGCGCATGAACTCCTTGATGATG. Founder mice were outcrossed to CD-1 mice to generate N1 mice. N1 animals were also verified by Sanger sequencing. Heterozygous mice were bred together to obtain homozygous animals. While each tagged allele was homozygous viable in a single mouse line, we were not able to obtain double Gata6-GFP, Nanog-mCherry homozygous mice. Extensive co-binding has been reported for NANOG and GATA6 in the preimplantation embryo.⁶⁶ Therefore, we speculate that when both transcription factors are tagged in a homozygous state, the tags may interfere with their binding.

Embryo isolation

For immunofluorescence experiments described in Figure S1C, CD-1 females were superovulated at 5–7 weeks old, and singly mated to Nanog-mCherry or Gata6-eGFP studs (8–40 weeks old). Briefly, females were injected with 7.5 IU PMSG (Biovendor RP1782725000) and 7.5 IU hCG (Sigma-Aldrich C1063), 47 hours apart. Embryos were collected in M2 media (CytoSpring, M2115) at the following times. Eight-cell-stage embryos were collected at E2.5 from the oviduct, early blastocysts were collected at E3.0 to E3.25 from the oviduct and uterus and late blastocysts were collected at E3.5 to E3.75 from the uterus.

For lineage composition experiments described in Figures S1E and S1F, Gata6-eGFP heterozygous females were superovulated at 5–7 weeks old and singly mated to Gata6-eGFP heterozygous males. Simultaneously, CD-1 females were superovulated at 5–7 weeks old and singly mated to CD-1 males. Nanog-mCherry homozygous females (7–12 weeks old) were naturally mated to single Nanog-mCherry homozygous males (7–12 weeks old) and plugs monitored daily. Simultaneously, CD-1 females (7–12 weeks old) were naturally mated to single CD-1 males and plugs monitored daily. For lineage composition experiments described in Figure S5, Nanog-mCherry homozygous;Gata6-eGFP heterozygous females were naturally mated to Halo-Sox2 heterozygous males and plugs monitored daily. Simultaneously, CD-1 females (7–12 weeks old) were naturally mated to single CD-1 males and plugs monitored daily. For all lineage composition experiments, embryos were collected by flushing oviducts at E2.5 in M2 media and cultured in EmbryoMax Advanced KSOM (Sigma Aldrich MR-101-D) under paraffin oil (Cooper Surgical, LGPO) for 48 hours until late blastocyst stage in an incubator at 37°C, with 5% O₂ and 6% CO₂. Controls were cultured and processed simultaneously with each experimental condition. For Nanog-mCherry; Gata6-eGFP; Halo-Sox2 triple embryos only, embryos were then briefly screened in media containing 1:1000 Janelia Fluor HaloTag Ligand JF646 MeO (100 uM stock, gift from L. Lavis)⁶⁴ via confocal microscopy for expression of all three markers before fixation.

For live imaging experiments females (7–12 weeks old) were naturally mated to studs (8–40 weeks old) and plugs monitored daily. For embryos NG-2, NG-3, NG-4, NG-5, and MEKi-3, Nanog-mCherry; Gata6-GFP males were mated to H2B-miRFP720 females. For embryos NG-1, NG-6, NG-7, MEKi-1 and MEKi-2, H2B-miRFP720 males were mated to Nanog-mCherry; Gata6-GFP females. For embryos NGS-1, NGS-2, NGS-3, NGS-4, and NGS-5, Halo-Sox2; H2B-miRFP720 males were mated to Nanog-mCherry;Gata6-GFP females. For embryos NS-1, NS-2, H2B-miRFP720; Halo-Sox2 males were mated to Nanog-mCherry females. Embryos were collected from naturally mated females by flushing oviducts at E2.5 in M2 media (CytoSpring, M2115).

Immunofluorescence staining and analysis to validate reporter lines

Embryos were fixed in 4% formaldehyde (Thermo Fisher Scientific Cat#28908, 16% stock solution diluted 1:4 in PBS) for 10 minutes and the zona removed by washing through several 10 uL drops of EmbryoMax Acidic Tyrode’s solution (Sigma-Aldrich Cat# MR-004-D). Embryos were then permeabilized using dPBS (Thermo Fisher Scientific Cat#14190136) + 0.2% Triton X-100 (Sigma Aldrich Cat# X100) for 10 minutes, blocked for at least 1 hour in blocking solution (PBS, 0.1% tween 20 (Sigma-Aldrich Cat# P7949), 2% BSA (Sigma-Aldrich Cat# A9647) and 5% normal donkey serum (Sigma-Aldrich Cat# D9663)) followed by primary antibody diluted in blocking solution overnight at 4°C. Embryos were washed 3× 15 minutes with washing solution (PBS, 0.1% tween 20 and 2% BSA), then incubated in secondary antibody diluted in blocking solution for 1 hour at room temperature. Embryos were finally washed 3× 15 minutes with washing solution, stained with Hoechst 33258 (Thermo Fisher Scientific Cat# 62249, 1:500 of 500 ng/uL working solution for a final concentration of 100 ng/mL) for 15 minutes then imaged using a spinning disk confocal microscope (W1, Nikon) in PBS. Primary antibodies used were mouse anti-Nanog (BD Biosciences Cat# 560259, RRID:AB_1645261, Clone: M55–312, Lot: 8250916, 1:200), rabbit anti-mCherry (Abcam Cat# ab167453, RRID:AB_2571870, Lot: GR3265215–3, 1:500), goat anti-Gata6 (R and D Systems Cat# AF1700, RRID:AB_2108901, Lot: KWTO417101, 1:100), chicken anti-GFP (Abcam Cat# ab13970, RRID: AB_300798, Lot: GR3190550–17, 1:500), goat anti-Sox17 (R&D Systems, AF1924, RRID:AB_355060, batch: KGA1223081, 1:100). Secondary antibodies used were donkey anti-chicken AlexaFluor488 (Jackson ImmunoResearch Labs Cat# 703–545-155, RRID: AB_2340375), donkey anti-goat AlexaFluor 568 (Thermo Fisher Scientific Cat# A-11057, RRID:AB_2534104), donkey anti-mouse AlexaFluor 488 (Thermo Fisher Scientific Cat# A-21202, RRID:AB_141607), donkey anti-rabbit AlexaFluor 555 (Thermo Fisher Scientific Cat# A-31572, RRID:AB_162543), donkey anti-mouse AlexaFluor555 (Abcam, ab150106, RRID:AB_2857373, batch: GR3220544–2, 1:500), donkey anti-goat AlexaFluor488 (Thermo Fisher Scientific, A-11055, RRID:AB_2534102, batch: 2059218, 1:500). All secondaries were used at 1:500.

For Gata6-eGFP homozygous experiment only, fixed and stained embryos were imaged singly, then immediately processed for gDNA extraction post fixation using the Red Extract-N-Amp kit (Sigma XNAT-100RXN). Briefly, individual embryos were collected at the blastocyst stage into 4 μl extraction buffer and 1 μl tissue preparation buffer. Embryos were processed at 56°C for 30 minutes to reverse crosslinking, incubated at 24°C for 5 minutes, then heat inactivated at 95°C for 5 minutes, after which 4 μl neutralization buffer was added. Embryos were immediately PCR genotyped using the primers for Gata6-eGFP. Homozygous embryos were identified by the presence of the transgenic band and absence of a WT band.

For quantification in Figures S1C and S1D, individual cells were hand segmented on the nuclear channel at the z slice where the largest diameter could be seen. Fluorescence intensity was measured in mean grey values in this area for each of the other channels (488, 555 and 641) using ImageJ.⁶⁷ For quantification in Figures S1E and S1F, cells were manually counted for presence or absence of NANOG and SOX17 and classified accordingly.

Live image acquisition

Time lapse movies were acquired on an InVi SPIM (Luxendo/Bruker). After embryo collection, embryos were washed in EmbryoMax Advanced KSOM (Sigma Aldrich MR-101-D). Embryos were imaged from 8–16 cells (E2.5) for 40–48 hours until the blastocyst stage in 75 uL of KSOM under 150 uL paraffin oil (Cooper Surgical, LGPO) in individual wells impressed into FEP foil lining a TruLive dish (Luxendo/Bruker L-TLD-96) in a humidified chamber at 37°C, with 5% O₂ and 6% CO₂. For embryos with the Halo-Sox2 reporter, media also contained 1:1000 Janelia Fluor HaloTag Ligand JF646 MeO (100 uM stock, gift from L. Lavis).⁶⁴ We acquired full 3D images with a z-step of 2.0 μm and a timestep of 15 minutes for histone and either 30 minutes or an hour for all lineage markers. For MEK inhibition experiments, embryos were imaged in media containing 0.5 uM PD0325901 (Biotechne 4192).

QUANTIFICATION AND STATISTICAL ANALYSIS

Image analysis pipeline

Acquired movies were processed through the pipeline described in Nunley et al. (2024)³⁷ with modified registration algorithms. In brief, image volumes acquired from the InVi SPIM were compressed using the Keller Lab Block file type (https://github.com/JaneliaSciComp/keller-lab-block-filetype), reducing storage space by a factor of ~0.25. Following compression, the H2B-miRFP720 channel was segmented using Stardist-3D,⁶⁸ and manual segmentation corrections were performed using a custom AnnotatorJ tool.⁶⁹ Registration algorithms were then used to correct for embryo rotation between consecutive time points. Using registered nuclear instances, lineages were tracked frame-by-frame using a semi-automated tool, based on nearest neighbor matching.³⁷ Finally, the constructed lineage tree was manually verified, and corrections to edges in the tree were made if necessary.

To correct for rotation between imaging time points, we employed a modified version of the coherent point drift (CPD)⁷⁰ frame-to-frame registration used in Nunley et al. (2024).³⁷ Conversely to the original method, each nucleus was represented by a best-fit sphere rather than a gaussian distribution as in CPD. To accommodate for this, we modified the mixture component probability in CPD to be the intersection over union (IoU) between two spheres:

$$p(s_i|s_j)=\frac{1}{N(s_i,s_j)}\frac{I(s_i,s_j)}{V_i+V_j-I(s_i,s_j)}$$ (1)

where $s_i$ is the best fit sphere for nucleus $i$, $I(s_i,s_j)$ is the volume of the intersection between spheres $s_i$ and $s_j$, $V_i$ is the volume of sphere $i$, and $N(s_i,s_j)$ is a normalization factor for the probability density function, computed numerically. The sphere centers of the frame being matched were parameterized as a rigid transformation consisting of a rotation and a translation. The transformation was found by minimizing the negative log-likelihood using the black-box minimization algorithm MA-ES.⁷¹ This modification to the CPD algorithm was performed to increase the accuracy of alignment in early stage embryos. In early embryos (~8 cell stage) multiple divisions occur at the same time point - a situation where the CPD performed poorly, getting stuck in local minima, likely due to the low number of cells.

Transcription factor (TF) intensity extraction

As the histone-based nuclear channel is used for cell segmentation, we needed to correct for possible misalignment before we could extract transcription factor intensities. Importantly, reliable segmentation of the transcription factor channel(s) was not possible, so the aforementioned point cloud registration could not be used on reporter channels alone. To do this correction, we performed the histone-to-reporter image registration by applying a rigid transformation to the transcription factor channel, minimizing the mean square deviation of the pixel intensities between it and the histone channel via MA-ES. After the histone and reporter channels were aligned, we extracted the average intensity of transcription factors using the MATLAB function ‘regionprops3’ over a given nuclear mask.

TF intensity processing

TFs were imaged at 30-minute or 1-hour intervals, corresponding to 2- or 4-frame skips, respectively, relative to the 15-minute imaging frequency nuclear marker. To accommodate these differing sampling frequencies, TF intensities were linearly interpolated between the skipped frames.

Extracted transcription factor intensities were min-max normalized per reporter and per movie, using the formula $I_{\text{norm}}\text{(}t\text{)}\text{=}\frac{I\text{(}t\text{)}-\text{min(}I\text{(}t\text{))}}{\text{max(}I\text{(}t\text{))}-\text{min(}I\text{(}t\text{))}}$, where $I\text{(}t\text{)}$ is the intensity of a given reporter (e.g. NANOG-mCherry(t)). The resulting normalized values are in the range [0, 1], where 0 was the minimum and 1 was the maximum intensity observed. Since min-max normalized NANOG-mCherry – GATA6-eGFP scatter plots failed to produce obvious clusters, we decided to analyze a measure of their relative abundance. To do so, we defined a value $\theta (t)=\text{arctan(}\frac{\text{NANOG}-{\text{mCherry}}_{\text{norm}}\text{(}t\text{)}}{\text{GATA6}-{\text{eGFP}}_{\text{norm}}\text{(}t\text{)+}ϵ}\text{)}$, where $ϵ=0.01$ to prevent division by 0. $\theta (t)$ appeared as an intuitive measurement as it represents the angle from the origin where a cell lies in a NANOG-mCherry – GATA6-eGFP scatter plot, as shown in Figure 2A. In that spirit, we decided to term the plots tracing $\theta (t)$ over lineage progression as ‘angle traces’. Angle traces for all ICM cells of embryo NG-2 can be seen in Figure 2B.

Cell fate classification

ICM and TE cells were identified by hand using their relative position within the blastocyst combined with NANOG-mCherry and GATA6-eGFP intensities. Additionally, Halo-SOX2 intensities were considered in NGS embryos.

To classify cell lineages into distinct fates, we made use of the unsupervised clustering technique, k-means.⁷² K-means partitions a set of observations into k clusters by minimizing the within-cluster sum of squares (WCSS):

$$\arg \underset{\text{S}}{\min }\sum _{i=1}^k{\sum _{\text{x}\in S_i}‖\text{x}-{\mu }_i‖}^2$$ (2)

where $\text{S}$ is the set of clusters, $S_i$ are the observations in cluster $i$, $\text{x}$ is a vector of features for an observation, and ${\mu }_i$ is the center of cluster $i$. Each observation is a time series of angle values where each time point is a separate feature in $\text{x}$.

We used the following procedure to choose which time points to use for clustering. First, since movies start and end at different time points, we rescaled and aligned them by the average 16-to-32 and 32-to-64 cell division time to produce a pseudo-time scale with those respective times being 0 and 1. Next, since clustering can be worsened by including pre-divergent data, we decided to cluster only on the portions of angle traces past a cutoff value t* where specification occurs. As can be seen in Figure 2B, the angle traces start diverging into a wider distribution around the 32-to-64 cell transition. Because of this increase we expect the period at which specification is beginning to occur, thus increasing variance of angle traces, to be after the 32-to-64 cell division, or the pseudo-time cutoff of t*=1. Increasing the t* restricts clustering to timepoints where more specification has occurred, thus lowering noisy pre-divergence information, but limits the amount of data to cluster on. We chose an alternative t* = 1.2 which was the latest cutoff that still allowed clustering on our shortest movie.

To evaluate the accuracy of our fate classification, two movies (NG-1 and NG-2) were tracked past the 64-cell stage until we observed a clear delineation between NANOG-mCherry positive nuclei and GATA6-eGFP positive nuclei. At such a time point, EPI and PE cell fates were assigned based on thresholds, as done previously.⁶ Using short variants of NG-1 and NG-2 trees, cut at time points similar to the length of the remaining NG-3, −4, −5, −6, and −7 tracks, we assigned cell fates using our clustering method. The accuracy was computed by comparing assigned fates at the early time point to the ground truth fates determined at the end of each long track (Figure 2E and Figure S4). Comparing accuracy between t* of 1 and 1.2 revealed a higher accuracy for t*=1.2 (Figure S2), which was then chosen as the default.

Comparison of TF dynamics in WT and MEK-inhibited embryos

NANOG-mCherry and GATA6-eGFP traces were cropped to between the 16/32 and 32/64 division based on the particular lineage’s division times. A line was fit to each cropped trace and the signs of the slope were saved for each cell. All untreated control (N=4) and MEK-inhibitor treated (N=3) embryos were pooled together for comparison and the significance was assessed via a two-sample binomial test: p=0.1 for NANOG-mCherry and p=2.97×10⁻⁵ for GATA6-eGFP. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001

Lineage tree fate backpropagation

Given cell fates at the final time point and the lineage tree, we determined the final fate distribution for each lineage over time using a depth first search. First, a directed graph was created from the lineage tree in which the undirected edges are converted to directed edges which point in reverse frame order. Then, starting with a cell at the final time point of each lineage, a depth first search is performed, returning a lineage trace ending at the first frame of the movie. The fate of the given cell for which the reverse trace was generated is appended to a list of progeny fates for each node in the reverse trace. After iterating through each cell at the final time point, the relative frequency of each fate in the progeny was computed for each node in the lineage tree from the stored list of back-propagated final cell fates. In the case of cell death, a separate death fate was back-propagated.

Average cell fate TF dynamics

At each time point, the mean and variance of a $\text{TF}$ for each cell fate were weighted by the relative frequencies of the back-propagated final cell fates (see Lineage tree fate backpropagation). The mean at a time point, ${\mu }_j$, for each fate $\text{j}\in \left\{\text{EPI,DP,PE,TE}\right\}$ was computed as:

$${\mu }_j=\frac{\sum _{i=1}^n{w}_{ij}{\text{TF}}_i}{\sum _{i=1}^n{w}_{ij}}$$ (3)

where $n$ is the number of cells at a time point, $w_{ij}$ are the relative frequencies of each cell fate for cell $i$ and fate $j$, and ${\text{TF}}_i$ is the normalized intensity of the $\text{TF}$ in cell $i$. The unbiased sample variance at a time point, ${\sigma }_j^2$, for each fate $j$ was computed as:

$${\sigma }_j^2=\frac{\sum _{i=1}^n{w}_{ij}{\text{(TF}}_i-{\mu }_j{)}^2}{V_{1j}-(V_{2j}/V_{1j})}$$ (4)

where $V_{1j}={\sum }_{i=1}^n{w}_{ij}$ and $V_{2j}={\sum }_{i=1}^n{w}_{ij}^2$.

Lineage fate bias

Back-propagated trees for wild-type embryos (NG-1, NG-2, NG-3, NG-4, NG-5, NG-6, and NG-7) were examined for bias of a given lineage to produce progeny of the same fate. To do so, sampling points at the end of the 16-, 32-, and 64-cell stages, within each lineage, were chosen. At the chosen sampling points, the fraction of progeny that become EPI, DP or PE were computed as a metric of bias for a given lineage (see schematic in Figure 3A). Lineages with ≥75% fraction of final progeny of a given fate were considered highly biased for that fate and counted for each cell stage. In order to ascertain statistical power of these biases, lineage trees of the 3 embryos were bootstrapped 10,000 times to produce a randomized distribution. Randomized trees were simulated by shuffling final cell fates of ICM cells within a tree such that the same cells were categorized as ICM and the number of EPI, DP and PE labels remained the same. Empirical p-values were calculated as the number of simulations where the fraction of highly biased (≥75%) lineages matched or exceeded empirical values.

SOX2 initiation and cell fate bias

ICM lineages in the NGS embryos were categorized manually as ‘‘early’’ or ‘‘late’’ referring to the presence or absence of Halo-SOX2 signal at the 16-cell stage, respectively. A binomial test was used to compare the fraction of EPI, DP, and PE cell fates in the early and late lineage groups relative to the expected fraction of EPI, DP, and PE cell fates in all ICM lineages (Figure 4E). * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001

Analysis of NANOG stochasticity at the 32-cell stage

To analyze stochasticity of NANOG-mCherry, embryos both containing those traces throughout the 32-cell stage and imaged early enough to observe the onset of NANOG-mCherry expression were pooled, and cells manually assigned as ‘ICM’ were combined into a dataset consisting of N=11 embryos (NG-1, NG-2, NG-3, NG-5, NG-7, NGS-1, NGS-2, NGS-4, NGS-5, NS-1 and NS-2) totaling n=248 ICM traces. Four key metrics were extracted and correlated for each lineage (Figures S7A and S7B): timing of NANOG-mCherry turning on, NANOG-mCherry average during the 32-cell stage, NANOG-mCherry peak at the 32-cell stage and cell cycle (or duration of the 32-cell stage). Timing of NANOG-mCherry turning on was defined as the first time point where the min-max normalized intensity crossed the threshold of 0.2, further normalized to the 32-cell stage such as the 32-cell division is t=0 and the 64-cell division is t=1. Note that these t=0 and 1 were specific to the given lineage and not the mean times of all appropriate divisions.

To probe the predictiveness of Halo-SOX2 traces on subsequent dynamics of NANOG-mCherry, we selected from the pooled embryos those that have both NANOG-mCherry and Halo-SOX2 data available, N=6 embryos (NGS-1, NGS-2, NGS-4, NGS-5, NS-1 and NS-2) totaling n=140 ICM traces. For a given cell, the 32-cell stage NANOG-mCherry trace was cropped out and correlated to the Halo-SOX2 trace or that of the nuclear marker (H2B-miRFP720) at the same frames. Then, the Halo-SOX2 and H2B-miRFP720 traces were re-extracted but shifted by 1 time point before (or delay of τ=1), and the correlation was recomputed. This operation was repeated for as far as the Halo-SOX2/H2B-miRFP720 traces were available. This generated a list of time-delays τ and correlation values that Halo-SOX2 or H2B-miRFP720 curves could produce (computed as described above). The same analysis was performed for all 90 lineages and the correlation values were pooled into distributions based on the time-delay τ. Average and standard deviations of these distributions as a function of τ for both correlation coefficient R and the coefficient of determination R² are shown in Figure S7C.

Limitations of datasets

We observed a global decrease in GATA6-eGFP fluorescence when embryos were incubated in HaloTag Ligand JF646 MeO. For the embryos in this study this phenomenon affects NGS-1, NGS-2, NGS-3, NGS-4 and NGS-5. While we were still able to analyze cell fates using min-max normalization, we note that there is a decreased dynamic range for GATA6-eGFP in these embryos (Potential reasons may include a reduction of GFP fluorescence in the presence of Halo dye or a technical issue with the 488 laser during the timeframe these data were acquired).

In the NG-3 dataset, the embryo temporarily moved slightly out of focus during frames 55–70 of acquisition (shortly after the 16/32 cell divisions). In the NG-4 dataset, the embryo temporarily orients with the ICM closer to the objective during frames 80–120 of acquisition (mid to late 32-cell-stage). For both datasets, this is noticeable in both transcription factor intensities (Figure S7) and the histone intensities, however, this does not affect any of the conclusions made in the manuscript.

Criteria to include/exclude imaging datasets for analysis: We included embryos that developed to the mid-blastocyst stage (64 or more cells) during the ~40 hour imaging period and did not show extensive cell death. Specifically, we considered embryos that had a maximum of 1 or 2 cell deaths in the ICM. We excluded embryos in which a number of ICM lineages did not divide to the 64-cell stage. Datasets were excluded on technical grounds if excessive rotation was found (meaning the embryo rotated while a Z stack was acquired), which our rigid alignment method was unable to correct. Occasionally, embryos had poor nuclear segmentation outputs which would have required extensive manual correction, and therefore these datasets were set aside. Overall, we estimate that we included ~30% of imaged embryos of the correct genotype in our analyses.

Embryos imaged beyond the mid blastocyst stage showed a wave of ICM cell death starting at the late 64-cell stage, consistent with the timing of a wave of cell death reported previously.^24,73,74 Cell death events made tracking challenging, and we only included datasets where we were highly confident in the accuracy of lineage tracks.

Image processing for figure and video presentation

For sample microscope images in Figures 1, 4, and 5, images are shown as max intensity projections cropped to the embryo diameter at that timepoint. For visualization purposes, brightness and contrast levels were adjusted using ImageJ⁶⁷ to best show timepoints without saturating pixels. All images for a single time series and a single channel were treated identically and compressed into jpg. All quantification and analysis described later were done on raw data. For sample microscope images in Figure S1, images are shown as single z slices cropped to the embryo diameter in that slice. Brightness and contrast levels were adjusted using ImageJ⁶⁷ to best show each slice without saturating pixels and compressed into jpg. All quantification and analysis described previously were done on raw data. For Videos S1–S3, movies are shown as max intensity projections cropped to the embryo diameter at its largest during the movie. Each channel was manipulated using the ‘‘Enhance Contrast’’ feature on ImageJ to best show intensity without saturating pixels. For Videos S1 and S3, GATA6-eGFP and NANOG-mCherry channels were then merged using Merge Channels on ImageJ.⁶⁷ Transcription factor and any composite channels were multiplied by 4 to match the timing of the histone channel and individual videos stitched together using ImageJ MultiStack Montage (https://github.com/BIOP/ijp-multi-stack-montage). Videos were compressed into jpg. Labels, scalebar and timestamps were added using ImageJ.⁶⁷

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.cub.2025.07.031.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-Nanog	BD Biosciences	Cat# 560259; RRID: AB_1645261
Rabbit polyclonal anti-mCherry	Abcam	Cat# ab167453; RRID: AB_2571870
Goat polyclonal anti-Gata6	R&D Systems	Cat# AF1700; RRID: AB_2108901
Chicken polyclonal anti-eGFP	Abcam	Cat# ab13970; RRID: AB_300798
Goat polyclonal anti-Sox17	R&D systems	Cat# AF1924; RRID: AB_355060
Donkey AF 488 anti-chicken	Jackson ImmunoResearch Labs	Cat# 703–545-155; RRID: AB_2340375
Donkey AF 568 anti-goat	Thermo Fisher Scientific	Cat# A-11057; RRID: AB_2534104
Donkey AF488 anti-mouse	Thermo Fisher Scientific	Cat# A-21202; RRID: AB_141607
Donkey AF 555 anti-rabbit	Thermo Fisher Scientific	Cat# A-31572; RRID: AB_162543
Donkey AF 555 anti-mouse	Abcam	Cat# ab150106; RRID: AB_2857373
Donkey AF 488 anti-goat	Thermo Fisher Scientific	Cat# A-11055; RRID: AB_2534102
Chemicals, peptides, and recombinant proteins
Pregnant mare serum gonadotropin (PMSG)	Biovendor	Cat# RP1782725000
Human chorionic gonadotropin	Sigma	Cat# C1063
NotI Enzyme	New England Biolabs	Cat# R3189L
M2	CytoSpring	Cat# M2115
EmbryoMax Advanced KSOM	Sigma Aldrich	Cat# MR-101-D
Paraffin Oil	Cooper Surgical	Cat# LGPO-500
Formaldehyde	Thermo Fisher Scientific	Cat# 28908
EmbryoMax Acidic tyrode’s solution	Sigma	Cat# MR-004-D
Triton X-100	Sigma	Cat# X100
Hoechst 33258	Thermo Fisher Scientific	Cat# 62249
Normal donkey serum	Sigma	Cat# D9663
Janelia Fluor HaloTag Ligand JF646 MeO	Gift from Luke Lavis64	N/A
PD0325901	Biotechne	Cat# 4192
Critical commercial assays
mMESSAGE mMACHINE SP6 Transcription Kit	Thermo Fisher Scientific	Cat# AM1340
MEGAshortscript T7 Transcription Kit	Thermo Fisher Scientific	Cat# AM1354
RNeasy Mini Kit	Qiagen	Cat# 74104
NucleoBond Xtra Maxi EF	Macherey-Nagel	Cat# 740424.50
Red Extract-N-Amp kit	Sigma	Cat# XNAT-100RXN
Experimental models: Organisms/strains
Mouse: CD-1 IGS	Charles River Laboratory	Strain #022
Mouse: Gata6-eGFP	This paper	N/A
Mouse: Nanog-mCherry	This paper	N/A
Mouse: Halo-Sox2	Gu et al.36	N/A
Mouse: H2B-miRFP720	Nunley et al.37	N/A
Oligonucleotides
Primer: Gata6-GFP forward: GAATGTTGCCAGAACCAGGAGGAT	This paper	N/A
Primer: Gata6-GFP reverse: CTGGACTGCTGGACAATATCAGACAC	This paper	N/A
Primer: Nanog-mCherry forward: CTATCTGGTGAACGCATCTGGAAG	This paper	N/A
Primer: Nanog-mCherry reverse: GAAGCGCATGAACTCCTTGATGATG	This paper	N/A
Recombinant DNA
Plasmid: pCS2+ Cas9	Gu et al.36,65	Addgene Plasmid #122948
sgRNA: Nanog C-term knock-in: TATGAGACTTACGCAACATCTGG	This paper	N/A
sgRNA: Gata6 C-term knock-in: GGTAGCACCAGCTCAGGCCAGGG	This paper	N/A
Plasmid: pX330-U6-Chimeric_BB0CBh-hSpCas9	Gu et al.36,65	Addgene Plasmid #42230
Plasmid: pBluescript-Nanog-mCherry	This paper	N/A
Plasmid: pUC57-Gata6-eGFP	This paper	N/A
Software and algorithms
FIJI	https://imagej.net/software/fiji/downloads	RRID: SCR_002285
BlastoSPIM 2.0-trained-Stardist-3D	Nunley et al.37	N/A
MATLAB 2024b	Mathworks	RRID: SCR_001622
Python (v3.9)	https://www.python.org/	RRID: SCR_008394
Keller Lab Block file type	https://github.com/JaneliaSciComp/keller-lab-block-filetype	N/A
Cell Fate Specification analysis	This paper	https://github.com/ddenberg/Cell-Fate-Specification
Other
TruLive dish	Luxendo/Bruker	Cat# L-TLD-96

Highlights.

• Live imaging endogenously tagged transcription factors in preimplantation embryos

• Early SOX2 expression biases toward primary epiblast lineage(s)

• FGF signaling initiates primitive endoderm differentiation in surrounding cells

• Decision to maintain trajectory or switch to secondary epiblast influenced by NANOG