Spatiotemporal sequence of mesoderm and endoderm lineage segregation during mouse gastrulation

During mammalian gastrulation, the pluripotent cells of the epiblast become lineage specified and form the three primary germ layers definitive endoderm (DE), mesoderm and (neuro-) ectoderm. Mesoderm and DE are generated at the posterior side of the embryo under the influence of elevated levels of the instructive signals of Tgfß/Nodal, Wnt and Fgf. These signals induce an epithelial-to-mesenchymal transition (EMT) of epiblast cells at the primitive streak (PS), leading to their delamination and the formation of the mesoderm and DE cell layer. The nascent mesoderm layer rapidly extends towards the anterior embryonic pole by cell migration between the epiblast and the visceral endoderm (VE) (reviewed by Arnold and Robertson, 2009; Rivera-Pérez et al., 2003). DE progenitors migrate from the epiblast together with mesoderm cells, before they eventually egress into the VE layer to constitute the DE (reviewed by Rivera-Pérez and Hadjantonakis, 2014; Viotti et al., 2014a).

Current concepts suggest that different cell fates are specified according to the time and position of cell ingression through the PS, reflecting different instructive signaling environments (Rivera-Pérez and Hadjantonakis, 2014). However, the precise morphogenetic mechanisms guiding the emergence of various cell types along the PS still remain uncertain. This is at least in part because of the lack of detailed knowledge about the precise timing and location of individual cells becoming lineage specified, and the challenge to exactly determine the signaling pathway activities during embryonic fate commitment. For example, it remains unclear what is the embryonic correlate of a suggested common mesendoderm progenitor, as described during embryonic stem cell (ESC) differentiation in vitro.

Clonal cell labeling and transplantation experiments have proposed the gross patterns and dynamics of cell specification during gastrulation, which have been represented in fate maps of the epiblast and the early germ layers (Tam and Behringer, 1997; Lawson, 1999). Accordingly, first mesoderm cells delaminate from the newly formed PS at the proximal posterior pole of the embryo and give rise to extra-embryonic mesoderm cells (ExM). These migrate proximally and anteriorly to contribute to the mesodermal components of the amnion, chorion and the yolk sac (Parameswaran and Tam, 1995; Kinder et al., 1999). Embryonic anterior mesoderm (AM) giving rise to cardiac and cranial mesoderm follows shortly after ExM (Kinder et al., 1999). As the PS elongates towards the distal embryonic pole, other mesoderm subtypes and DE are generated. The distal domain of the PS (referred to as anterior PS, APS) generates DE and axial mesoderm progenitors, giving rise to the node, notochord and prechordal plate mesoderm (Kinder et al., 2001; Lawson et al., 1991). Additional mesoderm subtypes, such as lateral plate, paraxial and intermediate mesoderm, are generated between the APS and the proximal PS (Lawson et al., 1991; Kinder et al., 1999; Tam et al., 1997; Parameswaran and Tam, 1995).

Tgfß/Nodal and Wnt signals are indispensable for gastrulation onset (Brennan et al., 2001; Conlon et al., 1994; Liu et al., 1999), and genetic experiments revealed that graded levels of Nodal and Wnt signaling instruct distinct lineage identities during gastrulation (Vincent et al., 2003; Dunn et al., 2004; reviewed by Robertson, 2014; Arkell et al., 2013). The T-box transcription factor Eomes is a transcriptional target of NODAL/SMAD2/3 signaling (Brennan et al., 2001; Teo et al., 2011; Kartikasari et al., 2013) and is crucial for the specification of all DE and AM progenitors (Arnold et al., 2008; Costello et al., 2011; Probst and Arnold, 2017). Another T-box transcription factor, Brachyury (T), is essential for the formation of posterior mesoderm starting from embryonic day (E)7.5. Thus, the specification of all types of mesoderm and endoderm relies on either of the two T-box factors Eomes or Brachyury (Tosic et al., 2019). Experiments using differentiating human ESCs showed that EOMES directly binds and regulates the expression of DE genes together with SMAD2/3 (Teo et al., 2011). Similarly, in the mouse embryo, DE specification relies on high NODAL/SMAD2/3 signaling levels (Dunn et al., 2004; Vincent et al., 2003). In contrast, in the presence of low or even absent NODAL/SMAD2/3 signals, EOMES activates transcription of key determinants for AM, including Mesp1 (Saga et al., 1999; Lescroart et al., 2014; Kitajima et al., 2000; Costello et al., 2011; van den Ameele et al., 2012).

Recently, single-cell RNA sequencing (scRNA-seq) analyses allowed for a more detailed view on the cellular composition of embryos during gastrulation stages, including the identification of previously unknown rare and transient cell types (Scialdone et al., 2016; Mohammed et al., 2017; Wen et al., 2017; Lescroart et al., 2018; Pijuan-Sala et al., 2019). Despite the insights into the molecular mechanisms of cell lineage specification, questions about the emergence of the two Eomes-dependent cell lineages, AM and DE, remain unresolved. It is still unclear whether both cell populations are generated simultaneously from a common progenitor, and when and where lineage separation occurs. Answers to these questions are required for a comprehensive view on how suggested differences in the signaling environment impact on lineage specification of mesoderm and DE identities that are generated in close proximity within the epiblast of early gastrulation stage embryos that consist of only a few hundred cells (Snow, 1977).

In this study, we used embryo imaging and genetic fate mapping approaches by novel reporter alleles, in combination with molecular characterization by scRNA-seq, to delineate the spatiotemporal patterns of Eomes-dependent lineage specification. We show that AM and DE progenitors segregate within the PS into distinct cell lineages. AM progenitors leave the PS earlier and at more proximal regions than DE, demonstrating a clear spatial and temporal separation of lineage specification. The analysis of scRNA-seq experiments suggests that AM and DE progenitors are immediately fate segregated and Eomes-positive progenitors that co-express DE and AM markers were not found. This suggests that bipotential Eomes-expressing progenitors rapidly progress into either AM or DE lineage-specified cell types preceding cell ingression at the PS.

We used our previously described Eomes^mTnG fluorescent reporter allele to observe the emergence of Eomes-dependent cell lineages during gastrulation. This reporter allele labels Eomes-expressing cells with membrane-bound Tomato (mT) and nuclear GFP (nG) (Probst et al., 2017; Fig. 1A-F). Embryos at stages shortly preceding gastrulation onset (E6.25) showed labeling within the cells of the posterior epiblast (Epi) before the formation of the PS (Fig. 1A), and reporter expression in the epiblast persisted until E7.5 (Fig. 1B-D). Importantly, all cells leaving the PS were Eomes-positive during these early gastrulation stages, as also seen by the complete reporter staining in the mesoderm layer (Fig. 1B-E,G,H; Movies 1, 2) that contains both mesoderm and DE progenitors (Viotti et al., 2014b). The maximum intensity projection (MIP) of z-stacks at E7.5 showed that the endoderm layer, which at this stage mainly consists of epiblast-derived DE cells, is composed of Eomes reporter positive cells (Fig. 1F), and only a few reporter negative cells could be detected. These most likely represent embryonic VE (EmVE) cells that express Eomes until E6.5 when it is downregulated (Fig. 1C,F; Movies 1, 2) (Nowotschin et al., 2013). As the fluorescent reporter proteins are more stable than the endogenous protein (Probst et al., 2017), we additionally performed immunofluorescence (IF) staining for EOMES at E7.25 and E7.5, showing the presence of EOMES protein in all cells of the posterior epiblast and in the mesoderm and endoderm layers (Fig. 1G,H). In conclusion, mesoderm and endoderm progenitors generated during the first day of gastrulation from E6.5 to E7.5 are exclusively descendants of Eomes-expressing cells (Fig. 1P). These constitute the progenitors of AM and DE, as also demonstrated previously by Eomes^Cre-mediated fate labeling (Costello et al., 2011).

To molecularly characterize Eomes-dependent cell types during early gastrulation, we performed scRNA-seq of cells collected from E6.75 and E7.5 embryos (Fig. 1I-O). A total of 289 handpicked cells from 14 E6.75 embryos, and 371 cells isolated by automated cell sorting from E7.5 pooled litters, were included in the scRNA-seq analysis. To identify transient progenitor populations, we clustered the cells using RaceID3 (Herman et al., 2018), an algorithm specifically developed for the identification of rare cell types within scRNA-seq data (Grün et al., 2015) (Fig. S1A,B). The tissue identities were assigned by the presence of differentially upregulated marker genes in each cluster compared with all other cells (Fig. 1I,J,L,M; Fig. S1A,B; Tables S1, S2). The heatmap representations indicate specifically expressed marker genes in different assigned cell types (Fig. 1J,M). At E7.5, RaceID identified rare cells, such as one single E7.5 primordial germ cell (PGC) (Fig. 1L,M; Table S2). The comparison of t-distributed stochastic neighbour embedding (t-SNE) maps at E6.75 and E7.5 (Fig. 1I,L) showed that, at E6.75, epiblast and PS/native mesoderm (NM) cells clustered closely to each other and only extra-embryonic tissues [extra-embryonic ectoderm (ExE) and VE] are clearly separated on the t-SNE maps (Fig. 1I). The epiblast subclusters at E6.75 (Fig. S1A) do not represent distinct trajectories towards mesoderm or endoderm progenitors. In contrast, at E7.5, separable clusters can be detected within the embryonic cell clusters, demonstrating the increase in transcriptome diversity of embryonic cell types between E6.75 and E7.5 (Fig. 1L).

At E6.75, Eomes is expressed in 209 of 289 analyzed cells (72% of all cells), showing highest expression in the PS/NM cluster (Fig. 1K; Fig. S1C) and weaker expression in the epiblast cluster, which is in agreement with the immunofluorescence staining (Fig. 1B; Movie 1). In addition, Eomes expression was found in the extra-embryonic tissues ExE and VE (Fig. 1K; Fig. S1C). At E7.5, Eomes transcripts were still present in a subset of epiblast cells, the PS, the NM, the node and the mesoderm (AM and ExM) and DE clusters (Fig. 1N; Fig. S1D). However, only 35% of cells showed RNA expression, whereas EOMES protein was still broadly detected (Fig. 1H,N). Thus, at E6.75 Eomes mRNA was expressed in relatively more cells and at higher levels than at E7.5 (Fig. 1O). In summary, scRNA-seq analysis showed that the embryonic Eomes-expressing cells at E6.75 cluster closely to each other, indicating that they are molecularly similar (Fig. 1I), and we could not identify separate clusters of lineage progenitors for AM and DE.

Mesp1 represents one of the earliest markers of mesoderm within the Eomes-positive cell population, and is a direct transcriptional target gene of EOMES (Costello et al., 2011). Lineage tracing with a Mesp1-Cre allele shows that it faithfully labels the ExM and the AM (Saga et al., 1999; Lescroart et al., 2014, 2018; Chan et al., 2013). To distinguish mesoderm from DE progenitors during the first day of germ layer formation, we generated a fluorescent Mesp1^mVenus reporter allele by inserting the sequence of membrane-bound Venus (mV) into the start codon of the Mesp1 locus followed by the Mesp1 coding sequence (Fig. 2A-C). Homozygous Mesp1^mVenus (Mesp1^mV) mice are viable and fertile, demonstrating sufficient Mesp1 expression from the reporter allele.

We analyzed the emergence of Eomes-dependent mesoderm progenitors in Mesp1^mV embryos and found Mesp1^mV reporter-expressing cells as early as E6.5 in the proximal epiblast during early PS formation (Fig. 2D,E). At this stage, the PS had not yet extended towards the distal part of the embryo and no Mesp1^mV-positive cells were present in distal portions of the epiblast (Fig. 2F). Notably, most cells leaving the early proximal PS showed Mesp1 reporter expression, identifying them as mesoderm progenitors (Fig. 2E, inset). Once Mesp1^mV-positive cells leave the PS they rapidly migrate proximally and anteriorly to their destinations of ExM and AM (Fig. 2G). Importantly, Mesp1^mV-reporter positive cells were detected in the epithelial portion of the PS (Fig. 2H inset, arrowheads), indicating that mesoderm fate specification takes place before cells delaminate from the epiblast. At E7.25, the mesodermal wings had migrated far anteriorly (Fig. 2J). Mesp1^mV-positive cells constituted the major population within the EOMES-positive mesodermal layer (Fig. 2I). Mesp1^mV-negative cells were found intermingled between Mesp1^mV-reporter expressing cells mostly towards distal regions (Fig. 2I,L, inset, arrowheads). At E7.25, nascent Mesp1^mV cells were still emerging from the proximal PS (Fig. 2K), whereas, more distally, no Mesp1^mV-expressing cells were present in the PS (Fig. 2L). Mesp1 therefore marks the earliest population of mesoderm progenitors that are continuously produced between E6.5 and E7.5 from Eomes-expressing cells. Mesp1-positive progenitors are present throughout the mesoderm layer but they are preferentially generated in the proximal domains of the PS.

Next, we investigated the spatial distribution of the Eomes-dependent cell lineages (Costello et al., 2011; Arnold et al., 2008). To simultaneously detect DE and AM progenitors, we used FOXA2 IF staining of embryos carrying the Mesp^mV reporter allele. Previous reports and our data show that Foxa2 is expressed in the VE and during gastrulation from E6.5 onwards in the epiblast, the APS/node, and its derivatives DE and axial mesoderm (AxM) (Fig. 3; Ang et al., 1993; Sasaki and Hogan, 1993; Monaghan et al., 1993; Viotti et al., 2014b). Previous lineage tracing by Cre-induced recombination and imaging by fluorescent reporters showed that Foxa2 expression faithfully labels DE progenitors (Park et al., 2008; Frank et al., 2007; Imuta et al., 2013).

The simultaneous analysis of Mesp1^mV and FOXA2 showed that AM and DE progenitors were generated at distinct levels along the PS (Fig. 3A-J). At E6.5, FOXA2-positive DE progenitors still remained within the epithelial epiblast and were located more distally in relation to proximally located Mesp1^mV-expressing AM cells, of which some have already delaminated from the PS, as observed in sagittal (Fig. 3A) or in consecutive transverse sections (Fig. 3C-F). Additionally, FOXA2 broadly marks VE cells (Fig. 3A-J). The distribution of proximally located AM and distal DE progenitors was also found at E6.75 (Fig. 3B) and E7.0 (Fig. 3G-J), at which point the most proximal sections showed Mesp1^mV expression in the PS and in cells of the mesoderm layer (Fig. 3G). More distal regions of the PS contained a mix of Mesp1^mV and FOXA2 single-positive cells (Fig. 3H,I). At the distal tip of the PS, only FOXA2-positive cells were found within the streak or the delaminated cells (Fig. 3J). Only rarely Mesp1^mV-positive cells that also showed a FOXA2 signal were found (Fig. 3H,I, arrowheads). Careful analysis of stained embryos showed that Mesp1^mV and FOXA2 double-positive cells were located mostly in the mesoderm layer and only few were found in the PS. These double-positive cells were found at an intermediate level between the proximal domain of Mesp1^mV-reporter positive cells and the distal domain of FOXA2-positive cells (Fig. 3H,I; Movies 3, 4). These cells most likely represent recently described FOXA2-positive progenitors that contribute to the cardiac ventricles and the outflow tract (Bardot et al., 2017; Ivanovitch et al., 2020 preprint). These double-positive cells could also represent a transient bipotential progenitor population for DE and ME. At both E6.5 and E7.0, we found Mesp1^mV single-positive cells in the proximal PS and mesoderm layer (Fig. 3C,G). Similarly, in distal PS only FOXA2 single-positive cells were found (Fig. 3E,J), suggesting that double-positive cells in the mid PS region represent a specific cell population that does not contribute to the majority of AM or DE cells. In summary, AM and DE progenitors are generated in mostly non-overlapping domains, and cells are already lineage separated when they are still located within the epithelial epiblast at the level of the PS.

Next, we employed scRNA-seq to analyze the segregation of Foxa2 expressing DE and Mesp1 expressing AM progenitors by their RNA expression profiles (Fig. 3K-N). At E6.75 and E7.5, Mesp1 and Foxa2 expression was found in distinct cells in the t-SNE maps (Fig. 3K-M; Fig. S2A-C). As our scRNA-seq analysis contained a limited amount of cells, we also employed a published scRNA-seq dataset that contains higher cell numbers (Pijuan-Sala et al., 2019). Here, we included and combined time points from E6.5 to E7.5 (E6.5, E6.75, E7.0, E7.25 and E7.5), and performed cell clustering using the Seurat package (Stuart et al., 2019). Similar clusters were identified between both datasets (Fig. S3A), and also in the larger dataset Mesp1 and Foxa2 expressing cells were largely non-overlapping on the Uniform Manifold Approximation and Projection (UMAP) representations (Fig. S3C,D). Interestingly, at E6.75, the Mesp1-positive cells cluster closely together on the t-SNE map, whereas Foxa2-expressing cells were found more scattered within the clusters of Epi/PS/NM (Fig. 3K-M; Fig. S2E). This suggests that Foxa2-positive cells have less homogenous expression profiles that are more similar to unspecified epiblast cells at these early time points of analysis, whereas at E7.5, Foxa2-positive cells form discrete clusters of node and DE cells (Fig. S2B,C,F).

Plotting cells for their expression of Eomes, Mesp1 and Foxa2 at E6.75 shows co-expression of Mesp1 or Foxa2 with Eomes in most cells (Fig. 3N, first and second plot). Mesp1 and Foxa2 expression was mostly exclusive, with the exception of 11 observed Mesp1/Foxa2 double-positive cells out of 324 Mesp1⁻ and/or Foxa2-positive cells (Fig. 3N, third plot; Fig. S3E). At E7.5, Eomes was rapidly downregulated and, consequently, increasing numbers of Mesp1 or Foxa2 single-positive Eomes-negative cells were found (Fig. S2D, Tables S3, S4). Mesp1-positive and Foxa2-positive cell populations remained mostly exclusive (Fig. S2D). Quantification of Mesp1/Foxa2 double-positive cells within both datasets showed that more than 95% of Mesp1⁻ or Foxa2-expressing cells were single-positive, and only between 1.7% to 5% of cells were double-positive at different time points (Fig. S3E). This analysis was confirmed by quantification of Mesp1^mV reporter-expressing and FOXA2-immunostained cells in whole sectioned embryos, which showed similar results (Fig. S3E).

In summary, the simultaneous analysis of early emerging AM and DE progenitors at E6.5 reveals the spatial separation of their sites of origin. Mesp1^mV mesoderm progenitors are generated from the proximal PS and Foxa2-expressing DE progenitors from the distal PS. ScRNA-seq analysis shows mostly exclusive lineage marker expression, suggesting that AM and DE progenitors are separated. Only a few Mesp1^mV/Foxa2 double-positive cells were detected, most likely representing a separate progenitor population.

Our analyses and published literature show that the generation of mesoderm and DE progenitors is spatially separated along the forming PS (Fig. 3; Lawson and Pedersen, 1987; Tam and Beddington, 1987; Lawson et al., 1991; Tam and Behringer, 1997). The fact that the PS elongates over time in a proximal to distal fashion suggests that mesoderm and DE progenitor specification is also temporally separated. To test the temporal sequence of lineage specification downstream of Eomes, we performed time-dependent genetic lineage tracing using a tamoxifen-inducible Eomes^CreER mouse line expressing CreER from the Eomes locus (Pimeisl et al., 2013) in combination with a Cre-inducible fluorescent reporter (Muzumdar et al., 2007; Fig. 4A). This Rosa26^mTmG reporter strain ubiquitously expresses membrane-bound Tomato that switches to membrane-bound GFP following Cre recombination. Short-term administration of tamoxifen (90 min) to dissected and morphologically staged embryos in culture was used to label Eomes-expressing cells at different developmental time points (Fig. 4A). Embryos were sorted into three groups according to the stage at the time of dissection (E6.25-E6.5, E6.75-E7.0 and E7.25-7.5), and were cultured for an additional 24 h. Whole embryos were imaged as z-stacks to evaluate whether the presence of GFP-labeled cells within the mesoderm and endoderm layers depends on the time point of Cre induction (Fig. 4A,B). Of note, in addition to the labeling of epiblast-derived cell types, this approach also marks Eomes-expressing VE cells (Fig. 4C).

In a first analysis, 53 of 55 embryos showed labeling of both endoderm and mesoderm, including embryonic mesoderm (EmM) and ExM (Fig. 4B). Interestingly, three E6.25-labeled embryos expressed GFP only within the ExM, supporting the notion that ExM is the first cell population generated in the PS (Fig. 4F) (Parameswaran and Tam, 1995; Kinder et al., 1999). As we were interested in the DE population within the labeled cells of the endoderm layer originating from Eomes-expressing cells in the posterior epiblast/PS, we needed to discriminate VE from DE cells. Therefore, we additionally stained embryos with the lectin dolichos biflorus agglutinin (DBA-lectin) that specifically labels VE cells but not DE (Fig. 4D) (Kimber, 1986). This revealed that the GFP-positive cells in the endoderm layer of 9 out of 13 (69%) E6.25-E6.5-labeled embryos were exclusively of VE origin, indicating that no DE was formed yet in most of the E6.25-E6.5 embryos (Fig. 4D). All E6.25-E6.5-labeled embryos showed GFP-positive cells in mesoderm cells (EmM 10/13) (Fig. 4B,F). The GFP positive cells in the endoderm layer of the remaining four E6.25-E6.5-labeled embryos were of mixed DE and VE origin. Thus, we confirmed the existence of a short time window before E6.5 during which Eomes-expressing cells in the posterior epiblast give rise to mesoderm (Fig. 4D,E,F). Starting from E6.5, progenitors of mesoderm and DE are both present (Fig. 3) and therefore embryos that were tamoxifen treated at E6.5-E6.75 or later showed GFP labeling both in mesoderm and DE cells (Fig. 4D,E,F). These experiments thus confirm that mesoderm and DE specification is also temporally separated so that mesoderm progenitor specification slightly precedes DE formation.

As the RaceID algorithm did not identify distinct progenitor populations for AM and DE within the Eomes-positive cells of the epiblast (Fig. 1I,L), we wanted to investigate scRNA-seq expression profiles during this lineage segregation in more detail. Thus, we analyzed the transcriptomes of Eomes-expressing cells (expression cutoff, 0.3 normalized transcript counts) from the published dataset (Pijuan-Sala et al., 2019) at time points from E6.5 to E7.5. According to our analysis (Figs 1, 3), the Eomes-positive population should include the unspecified progenitors, as well as early AM and DE progenitors. VE and ExE cells were excluded from the analysis. VarID (Grün, 2020) identified cell clusters representing the posterior epiblast and two branches consisting of the proximal PS, NM, AM and ExM, and of the distal PS, AxM, node and DE (Fig. 5A; Fig. S4A).

Eomes-positive cells were categorized into three groups of Eomes/Mesp1 double-positive cells (blue), Eomes/Foxa2 double positive cells (red), and Eomes single-positive cells (grey) (expression cut-off 0.3 normalized transcript counts for all three genes) (Fig. 5B). Differential gene expression analysis between the Eomes/Mesp1 or Eomes/Foxa2 double positive cells and Eomes single-positive cells showed that Foxa2-positive cells expressed higher endoderm and axial mesoderm marker genes (e.g. Sox17, Cer1 and Gsc), and Mesp1-positive cells showed increased expression of mesodermal/mesenchymal/EMT genes (e.g. Fn1, Lefty2, Myl7 and Snai1) (Fig. S4B; Table S5). Both Mesp1-expressing and Foxa2-expressing cells showed a downregulation of anterior epiblast markers (e.g. Pou3f1, Utf1 and Slc7a3) (Fig. S4B; Table S5), indicating the differentiation of these two cell populations towards their respective fates. Overall, more genes were differentially regulated in Mesp1-expressing cells than in Foxa2-expressing cells (126 versus 43 genes were more than twofold changed, respectively), suggesting that DE progenitors are transcriptionally more similar to cells of the epithelial epiblast or that AM progenitors are further differentiated. Foxa2-positive DE progenitors delaminate from the PS and migrate anteriorly together with the mesoderm cells before they intercalate into the VE layer to form the DE layer (Viotti et al., 2014b). To analyze whether EMT regulation of DE progenitors differs from EMT in mesoderm progenitors, we compared the expression of EMT- and migration-associated genes during gastrulation. Several EMT (including Zeb2, Twist1 and Snai1) and migration genes (e.g. Itga5 and Rasgrp3) were expressed at lower levels in Eomes single-positive and Eomes/Foxa2 double positive cells compared with Eomes/Mesp1 double-positive cells (Fig. 5D,E). This indicates that even though DE progenitors delaminate from the PS, the regulation of EMT is different from mesoderm cells at the transcriptional level.

To further analyze the rare Mesp1/Foxa2 double-positive cells found at the intermediate level between Mesp1^mV and FOXA2⁺ PS regions, we plotted Mesp1/Foxa2 double-positive cells onto the Eomes-positive population. These cells are distributed across different clusters, but are enriched in the distal PS cluster close to the branch point of DE and AM (Fig. 5C). Differential gene expression analysis shows that Eomes/Mesp1/Foxa2 triple-positive cells express both mesoderm and DE gene markers at low levels, which would fit with a role as very transient bipotential progenitors (Fig. S4C; Table S6). However, clustering of Eomes/Mesp1/Foxa2 triple-positive cells and visualization of gene signatures of AM, DE and epiblast clusters in these cells suggests that they are not a particularly short-lived population. They rather represent a continuum of states from the epiblast to lineage-specified progenitors (Fig. S4D-F). Finally, lineage tracing with a Mesp1^Cre mouse line (Saga et al., 1999) indicated that Mesp1-expressing cells generally do not give rise to DE progenitors (Fig. S4G) (Yoshida et al., 2008; MacDonald et al., 2008; Park et al., 2006; Zhang et al., 2005; Saga et al., 2000). Thus, the rare Foxa2/Mesp1 double-positive cells are unlikely bipotential progenitors for the majority of AM and DE progenitors.

Next, we investigated whether Eomes single-positive cells are differentiation biased towards either AM or DE progenitors by using FateID, which uses a random forests-based approach to assign a fate bias (on a scale of 0 to 1) to all the cells included in the analysis. It requires the input of target cells, i.e. the differentiated cells, which are used to train an iterative random forests classifier for the inference of the fate probabilities of the remaining cells included in the analysis based on the characterized transcriptome (Herman et al., 2018). To avoid artefacts originating from different developmental stages of cells, we analyzed each time point separately, with the exception of E6.5 and E6.75 cells that were combined to increase cell numbers (Fig. 6A,E; Fig. S5B). We defined early Eomes/Mesp1 and Eomes/Foxa2 double-positive cells as target cells and excluded more differentiated clusters (target cells, shown in red cells in Fig. 6C,D,G,H; Fig. S5D,E). On the respective UMAP representations of E6.5/E6.75 cells, early Mesp1-expressing cells were grouped and Foxa2-expressing cells were more scattered (as described in Fig. 3K,L; Fig. 6A), whereas at later time points Eomes/Mesp1 and Eomes/Foxa2 double-positive cells formed two distinct branches (Fig. 6E; Fig. S5B). Utf1 expression is shown to indicate the undifferentiated epiblast population (Fig. 6B,F; Fig. S5C) (Tosic et al., 2019; Galonska et al., 2014). We then calculated the fate bias probabilities (between 0 and 1) of Eomes single-positive cells for each target group, i.e. Eomes/Mesp1 or Eomes/Foxa2 double-positive cells in red (fate bias probability of 1). The target cells of the respective other fate appear blue (fate bias probability of 0). This analysis revealed that at E6.5/E6.75 Eomes single-positive cells have a similar fate bias probability towards both lineages of AM and DE (yellow cells, Fig. 6C,D), and thus are not fate biased towards either lineage. Accordingly, only very few differentially expressed genes could be found when we compared expression values in mesoderm- and endoderm-biased cells (fate bias probability cutoff was set to 0.6 for each of the respective lineages) (12 genes were more than twofold changed, Fig. S5A; Table S7). At E7.25, the Eomes single-positive cells of the undifferentiated epiblast (Fig. 6F, Utf1⁺) were biased towards Eomes/Foxa2 double-positive target cells (orange cells, Fig. 6H). The cells closer to the branching point mostly did not display a clear fate bias showing an intermediate probability for both lineages (yellow cells, Fig. 6G,H). This indicates that at E7.25 most mesoderm downstream of Eomes has already been generated and the majority of the remaining Eomes single-positive epiblast cells will give rise to DE/axial mesoderm. Analysis of differentially expressed genes between endoderm- and mesoderm-biased cells at E7.25 revealed that epiblast markers were more strongly expressed in endoderm-fated cells, whereas mesoderm-fated cells already show a pronounced mesodermal expression profile (52 genes were more than twofold changed, Fig. S5A; Table S7). At E7.0, FateID analysis showed the progressively increasing fate bias towards the Foxa2-expressing population in Eomes-expressing epiblast cells (Fig. S5D,E).

In conclusion, until E7.0, the Eomes single-positive posterior epiblast cells are not fate biased towards either lineage before the onset of Mesp1 or Foxa2 expression, and Mesp1 and Foxa2 are mostly expressed in distinct cell populations. We therefore propose that cells differentiate directly from a posterior Eomes-positive epiblast state to either mesoderm or endoderm lineages, without passing through an intermediate mesendoderm progenitor state.

To date, the understanding of lineage specification on the level of individual cells within the gastrulation stage embryo remains limited. It is still unclear how cells in close proximity acquire different fates according to local signaling environments and how these specification events are regulated in a temporal manner. In this study, we have analyzed the emergence of AM and DE populations that are both dependent on the transcription factor Eomes (Arnold et al., 2008; Costello et al., 2011; Teo et al., 2011). Eomes-expressing cells give rise to the mesoderm derivatives of the anterior embryo and the entire DE (Costello et al., 2011; Arnold et al., 2008). The Eomes-expressing population in the early PS was thought to be one of several populations leaving the PS between E6.5 and E7.5 (Robertson, 2014). However, our data indicate that early posterior epiblast cells uniformly express Eomes and all cells passing through the PS during the first day of gastrulation (E6.5 to E7.5) are positive for EOMES. Thus, between E6.5 and E7.5, only cells that will contribute to the mesoderm of the anterior embryo and the DE progenitors leave the PS, and posterior mesodermal tissues are generated after E7.5. Accordingly, spatial gene regulatory network analysis of gastrulation stage embryos indicates that separate anterior and posterior mesoderm populations exist at E7.5 (Peng et al., 2019). The FateID analysis further indicates that AM downstream of Eomes is mainly generated until E7.25. During the following stages, mesoderm formation is most likely regulated by other factors, such as the related T-box factor Brachyury and Wnt signaling (Koch et al., 2017; Wymeersch et al., 2016). The existence of distinct anterior and posterior mesoderm populations downstream of different T-box factors has been suggested previously, such as in the zebrafish (Kimelman and Griffin, 2000). However, the molecular details of this transition in the regulation of gastrulation are currently incompletely understood.

The first lineage decision following Eomes expression in the epiblast segregates AM and DE. Here, we show that the AM and DE marker genes, Mesp1 and Foxa2, are already expressed in epithelial epiblast cells at the PS. Therefore, we can place the event of lineage specification within the PS before cells migrate to form the mesoderm layer. Earlier cell tracing experiments have shown that cells are restricted in their potency after their passage through the PS (Tam et al., 1997). Our simultaneous marker analysis shows that mesoderm and DE are produced at distinct places along the PS (Fig. 7A,B). The proximal domain of the PS generates only mesoderm from the initiation of gastrulation. With a slight temporal delay, the most distal tip of the PS produces only Foxa2-positive DE and axial mesoderm progenitors. However, this study did not address the generation and distribution of the axial mesoderm that is also derived from Foxa2-expressing progenitors. At intermediate levels, the PS generates both mesoderm and endoderm progenitors, and here we also find rare Mesp1/Foxa2 double-positive cells. The separation of BRACHYURY-positive and FOXA2-positive domains in the PS of E6.5 embryos (Burtscher and Lickert, 2009) further suggests the existence of distinct areas of progenitor specification within the PS.

Experiments with ESCs indicate that cells go through a transient state in which they co-express mesoderm and DE lineage markers (often referred to as mesendoderm cells) (Tada et al., 2005; Kubo et al., 2004). This might represent a transient bipotential state that all AM and DE progenitors pass through. However, in the embryo, this co-expression of lineage markers is found in only a few cells within a restricted domain of the PS and mesoderm layer. Their scRNA expression profiles and localized presence of these cells in the PS suggest that they do not represent a bipotential progenitor for all the AM and DE. Cells within the PS are lineage specified but not lineage determined (Tam et al., 1997). In culture conditions, cell state changes might take place that will not occur in the embryo. Similarly, changes in signals delivered by the culture conditions (medium composition and/or other signals) might contribute to the appearance of cell signatures during ESC differentiation that are only rarely found in an undisturbed embryo or in limited cells. For example, DE differentiation was described to generally pass through a Brachyury-positive state during ESC differentiation (Kubo et al., 2004; D'Amour et al., 2005), whereas lineage tracing in the embryo suggests that only the hindgut is generated from Brachyury-expressing cells (Imuta et al., 2013; Perantoni et al., 2005; MacDonald et al., 2008). Therefore, ESC differentiation through a Brachyury-positive state might produce mostly hindgut DE and does not represent the specification events for all DE progenitors in the embryo.

Live embryo imaging analysis has shown that DE and mesoderm progenitors leave the PS and migrate together within the mesodermal wings before DE cells insert into the outer VE layer (Viotti et al., 2014b). Our data show that DE cells are already specified as they leave the PS. Single-cell transcriptome analysis of EMT and migration genes indicates that they are differentially regulated between Eomes-dependent AM and DE progenitors. It will be interesting to further investigate how DE cells behave within the mesoderm population and which mechanisms are used to separate them during the anterior-ward migration.

Analysis of the fate bias of Eomes-positive cells, which are not yet expressing Mesp1 or Foxa2 markers, indicates that an unbiased posterior epiblast state directly progresses to either AM or DE, arguing for fast-acting control mechanisms that independently promote AM or DE programs (Fig. 7B). However, we cannot rule out the existence of different already lineage-restricted progenitors for AM and DE within the Eomes single-positive population, as we demonstrate that these are spatially separated cell populations. Embryonic clonal lineage analyses had suggested that the common posterior epiblast progenitor for mesoderm and DE represents only a very transient cell population as clones containing both AM and DE cells were only very rarely detected by genetic or labeled lineage tracing (Tzouanacou et al., 2009; Lawson et al., 1991). Novel approaches using a combination of scRNA-seq and molecular recording of cell lineage might be able to provide information about the lineage segregation and relationship of AM and DE in more detail, which was not explored in existing datasets to date (Chan et al., 2019).

In conclusion, this study demonstrates that the generation of the Eomes-dependent lineages of AM and DE is spatiotemporally separated during early gastrulation. These cells are molecularly separated early during the differentiation process and share as last common progenitor the Eomes-expressing posterior epiblast cells.

To generate a fluorescent allele to follow Mesp1-expressing cells during gastrulation, we targeted the Mesp1 locus by homologous recombination to insert a membrane-bound Venus fluorescent protein reporter (mVenus) into the locus. Following the mVenus coding sequence, the Mesp1 coding sequence, including a 3xFLAG C-terminal tag, was inserted. These two coding sequences are linked by a T2A peptide, which leads to co-translational cleavage of the two proteins, resulting in independent mVenus and Mesp1-3xFLAG proteins (Fig. 2A). The start site of mVenusT2AMesp1-3xFLAG was inserted at the translational start site of the Mesp1 gene. The 5′ homologous arm spans from an AfeI site upstream of Mesp1 exon 1 to the Mesp1 translational start site. The mVenusT2AMesp1-3xFLAG sequence followed by a bGH PolyA signal was then inserted via the 5′ untranslated region EcoRI site and a FspA1 site within Mesp1 exon 1, thereby deleting 337 bp of the Mesp1 coding sequence (CDS). A PGK-neomycin resistance (neoR) cassette flanked by loxP sites was inserted downstream of mVenusT2AMesp1-3xFLAG. Between the mVenusT2AMesp1-3xFLAG insert and the neoR cassette, a NdeI site was introduced for screening by Southern blot analysis. The 3′ homology region spanned to an NsiI site downstream of Mesp1 exon 2 and was flanked by a pMCI-TK negative selection cassette.

Linearized targeting vector was electroporated into CCE mouse ESCs (derived from 129/Sv), and neomycin-resistant and fialuridine-insensitive ESC clones were screened by genomic Southern blot. Genomic DNA was digested with NdeI and probed with an external 3′ probe (wild type allele, 8.4 kb; mutant allele, 4.7 kb; Fig. 2B). Two independent positive clones were injected at the morula stage for chimera generation. Mesp1^mVenus mice were genotyped by PCR at 62°C annealing temperature to detect the wild-type allele (334 bp) and the knock-in allele (419 bp) using the following primers: wt forward primer 5′-CGCTTCACACCTAGGGCTCA-3′; wt reverse primer 5′-TGTGCGCATACGTAGCTTCTCC-3′; ki forward primer 5′-GCCAATGCAATCCCGAAGTCTC-3′; and ki reverse primer 5′-GCCCTTGGACACCATTGTCTTG-3′ (Fig. 2C). The neomycin cassette was removed by crossing Mesp1^Venus-positive males to females carrying the Sox2::Cre transgene (Vincent and Robertson, 2003).

Mesp1^Venus mice were backcrossed to the NMRI strain and otherwise kept as homozygotes, as they were viable and fertile, and showed no obvious phenotypic differences to wild type. Eomes^mTnG mice (Probst et al., 2017) were also kept on a NMRI background and were kept as heterozygotes. Mesp1^Cre mice (Saga et al., 1999) and R26^R-reporter mice (Soriano, 1999) were kept on a mixed background. Mice were maintained as approved by the Regierungspräsidium Freiburg (license numbers G11/31 and X19/O2F).

Embryos were dissected in PBST (PBS with 0.1% Tween 20), fixed for 1 h in 4% paraformaldehyde (PFA) in PBS at 4°C or on ice, and washed twice in PBST. At this point embryos could be kept at 4°C for at least 1 month. To perform the staining, embryos were permeabilized in 0.3% Triton X-100 in PBST at room temperature for 30 min. Embryos were blocked for 2 h at room temperature in blocking solution [1% bovine serum albumin (BSA) in PBST]. The primary antibodies (antibodies are listed in the next section) were incubated in blocking solution at 4°C overnight. Embryos were washed four times for 5 min each time in PBST at room temperature and then incubated with the secondary antibodies in blocking solutions for 3 h at room temperature. Embryos were washed two times for 5 min each time in PBST at room temperature, stained with DAPI for 30 min at room temperature and washed with PBST. Embryos were stored and imaged in PBST. Imaging was performed using a Zeiss inverted laser-scanning microscope or a Zeiss spinning disk inverted microscope in glass bottom dishes.

Embryos were fixed in 4% PFA for 1 h at 4°C in the deciduae that were opened to expose the embryo. Deciduae were washed with PBST and then processed through 15% and 30% sucrose/PBS at 4°C, and incubated for at least 1 h in embedding medium (15% sucrose/7.5% gelatin in PBS) at room temperature before embedding. Sections (6-7 μm) were cut with a Leica cryotome. To perform the immunofluorescence staining, sections were washed three times for 5 min each time in PBS and permeabilized in PBST containing 0.2% Triton X-100. Sections were blocked in blocking solution (1% BSA in PBST) for 1 h at room temperature. The primary antibodies were added in blocking solution at 4°C overnight. The slides were washed three times for 5 min each time with PBS and then incubated with the secondary antibody in blocking solution for 1 h at room temperature. After washing the antibody away with PBS, the sections were stained with DAPI in PBST for 5 min and then mounted with ProLong Diamond Antifade Mountant (Life Technologies, P36970), and imaged using an inverted Zeiss Axio Observer Z1 microscope. The following primary antibodies were used: GFP (1:1000, Abcam, ab13970), RFP (1:500, Rockland, 600-401-379), EOMES (1:300, Abcam, ab23345) and FOXA2 (1:500, Cell Signaling Technology 8186). Secondary Alexa Fluor-conjugated antibodies (Life Technologies) were used at a dilution of 1:1000. LacZ staining was performed as described previously (Nagy et al., 2003).

For E6.75 and E7.25, all Mesp1^mV, FOXA2, and Mesp1^mV/FOXA2 double-positive cells were counted in two embryos each. Entire embryos were imaged from 7 µm transversal sections, and cells were counted from every second section to avoid double counting of same cells. The outer layer (VE and DE) was not counted to avoid the inclusion of VE cells in the analysis. This might lead to a slight underrepresentation of FOXA2-positive cells within the data. In addition, the Mesp1^mV reporter expression might not entirely reflect endogenous MESP1 protein.

Embryos were isolated at E6 or E7 in prewarmed dissection medium [10% fetal calf serum (FCS) in Dulbecco's modified Eagle's Medium (DMEM)/F12 containing Glutamax) and were then placed in embryo culture medium (50% DMEM/F-12 containing GlutaMAX and 50% rat serum) containing 10 μM of 4-OH-tamoxifen (Sigma-Aldrich, H7904; dissolved at 10 mM in DMSO) for 90 min. Embryos were washed three times in dissection medium and placed individually in ibidi eight-well slides in embryo culture medium without 4-OH-tamoxifen. Embryos were cultured for 24 h in regular tissue culture incubators at 37°C with 5% CO₂. A picture was taken of each individual embryo before and after this 24 h period. After the incubation, embryos were fixed and stained with GFP and RFP antibodies as described above, and imaged using a Zeiss spinning disk inverted microscope in glass bottom dishes.

Embryos from the lineage-tracing experiments were re-stained with the biotinylated DBA-lectin (Sigma-Aldrich, L6533). Because embryos were already stained with GFP and RPF antibodies, no extra blocking step was performed. Embryos were washed in PBST and then the DBA-lectin was added at a dilution of 1:1000 in PBS with 1% BSA at 4°C overnight. The next day embryos were washed three times for 10 min each time with PBST and then incubated with Alexa Fluor-647-streptavidin (Molecular Probes, S21374; dissolved in PBS at 1 mg/ml) in PBS containing 1% BSA at a dilution of 1:500 for 1 h at room temperature. Before adding the streptavidin, the tube was briefly centrifuged. Finally, embryos were washed three times in PBST and imaged using a Zeiss inverted laser-scanning microscope in glass bottom dishes.

Embryos were dissected in pre-warmed dissection medium (10% FCS in DMEM/F12 containing GlutaMAX) and washed in pre-warmed PBS. For the E6.75 time point, the extra-embryonic part was cut off and a picture of each embryo was taken, and single embryos were transferred into the wells of a pre-warmed non-adhesive 96-well plate containing 40 μl of TrypLE Express (Gibco, 12604013). The wells were coated with FCS before adding the TrypLE. Embryos were incubated at 37°C for 10 min with pipetting up and down once during incubation and at the end to make a single-cell solution. Dissociation was stopped with 120 μl of dissection medium, and cells were centrifuged for 2 min at 200 g in the 96-well plate. The supernatant was removed and cells from one embryo were resuspended in 200 μl ice-cold PBS. For handpicking, the drop containing the cells was placed in a plastic Petri dish. Cells were picked under a Leica M165 FC binocular using ES-blastocyst injection pipettes (BioMedical Instruments, blunt, bent ID 15 μm, BA=35°) and placed into 1.2 μl lysis buffer containing polyT primer with unique cell barcode. Embryos from the E7.5 time point were cut under the chorion to include the extra-embryonic mesoderm in the analysis. The embryos were imaged and the embryos of one or two litters were pooled and processed in an FCS-coated Eppendorf tube in the same way as the E6.75 embryos. After centrifugation the cells were resuspended in 200 μl PBS and kept on ice until flow sorting.

Single-cell RNA sequencing of 576 handpicked cells (E6.75) was performed using the CEL-Seq2 protocol, whereas sequencing of 1152 flow-sorted cells (E7.5) was performed using the mCEL-Seq2 protocol (Hashimshony et al., 2016; Herman et al., 2018). Eighteen libraries with 96 cells each were sequenced per lane on Illumina HiSeq 2500 or 3000 sequencing systems (pair-end multiplexing run) at a depth of ∼200,000-250,000 reads per cell.

Paired-end reads were aligned to the transcriptome using BWA (version 0.6.2-r126) with default parameters (Li and Durbin, 2010). The transcriptome contained all gene models based on the mouse ENCODE VM9 release downloaded from the University of California Santa Cruz genome browser comprising 57,207 isoforms, with 57,114 isoforms mapping to fully annotated chromosomes (1 to 19, X, Y, M). All isoforms of the same gene were merged to a single gene locus. Furthermore, gene loci overlapping by more than 75% were merged to larger gene groups. This procedure resulted in 34,111 gene groups. The right mate of each read pair was mapped to the ensemble of all gene loci and to the set of 92 External RNA Controls Consortium spike-ins in sense direction (Baker et al., 2005). Reads mapping to multiple loci were discarded. The left read contained the barcode information: the first six bases corresponded to the unique molecular identifier (UMI), followed by six bases representing the cell specific barcode. The remainder of the left read contained a polyT stretch. For each cell barcode, the number of UMIs per transcript was counted and aggregated across all transcripts derived from the same gene locus. Based on binomial statistics, the number of observed UMIs was converted into transcript counts (Grün et al., 2014).

Clustering analysis and visualization of the data generated in this study were performed using the RaceID3 algorithm (Herman et al., 2018). The numbers of genes quantified were 19,574 and 20,108 in the E6.75 and E7.5 datasets, respectively. Cells with a total number of transcripts of less than 3000 were discarded, and count data of the remaining cells were normalized by downscaling. Cells expressing more than 2% of Kcnq1ot1, a potential marker for low-quality cells (Grün et al., 2016), were not considered for analysis. Additionally, transcripts correlating to Kcnq1ot1 with a Pearson's correlation coefficient of more than 0.65 were removed. The following parameters were used for RaceID3 analysis: mintotal=3000, minexpr=5, outminc=5 and probthr=10⁻⁴. Mitochondrial genes, ribosomal genes, as well as genes starting with ‘Gm’, were excluded from the analysis. We observed strong batch effects in the E6.75 dataset based on the day of the handpicking. Batch effects were corrected by matching mutual nearest neighbors (MNNs) as described previously (Haghverdi et al., 2018). mnnCorrect function from the scran package was used for the batch correction (Lun et al., 2016). MNN-based batch correction was also performed on the combined Eomes-positive dataset used for FateID analysis.

Processed atlas data on mouse organogenesis from Pijuan-Sala et al. (2019) were downloaded from ArrayExpress (accession number: E-MTAB-6967). The following time points and sequencing batches were analyzed: E6.5 (sequencing batch 1), E6.75 (sequencing batch 1), E7 (sequencing batches 1, 2 and 3), E7.25 (sequencing batch 2) and E7.5 (sequencing batches 1 and 2). Cells defined as doublets in the study were removed from the analysis. Integration of datasets from different time points and sequencing batches was performed using Seurat version 3 with default settings (Stuart et al., 2019). Ribosomal genes (small and large subunits), as well as genes with Gm-identifiers were excluded from the data before integration. The integrated dataset contained 45,196 cells. A focused analysis of Eomes-expressing cells was performed using VarID (Grün, 2020). From the complete dataset containing 45,196 cells, cells with a total number of transcripts of less than 6000 were discarded, and count data of the remaining cells were normalized by downscaling. Cells having normalized Eomes transcript counts of more than 0.3 were considered as Eomes-positive (14,329 cells) and further clustered and visualized using VarID with the following parameters: large=TRUE, pcaComp=100, regNB=TRUE, batch=batch, knn=50, alpha=10 and no_cores=20. Each batch contained cells from different time points and sequencing libraries. Dimensionality reduction of the datasets was performed using UMAP.

In order to investigate the transcriptional priming of single Eomes-expressing cells towards the mesodermal and DE fates, FateID (Herman et al., 2018) was run on the mouse gastrulation data (Pijuan-Sala et al., 2019) separately at the following different time points: E6.5/E6.75, E7.0 and E7.25 with cells having normalized Eomes transcript counts of more than 0.3 using default parameters. Mesp1-positive (mesoderm specified, normalized transcript count of more than 0.3) and Foxa2-positive (DE specified, normalized transcript count more than 0.3) cells were used as target cells. Extra-embryonic cells were excluded from the FateID analysis and t-distributed stochastic neighbor embedding was used for dimensional reduction and visualization of the results. Differential gene expression analysis was performed between cells biased towards one of the lineages with a fate bias probability of more than 0.5 using the diffexpnb function. UMAP coordinates from the VarID analysis were used for the visualization of the results.

Differential gene expression analysis was performed using the diffexpnb function of the RaceID3 algorithm. Differentially expressed genes between two subgroups of cells were identified in a similar way to a previously published method (Anders and Huber, 2010). First, negative binomial distributions reflecting the gene expression variability within each subgroup were inferred based on the background model for the expected transcript count variability computed by RaceID3. Using these distributions, a P-value for the observed difference in transcript counts between the two subgroups was calculated and multiple testing corrected by the Benjamini-Hochberg method.

We thank Yumiko Saga for the Mesp1^Cre mouse line; Thilo Bass for technical assistance; the staff of the Life Imaging Center of the Albert-Ludwigs-Universität Freiburg for help with microscopy; Shankar Srinivas for teaching Simone Probst mouse imaging techniques; Matthias Weiß and the CEMT team for support with animal care; and Katrin Schüle for critical reading of the manuscript.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.193789.reviewer-comments.pdf