An epiblast stem cell-derived multipotent progenitor population for axial extension

The anteroposterior axis of a vertebrate can be subdivided into three anatomically distinct regions: the head, the trunk and the tail. The trunk starts at the end of the hindbrain, runs to the anus and comprises derivatives of the mesoderm and the ectoderm, such as the thoracic cage, muscles, kidneys and spinal cord. The thoracic tract has different origins in different organisms: in anamniotes, e.g. fish and frogs, it is inferred to arise during gastrulation from a pool of pre-existing cells within the ectoderm, while in amniotes, e.g. chickens and mice, it is derived from the expansion of the caudal epiblast (CE), a proliferative region located at the caudal end of the embryo, where the primitive streak persists, that acts as a source for paraxial, intermediate and lateral plate mesoderm, as well as for the spinal cord (Henrique et al., 2015; Stern, 2005; Steventon and Martinez Arias, 2017; Sweetman et al., 2008; Wilson et al., 2009). Lineage-tracing studies have shown that the CE harbours a population of bipotential progenitors located behind the node, at the node streak border (NSB), and extending laterally into the caudal lateral epiblast (CLE), that give rise to neural and mesodermal precursors. These cells, called neural mesodermal progenitors (NMPs), are often characterized by simultaneous expression of T (brachyury, also known as Bra) and Sox2 (Cambray and Wilson, 2007; Wymeersch et al., 2016), and are capable of limited self-renewal (Cambray and Wilson, 2002; McGrew et al., 2008; Tzouanacou et al., 2009).

Recently, a few studies have claimed the generation of NMP-like cells in adherent cultures of mouse and human embryonic pluripotent stem cells (PSCs) (Denham et al., 2015; Gouti et al., 2014; Lippmann et al., 2015; Turner et al., 2014). In these studies, embryonic stem cells (ESCs) are coaxed into a transient T and Sox2 co-expressing state that, depending on the culture conditions, can be differentiated into either paraxial mesoderm (PXM) or spinal cord progenitors and their derivatives. However, there is no evidence that these NMP-like cells are propagated in vitro as they are in the embryo (Tsakiridis and Wilson, 2015). Furthermore, co-expression of T and Sox2 might not be a unique characteristic of NMPs as it is also a signature of EpiSCs (Kojima et al., 2014), which are pluripotent, and this does not imply that EpiSCs are NMPs. Although other markers have been used to refine the molecular identity of NMPs in vitro, e.g. Nkx1-2, Cdx2, Cdh1 and Oct4, these are also expressed in the epiblast and in the primitive streak during gastrulation (see Table S1 and Fig. S1), emphasizing the notion that these gene expression signatures are not uniquely associated with NMPs. Altogether, these observations raise questions about the identity of the T/Sox2 co-expressing cells derived from ESCs and about the signature of the NMPs.

Here, we show that T/Sox2 co-expressing cells derived from ESC- and EpiSC-based differentiation protocols display differences at the level of gene expression and represent collections of different developmental stages of the transition between naïve, primed pluripotency and neuro-mesodermal fate choices. Furthermore, we find that, in adherent culture, all available protocols generate a multipotent population where, in addition to an NMP signature, we also find signatures for lateral plate and intermediate mesoderm (LPM and IM), as well as the allantois. We report on a new protocol, based on EpiSCs, that sequentially generates, at a high frequency, the multipotent population and an NMP-like population with many of the attributes of the embryonic NMPs. In particular, these cells can be maintained in vitro for a limited period of time and contribute to posterior neural and mesodermal regions of the embryonic body in a xenotransplantation assay. Our study leads us to propose that, in vitro and in vivo, NMPs are derived from a multipotent population that emerges in the epiblast at the end of gastrulation and gives rise not only to the elements of the spinal cord and PXM but also to all elements of the trunk mesoderm.

Several protocols allow the differentiation of ESCs into an NMP-like population, defined as cells that co-express T and Sox2 that can be further differentiated into neural and mesodermal progenitors (summarized by Henrique et al., 2015). However, it is not clear whether these NMP-like cells derived through different protocols are similar to each other and, importantly, how each relates to the NMPs in the embryo. To begin to answer these questions, we compared NMP-like cells obtained from three different protocols: ES-NMPs (Turner et al., 2014) and ES-NMPFs (Gouti et al., 2014), derived from ESCs, as well as Epi-NMPs, derived from a new protocol that we have developed from EpiSCs (Fig. 1A,B; see Materials and Methods). All protocols yield cells co-expressing T and Sox2 at the level of both mRNA and protein (Fig. 1C,D and Fig. S2A), but differ in the numbers of cells with this signature as well as in the levels and degree of correlated expression of the two genes (Fig. S2). At the protein level, all the conditions exhibit high percentage of cells co-expressing Sox2 and T (Fig. 1D) and a significant positive correlation between the two genes is observed only in the ES-NMP condition, whereas there is a negative correlation in the EpiSC population. Across all the conditions, Sox2 shows the same degree of variability both in the protein and mRNA levels; T, however, exhibits greater variability at the protein level in the EpiSCs and EpiSC-derived populations in comparison with the ESC-derived NMP-like populations, which is not the case at the mRNA level (Fig. 1D and Fig. S2B). The Epi-NMP population has the highest number of T/Sox2-positive cells with low variance in comparison with the other conditions (Fig. S2B).

To characterize the different NMP-like populations further, we investigated the expression of a total of 27 genes associated with the epiblast, the CE and the NSB/CLE region, where NMPs are thought to reside, as well as of genes associated with neural and mesodermal differentiation between stages E7.0 and E8.5 (see Table S1 and Fig. S1 for the criteria we followed to select these genes). Both of the ESC-derived NMP-like populations exhibit expression of Cdh1, Oct4, Fgf5 and Otx2, an epiblast signature (Figs 1C,D, 2A and 3). Surprisingly ES-NMPF cells also display high levels of genes associated with mesendoderm differentiation, e.g. Mixl1 (endoderm), Tbx6 (paraxial mesoderm) and Evx1 (extra-embryonic mesoderm) (Fig. 2A, Fig. S3 and Gouti et al., 2014); this suggests that ES-NMP and ES-NMPF are overlapping populations at various stages of differentiation, including cells in the early epiblast/gastrula-like stages. In contrast, Epi-NMPs are in a different state: in addition to the accepted NMP signature (T, Sox2 and Nkx1-2), these cells express significant levels of Nodal, Fgf8, Fgf5, Foxa2, Otx2 and Oct4, together with Cyp26a1 (Figs 1C,D, 2A, Fig. S3 and Fig. 3). This is a profile associated with the late epiblast (about E7.5), around the time of the appearance of the CE, before NMPs can be detected (Table S1 and Fig. S1).

To correlate the gene expression profiles with the state of the cells in the different culture conditions, we defined two measures of the degree of differentiation based on the average expression z-score values for mesodermal and neural genes (see Materials and Methods). These measures are indexes that define a global value of the state of the cells in a particular condition. The ‘epiblast index’ reflects the degree of differentiation of the population based on the expression of the chosen 27 genes and is defined by the ratio between the non-differentiated (epiblast) and differentiated stages. On the other hand, the ‘NMP index’ is defined as the degree of neural or mesodermal differentiation, i.e. whether the population exhibits any bias towards either fate (Fig. 2B,C, Materials and Methods). In both cases, the distance of the cells from the diagonal and from the origin of the plot reflects the average phenotype of the cell population. The closer to both, the more the population is in a progenitor uncommitted epiblast state, as in the case of the epiblast index; this means that the cells are in low differentiation state. In the case of the NMP index, it means that they do not exhibit a differentiation bias (see Materials and Methods). The indexes show that Epi-NMPs have a high undifferentiated epiblast identity with some degree of differentiation towards mesoderm, whereas ES-NMPs exhibit low epiblast identity and a degree of differentiation towards the neural fate. In contrast, ES-NMPFs exhibit high epiblast identity but also a strong mesodermal differentiation bias. The differences between the three populations are further emphasized by an examination of the protein levels of some of these markers (Fig. 1C,D and Fig. 3). NMP-like populations derived from ESCs exhibit high levels of Sox2, Oct4 and Cdh1 expression with some cells expressing Otx2, but little or no expression of Cdh2, a signature characteristic of early undifferentiated epiblast (Morgani et al., 2018). ES-NMPFs exhibit a combination of Cdh1 and Cdh2 at the level of single cells (Fig. 3), a situation rarely seen in vivo. In contrast, the Epi-NMPs exhibit lower level of Sox2 and Oct4 (Fig. 1C,D), and a mutual exclusive expression of Cdh1 and Cdh2 (Fig. 3), which is a characteristic of the late epiblast (Corsinotti et al., 2017; Morgani et al., 2018).

Exposure of the different NMP-like populations to neural and mesodermal differentiation environments reveals their potential (Fig. 2A, Fig. S3 and Materials and Methods). In all cases, the cells differentiated into neural and mesodermal progenitors but with different biases depending on their origin (Fig. 2). ES-NMPFs and their differentiated progeny exhibit a mesodermal bias, while ES-NMPs exhibit a slight bias towards the neural fate, both in agreement with their indexes. In contrast, Epi-NMPs, which have a high epiblast index, differentiate equally into neural and mesodermal cell types (Fig. 2C).

Altogether, these results suggest that different protocols yield related, but different, NMP-like populations that might have different functional properties. Furthermore, these populations are differently biased in their differentiation potential. The NMP-like population derived from EpiSCs, appears to be the closest to an uncommitted epiblast state and to harbour the most unbiased state.

The differences between the candidate NMP-like populations derived in vitro suggest that they might represent different stages of the transition between the early postimplantation epiblast and the CLE. To test this, we created a developmental stage reference using a previously published microarray study of the epiblast at different embryonic stages between early postimplantation (E5.5) and early CLE (E7.5 (Kojima et al., 2014), and mapped the NMP-like populations, as well as their differentiated derivatives, onto it (see supplementary Materials and Methods and Fig. S4). Using this as a reference, we observe that, in a three-dimensional principal component analysis space, ES-NMP, Epi-NMP and their derivatives mapped closely to the different embryonic stages, whereas the ES-NMPF and its differentiated populations lie separate from these trajectories and from the embryonic stages (Fig. S4C). Furthermore, the Epi-NMP and its derivatives projected closely to each other within the embryo trajectory between the LMS and EB stages.

We also used our developmental reference to explore the proximity of the in vitro-derived populations to specific epiblast states in vivo. To do this, we used the microarray epiblast analysis of Kojima et al. (2014) as a reference and calculated the cosine similarity between the in vitro population and the different stages of the embryo as a metric for the proximity of each in vitro population to a particular embryonic stage (Fig. 4A, Fig. S4B and see Materials and Methods). Using this measure, and keeping with the indexes discussed above, we find that ES-NMPs correlate with an early epiblast state, whereas ES-NMPFs appear to be a broad population associated with several late-differentiating epiblast stages, mostly late epiblast, confirming that ES-NMPFs represent a heterogeneous population of differentiating and non-differentiating cells. On the other hand, the Epi-NMPs exhibit two peaks that are associated with the early epiblast and the primitive streak. This analysis also reveals that Epi-meso, a population derived from the Epi-NMPs, resembles the LB stages, which correspond to E7.5/8.25 – the stage where the CLE can be seen to harbour the NMP for the first time (Wymeersch et al., 2016).

Altogether, these results support the notion that different starting conditions and differentiation protocols lead to populations with different identities. ES-NMPs seems to resemble early epiblast state and ES-NMPFs represent heterogeneous populations with representations of differentiated cells. On the other hand, Epi-NMPs appear to represent a tighter population resembling an epiblast stage.

In the course of our survey of markers distinguishing the embryonic regions of the CE in the different populations, we noticed that all protocols lead to co-expression of T and Sox2, together with the expression of genes that are not associated with NMPs, e.g. Mesp1, Evx1, Mixl1, Gata6, Bmp4, Msx1, Msx2, Osr1, Pax2 and Tbx2 (Fig. 2, Fig. S3 and Amin et al., 2016; Gouti et al., 2014). A survey of the literature shows that in the embryo between E7.0 and E8.5, roughly the stage of the differentiating in vitro cells, these genes are expressed in the posterior domain of the CE, in the progenitors of the allantois (Tbx2, Tbx4, Mixl1 and Evx1), the LPM (Msx1 and Msx2) and the IM (Pax2 and Osr1) (see Table S1 and Fig. S1). The in vitro-derived populations can be mapped to this stage interval, thus suggesting that they are not restricted to harbour NMPs only, but rather that they represent a multi-potential population that includes progenitors of LPM, IM and allantois.

In the embryo, the further differentiation of the CE is under the control of BMP signalling, which favours more posterior fates (LPM, IM and allantois progenitors) at the expense of more anterior ones (NMPs) (Wymeersch et al., 2016). To test this dependence of fate on BMP signalling, we altered the levels of BMP in the Epi-meso population (Fig. 4B,C and Fig. S5A,C), which appears to be the closest to the embryo CLE, and applied the NMP index to ascertain the differentiation bias of the resulting population. In our cultures, inhibition of BMP signalling elevates the expression of NMP markers, e.g. T, Sox2 and Cdx2 [Epi-meso condition supplemented with the BMP inhibitor DMH-1 (Epi-mesoFCD) in Fig. 5C] and thus increases its NMP identity (Fig. 4C). On the other hand, addition of BMP to the derivatives of Epi-meso population (EM2-FCB) elevates dramatically its mesodermal state (Fig. 4C) and specifically increases the expression of genes associated with posterior fates: Bmp4, Msx1, Msx2 and Tbx2, together with Cdx2 and Snail (Fig. S5A,C). Similar to Epi-mesoFCD, inhibition of BMP in the Epi-meso2 population sample (EM2-FCD, Fig. 4B,C) slightly improves its NMP index in comparison with Epi-meso2.

When Epi-meso cells are grown in N2B27 supplemented with Chiron alone (Fig. 4B,C and Fig. S5B,C), we observe an increase in the levels of expression of neural markers (Sox1, Sox2 and Hes5) with a concomitant shift of its NMP index to neural identity and a loss of mesodermal identity. Furthermore, inhibition of Wnt in the Epi-meso2 state (EM2-FP, Fig. 4B,C and Fig. S5B,C) leads to a reduction in the expression of neural progenitor markers and an elevation in the expression of mesodermal markers (Gata6 and Snail1), which appropriately shift the NMP index of the Epi-meso2 state to the mesodermal side, with low neural averaged value in comparison with Epi-meso2. This is surprising, as it is often thought that Wnt signalling suppresses neural development during the early stages of gastrulation (E6.0-E7.0). However, although this is the case for anterior neural fates early in development, the expansion of neural progenitors requires Wnt signalling (Garriock et al., 2015; Zechner et al., 2003). Therefore, these observations support the suggestion that the Epi-meso populations are related to the NMPs, rather than to an early epiblast population.

In the embryo, as the posterior region of the CE is dominated by BMP signalling, the differentiation of the anterior domain is dependent on its proximity to the NSB. We observe that the Epi-NMP population expresses Nodal and Foxa2 genes (Fig. 2A and Fig. S3A), which are associated with the NSB. Inhibition of Nodal signalling reduces T-Sox2 co-expressing cells (Turner et al., 2014) and this led us to test whether Nodal signalling influences the NMP signature and differentiation potential of the NMP-like cells. To do this, we cultured Epi-NMP from Nodal mutant EpiSCs (Nodal^−/− Epi-NMP, see Materials and Methods, Fig. 5A) and compared them with Nodal mutant Epi-NMPs supplemented with two different doses of Nodal in the presence of FGF and Chiron: 100 ng/ml of Nodal (Nodal^−/− Epi-NMP+0.1×Nodal) or 1 μg/ml of Nodal (Nodal^−/− Epi-NMP+1×Nodal). Addition of Nodal to Nodal^−/− Epi-NMP cells lifted their levels of T and lowered their levels of Sox2, together with increasing of Cyp26a1 and Fgf8 expression. These results suggest that Nodal signalling is necessary to maintain the relative levels of Sox2 and T, and significant levels of Fgf8 and Cyp26a1, which are characteristic of the CE.

In summary, our results indicate that, in all protocols tested, differentiation of PSCs towards a caudal population does not result in the specification of NMPs only, but rather in a multipotent population for all axial derivatives; different protocols appear to exhibit different representations of this population, which is further differentiated by BMP and Nodal. The differences between the different protocols might not only result in different stages of development, but also in different proportions of the different mesodermal populations.

In the embryo, the initial NMP population needs to be amplified, together with the progenitors of the LPM and IM, if it is to account for the cellular mass along the length of the region posterior to the brain (Steventon and Martinez Arias, 2017; Wymeersch et al., 2016); if this were not the case, the initial population would be exhausted before completing axial elongation. We suggest that this amplification should be an additional criterion to identify NMPs in vitro.

Earlier studies have shown that ESC-derived NMPs are not able to maintain the T/Sox2 co-expressing cells when they are passaged in the conditions in which they were generated, namely FGF and Chiron or Chiron alone (Gouti et al., 2014; Turner et al., 2014). Surprisingly, we noticed that when Epi-NMPs are induced to differentiate into mesoderm by exposure to FGF and Chiron, they maintain T and Sox2 expression for at least two passages (Epi-NMP to Epi-meso and Epi-meso to Epi-meso2) with a low differentiation index (Fig. 5B,C and Fig. S6). By passaging Epi-meso, the levels of T decrease, but unlike other situations, do not disappear (the levels of T are detectable by RT-qPCR, see Fig. S6). Furthermore, in the transition from Epi-NMP to Epi-meso, cells lose the expression of epiblast markers, e.g. Fgf5, Nodal, Otx2, Oct4 and Cdh1 (Fig. 2A and Fig. S3A and protein expression of Oct4, Otx2, Cdh1 and Cdh2 in Fig. 1C,D and Fig. 3). This suggests that the Epi-meso population state is distinct from the epiblast and contains many features of the NMPs that are a subset of the CLE; however, this differs from the epiblast characteristic that we describe as Epi-NMP.

During the passages of Epi-meso, we observe a progressive decrease, but not an extinction, in the expression of NMP markers (Cyp26a1, Fgf8 and Nkx1-2), similar to the case of T. This decrease is accompanied by a slow increase in the expression of differentiation genes associated with neural fate: Cdh2, Sox2 and Hes5 (Fig. S6). This expression pattern is likely to reflect a decrease in the NMP state of the population, and mirrors a similar decrease in the embryo (Wymeersch et al., 2016). Another explanation for this magnitude of loss, in the context of culturing cells in vitro, is the lack of mechanical support in the in vitro system and the progressive differentiation to neural fates as a default differentiation programme. It is important to emphasize that our current understanding of the NMPs does not determine the levels of T or Sox2 that are required and that low levels are still likely to be sufficient to maintain this state.

During the elongation of the posterior body axis, NMPs progressively exit from their niche in the epiblast adjacent to the node and enter either the primitive streak, where they ingress and integrate with the presomitic mesoderm, or are retained in the epiblast and enter the posterior region of the emerging spinal cord. Depending on the time of exit from the niche, they will contribute to different anteroposterior axial levels. With this in mind, we tested the ability of the in vitro-derived cells to display these behaviours by transplanting differentiating ESCs and EpiSCs into the elongating region of chicken embryos. Previous experiments have demonstrated that these embryos are good hosts for these experiments and that transplanted mammalian cells integrate with the host and produce functional neural and mesodermal derivatives (Fontaine-Perus et al., 1997, 1995; Gouti et al., 2014). However, as pluripotent stem cells can also contribute to spinal cord and somitic mesoderm upon transplantation (Baillie-Johnson et al., 2018), we have focussed our assessment of NMP behaviour on the length of time that the transplanted cells remain within the NMP niche and continually generate these progenitor populations. This is, in turn, reflected by the length of labelled cells distributed along the anteroposterior axis (Baillie-Johnson et al., 2018).

We focussed our experiments on the EpiSC-derived NMP-like populations as the different tests discussed above suggest that they are the closest to the embryonic NMPs. In our experiments, we transplanted cells from the EpiSC, Epi-NMP and Epi-meso conditions into a small region caudal and lateral to the node of the developing chick embryo. This region corresponds to a region that has been shown to contribute to spinal cord and paraxial mesoderm progenitors by fate mapping (Baillie-Johnson et al., 2018) and is outlined by dotted boxes in Fig. 6 (for details see Fig. S7). After around 15 h of incubation, we measured the contributions of grafted cells to the somites and the developing neural tube, but most importantly the extent of their contribution to axial length. As shown in Fig. 6A-D, EpiSCs contributed only to short axial extensions, and their descendants were mainly located in the mesodermal compartments. This result suggests that the EpiSCs exit early from the NMP domain and, at that stage, their most likely fate is mesodermal. This is further reflected in the contribution of labelled EpiSCs to the anterior axial levels shown in Fig. 6D (i.e. at the level of somites 1-6; see also Fig. S7 and Fig. S8). On the other hand, Epi-NMP and Epi-meso populations contributed to increasingly more posterior regions (compare Fig. 6H,L with D; see also Fig. S8) with mixed neural and mesodermal contributions; in particular, we noticed that the Epi-meso grafts made more frequent dual neural and mesodermal contributions than did the Epi-NMPs and that they showed a significant bias towards more posterior positions (Fig. 6L, Fig. S7 amd Fig. S8). As the Epi-meso population is derived from Epi-NMPs, these results suggest that their temporal sequence in vitro results in cells with an ability to colonise more posterior axial levels after transplantation. Perhaps this reflects the fact that Epi-meso cells express more posterior Hox genes than Epi-NMPs (Fig. 2A and Fig. S3); this might contribute to their ability to colonize more posterior regions of the embryo (see Denans et al., 2015). This may additionally explain why the Epi-meso population less frequently produced long contributions to the embryonic axis when compared with the Epi-NMP population, if it has a bias towards these more-posterior axial levels. On comparing the contributions of all three populations to the axial length, contributions of Epi-NMPs and Epi-meso cells were found to be significantly greater than those from the EpiSC condition (see Fig. S7 and Fig. S8). Taking the above result together with our gene expression analyses, we conclude that the continued propagation of the Epi-NMP population in culture can produce a population that closely resembles the newly arisen embryonic NMPs at E8.25.

We have used and compared three PSC-based differentiation protocols to study the emergence in vitro of a population of bipotential progenitors, NMPs, that, in the mammalian embryo, give rise to the paraxial mesoderm and spinal cord of the thoracic tract. Our results show that each of these protocols produces populations of cells with different gene expression signatures and abilities to contribute to axial elongation but with two common denominators: co-expression of T and Sox2, as well as of genes associated with LPM, IM and allantois. These results suggest that: co-expression of T and Sox2 is not a univocal criterion to identify NMPs; the populations generated in vitro are not restricted to NMPs; and, therefore, the identification of these progenitors in vitro requires additional criteria, in particular an ability to self-renew and to make large contributions to axial extension, as well as an association with the node (Gouti et al., 2015; Henrique et al., 2015; Steventon and Martinez Arias, 2017; Wilson et al., 2009). By applying these criteria to differentiating PSC populations, we identify a specific protocol that, starting with EpiSCs, yields a population, Epi-meso, that is similar to the NMPs in the embryo in terms of cellular function, gene expression, limited maintenance over time, long axial contributions and the exit timing of the progenitors from the caudal domain of the embryo. We surmise that this population emerges from a late epiblast-like state, Epi-NMP, that can also give rise to LPM, IM and extra-embryonic mesoderm in a signalling-dependent manner. Our observations are in agreement with recent descriptions of the development of the tail bud (Wymeersch et al., 2019) and suggest that such a multipotent population might be an obligatory intermediate for the emergence of the NMPs in vitro. ESC-based protocols yield similar populations that can be differentiated into mesodermal and neural progenitors but lack several features characteristic of NMPs, in particular their ability to self-renew and to contribute significantly to axial extension (Baillie-Johnson et al., 2018; Gouti et al., 2014; Turner et al., 2014). Furthermore, as we have shown here, these populations represent highly heterogeneous populations with a low representation of NMPs. A similar multipotent population is likely to exist in the embryo; analysis of lineage-tracing data at the single cell level reveals the existence of clones that span the spinal cord and, at least, two mesodermal derivatives (see Tzouanacou et al., 2009).

Upon further exposure to Wnt and FGF signalling in vitro, this multipotent population, Epi-NMP, evolves and generates cells with many of the hallmarks of the NMPs, including a limited ability to maintain in culture the co-expression of Sox2 and T over few passages, and the ability to differentiate into neural and mesodermal progenitors in a Wnt dependent manner and to make long and more posterior contributions to axial extension in a xenotransplantation assay. For these reasons, we propose naming this Epi-NMP population Epi-CE, and the Epi-meso population Epi-NMP. In the embryo, the emergence of the NMPs from the multipotent population is likely to respond to a regionalization of signalling with BMP and Wnt signalling in the posterior domain, favouring LPM and IM (Sharma et al., 2017), and Nodal and Wnt signalling in the anterior region, favouring NMPs. In agreement with this, we find that the fate of the Epi-CE cells is dependent on a balance between BMP and Nodal signalling, and has a strict requirement for Wnt signalling in both neural and mesodermal lineages (see also Edri et al., 2018 preprint).

There are many studies in which Wnt signalling can caudalize epiblast-like populations (Amin et al., 2016; Mazzoni et al., 2013; Neijts et al., 2016; Nordström et al., 2002, 2006); in these cases, which are mostly ESCs based, the NMP-like cells fail to self-renew as they do in vivo. In contrast to these ESC-based protocols, here we have shown that exposure of pre-treated EpiSCs to FGF and chiron generates a population with a gene expression signature that is characteristic of a late CE, around the time of the appearance of the node, i.e. a multipotent population. The importance of Nodal in the establishment of the multipotent population, and perhaps also in the definition of the NMP domain, is underscored by our studies with Nodal mutant cells, in which the rescue of a population with a disrupted relative level between Sox2 and T, is crucially dependent on the levels of Nodal signalling. Consistent with a role of the node and of Nodal in the establishment of this population, embryos mutant for Foxa2 that lack a node, exhibit deficiencies in the organization of the CE and axial elongation (Ang and Rossant, 1994; Weinstein et al., 1994); the same can be observed in embryos mutant for Smad2 and Smad3 (Vincent et al., 2003).

In vitro, the transition between the Epi-CE and Epi-NMP is linked to the loss of expression of several genes that are associated with the epiblast, e.g. Fgf5, Otx2 and, especially, Oct4 (a POU domain transcription factor that, together with Sox2, maintains pluripotency). A similar transition can be observed in the embryo where Oct4 expression ceases at around E8.5/9.0 (Downs, 2008; Osorno et al., 2012), the time at which cells start differentiating. It is possible that the combination of Oct4 and Sox2 promotes multipotency and that only when Oct4 expression ceases can Sox2 implement a pro-neural role. A function for Oct4 in axial elongation can be gauged from the severe axial truncations that follow loss of Oct4 activity from E7.0/7.5 (DeVeale et al., 2013) and the extended axial elongations associated with overexpression of Oct4 (Aires et al., 2016). This may reflect an increase in the initial size of the multipotent CE pool rather than a specific alteration in the NMP population.

During the passage of the Epi-NMP population in the presence of Wnt and FGF signalling, we noticed that cells progressively lose T expression and increase Sox2 expression. This is surprising because a widespread notion suggests that Wnt signalling suppresses neural differentiation and promotes mesoderm formation. However, in the embryo, this is true during the first phase of gastrulation, before the appearance of the node at E7.5, and reflects the maintenance of an anterior neural fate away from mesendoderm (Arkell and Tam, 2012). This might not be the case during the development of the caudal region of the embryo, as there is a clear evidence that, during this period, Wnt/β-catenin signalling is required for the expression of Sox2 (Takemoto et al., 2006, 2011) and for the expansion of the neural progenitors in the spinal cord (Zechner et al., 2003). Furthermore, during this period, increases in Wnt/β-catenin signalling do not suppress neural development (Garriock et al., 2015). A requirement for Wnt signalling in the development of the spinal cord is further emphasized by the observation that the Sox2 gene has a Tcf response element and responds to Wnt signalling (Takemoto et al., 2006). Thus, we suggest that the response of neurally specified cells to Wnt signalling is a measure of the stage and position of the cells generated by the in vitro protocols.

In summary, using a specific experimental protocol, we have shed light on the origin of the NMP population in vivo and in vitro. Our work highlights the importance of the starting population in the differentiation of specific cell types, as well as the relationship between the state of the in vitro-produced cells and the embryo.

E14-Tg2A ESCs were grown in tissue-culture plastic flasks coated with 0.1% gelatine (Sigma-Aldrich, G1890-100G) in PBS (with calcium and magnesium, Sigma-Aldrich, D8662) filled with GMEM (Gibco) supplemented with non-essential amino acids, sodium pyruvate, GlutaMAX, β-mercaptoethanol, foetal bovine serum and LIF. Culture media were changed daily and cells passaged every other day. The differentiation protocols are as follows.

Cells were plated at a density of 4.44×10³ cells/cm² in a 0.1% gelatine-coated flask with a base medium of N2B27 (NDiff 227, Takara Bio) for 2 days. After 48 h, the N2B27 medium was supplemented with 3 μM of CHIR99021 (chiron 10 mM, Tocris Biosciences) for an additional 24 h, to a total of 72 h.

ES-NMP cells were detached from the culture flask using Accutase (BioLegend 0.5 mM) and divided into 2 flasks coated with 0.5% Fibronectin at a density of 7.5×10³ cells/cm². For ES-neuro and ES-meso differentiation, the cells were grown for 2 days in N2B27 or N2B27 supplemented with 20 ng/ml FGF2 (R&D systems, 50 µg/ml) and 3 μM chiron, respectively.

These cultures were based on the methods of Gouti et al. (2014). Cells were plated at a density of 5×10³ cells/cm² in a 0.1% gelatine-coated CellBINDSurface dish (Corning) with a base medium of N2B27 supplemented with 10 ng/ml FGF2. After 48 h, the N2B27 was supplemented with 10 ng/ml FGF2 and 5 μM chiron for an additional 24 h, to a total of 72 h. To induce neural spinal cord identity (ES-neuroF) or mesodermal identity (ES-mesoF), cells were grown from day 3 to day 5 in either N2B27 supplemented with 100 nM RA (Sigma) or N2B27 supplemented with 5 μM chiron, respectively.

E14-Tg2A ESCs were grown in tissue-culture plastic flasks coated with 0.5% plasma fibronectin (FCOLO, 1 mg/ml, Temecula) in PBS (with calcium and magnesium). ESCs were grown in Epi-medium [N2B27 supplemented with 12 ng/ml FGF2 and 25 ng/ml activin A (Stem Cells Institute, Cambridge, UK 100 μg/ml stock), with or without 20 μM XAV939 (XAV Tocris Biosciences, 10 mM)] for at least four passages to generate EpiSCs (or EpiXAV when the β-catenin inhibitor XAV is used). These cells were tested for EpiSC characteristics by seeding them at a clonal density (67 cells/cm²) in restricted medium [2i; N2B27 supplemented with 3 μM chiron and 1 μM PD0325901 (PD03, Tocris Biosciences, 10 mM)]. On observing no colony formation, it was concluded that the cells had exited naïve pluripotency and had entered the primed pluripotent state (data are not shown).

EpiSCs (treated with or without XAV) were plated at a density of 5×10⁴ cells/cm² in a 0.5% fibronectin pre-coated flask with Epi-media for the first day. The concentration of FGF2 was increased after 1 day to 20 ng/ml in the base medium of N2B27 and activin A or XAV (if used) were removed. On day 3, the N2B27 was supplemented with 3 μM chiron, which was added to the 20 ng/ml FGF2. After 72 h, the resulting population was known as Epi-NMP or EpiXAV-NMP (if XAV was used in the Epi-medium). This protocol is a variation of one previously used to derive NMP-like cells from human ESCs (Lippmann et al., 2015).

Epi-NMP cells (cultured without XAV) were detached from the culture flask using accutase and divided into two flasks coated with 0.5% fibronectin at a density of 5×10⁴ cells/cm². To derive Epi-neuro and Epi-meso populations, the cells were grown for 2 days in N2B27 alone or N2B27 supplemented with 20 ng/ml FGF2 and 3 μM chiron, respectively.

Epi-meso (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 20 ng/ml FGF2 and 3 μM chiron. The first passage from Epi-meso is named Epi-meso2 (EM2), the second passage is named Epi-meso3 (EM3), and so forth.

Epi-NMP (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 20 ng/ml FGF2, 3 μM chiron and 500 nM dorsomorphin-H1 (DMH-1 5 mM, Tocris Biosciences), which is a BMP inhibitor.

Epi-meso (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 20 ng/ml FGF2, 3 μM chiron and 500 nM DMH-1.

Epi-meso (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 20 ng/ml FGF2, 3 μM chiron and 1 ng/ml BMP4 (R&D Systems, 100 μg/ml).

Epi-meso (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 3 μM chiron alone.

Epi-meso (cultured without XAV) cells were detached from the culture flask using accutase and plated back to a 0.5% fibronectin-coated flask at a density of 5×10⁴ cells/cm² for 2 days in N2B27 supplemented with 20 ng/ml FGF2 and 1 μM IWP-2 (PIN 5 mM, STEMGENT), which is a Wnt pathway inhibitor.

ESCs mutant for Nodal (Nodal^−/−) were provided by J. Collignon (Institut Jacques Monod, Paris, France) following derivation from the 413.d mutant mouse line (Conlon et al., 1991). They were grown on a 0.5% fibronectin-coated culture flask with Epi-medium: N2B27 supplemented with 12 ng/ml FGF2 and 25 ng/ml activin A for at least four passages. The Nodal-null EpiSCs were plated at a density of 5×10⁴ cells/cm² on a 0.5% fibronectin pre-coated flask with Epi-medium for the first day. The concentration of FGF2 was increased after 1 day to 20 ng/ml in the base medium of N2B27 and activin A was removed. On day 3, N2B27 was supplemented with 3 μM chiron, which was added to the 20 ng/ml FGF2. After a total of 72 h, the resulting population was known as the Nodal^−/− Epi-NMPs. In order to examine the role of Nodal in establishing the NMPs, the Nodal mutant Epi-NMPs were supplemented with two different doses of Nodal in the culture medium on the 3rd day: 20 ng/ml FGF2, 3 μM chiron and either 100 ng/ml of Nodal (R&D Systems: Nodal^−/− Epi-NMP+0.1×Nodal) or 1 μg/ml of Nodal (Nodal^−/− Epi-NMP+1×Nodal) in the base medium of N2B27.

Total RNA was isolated from cells using Trizol. First-strand cDNA synthesis was performed with the Superscript III system (Invitrogen). The quantification of double-stranded DNA obtained for primer-specific genes was achieved with QuantiFast SYBR Green PCR Master Mix (Qiagen) and the standard cycler program (Qiagen RotorGene Q). The qPCR was carried out in technical triplicates. The primers that have been used are available in Table S2. All experiments were performed in biological duplicate or triplicate. Expression values were normalized against the housekeeping gene Ppia. To enable comparison between different qRT-PCR experiments, each run of the qPCR included one common condition (Epi-meso in this case). Each condition in every run was normalized to Epi-meso and averaged across biological replicates. The steps to calculate the normalized gene expression values are as follows: (1) identify the C_t (threshold cycle) for each gene (technical triplicates) and calculate the expression values (2^−Ct); (2) calculate the average and the standard deviation (s.d.) for each gene from the triplicate expression values; divide the average and the s.d. results for each gene by the expression value for Ppia. The normalized gene expression values in condition x are divided by the normalized gene expression values in the common condition for every qRT-PCR experiment (Epi-meso) as follows:

F=x/y,

• F₁ and F₂ are biological replicates of the same gene in the same condition and their expression was normalized as above. The average of the normalized expression and the error is calculated as the standard error (SE):

• Standardizing the normalized expression values of a gene to Z-score values across conditions was carried out using:

  

where µ and σ denote the average and the s.d., respectively, of the normalized expression of a gene across all the conditions examined in this work (the average and s.d. of for a specific gene across all the conditions).

The genes were sorted as follows: neural genes, Sox1, Pax6, Hes5 and Sox2; mesodermal genes, Bmp4, Evx1, Gata6, Meox1, Mesp1, Mixl1, Raldh2, Tbx2, Tbx6, Msx1, Msx1, Pax2, Osr1, Snai1 and T; epiblast genes, Cdh1, Fgf5, Oct4, Otx2, Cdx2, Fgf8, Nodal, Wnt3a, Cyp26a1, Nkx1-2, Hoxc6, Cdh2 and Foxa2.

The NMP index was calculated as follows. In all the conditions (17 in total) the average expression Z-score value of the neural genes and the mesodermal genes was obtained and scaled between 0 and 1 across the 17 conditions. This resulted in two values for each condition: the neural averaged value and the mesodermal averaged value. The epiblast index was calculated in a similar manner. The average of the Z-score expression values of the epiblast genes was calculated versus the differentiation genes (neural and mesodermal) and scaled between 0 and 1 across the 17 conditions, resulting in an epiblast averaged value and a differentiation averaged value for each condition.

Single-molecule RNA fluorescence in situ hybridization was carried out as described previously (Nair et al., 2015). Cells were dissociated using accutase, washed in PBS, fixed in 37% formaldehyde at room temperature, permeabilized and stored in 70% ethanol at 4°C. All washes and hybridizations were carried out in suspension. Wash buffers included 0.1% Triton X-100 to minimize losses of cells sticking to the tube walls. Samples were mounted between a slide and #1 cover glass, in the glucose oxidase-based 2×SSC anti-fade buffer. The probes for the genes (Table S3) were designed using Stellaris website (www.biosearchtech.com/support/education/stellaris-rna-fish) and bought via Stellaris FISH probes (Biosearch Technologies) (Raj et al., 2008). Additional information about how the probes were designed, prepared and used can be found in Raj et al. (2008). Cells were imaged within 24 to 48 h of fixation on a Nikon Ti-E wide-field microscope, using a 60× oil-immersion objective and a cooled camera (Orca flash 4.0, Hamamatsu). The cells in the images were segmented manually and the spot detection was carried out semi-automatically using a MATLAB graphic user interface (GUI) developed by Marshall J. Levesque and Arjun Raj (University of Pennsylvania) or with custom-made protocols written in ICY (icy.bioimageanalysis.org) (de Chaumont et al., 2012).

Principal component analysis (PCA) involves the assignment of data, in our case gene expression, to new coordinates named principal components or PCs. The variance of observed coordinates in each PC occurs in a decreasing order, observations (the samples) projected on PC1 have a greater variance than the same observations projected on PC2, and so on. The PCs were calculated according to the Z-score expression values of the 27 genes measured (Fig. 2A and Fig. S3B) at different stages of mouse embryo epiblast/ectoderm and in the three in vitro protocols and their neural and mesodermal differentiation: ES-NMP, ES-NMPF and Epi-NMP.

Samples of the different cell cultures were grown in four-well (Ibidi) plastic tissue-culture dishes. Samples were washed in BBS+CaCl₂ (50 mM BES sodium salt, 280 mM NaCl, 1.5 mM Na₂HPO₄, 1 mM CaCl₂ adjusted to pH 6.96 with 1 M HCl) and fixed for 15 min in 4% paraformaldehyde. Samples were washed and permeabilized with BBT (BBS+CaCl₂ supplemented with 0.5% BSA and 0.5% Triton X-100) before overnight incubation with primary antibodies. The following day, the samples were washed with BBT and incubated for 2 h with the desired fluorescently conjugated secondary antibodies. Prior to imaging, samples were washed with BBS+CaCl₂ and covered in a mounting medium (80% spectrophotometric grade glycerol, 4% w/v n-propyl-gallatein in BBS+CaCl₂).

Antibodies against the following proteins were used: T (brachyury) N19 (goat; Santa Cruz Biotechnologies, sc17743, 1:100), Oct3/4 (mouse; Santa Cruz Biotechnologies, sc5279, 1:100), Sox2 (rabbit; Millipore, AB5603, 1:200), Otx2 (goat; R&D, AF1979, 1:200), Cdh2 (mouse; BD Biosciences, 610920, 1:200) and Cdh1 (rat; Takara, M108, 1:100). Secondary antibodies (goat-A488, rabbit-A633, mouse-A568 and rat-A633; Molecular Probes) were raised in donkey and used at a 1:500 dilution with Hoechst 33342 (H3570, 1:1000; Invitrogen). Samples were imaged using a LSM700 on a Zeiss Axiovert 200M with a 63× EC Plan-NeoFluar 1.3 NA DIC oil-immersion objective. Hoechst, Alexa488, Alexa568 and Alexa633 were sequentially excited with a 405, 488, 555 and 639 nm diode lasers, respectively. Data capture was carried out using Zen2010 v6 (Zeiss). The cells from fluorescence microscopy images were segmented manually and the quantification of the cellular fluorescence level subtracting the background was carried out using the open source FIJI ImageJ platform (Schindelin et al., 2012).

Fertilised chicken eggs were stored in a humidified 10°C incubator for up to 1 week until required. Eggs were transferred to a humidified, rocking 37°C incubator for 24 h prior to the preparation of embryo cultures, which was carried out according to a modified version of New's culture (New, 1955). Embryo cultures were incubated at 37°C prior to grafting and were fixed in 4% paraformaldehyde within 24 h.

Cell cultures were prepared as described for chicken embryos. Adherent cell cultures were detached mechanically using a cell scraper in PBS (with calcium and magnesium) to lift intact colonies with minimal sample dissociation. The tissues were labelled by transferring them to a FBS-precoated FACS tube and were centrifuged at 170 g for five minutes. The supernatant was discarded and the colonies washed by gentle resuspension in PBS (with calcium and magnesium) before the centrifugation step was repeated. The tissues were then resuspended gently in PBS (without calcium and magnesium) for labelling with DiI (ThermoFisher Scientific Vybrant, V22885, 1% v/v) for 25 min in the dark, on ice. The labelled tissues were centrifuged at 170 g for five minutes and the pellet was gently resuspended in PBS (with calcium and magnesium) for grafting.

Labelled tissues were grafted into the region of the chick caudal lateral epiblast as described in Fig. S7 and (Baillie-Johnson et al., 2018; Gouti et al., 2014), using an eyebrow knife tool. Embryo cultures were imaged at single time points and as time-lapses for 15-18 h after grafting.

Wide-field images were acquired with a Zeiss AxioObserver Z1 using a 5× objective in a humidified 37°C incubator, with the embryo cultures positioned on the lid of a six-well plate. An LED white light illumination system (Laser2000) and a Filter Set 45 filter cube (Zeiss) was used to visualize red fluorescence. Emitted light was recorded using a back-illuminated iXon800 Ultra EMCCD (Andor) and the open source Micro-Manager software (Vale Lab, UCSF, USA). The open source FIJI ImageJ platform (Schindelin et al., 2012) and the pairwise stitching plug-in (Preibisch et al., 2009) were used for image reconstruction and analysis.

MATLAB and the Statistics Toolbox (2018b release, MathWorks) were used to produce the histograms in Fig. S8 and to compute the statistical tests used to compare the resulting distributions (see supplementary Materials and Methods). A two-tailed Wilcoxon Rank Sum test was used to compare the axial lengths of contributions of the labelled cells after grafting. The Kruskal–Wallis one-way analysis of variation was used to make an initial comparison and was followed by a test for unequal medians (the multcompare function).

We are grateful to J. Collignon for the Nodal mutant cells; to James Briscoe, Meritxell Vinyoles, Vikas Trivedi and Valerie Wilson for discussions; and to anonymous reviewers for discussions and comments.