Understanding axial progenitor biology in vivo and in vitro

Lineage specification in the early vertebrate embryo is initiated during gastrulation (see Glossary, Box 1), which involves the formation of the three germ layers (ectoderm, mesoderm and endoderm) via cell ingression and epithelial-to-mesenchymal transition (EMT) from a localised region of the primitive, pluripotent epithelium. This region is known as the primitive streak (PS) in amniotes (see Glossary, Box 1). Following gastrulation, embryonic anteroposterior (A-P) axis elongation lasts until the end of somite production (somitogenesis). In all vertebrates, the arrangement of the axial tissues depends on a structure present during gastrulation known as the organiser (see Glossary, Box 1; reviewed by Anderson and Stern, 2016; Martinez Arias and Steventon, 2018). The organiser itself gives rise to midline axial tissues such as the notochord (see Glossary, Box 1) and ventral neural tube (the progenitor of the floor plate). Signals from the organiser also pattern the surrounding cells such that those closer to the organiser differentiate as more medial (midline) structures. In the ectoderm, the medial-to-lateral axis is made up of successively more dorsally-fated neural tissues flanked by surface ectoderm. Meanwhile, mesoderm precursors produce the paraxial, intermediate and lateral plate/ventral mesoderm (see Glossary, Box 1; Fig. 1). Despite a reorganisation of cells around the organiser during axial elongation to form the growing posterior (or caudal) end of the embryo termed the tail bud (Fig. 1), the progenitors for further axial elongation remain in this caudal location until the end of axis elongation. These progenitors generate sequentially more posterior neural and mesodermal components of the axis, starting from around the base of the future hindbrain, and ending at the tip of the tail. Collectively known as axial progenitors (see Glossary, Box 1), they are responsible for producing a large fraction of the spinal cord and musculoskeleton, the notochord, as well as the body wall and mesodermal organs, such as the kidneys and gonads (Fig. 1).

Research in the last two decades has clarified the number of progenitor types, their locations, fate and potency (see Glossary, Box 1), and how construction of the A-P axis varies between vertebrates. Moreover, recent pluripotent stem cell (PSC)-based models have provided novel insights into various aspects of this process. The link between PSC-based models and disease modelling/regenerative medicine applications has stimulated a wider interest in axial progenitor biology, coming from a diverse range of disciplines and perspectives (Box 2). In this Review, we provide an overview of the field, focusing particularly on the progenitors of the spinal cord and trunk skeletal muscle/vertebral column and their role in the conceptual development of novel PSC differentiation strategies.

Anatomical studies, grafting and lineage-tracing experiments have provided compelling evidence about the existence of multipotent axial progenitors, their location and contribution to different tissues. First, we discuss data from amniotes, where several key advances have been made, and compare these with data from anamniotes.

Lineage tracing in gastrulation-stage chick embryos provided some of the earliest indications that multi-fated axial progenitors are located within defined PS regions. Selleck and Stern showed that descendants of single node cells could be found in more than one germ layer, specifically the neural tube and notochord. Moreover, the descendants of these labelled cells were retained in the progenitor region, leading to the hypothesis that some cells in the node area exhibit stem cell-like properties (Selleck and Stern, 1991). This has been further supported by experiments in mouse embryos, showing that labelling of individual node/streak cells at various gastrulation stages (Forlani et al., 2003; Lawson et al., 1991), or groups of node cells in early somite-stage embryos (Wilson and Beddington, 1996), produces descendants in both the differentiated notochord and the node itself, even after relatively long periods of axis elongation [48 h, forming half (∼30) of the total somite number].

Retrospective clonal analyses in the myotome (see Glossary, Box 1) (Nicolas et al., 1996), neural tube (Mathis and Nicolas, 2000) or in the entire embryo (Tzouanacou et al., 2009) each produced clones that contributed to tissues over large anteroposterior distances. These clones contributed from a variable anterior point as far as the posterior end, indicating an increased probability of recombination in a long-lived progenitor over time. This suggests a continuing stem cell-like progenitor at the posterior end of the embryo. Clones contributing to both neural and (principally paraxial, i.e. somitic) mesodermal tissues, but no other tissue types, indicate a dual-fated neuromesodermal progenitor (NMP). However, retrospective clonal analysis infers the properties of a progenitor from its descendants and requires prospective analyses to identify potential locations for these progenitors.

Prospective fate mapping (see Glossary, Box 1) via dye injection or grafting of small tissue pieces in the PS area of cultured mouse embryos has identified two such regions of neuromesodermal (NM) fate: the node-streak border (NSB) and the anterior caudal lateral epiblast (CLE) on either side of the PS (Fig. 2A). These two areas differ in their contribution to the two lineages: the NSB gives rise to medially-located cells in the somites and the ventral neurectoderm, whereas the CLE produces more lateral somitic cells and mainly lateral neurectoderm (Cambray and Wilson, 2007; Mugele et al., 2018 preprint; Wymeersch et al., 2016). Only the dorsal part of the NSB contributes to both neurectoderm and mesoderm. By contrast, the ventral part of the NSB contains notochord progenitors (NotoPs) (Cambray and Wilson, 2007; Kinder et al., 1999; Wilson and Beddington, 1996) (Fig. 2A,B). Interestingly, the relative quiescence of the ventral node has been suggested to indicate a stem cell-like character of these progenitors (Bellomo et al., 1996; Ukita et al., 2009). However, the ventral node cells at the NSB itself, termed the ‘crown’, are more proliferative (Wymeersch et al., 2019). At embryonic day (E) 8.5, other areas of distinct fate include the anterior half of the PS, which harbours predominantly paraxial mesoderm-fated progenitors. In contrast, the posterior-most PS and CLE contains progenitors of lateral plate mesoderm (termed lateral and paraxial mesoderm progenitors, LPMPs, after their lineage potential), which exit the PS as the tail forms (Fig. 2A) (Cambray and Wilson, 2007; Kinder et al., 1999; Mugele et al., 2018 preprint; Wymeersch et al., 2016). During trunk-to-tail transition (see Glossary, Box 1), the NSB gives rise to a region within the tail bud called the chordoneural hinge (CNH; see Glossary, Box 1) (Cambray and Wilson, 2002; Wilson and Beddington, 1996). This region also contains axial progenitors with dual NM fate (Catala et al., 1995; McGrew et al., 2008; Wilson and Beddington, 1996). The CNH itself comprises the notochord end, derived mainly from the ventral NSB cells (Wymeersch et al., 2019) and overlying epithelial tissue continuous with the neurectoderm, in which NMPs are likely to reside (Box 3; Fig. 2C) (Cambray and Wilson, 2002).

More recent work has shown that, although NSB cells contribute fairly equally to both neural and mesodermal derivatives, the fate of NM-potent cells in the CLE is dependent on their exact location in the epiblast layer: anterior and lateral cells contribute largely to neural tube, whereas posterior and medial cells contribute more to mesodermal tissues (Wymeersch et al., 2016) (Fig. 2D). Furthermore, the observation that the majority of the dorsal CNH derives from NSB cells, whereas only a minor proportion of cells derive from the CLE, adds further diversity to NMP fates (Cambray and Wilson, 2007; Catala et al., 1995). This suggests that NMP subpopulations are biased towards neural or mesodermal differentiation outcomes, and towards long- or short-term contribution to the axis depending on their physical location. Alternatively, different NMP subtypes might occupy these separate locations (see ‘Axial progenitor fate versus potency’ sections below).

The presence of dual NM-fated cells in posterior locations equivalent to those in mouse has been demonstrated during chick embryonic axis elongation (Brown and Storey, 2000; Guillot et al., 2020 preprint; Iimura et al., 2007; McGrew et al., 2008; Wood et al., 2019 preprint) (Fig. 3A,B). More recently, the early origin and later location of individual dual-fated cells has been described using live imaging (Guillot et al., 2020 preprint; Wood et al., 2019 preprint), which revealed that cells that produce neural and mesodermal descendants lie at the interface between cells of purely neural or mesodermal fate. These NM progenitors form an arc around the node at the end of gastrulation, which gradually extends posteriorly to form an inverted ‘U’ shape, similar to that seen in mouse (Figs 2A and 3A).

Although evidence in chick and mouse supports the existence of NM-fated progenitors that contribute over large stretches of posterior axis, NMP contribution to anamniote axial elongation is unclear. Work in the early 2000s in the Xenopus tail showed that multi-fated descendants were observed after focal labelling the earlier-stage tail bud (Davis and Kirschner, 2000). However, the photoactivation method employed in that study labelled small groups of cells, rather than individual ones. Careful anatomical studies in frog identified a region present after gastrulation in the elongating axis, in which the ventral neural tube was continuous with the underlying notochord. This region was termed the CNH and is delimited posteriorly by a small continuous cavity between the neural tube and the gut, termed the neurenteric canal (see Glossary, Box 1; Fig. 3C,D). Interestingly, although this cavity is absent in the mouse, it is present during gastrulation in many other species, including early-stage human embryos (Rulle et al., 2018) (Fig. 3G,H). This indicates a well-conserved discontinuity between the dorsal blastopore lip and CNH on the one hand, and the lateral part of the blastopore and posterior wall of the tail bud on the other (Fig. 3C,D). Lineage tracing further showed that the CNH is derived from the organiser and is fated for the midline (ventral neural tube and notochord), whereas the cells on the posterior side of the neurenteric canal produce paraxial mesoderm (Gont et al., 1993), as does the PS (Beck and Slack, 1998, 1999; Tucker and Slack, 1995). The lineage continuity of cells from the late blastopore to the CNH suggests that the organisation of axial progenitors and their progression to the tail bud is similar to that in amniotes (Gont et al., 1993). However, the Xenopus tail-forming region includes a large section of tissue anterior to the blastopore as trunk cells are incorporated into the tail by anterior displacement of the anus during body extension (Tucker and Slack, 1995). In contrast, recent grafting in the axolotl embryo has shown that the posterior third of the trunk and the entire tail is derived from a region of the posterior neural plate (which co-expresses Sox2 and Brachyury, see below), suggesting that the extent to which tissue rearrangement contributes to axis elongation varies between species (Taniguchi et al., 2017).

In the zebrafish, single-cell fate mapping has shown that a region of overlapping neurectoderm and mesoderm fate exists near the organiser at early gastrulation stages (Kimmel et al., 1990). In contrast, single-cell injections by Kanki and Ho produced no multi-fated progenitors during later tail development (Kanki and Ho, 1997). More recently, Martin and Kimelman confirmed that – consistent with amniotes – zebrafish embryos harbour bipotent axial progenitors throughout posterior body formation, because mesoderm-fated cells can switch fate and give rise to neural tissue upon depletion of β-catenin signalling (Martin and Kimelman, 2012) (Fig. 3E,F). As the presence of bipotent cells does not necessarily imply dual fate, these findings can be reconciled with those of Kanki and Ho if the majority of cells with NM potency do not actually give rise to both neural and mesodermal lineages. Indeed, single-cell tracking in the zebrafish tail bud has shown that only a minority of cells exhibit both fates (Attardi et al., 2018) suggesting that, despite a localised region of NM fate, the number of individual dual-fated cells might be low in fish. In fact, NM contributions are restricted to the last seven to nine tail segments (Attardi et al., 2018) with the largest volumetric increase of trunk tissue in fish originating from the displacement of lateral cells to the posterior, instead of being laid down by the tail bud (Steventon et al., 2016) (Fig. 3E,F). Thus contributions of NM-fated axial progenitors in fish (and possibly also in frog) appear more limited to the posterior-most regions of the axis than those in chick and mouse, and instead the majority of axis elongation relies more on convergence and extension of pre-existing neural and mesodermal tissue formed during gastrulation.

Further volumetric comparisons between lamprey, dogfish and mouse embryos have revealed that the latter two species, which have relatively long anteroposterior axes, initially increase the volume of their unsegmented mesoderm (producing anterior somites), after which it decreases during later elongation to produce the posterior somites (Steventon et al., 2016). This volumetric expansion and contraction has also been observed for a number of organisms, including chicken, mouse and snake (Gomez et al., 2008). Interestingly, in the mouse embryo, the number of putative NMPs shows a similar trajectory of expansion and contraction during elongation (Wymeersch et al., 2016) and a reorganisation in the progenitor pool around tail bud formation (Tzouanacou et al., 2009), both of which are likely to precede changes in axial tissue volume. Although recent studies have elucidated some of the mechanisms underlying trunk-to-tail transition and their effects on axial progenitors (see below), it remains to be investigated how axial progenitor numbers and dynamics shape diverse body plans during elongation across different species.

As mentioned above, in the NM-fated zone of the early somitogenesis stage mouse (E8.0-E8.5), the anterior/lateral and posterior/medial regions of the CLE tend to produce neurectoderm and mesoderm fates, respectively (Wymeersch et al., 2016). As in fish, this suggests that NM fate may not be exhibited by all cells with NM potency. Consistent with this idea, heterotopic grafts of these differentially-fated regions have shown that it is the environment, rather than their level of commitment, that dictates these differential fates (Wymeersch et al., 2016). Indeed, NMPs can also form exclusively lateral plate mesoderm upon transplantation to the posterior PS, although they do not generally exhibit this fate in undisturbed embryos (Cambray and Wilson, 2007; Row et al., 2018; Tzouanacou et al., 2009; Wymeersch et al., 2016). Wider potency than fate appears to be a general property of the region surrounding the PS, as lateral mesoderm-fated PS cells can form paraxial mesoderm upon transplantation to paraxial mesoderm-fated regions of the PS (and are therefore termed LPMPs) (Wymeersch et al., 2016). Similar environment-driven cell fate changes have been also reported in chick (Wood et al., 2019 preprint), suggesting that extrinsically-imposed neural or mesodermal differentiation on NM potent cells is essential to ensure correctly balanced production of neural and mesodermal subtypes during axis elongation. Despite this developmental plasticity, kidney capsule grafts have shown that cells in NM-fated regions are restricted to neural and mesodermal fates from the beginning of somitogenesis until at least E10.5 (Osorno et al., 2012; Wymeersch et al., 2016). In contrast, LPMPs do not give rise to neurectoderm in kidney capsule assays. Thus, the fate of NMPs and LPMPs is restricted both by their location and by their intrinsic potential to differentiate.

Surprisingly, the differentiation of axial progenitors is not only affected by their spatial but also their temporal environment. CNH cells contribute to anterior neurectoderm and paraxial mesoderm when grafted heterochronically into the E8.5 NSB (Cambray and Wilson, 2002). Indeed, NMP descendants in the tail bud mesoderm up to E13.5 also demonstrate this property (Tam and Tan, 1992). Furthermore, NMPs can be serially passaged from CNH to NSB through multiple generations of embryos and – at least in chick – descendants of these ‘late’ NMPs grafted to early stage embryos can adopt the A-P identity of the host environment (Cambray and Wilson, 2002; McGrew et al., 2008). The adaptability of NMPs to temporally unmatched environments, together with the finding that their transcriptome changes dramatically over time (Wymeersch et al., 2019), suggest that the temporal NMP transcriptional state is set (at least in part) by extrinsic signals. Nevertheless, cells grafted heterochronically integrate less efficiently into host tissues than isochronic grafts (whether heterotopic or homotopic), suggesting that this temporal resetting may be less efficient than their acutely sensitive response to spatial cues.

Despite extensive transcriptome analysis, mainly of amniote embryos at both the single-cell and bulk-population level (de Lemos et al., 2019 preprint; Gouti et al., 2017; Guillot et al., 2020 preprint; Koch et al., 2017; Olivera-Martinez et al., 2014; Wymeersch et al., 2019), no unique, exclusive markers that detect NMPs at all stages have been identified to date. In mice, the best readout of NM bipotency appears to be the co-expression of definitive neural and mesodermal genes, such as those encoding the transcription factors Sox2 and brachyury (T) (Henrique et al., 2015; Tsakiridis et al., 2014). T⁺Sox2⁺ double-positive cells emerge at the end of gastrulation at E7.5 and persist in NM-potent regions until axis elongation ends at E13.5 (Wymeersch et al., 2016). T⁺Sox2⁺ cells have been identified in analogous regions of the developing tail bud in zebrafish, chick and human embryos (Guillot et al., 2020 preprint; Martin and Kimelman, 2012; Olivera-Martinez et al., 2012) (Figs 3 and 4). Lineage-tracing experiments in mice have also confirmed that posterior neurectoderm cells in the spinal cord are derived from T⁺ progenitors (Anderson et al., 2013; Mugele et al., 2018 preprint; Perantoni et al., 2005). However, more recent studies have raised doubts about the extent of the contribution of T⁺ and Sox2⁺ cells into the neural tube and paraxial mesoderm, respectively (Mugele et al., 2018 preprint; Serizawa et al., 2019). These findings may reflect differences in the efficiency of the recombinase systems employed and their relative dependencies on the activity levels of the promoters driving their expression (Araki et al., 1997). Therefore, further experiments are required to clarify the discrepancy with published lineage-tracing studies and expression data on NM-potent regions.

In addition to T and Sox2 co-expression, transcriptome analyses and lineage-tracing experiments have revealed a number of other posteriorly expressed genes, such as Nkx1-2, Cdx2, Epha1, Tbx6 and Hox family members, that mark NMPs or NMP subsets in the mouse (Garriock et al., 2015; Javali et al., 2017; Rodrigo Albors et al., 2018) (Tables 1 and 2; Fig. 5A). These markers also include components of various signalling pathways, predominantly Wnt and Fgf. Loss-of-function studies have demonstrated that many of these genes (e.g. T, Cdx2 and trunk Hox genes) are also key regulators of NMP ontogeny (Tables 1 and 2) acting downstream of Wnt and Fgf signalling inputs to control both mesoderm production and progenitor maintenance (Ciruna and Rossant, 2001; Takemoto et al., 2011; Wymeersch et al., 2016; Yamaguchi et al., 1999). Conversely, Nodal and Shh components are found in notochord precursor regions in the node, and BMP-associated transcripts are enriched in LPMPs in the posterior PS (Table 2; Fig. 5A).

Transcriptome analysis has shown that mouse NMPs at early (E8.0-E8.5) and later (tail bud) stages of axial elongation are substantially distinct, despite their shared ability to produce neurectoderm and mesoderm (Dias et al., 2020; Gouti et al., 2017; Wymeersch et al., 2019). Interestingly, components of the Wnt, Fgf and Notch signalling pathways are some of the most differentially expressed genes between late and early NMPs, showing a peak of expression mid-axial elongation, when NMP numbers peak and lead to maximal production of both presomitic mesoderm (see Glossary, Box 1) and somites. Wnt signalling is essential to drive this increase in NMP numbers during E8.5-E9.5 (Wymeersch et al., 2016). Together with the observation that cells lacking the Wnt response gene Cdx2 (a crucial axis elongation factor) can be rescued by transplantation to a wild-type environment (Bialecka et al., 2010), this indicates that NM bipotency is not intrinsically determined, but rather relies on the appropriate combination of extrinsic cues that may include modulation of at least Wnt, but possibly also Fgf and Notch signalling levels. This is in line with evidence that the Wnt-expressing milieu is sufficient to maintain axial progenitors lacking zebrafish ntl (the zebrafish equivalent of T; also known as tbxta), another crucial Wnt response gene required for axis elongation (Martin and Kimelman, 2010). Intriguingly, the transcriptome of the adjacent notochord progenitors remains relatively stable throughout embryonic axis elongation (Wymeersch et al., 2019). Removal of these cells indicates a crucial role in trunk elongation (Ang and Rossant, 1994; Wymeersch et al., 2019) and suggest that they may also be central to NMP maintenance.

NMP differentiation appears to be driven by signal-induced potentiation of cross-repressive, lineage-specific gene regulatory network activities, which co-exist in bipotent cells (Tables 1 and 2; Fig. 5). In vivo and in vitro data show that the transition of NMPs into paraxial mesoderm is mediated by elevated canonical Wnt/β-catenin and Fgf signalling levels, which trigger T upregulation, downregulation of the pro-neural transcription factor Sox2, and the induction of downstream pro-mesodermal transcription factors, such as Tbx6 and Msgn1, which are associated with the somitogenesis clock together with activation of Notch signalling components (Chalamalasetty et al., 2011, 2014; Garriock et al., 2015; Gouti et al., 2017; Hofmann et al., 2004; Javali et al., 2017; Koch et al., 2017; Takemoto et al., 2011; Yasuhiko et al., 2006) (Fig. 5B). Interestingly, this process involves the induction of an intermediate Sox2⁺Tbx6⁺ NMP-like cell population resident in the anterior PS and tail bud (Javali et al., 2017). This entity may correspond to the recently described mesoderm-fated NMPs shown to undergo an incomplete EMT (termed ‘tbEMT’) during the formation of the tail bud (Dias et al., 2020) (Fig. 5B). BMP signalling also appears to play a role in the maintenance of early paraxial mesoderm progenitors at tail bud stages (Sharma et al., 2017), in addition to its function as an inducer of LPMP and intermediate mesoderm fates (Edri et al., 2019a; Row et al., 2018; Wymeersch et al., 2019).

The generation of neural derivatives of NMPs requires downregulation of pro-mesodermal transcription factors and signals (e.g. Wnt/Fgf/T/Tbx6) by retinoic acid (RA) from somites (Diez del Corral et al., 2002; Gouti et al., 2017; Martin and Kimelman, 2010, 2012; Molotkova et al., 2005; Olivera-Martinez et al., 2012). This involves the initial production of posterior neural progenitors from NMPs, marked by the upregulation of Sox2 and concomitant restriction of T/Tbx6 activities, likely due to lower levels of or shorter exposure to Wnt and Fgf signalling, which control Sox2 expression in the caudal epiblast through activation of the N-1 enhancer (Delfino-Machin et al., 2005; Diez del Corral et al., 2002, 2003; Javali et al., 2017; Koch et al., 2017; Takemoto et al., 2011). Moreover, Wnt and Notch signals have been shown to further regulate early posterior neural progenitor maintenance and differentiation (Akai et al., 2005; Olivera-Martinez and Storey, 2007; Takemoto et al., 2006) (Fig. 5B).

The production of axial tissues is a highly coordinated process that involves the generation of NMP descendants in nascent mesoderm or neurectoderm and their eventual allocation and patterning at a specific location along the A-P axis (Fig. 6). The terminal positional identity/patterning of NMP derivatives is determined by the timing of their emergence within the posterior progenitor niches and is largely down to the actions of Hox family members, which are arranged as paralogous groups (PG1-13) in four chromosomal clusters (A,B,C and D). In vertebrates, Hox gene expression is initiated in the posterior of the embryo, in a temporally progressive (collinear) fashion that reflects their 3′-5′ genomic order (i.e. members of PG1 are activated first and PG13 last) (Denans et al., 2015; Wacker et al., 2004; reviewed by Deschamps and van Nes, 2005; Mallo et al., 2010). As mentioned above, the ‘resetting’ of chick tail bud NMPs to an ‘earlier’ state following their engraftment to the NSB of younger hosts is accompanied by the reprogramming from a Hox PG10⁺ to a Hox PG8⁺ identity (McGrew et al., 2008). This suggests that the Hox code in axial progenitors is labile and sensitive to environmental cues. Based on molecular, genetic and genome-wide analyses, we outline below the potential dynamic interplay between Hox PG member activity and key axial progenitor transcription factors/signals driving morphogenesis during axial elongation. In the mouse embryo, in which targeted mutations have elucidated some of the genetic circuitry of axis elongation, the process can be divided into three major phases, with two intervening, ‘transitional’ phases (Fig. 7).

Wnt and Fgf signalling delimit the posterior region, triggering the induction of T and Cdx2 and activating early Hox genes belonging to PG1-4 in a Cdx-independent manner (Amin et al., 2016; Neijts et al., 2017). RA is present (Ribes et al., 2009) and can induce early Hox expression, also stimulating expression of the RA-catabolising enzyme gene Cyp26a1 (Cunningham et al., 2016).

Cyp26a1 expression becomes established, RA is cleared from the PS (Ribes et al., 2009) and the principal regulator of embryonic pluripotency Oct4 (encoded by Pou5f1) begins to be downregulated (Osorno et al., 2012). PG5-9 Hox genes are activated mainly by Wnt/Fgf, rather than RA, via the action of Cdx2/4 (Amin et al., 2016; Gouti et al., 2014; Hackland et al., 2019; Lippmann et al., 2015; Mazzoni et al., 2013; Metzis et al., 2018; Young et al., 2009) and will take at least 24 h to reach maximal expression levels (Gouti et al., 2017; Wymeersch et al., 2019). Wnt/Fgf activities, acting via T/Cdx2 and the simultaneous lowering of RA signalling, lead to the induction of NMPs (Amin et al., 2016; Garriock et al., 2015; Gouti et al., 2017, 2014; Turner et al., 2014; Wymeersch et al., 2016). At this stage, the pluripotency factors Oct4 and Sall4 are re-purposed to act in maintenance and/or differentiation of axial progenitors (Aires et al., 2016; Tahara et al., 2019). Oct4 in particular, plays a subsequent role in stimulating trunk growth by delaying the activation of posterior Hox PG10-13 members, which are associated with the trunk-to-tail transition (Aires et al., 2016) (see below).

This stage coincides with an increase in the numbers of axial progenitors/NMPs under the influence of Fgf and Wnt3a/β-catenin signalling (Garriock et al., 2015; Wymeersch et al., 2016), which are also responsible for a maximal increase in PG4-9 gene expression. Interestingly, expression of PG4-9 members Hoxa5 and Hoxb8 (in combination with Wnt signalling) stimulate trunk elongation (Young et al., 2009), suggesting that trunk growth is linked to the increase in PG4-9 expression between the 5-10-somite stage. Expression of Gdf11, a BMP/TGFβ family member that is crucial for the relocation of axial progenitors from the NSB/CLE/PS regions to the tail bud (Jurberg et al., 2013), is upregulated in the CLE (Nakashima et al., 1999), leading to the downregulation of Pou5f1 and further upregulating of Cyp26a1 (Aires et al., 2016; Jurberg et al., 2013). RA activity is almost absent from the posterior end (Sirbu and Duester, 2006). Hox PG10-12 genes, expression of which begins at this stage, are also Wnt/Fgf responsive, require the absence of RA and Oct4, and are stimulated by Gdf11 (Aires et al., 2016; Lippmann et al., 2015; Mazzoni et al., 2013). Moreover, Wnt3a induces Wnt5a via Cdx binding (Amin et al., 2016; Takada et al., 1994).

This is the stage at which Oct4 is absent (Osorno et al., 2012) and RA activity is undetectable (Cunningham et al., 2011; Sakai et al., 2001), potentially facilitating maximal expression of PG10-12 genes. NMP numbers peak at this stage (Wymeersch et al., 2016). Hox PG13 genes are induced at this point, possibly driven by Gdf11 (Aires et al., 2019) and Fgf signalling activity (Hackland et al., 2019). Canonical Wnt/β-catenin activity decreases, whereas Wnt5a levels remain high, promoting tail outgrowth (in collaboration with Wnt11), gradually releasing NMPs from the progenitor region EMT (Andre et al., 2015).

This phase is marked by an increase in the expression of Hox PG13 members, which bind to Cdx2 target sequences, reduce Wnt-driven T expression and consequently NMP numbers (Amin et al., 2016; Denans et al., 2015), leading to the eventual exhaustion of progenitors for axial elongation. Genes upregulated at E13.5 (Wymeersch et al., 2019) may affect the final extinction of axial progenitors, as well as tail bud remodelling to remove the neural tube, notochord and tail bud mesoderm.

In other vertebrates, including those with highly divergent somite numbers such as snakes, the size of the presomitic mesoderm and somites has been shown to peak midway through axis elongation (Gomez et al., 2008). This argues that, as in mouse, a stepwise mechanism dependent on multiple temporal feedbacks is likely to operate in animals with diverse body plans. For example, the onset of Gdf11 expression has been correlated with the position of the hindlimb primordia in various species (Matsubara et al., 2017) and extended Oct4 activity in snake, with a long trunk and relatively short neck and tail, may support trunk expansion (Aires et al., 2016).

The limited numbers of microdissected embryonic axial progenitors is a major obstacle to the study of their genetic and epigenetic regulation, for example through biochemical approaches requiring large amounts of starting material. Moreover, obvious ethical and practical reasons preclude the study of axial progenitors in human embryos. The in vitro generation of large numbers of axial progenitors from PSCs offers an attractive alternative approach to addressing these issues. Furthermore, in vitro-derived axial progenitors can act as a promising source of posterior spinal cord and mesodermal cell types, specifically at thoracic and lumbosacral levels, which could be employed for disease modelling and cell replacement applications. Early hallmarks of in vitro-derived NMP-like cells, such as the emergence of T⁺Sox2⁺ cells and upregulation of transcripts marking NMP-containing regions (e.g. Cdx2, Fgf8 and Hox genes), were first reported in mouse epiblast stem cell (mEpiSCs) cultures treated with the Wnt agonist/glycogen synthase kinase 3 (GSK3)-inhibitor CHIR99021 (CHIR) in combination with FGF2 and activin A (Tsakiridis et al., 2014). This subpopulation formed a distinct minor entity within cultures dominated by T⁺Foxa2⁺ mesendodermal progenitors. Interestingly, when cells from these cultures were grafted into the NSB of E8.5 mouse embryos they colonised both the neural tube and paraxial mesoderm, indicating their bipotency (Tsakiridis et al., 2014).

A number of reports have been subsequently published describing improved protocols for the directed differentiation of both human (hPSCs) and mouse PSCs (e.g. mouse embryonic stem cells; mESCs and mEpiSCs) into T⁺Sox2⁺ cultures (Amin et al., 2016; Cooper et al., 2020 preprint; Cunningham et al., 2016; de Lemos et al., 2019 preprint; Denham et al., 2015; Diaz-Cuadros et al., 2020; Edri et al., 2019a; Edri et al., 2019b; Frith et al., 2018; Gouti et al., 2014; Hackland et al., 2019; Kirino et al., 2018; Kumamaru et al., 2018; Lippmann et al., 2015; Row et al., 2018; Tsakiridis and Wilson, 2015; Turner et al., 2014; Verrier et al., 2018) (Table 3). Despite variations in terms of treatment duration, media composition and choice of signalling agonists/coating substrates, all protocols for the in vitro generation of axial progenitors rely on the stimulation of the Wnt and, often, Fgf signalling pathways (Fig. 8), thus recapitulating the signalling environment of progenitor niches in vivo (Table 2; Fig. 5). Resulting PSC-derived cell populations expressed caudal markers (Edri et al., 2019b; Frith et al., 2018; Gouti et al., 2017, 2014; Verrier et al., 2018), exhibited the ability to generate neural and mesodermal cell types in vitro (Frith et al., 2018; Gouti et al., 2014; Turner et al., 2014) and/or contributed to both the neural tube and paraxial mesoderm after engraftment into host chick or mouse embryos (Baillie-Johnson et al., 2018; Edri et al., 2019a; Gouti et al., 2014) (Table 3). Embryo grafting in these cases has provided a useful assay for the developmental potential of in vitro-derived axial progenitors, although the early somite mouse embryo appears to offer a more stringent host environment for distinguishing between NM bipotency versus pluripotency compared with their late gastrula chick counterparts (Baillie-Johnson et al., 2018; Gouti et al., 2014; Huang et al., 2012; Tsakiridis et al., 2014).

NMP-like cells have also been reported to arise in a regionalised manner, in 3D self-organising aggregates of PSCs following a short timed pulse of CHIR (Beccari et al., 2018; Faustino Martins et al., 2020; Libby et al., 2020 preprint; Turner et al., 2014; van den Brink et al., 2020; Veenvliet et al., 2020). These organoid-like structures show signs of A-P and dorsoventral axis formation, polarised collinear Hox gene activation and somitogenesis (Beccari et al., 2018; van den Brink et al., 2020; Veenvliet et al., 2020), and thus make an attractive model to study early morphogenetic axis elongation events. These data collectively suggest NMP-like cells can be generated in vitro from PSCs and these can act as a source of differentiated axial tissue derivatives that resemble their in vivo counterparts. However, the various in vitro-derived NMP-like populations that have been reported appear to be quite different, both at a transcriptome level and in terms of their differentiation potential. This is likely because of the use of differing culture regimes and starting PSC populations (mESCs, mEpiSCs, hPSCs) resulting in the generation of progenitor populations corresponding to a continuum of developmental states, ranging from the late gastrula caudal epiblast up to tail bud stages (Edri et al., 2019a,b; Frith et al., 2018; Gouti et al., 2017; Gouti et al., 2014; Verrier et al., 2018). Furthermore, the culture conditions that induce T⁺Sox2⁺ cells from PSCs also appear to promote the emergence of LPMPs and intermediate mesoderm progenitors, as revealed by the expression of associated markers (Edri et al., 2019a; Row et al., 2018), indicating the induction of heterogeneous cultures that contain a mixture of posterior axial progenitor subpopulations.

It should be noted that, thus far, none of these protocols has demonstrated the existence of truly bipotent cells, as defined by the ability of a single cell to generate both neural and mesodermal derivatives. Previously, we attempted to interrogate the potency of single cells via the clonal plating of FACS-sorted T-GFP reporter-positive mEpiSC-derived NMP-like cells in the presence of Wnt and Fgf agonists. Through this approach we showed that mixed neural and mesodermal colonies could be generated from single T-GFP⁺ cells, although the majority of the resulting colonies consisted exclusively of Sox2⁺ neural cells, suggesting either a neural bias of NMPs or neural-favouring plating/culture conditions (Tsakiridis and Wilson, 2015). Moreover, in vitro-derived NMP-like cells are a transient entity both in vivo and in vitro, and attempts to maintain homogeneous cultures of T⁺Sox2⁺ cells over extensive time periods (>3 passages) in the presence of Wnt and Fgf agonists (i.e. the conditions promoting their induction) have thus far been unsuccessful. In all cases, cultures tend to gradually downregulate T and other NMP-specific markers and differentiate into their downstream derivatives (Cooper et al., 2020 preprint; Edri et al., 2019a; Gouti et al., 2017; Tsakiridis and Wilson, 2015; Wind et al., 2020 preprint).

Interestingly, single-cell RNA-sequencing data have revealed a subpopulation of cells expressing node-associated transcripts (T, Foxa2, Noggin, Chordin, Shh) that co-emerges with NMP-like cells during their induction from mEpiSCs (Edri et al., 2019b). Some of these markers have also been detected in hPSC-derived NMP-like cultures (Denham et al., 2015; Frith et al., 2018). It is tempting to speculate, based on evidence from co-culture experiments (Edri et al., 2019b), that these node-like cells facilitate the maintenance of a T⁺Sox2⁺ NMP identity in vitro and their progressive elimination upon prolonged culture may be responsible for the exhaustion of NMP-like cells. Additional experiments are required to properly dissect cell-autonomous versus non-autonomous aspects of NMP self-renewal in vitro.

The A-P axial identity of PSC derivatives is likely to shape their functionality and developmental/regenerative potential, as demonstrated by xenotransplantation experiments in animal models (Kadoya et al., 2016; Peljto et al., 2010). As various congenital birth defects and neurodegenerative conditions affect certain cell types in an axial level-specific manner (Box 2), their in vitro modelling relies not only on generating from PSCs cell populations of the correct lineage, but also the appropriate axial identity (Allodi et al., 2019; An et al., 2019; Gordon et al., 2014; Vega-Lopez et al., 2018).

Until recently, the directed differentiation of PSCs towards neural derivatives has been heavily influenced by Nieuwkoop's ‘activation-transformation’ model and the idea that anterior neurectoderm is induced ‘by default’ in the absence of posteriorising signals, which can be further patterned by RA/Wnt signals to induce progressively more posterior neural cells (Chambers et al., 2009; Nieuwkoop and Nigtevecht, 1954). However, this approach predominantly gives rise to hindbrain or anterior (cervical/brachial) spinal cord derivatives and fails to efficiently generate cells of a more caudal identity (e.g. thoracic/lumbosacral) in vitro (Imaizumi et al., 2015; Kirkeby et al., 2012; Maury et al., 2015; Meinhardt et al., 2014). The in vitro derivation of NMP-like cells paved the way for a new generation of differentiation protocols that instead rely on the production of a Wnt/Fgf-induced T⁺Sox2⁺ NMP/posterior pre-neural intermediate state from PSCs, and its subsequent ‘neuralisation’ via addition of RA and inhibition of TGF/Nodal and/or BMP signalling pathways (Fig. 8). This strategy has led to the production of neurectodermal cells corresponding to the thoracic and lumbosacral regions (Cooper et al., 2020 preprint; Frith et al., 2018; Gouti et al., 2014; Kumamaru et al., 2018; Lippmann et al., 2015; Mouilleau et al., 2020 preprint; Verrier et al., 2018; Wind et al., 2020 preprint). Interestingly, these cells can be passaged and expanded as spinal cord neural stem cells over extended time periods (Cooper et al., 2020 preprint; Kumamaru et al., 2018) and further directed to produce neurons of distinct dorsoventral identities (Cooper et al., 2020 preprint; Kumamaru et al., 2018; Lippmann et al., 2015; Verrier et al., 2018; Wind et al., 2020 preprint). More recent reports have combined micropatterning and 3D organoid-based approaches with the induction of NMP-like cells in order to increase reproducibility, dissect signalling dynamics in spinal cord cells, or study interactions between NMP derivatives such as skeletal muscle and spinal cord neurons in neuromuscular organoids (Duval et al., 2019; Faustino Martins et al., 2020; Knight et al., 2018; Mouilleau et al., 2020 preprint). Evidence from transplantation experiments in a rat spinal cord injury model has suggested that posterior spinal cord cells derived from PSCs via an NMP intermediate may offer a promising avenue toward the development of cell replacement therapies (Kumamaru et al., 2018), yet further characterisation of these cells is required to better define their value in a clinical context.

Wnt/Fgf-triggered induction of NMP-like cells is accompanied by sequential upregulation of Hox family members up to PG9, thus reflecting a predominantly posterior brachial and thoracic character (Lippmann et al., 2015) (Fig. 6). A shift to more posterior Hox PGs, corresponding to lumbosacral levels, can only be achieved through increased Fgf signalling activity and/or treatment with GDF11, a key regulator of trunk-to-tail transition during embryonic axis elongation, as well as caudal Hox expression (Aires et al., 2019; Hackland et al., 2019; Jurberg et al., 2013; Lippmann et al., 2015; Liu et al., 2001; Mouilleau et al., 2020 preprint). However, despite the reported production of lumbar motor neurons through this approach (Lippmann et al., 2015; Mouilleau et al., 2020 preprint), there is still limited quantitative/functional evidence regarding its efficiency, and more research is needed to understand the optimal conditions for generating tissues of defined sections of the postcranial axis.

PSC-derived T⁺Sox2⁺ cells are also an efficient source of posterior neural crest (NC), the multipotent embryonic cell population that arises in the dorsal neural plate/non-neural ectoderm border region and acts as a source of various cell types including peripheral neurons. Similar to central nervous system derivatives, the conventional route to obtain NC cells in vitro has relied on the production of an anterior neurectodermal intermediate that is subsequently steered towards NC fate through stimulation of Wnt/BMP signalling (to recapitulate the neural plate border environment). However, this method has been inefficient in producing trunk or lumbar NC (Hackland et al., 2017; Lee et al., 2007; Menendez et al., 2011). Wnt/Fgf-treated hPSC cultures contain a T⁺ subpopulation that exhibits features of early NC precursors arising at the border of the caudal neural plate in vivo. These findings have led to the establishment of differentiation protocols for the efficient generation of trunk NC cells and their sympathoadrenal derivatives in vitro (Abu-Bonsrah et al., 2018; Cooper et al., 2020 preprint; Denham et al., 2015; Frith et al., 2018; Frith and Tsakiridis, 2019; Gomez et al., 2019; Hackland et al., 2019). The resulting trunk NC cells are derived from T⁺ cells within hPSC-derived NMP-like cultures (Frith et al., 2018). Moving forward, single-cell lineage-tracing experiments are required to map more precisely the segregation dynamics of posterior NC, spinal cord neurectoderm and paraxial mesoderm, as well as the origin of their progenitors both during PSC differentiation and in vivo.

A number of recent PSC-differentiation protocols have also been shown to efficiently produce paraxial mesoderm and skeletal muscle via the early induction of T⁺Sox2⁺ NMP-like cells (Fig. 8) (Chal et al., 2016; Diaz-Cuadros et al., 2020; Loh et al., 2016; Sudheer et al., 2016; Xi et al., 2017). These studies have provided valuable insights into the molecular and signalling basis of early paraxial mesoderm specification (Diaz-Cuadros et al., 2020; Loh et al., 2016; Sudheer et al., 2016) opening new avenues toward cell replacement and disease modelling applications (Chal et al., 2016). As axial progenitors closer to the node (e.g. NMPs) give rise preferentially to medial somites, whereas their more caudal PS counterparts tend to exhibit more lateral somite contribution (Cambray and Wilson, 2007), it would be interesting to exploit such PSC-based systems in order to dissect the effect of various signals on the propensity of axial progenitors to generate medial versus lateral somite derivatives.

PSC differentiation studies have also indicated that endogenous BMP signalling is crucial in regulating the balance between paraxial mesoderm (requires BMP inhibition) and lateral plate mesoderm/NC cell (BMP-dependent) fates within in vitro-derived NMP-like cultures (Frith et al., 2018; Loh et al., 2016; Row et al., 2018; Xi et al., 2017). These data are in line with findings from in vivo studies showing that the balance between Fgf and BMP signalling (and their downstream bHLH effectors) in mesodermal progenitors is crucial in assigning a mediolateral fate (Miura et al., 2006; Row et al., 2018). Moreover, T⁺ axial progenitors, induced from PSCs following Wnt stimulation, appear to be competent to generate intermediate mesoderm and, subsequently, Wolffian duct cells and nephric mesenchyme (Taguchi et al., 2014; Taguchi and Nishinakamura, 2017; Takasato et al., 2015). These data are consistent with lineage-tracing experiments in mouse embryos showing that E7.5 and E8.5 T⁺ axial progenitors give rise to the Wolffian duct and meso/metanephric mesenchyme, respectively (Taguchi et al., 2014), as well as embryo grafting studies in which only contribution to the nephrogenic mesenchyme was observed from the CLE (Cambray and Wilson, 2007; Wymeersch et al., 2016).

Embryonic axis elongation has been in the research spotlight for a long time, being a valuable paradigm of a crucial, evolutionarily-conserved developmental process. The discovery that it is driven by stem cell-like progenitors with the ability to generate both neural and mesodermal derivatives has challenged conventional views about the nature of cell lineage, germ layers and multipotency. The combination of classic embryology techniques with recently established 2D and 3D PSC-based in vitro systems and cutting-edge imaging/single-cell sequencing approaches has provided unprecedented insights into the cellular and molecular basis of axial progenitor fate decisions and behaviour. These will continue to be powerful tools in future efforts to increase our understanding of axis elongation and associated birth defects, especially when used in conjunction with genetic perturbation experiments, controlled artificial microenvironments (e.g. through the use of micropatterning/microfluidics platforms), and may include input from other disciplines such as mathematics, physics and engineering. The next obvious research direction will involve the implementation of spatial transcriptomics and quantitative hybridization chain reaction methods (Choi et al., 2018; Eng et al., 2019; Junker et al., 2014; Rodriques et al., 2019) to record gene expression profiles of single embryonic axial progenitor cells in situ. Such studies are crucial for the definitive spatiotemporal mapping of the interplay between niche signals, downstream transcriptional read-outs and cell behaviour (fate/potency) within the native context of the embryo. Moreover, these could be expanded to include also the analysis of chromatin landscape/higher order chromosomal organisation changes associated with distinct phases of axis elongation. To further address key questions on how tissue scaling, de-coding of positional information and homeostasis of axial progenitor pools are coordinated during the construction of the body plan, the comparative study of a wide range of diverse and novel animal models is essential, extending earlier findings from previous cross-species studies (Aires et al., 2016; Gomez et al., 2008; Steventon et al., 2016). Finally, additional work building on key findings from pioneering studies that indicate the role of tissue mechanics (Mongera et al., 2018), signalling-gene expression dynamics (Lauschke et al., 2013; Matsuda et al., 2020; Sonnen et al., 2018) and metabolism (Bulusu et al., 2017; Oginuma et al., 2017) in embryonic axis elongation is required.

We would like to thank Kate G. Storey and David Kimelman for sharing immunostaining images and Anahi Binagui-Casas for critical reading of the manuscript. We apologise to colleagues whose work could not be cited owing to space constraints.