Nodal signaling has dual roles in fate specification and directed migration during germ layer segregation in zebrafish

Gastrulation is central to animal development and involves the specification of three different germ layers (endoderm, mesoderm and ectoderm) and their segregation to different locations in the embryo (Wolpert, 1992). In contrast to the mechanisms underlying cell fate specification, the mechanisms used to drive segregation of the three germ layers are much less well understood. In this work, we focus on endodermal cells, which are initially specified on the surface of the embryo but must segregate to the interior, where they give rise to the gut and associated tissues. Endoderm migration is crucial for the formation of the gut tube and digestive tract across the animal kingdom. The in-folding of surface blastoderm cells to form the endoderm is well-documented in a wide range of species (Stern, 2004; Wolpert, 1992). However, it has been experimentally difficult to separate the initiation of migration events from cell fate specification. Thus, the molecular logic of the cell internalization, including which signals trigger this migration and how cell fate and migration are related, still remain unclear.

Several models have been proposed for how the germ layers segregate during embryogenesis. Most prominently, the differential adhesion hypothesis proposed that differences in intercellular adhesion among the different germ layers drives sorting (Steinberg, 1962). However, although differential adhesion and cortex tension have been observed in vitro, in vivo measurements of tissue surface tension were indistinguishable among the three germ layers. Thus, differential adhesion is unlikely to fully account for the ability of the germ layers to sort in the embryo (Krieg et al., 2008; Maître et al., 2012; Krens et al., 2017).

Previous studies have shown that directed cell migration appears to be the driving force for endoderm segregation in vivo for zebrafish (Montero et al., 2005; Krens et al., 2017; Giger and David, 2017). At the onset of zebrafish gastrulation, the blastoderm consists of several thousand cells positioned above the yolk cell. Internalization begins on the dorsal side where inward-moving cells form the hypoblast (mesoderm and endoderm) in contrast to the cells remaining on the outside as epiblast (ectoderm) (Warga and Kimmel, 1990). A germ ring forms at the boundary of hypoblast and epiblast and the embryonic shield is formed on the dorsal side of the margin. Early dye labeling experiments showed that cells relocate to deeper levels within the germ ring by inverting their order relative to the margin as they internalize (Kimmel and Warga, 1987). Initially, an involution model was proposed to describe the population flow as a cellular sheet (Trinkaus, 1984). Later, time-lapse tracking showed that individual cells within the germ ring transiently move out of the epiblast and relocate into the hypoblast (D'Amico and Cooper, 1997; Concha and Adams, 1998). More recent studies have shown that such cells extend protrusions inward and exhibit active directed migration (Montero et al., 2005; Krens et al., 2017; Giger and David, 2017).

Nodal, as a member of the TGFβ superfamily, is essential for germ layer patterning in zebrafish. Nodal ligands are expressed at the margin and yolk syncytial layer (YSL) during the blastula stage, where it forms a morphogen gradient (Chen and Schier, 2001; Dougan et al., 2003). The signaling pathway is activated by Nodal binding to a type II TGFβ receptor, inducing interaction with an EGF-CFC co-receptor, Teratocarcinoma-derived growth factor 1 (Tdgf1; also known as One-eyed-pinhead), and the type I TGFβ receptor Acvr1ba (Weng and Stemple, 2003; Gritsman et al., 1999; Aoki et al., 2002b). Subsequent phosphorylation of the transcription factors Smad2 and Smad3 facilitates the formation, together with Smad4, of a Smad complex that translocates into the nucleus to regulate the expression of target genes (Weng and Stemple, 2003; Jia et al., 2008). One of the key downstream targets is gene encoding Sox32, which plays an essential cell-autonomous role in endoderm formation (Kikuchi et al., 2001).

Compared with endoderm specification, the signals that initiate and direct endoderm migration are not as well understood. From previous studies, it is known that endodermal cells initially undergo random walk migration but switch to convergence movements at mid-gastrulation (Pézeron et al., 2008). Endodermal migration is also regulated by chemokine signaling downstream of the Nodal pathway (Nair and Schilling, 2008; Mizoguchi et al., 2008). We recently demonstrated that Nodal signaling regulates endodermal cell motility and actin dynamics via Rac1 and Prex1 (Woo et al., 2012). However, it is not known whether these migration patterns arise due to endodermal cell fate alone or whether additional cues in their morphogenic field are required.

Here, we utilized an in vivo system to study germ layer segregation in zebrafish embryos. In the early zebrafish embryo, an initially mixed mesendodermal population ultimately resolves into distinct mesodermal and endodermal cell layers, but these complex morphogenetic movements occur simultaneously with fate specification (Ho, 1992). To disentangle the endodermal specification program from the migration program, we used a constitutively active version of the Nodal receptor, acvr1ba*, to predispose cells into an endodermal fate (Renucci et al., 1996; Aoki et al., 2002b; David and Rosa, 2001). By transplanting these ectopically induced endodermal cells into the animal pole of the embryo, we removed them from the endogenous signals that normally orchestrate endodermal development as well as the effects of nearby ingressing cells. We found that these ectopically introduced endodermal cells do not take the normal path of endogenous endoderm migration by internalizing at the germ ring; instead, they radially ingress into the inner layer. Nodal signaling is necessary and sufficient to initiate this process, and the ectopic endodermal cells (but not the surrounding cells) need to receive the Nodal ligand in an autocrine fashion to trigger ingression. Our results suggest that Nodal signaling plays dual roles in specifying endodermal fate and initiating the sorting of these cells to the interior of the embryo. As these migration events are not observed for in vitro culture conditions, this in vivo approach for endodermal sorting should be a powerful system for continued dissection of the logic of germ layer segregation during gastrulation.

To determine the requirements for initiating and directing endoderm migration, we developed a cell transplantation model that allowed us to directly query endodermal sorting while disentangling the endodermal specification program from the migration program. We generated ectopic endodermal cells by expression of the constitutively activated Nodal receptor acvr1ba*. We then transplanted these cells (Fig. 1A) into the animal pole of a wild-type host embryo, far from the marginal location of endogenous endodermal cells, and determined whether these misplaced ectopic endodermal cells could sort into the correct endodermal layer (Fig. 1B). First, ectopic endoderm production by acvr1ba* was confirmed by qPCR analysis of sox17 and sox32 expression (Fig. S1A), markers for endodermal cell fate (Kikuchi et al., 2001; Shivdasani, 2002). Although Nodal signaling can also induce mesoderm fate (Peyriéras et al., 1998; Aoki et al., 2002b), we found that acvr1ba* expression upregulated the mesodermal markers gsc and tbxta (also known as ntl) to a lesser extent than sox17 and sox32 (Fig. S1B), demonstrating that acvr1ba*-expressing cells are biased to an endoderm fate. Next, we found that, after transplantation to the animal pole, these ectopically introduced endodermal cells accumulated in endoderm-derived tissue by preferentially migrating to the correct endodermal layer (Fig. 1B-E). When induced endodermal cells were transplanted together with non-endodermal cells (Fig. 1F), these cell types separated into two layers from an originally mixed population (Fig. 1G, Movie 1). Visualizing the migration path of these cells by time-lapse microscopy showed that induced endodermal cells did not move towards the margin and then involute to form endodermal layer (the normal path of endogenous endoderm migration); instead, they radially ingressed into the inner layer (Fig. 1G), consistent with a recent report (Giger and David, 2017). Additional single-cell tracking analysis revealed that the trajectories of transplanted cells did not exhibit random walk or sample both inward and outward directions; instead, the ingression was highly unidirectional (Fig. 1H, Movie 2). These data indicate that endodermal cells produced by acvr1ba* expression, if placed ectopically, can initiate ingression via highly polarized and unidirectional migration.

We next sought to define the molecular logic of ectopic endodermal cell sorting. In addition to expressing the constitutively activated Acvr1ba* receptor, ectopic endodermal cells can also be produced by overexpression of the transcription factor Sox32, a target of Nodal signaling (Dickmeis et al., 2001; Kikuchi et al., 2001; Sakaguchi et al., 2001). However, previous work suggested that, unlike Acvr1ba* expression, overexpression of sox32 is not sufficient to drive the sorting process (Kikuchi et al., 2001). We too observed that cells overexpressing sox32 could preferentially segregate to endoderm-derived tissues when placed near the dorsal margin but not when transplanted to the animal pole (Fig. 2A,B, Fig. S2).

Notably, this means that these two different means of generating endodermal cells (acvr1ba* versus sox32) are not equivalent in their ability to drive internalization movements when transplanted far from the normal endodermal domain (Fig. 2C); only acvr1ba*-induced endodermal cells are capable of ingression when placed at the animal pole. These data suggest that Nodal signaling initiates endodermal sorting in addition to specifying endodermal fate. acvr1ba* likely activates additional pathways that are absent when sox32 is overexpressed, thus allowing cells to sort regardless of the location within the embryo. In contrast, cells overexpressing sox32 may require extrinsic factors present at the margin to activate these additional ‘sorting’ pathways, explaining why they can only sort in regions close to the margin (Kikuchi et al., 2001). These observations suggest that the triggering of sorting involves an ‘AND’ gate consisting of sox32-dependent endodermal specification and additional signaling downstream of acvr1ba* (Fig. 2D).

To confirm the necessity of sox32-dependent arm of the putative AND gate, we injected acvr1ba*-induced endodermal cells with a sox32 morpholino (MO), which blocks the transcriptional program that initiates endodermal specification (Sakaguchi et al., 2001; Dickmeis et al., 2001) (Fig. S4B). We found that these cells lost the ability to ingress into the inner layer of the embryo after transplantation to the animal pole of a wild-type host (Fig. 2G,H). Single-cell tracking revealed that morpholino against sox32 (Fig. S5) inhibits the ability of acvr1ba* cells to migrate to the interior of the embryo. These results suggest that both autocrine production of Nodal ligands and sox32-dependent endodermal specification are necessary to trigger ectopic endodermal sorting.

To better understand the differences between these two methods of generating ectopic endoderm, we compared the signaling and transcriptional networks activated by acvr1ba and sox32. A recent report (Giger and David, 2017) suggested that N-cadherin (Cdh2) expression triggers endoderm ingression. However, we found that both acvr1ba* and sox32 overexpression induced cdh2 expression (Fig. S1C) to similar extents. This suggests that N-cadherin expression alone does not account for the difference of ingression capability between these two types of endodermal cells.

Nodal signaling through acvr1ba, but not sox32 alone, is capable of inducing expression of Nodal ligands ndr1 and ndr2 (Chan et al., 2009; Feldman et al., 1998; Dougan et al., 2003). In wild-type embryos, these ligands are expressed at highest levels near the margin, which could explain the observation that sox32-induced endodermal cells only sort in this location. In contrast, the autocrine production of Nodal ligands downstream of acvr1ba* could enable these cells to sort regardless of location in the embryo. To confirm that ndr1/2 is secreted by acvr1ba*-expressing cells but not by cells expressing sox32 alone, we quantified the ndr1/2 expression profile under all experimental conditions (Fig. S3). When transplanted to wild-type embryos, we saw a 2- to 3-fold increase of ndr1/2 expression in acvr1ba*-expressing cells compared with the sox32-expressing cells. However, this is likely to be an underestimate of the difference in Nodal ligand expression induced by acvr1ba* or sox32 expression because of the presence of maternally deposited ndr1/2 in the host embryo that can also trigger a Nodal positive-feedback loop. To address this complication, we expressed acvr1ba* and sox32 in maternal-zygotic (MZ) tdgf1 mutant embryos, which lack sufficient Nodal signaling (Gritsman et al., 1999) (Fig. S3B). In this background, we found that acvr1ba* expression increased ndr1 expression to a level 25-fold higher than sox32 expression. Finally, injection of ndr1/2 mRNA increased the expression of ndr1/2 in sox32-expressing cells to levels similar to those seen in acvr1ba*-expressing cells (Fig. S3C). Taken together, these data confirm that acvrb1a*, but not sox32, induces Nodal ligand expression.

Next, we sought to test whether the excess production of Nodal ligands is the driver of ingression. We first examined the necessity of Nodal ligand expression for endodermal sorting by using morpholinos to knockdown ndr1 and ndr2 in the acvr1ba*-induced endodermal cells, which were then transplanted into the animal pole of wild-type embryos (Fig. 2E). We verified the functionality of the morpholinos by demonstrating the inhibition of sox17 expression (Fig. S4). At 20 hours post-fertilization (hpf), whereas cells expressing acvr1ba* only preferentially localized to endodermal tissues such as the pharynx, the acvr1ba* cells with ndr1 and ndr2 MO knockdown primarily localized in non-endodermal tissue, particularly in the head region (Fig. 2F,H). Single-cell tracking revealed that the acvr1ba* cells with ndr1 and ndr2 MO remain in the ectoderm and move near the surface of the embryo (Fig. 2I). Notably, the failure to ingress caused by knockdown of ndr1 and ndr2 is unlikely to be due to effects on endoderm specification as cells maintained endoderm identity as assessed by Tg(sox17:GFP) reporter expression (Fig. 2J,K).

To further test whether ndr1/2 secretion is necessary for sorting, we analyzed the internalization dynamics of acvr1ba*- or sox32-overexpressing cells in host embryos injected with ndr1/2 MO. After transplantation into the margin of a ndr1/2-depleted host, acvr1ba*-expressing cells localized to the endoderm-derived tissue (Fig. S6). In contrast to our animal pole transplantation experiments, sox32-overexpressing cells transplanted to the margin of ndr1/2-depleted hosts were able to contribute to both endoderm and ectoderm-derived tissues, although the extent of endoderm contribution was significantly less compared with acvr1ba* cells (Fig. S6). Because there are still global morphogenesis movements happening at the margin of the host embryos, it is possible that some transplanted cells are internalized along with their host cell neighbors. Such community effects were previously observed for transplanted MZ tdgf1 mutant cells that could initially internalize with their wild-type neighbors (Carmany-Rampey and Schier, 2001). This might account for the increased percentage of endoderm contribution for marginally versus animal pole transplanted sox32-overexpressing cells.

In zebrafish, the two Nodal ligands Ndr1 and Ndr2 are known for their functional redundancy in inducing mesendoderm fate (Feldman et al., 1998; Erter et al., 1998; Jing et al., 2006). But it is not known whether they behave redundantly to induce ingression. To address this question, we performed transplantation experiments of acvr1ba*-expressing cells with either ndr1 MO or ndr2 MO alone. Our results showed that neither ndr1 MO or ndr2 MO abolished the ingression behavior of acvr1ba*-expressing cells (Fig. S7). These data suggest that these Nodal ligands act redundantly to support the ingression of acvr1ba*-expressing cells.

Next, we investigated whether addition of Nodal ligands could trigger ectopic endodermal cell sorting in sox32-induced endodermal cells, which were not able to ingress into the inner layer when they are transplanted to the animal pole (Fig. 3A). To test this, we injected donor cells with mRNA for both sox32 and Nodal ligands (ndr1, ndr2) prior to transplantation into wild-type embryos, using acvr1ba*-induced endodermal cells as a positive control (Fig. 3B). When examined at the 18-somite stage, sox32-injected cells also expressing Nodal ligands accumulated in endoderm-derived tissue significantly better than cells expressing sox32 alone (Fig. 3D,F,G). These data indicate that the expression of Nodal ligands is sufficient to confer sorting ability in sox32-induced endodermal cells.

Can any cell expressing Nodal ligands sort to endodermal tissues, or do cells require both endodermal specification and Nodal ligand expression to support sorting? Because Nodal ligands themselves can drive endodermal fate (Chen and Schier, 2001; David and Rosa, 2001; Dougan et al., 2003), we addressed this question by overexpressing Nodal ligands in conjunction with sox32 MO (Fig. 3C). We found that Nodal ligands cannot support sorting in the sox32 MO background (Fig. 3E,G). These data suggest that Nodal ligands can only trigger sorting in conjunction with sox32-dependent endodermal specification. Together, our necessity and sufficiency experiments demonstrate that Nodal ligands and sox32 constitute an ‘AND’ gate to initiate internalization in the early embryo.

So far, we have shown that Nodal ligand production is necessary and sufficient to trigger the ingression-based cell sorting of ectopic endodermal cells. We next determined which cells are responding to the Nodal ligands to support sorting. The Nodal ligands could either be acting on the same endodermal cells that undergo sorting to form an autocrine circuit or on the surrounding cells in a paracrine circuit, possibly by orchestrating endodermal extrusion by the surrounding ectoderm (Fig. 4A,B). We blocked the autocrine reception of Nodal ligands pharmacologically by applying the Nodal receptor inhibitor SB505124 to acvr1ba*-expressing cells. We found that 25-50 µM SB505124 inhibited endodermal cell fate specification, even in embryos expressing acvr1ba*, suggesting that this compound interfered with the signaling circuit upstream of acvr1ba (Fig. S8A-C). Following pharmacological Nodal inhibition, acvr1ba*-expressing cells failed to internalize after being transplanted to the animal pole (Fig. S8D). To inhibit only the autocrine reception of the Nodal ligands while maintaining endodermal cell fate, we used the MZ tdgf1 mutant to block Nodal signal reception; this mutant lacks the EGF-CFC co-receptor essential for the ability to respond to Nodal ligands (Gritsman et al., 1999). When we transplanted MZ tdgf1 donor cells expressing acvr1ba* into wild-type recipient hosts, the donor cells were incapable of ingressing into the inner layer of the host embryo and did not contribute to endoderm-derived tissue at the 18-somite stage (Fig. 4C,F). In contrast, when we injected acvr1ba* into wild-type donor embryos and transplanted these cells into MZ tdgf1 mutant host embryos, these transplanted cells still successfully ingressed into the inner layer, indicating that ectopic endodermal cells retained their ability to sort irrespective of the Nodal signaling state of the surrounding cells (Fig. 4D,G). Together, these results suggest that an autocrine circuit of Nodal ligand reception is required to support sorting of ectopic endodermal cells (Fig. 4E). Furthermore, because the MZ tdgf1 mutant host embryos lack endogenous endoderm (Gritsman et al., 1999; David and Rosa, 2001) but still supported ingression of ectopic endodermal cells, these experiments further suggest that signals released by endogenous endodermal cells are not required for ectopic endodermal cell sorting.

It is possible that blocking autocrine Nodal reception, either by pharmacological treatment or loss of tdgf1, would also decrease Ndr1/2 production and inhibit ingression. However, given that nuclear accumulation of Smad2 is not significantly different in acvr1ba*-expressing embryos either with or without ndr1/2 MO (Fig. 2K), a small decrease in Ndr1/2 production is unlikely to affect the magnitude of Nodal signal to an extent that would impact migration. Therefore, altered levels of Nodal activation do not account for the role of autocrine Ndr1/2 in endodermal internalization. It is more likely that autocrine reception of Nodal ligands may activate other signaling pathways downstream of acvr1ba* as an input to the ‘AND’ gate for sorting.

In our transplant experiments, we observed ectopic endodermal cells moving from the outer layer of the embryo radially to the inner layer, but we never observed cells moving in the opposite direction (i.e. extruded from the embryo). A recent study has shown that during normal gastrulation movements, endodermal cells at the margin extend polarized protrusions toward the yolk syncytial layer and appear to internalize by active migration (Giger and David, 2017). Therefore, we investigated whether ectopically placed endodermal cells similarly undergo active, directed migration to enter the interior of the embryo. To visualize actin dynamics in the ectopic endodermal cells during sorting, we expressed GFP-UTRN (Burkel et al., 2007), an actin reporter that we have previously used to analyze endodermal actin dynamics in zebrafish (Woo et al., 2012). We transplanted GFP-UTRN-labeled and acvr1ba*-induced endodermal cells to the animal pole of wild-type hosts and imaged actin dynamics during ingression. We divided each single transplanted cell into two sectors, one facing towards the interior of the embryo and the other facing towards the embryo surface and then quantified the accumulation of actin in each sector. We observed a significant accumulation of actin in the interior-facing sector of ectopic endodermal cells as well as actin-based protrusions extending towards the interior of the embryo. Control transplanted cells not expressing acvr1ba* lacked this polarity of actin enrichment and protrusions (Fig. 5A,B). In contrast, acvr1ba*-expressing cells injected with ndr1/2 MOs exhibit non-polarized protrusions, suggesting that protrusion generation may be normal but their orientation may be defective in the absence of ndr1/2 (Fig. S9). Together, these data indicate an asymmetry in protrusion polarization in ectopic endodermal cells but in those with ndr1/2 knockdown. This indicates that ndr1 and ndr2 are necessary to direct the actin-enriched protrusions, consistent with sorting based on active migration.

Which spatial cues are ectopic endodermal cells reading to achieve their directional migration? Such cues are unlikely to arise from the endogenous endodermal cells, as ectopic endoderm can still sort in an MZ tdgf1 host that lacks endogenous endoderm (Fig. 4E). Might the endogenous Nodal gradient of the host embryo set the direction for ectopic endodermal cell migration? To investigate this hypothesis, we used ndr1/2 MO to knock down endogenous Nodal ligands in host embryos (Fig. 5C). We transplanted acvr1ba*-expressing donor cells to the animal pole and found that they maintained their ability to ingress into the inner layer of the embryo (Fig. 5D,G). These data suggest that the endogenous Nodal is not necessary to trigger the sorting behavior. Conversely, we saturated the endogenous Nodal gradient by overexpressing ndr1 and ndr2 ligands in the host embryo (Fig. 5E). As before, the acvr1ba*-induced endodermal cells retained their ability to ingress into the inner layer (Fig. 5H,I) after transplantation. Together, these data suggest that although autocrine Nodal reception is essential for initiating internalization, neither the endogenous endodermal cells nor the endogenous Nodal ligands spatially direct ingression.

In this work, we investigated the molecular signals that initiate the movement of endodermal cells from the surface to the interior of the embryo during zebrafish development. By leveraging the ability of ectopically induced endodermal cells to sort to the endogenous endodermal domain, we dissected the molecular logic of sorting without the confounding influences of fate specification and global morphogenetic movements at the margin. Our work shows that an autocrine circuit of Nodal activated by acvr1ba* is both necessary and sufficient to trigger internalization of endodermal cells (Figs 2H, 3D and 4E). Neither the endogenous Nodal gradient nor endogenous endodermal cells are required to direct the sorting process. Our work defines an ‘AND’ gate consisting of sox32-dependent endodermal specification and Nodal ligand reception that initiates the internalization process (Fig. 3A).

Most of the focus on Nodal signaling during endoderm development has centered around its role in fate specification (Aoki et al., 2002a; Hagos and Dougan, 2007; Dubrulle et al., 2015). Here, we identified an additional role for Nodal as a signaling molecule that regulates endodermal sorting. Endodermal cells that either lack the ability to generate Nodal ligands or to receive Nodal ligands fail to undergo internalization when transplanted to the animal pole (Figs 2H and 4E). Normally, both endoderm specification and Nodal ligand reception occur in the same location in the embryo near the margin, and activation of this ‘AND’ gate (endodermal specification+Nodal ligand reception) could help specify when and where the internalization process occurs. The requirement for both Nodal ligands and endodermal specification could prevent non-endodermal cells that transiently receive Nodal ligands from internalizing. The autocrine nature of this circuit could help control the timing of internalization, which could be triggered when differentiation has proceeded sufficiently to drive this positive-feedback loop. This positive-feedback loop, in which cells that receive Nodal ligand input release more Nodal ligand, has previously been implicated in the large-scale self-organization of the Nodal field (Chan et al., 2009), and our work demonstrates an additional role for this feedback loop in coordinating endodermal cell sorting. This circuit could also enable the multicellular coordination of internalization. In chick embryos, single-cell ingression can be amplified to induce more of the epiblast to undergo ingression (Voiculescu et al., 2014). Such community effect is Nodal dependent and underlies the formation of primitive streak.

Nodal ligands are received through the Nodal receptor Acvr1ba and its co-receptor Tdgf1 (the zebrafish homolog of TDGF1/Cripto). The constitutively active Nodal receptor Acvr1ba* has frequently been used to investigate the Nodal signaling pathway (Schier and Shen, 2000; Gritsman et al., 1999; Warga and Kane, 2003). Surprisingly, we find that acvr1ba* requires additional autocrine production and reception of Nodal ligands to support endodermal sorting (Fig. 4). Why might this be? One possibility is that internalization is only triggered above a certain threshold of Nodal signaling. For wild-type cells, this signaling threshold might only be achieved at the margin, where Nodal expression is highest, whereas cells expressing acvr1ba*, in which Nodal signaling is activated beyond wild-type levels, can reach the thresholds needed for internalization even at positions far from the margin. However, in either case (wild type or acvr1ba*), these high signaling levels are achieved by a Nodal-induced Nodal expression positive-feedback loop. In acvr1ba*-induced endodermal cell, Smad2 activation levels are comparable with and without ndr1/2 knockdown (Fig. 2J,K), suggesting that acvr1ba* can initiate Nodal signaling independent of Nodal ligands. In future experiments, this model could be tested by perturbing Smad function to varying degrees in the presence of acvr1ba* and assessing the effects on ingression ability. An alternative explanation for the role of a Nodal autocrine circuit could be activation of tdgf1, which may have signaling roles that are independent of acvr1ba. This model would be consistent with previous literature showing that acvr1ba* can only partially rescue tdgf1 loss of function (Warga and Kane, 2003). In future work, it will be interesting to examine further the differential signaling engaged by acvr1ba* in the absence and presence of tdgf1 to identify the tdgf1-specific effectors that could participate in endoderm migration.

We dissected the role of Nodal as a trigger for endodermal cell internalization. Through experiments with MZ tdgf1 as a background for donor and host, we found that ectopic endodermal cells trigger sorting in an autocrine manner. By labeling the actin dynamics, we observed basal enrichment of actin-based protrusions, consistent with other reports suggesting that endodermal cells internalize through active migration (Giger and David, 2017). Previous work in hydra also demonstrated the ability of individual endodermal cells to migrate towards the center of ectodermal aggregates, suggesting that invasion of endodermal cells into ectoderm may represent an ancient morphogenetic behavior (Takaku et al., 2005).

But which spatial cues are these cells reading to migrate towards the interior of the embryo? We ruled out endogenous endodermal cells as an attractive positional cue because acvr1ba*-expressing cells can ingress in the MZ tdgf1 background, which lacks endogenous endodermal cells. Moreover, a functioning Nodal gradient does not exist in the MZ tdgf1 host embryos, suggesting that Nodal itself is not providing positional cues either. This latter point was further demonstrated in our experiments either knocking down the endogenous Nodal gradient or flooding the embryo with uniform Nodal level, both of which failed to block ingression. The intersection of endodermal specification and Nodal ligand reception could unlock the ability of these cells to read other extracellular cues that are polarized from the outside to the inside of the embryo, such as soluble ligands, ECM components and mechanical cues (Piccolo, 2013; Brunet et al., 2013). Apela (also known as Toddler and Elabela) functions as a motogen and enhances the movement of mesodermal and endodermal cells through Apelin receptor signaling, and Nodal is known to activate Apelin receptor expression (Pauli et al., 2014). However, we found that morpholino-directed knockdown of Apelin receptors a and b did not affect the ability of acvr1ba*-induced cells to ingress into the interior of the embryo (Fig. S10), suggesting that Apela is unlikely to be the spatial cue. Alternatively, the cells could be responding to intrinsic polarity cues, such as an oriented apical-basal polarity followed by apical constriction. Consistent with this idea, Xenopus bottle cells and Caenorhabditis elegans endodermal progenitor cells have apical-basal polarity and activate apical constriction to initiate gastrulation movements (Nance and Priess, 2002). Clearly, additional work is needed to resolve this question.

This work aims to understand the molecular cues that initiate endodermal internalization and germ-layer sorting. In addition to laying the foundation for coordinated cell movement at the primary site of endodermal cell internalization during normal development, single-cell ingression may also function as a backup plan to ensure that endodermal cells that are specified late or otherwise miss initial ingression can still find a path into the inner layer. Given that this sorting behavior is based on an autocrine circuit, endodermal cells can still ingress even if they are no longer adjacent to the margin, and this could increase the precision of the first step of endoderm morphogenesis.

From previous work on zebrafish morphogenesis, it is known that dorsal endodermal cells migrate highly asynchronously, which could lead to challenges in germ layer segregation (Keller et al., 2008). Cell sorting is thought to enable systems with initially noisy fate specification to generate robust final patterns. One extreme example is Dictyostelium, in which the initial differentiation decision into prestalk or prespore cell is random, and differential migration is responsible for the final pattern (Dormann et al., 2000). Similarly, during neural tube formation in zebrafish, heterogeneous Sonic hedgehog responsivity is sharpened by neural progenitor cells sorting into discrete domains (Xiong et al., 2013). If migration were random, it would be expected to blur the boundaries between different germ layers for cells responding to a source of morphogen such as Nodal. In contrast, by linking directed migration to cell fate specification and signaling, this movement may instead improve the precision of the overall process. By establishing the necessary and sufficient triggers for endodermal sorting in vivo, our approach should be useful for continuing to define the logic of endodermal sorting during zebrafish gastrulation.

Zebrafish maintenance was carried out under standard laboratory conditions in the zebrafish facility at Smith Cardiovascular Research Institute. Embryos were grown at 28-31°C in egg water and staged as described previously (Kimmel et al., 1995). The following wild-type, mutant and transgenic lines were used: (wild type) AB/TL; (mutant) tdgf1^tz57/+ (a generous gift from Lilianna Solnica-Krezel’s lab in Washington University in St. Louis, MO, USA); (transgenic) Tg(sox17:GFP)^s870, Tg(sox17:DsRed)^s903Tg(h2afva:h2afva-mCherry)^tud7, Tg(ubb:GFP-Smad2)^sfc16. Tg(sox17:GFP)^s870 and Tg(sox17:DsRed)^s903 (Chung and Stainier, 2008; Mizoguchi et al., 2008) and Tg(h2afva:h2afva-mCherry)^tud7 (Knopf et al., 2011) have been previously described. To construct Tg(ubb:GFP-Smad2)^sfc16, transgene plasmid mTol2-ubiq:GFP-Smad2 was created by separate PCR amplification of the ubiquitin promoter and GFP ORF and then cloned into pmTol2-ef1a:Venus-Smad2 (gift from Steve Harvey, Wellcome Trust/CR-UK Gurdon Institute and Department of Zoology, The University of Cambridge, Cambridge, UK) cut with EcoRV and AgeI to remove the ef1a promoter and Venus ORF. Tg(ubb:GFP-Smad2)^sfc16 was created by injecting 20 pg of the transgene plasmid DNA along with 100 pg of Tol2 transposase mRNA at the one-cell stage. Injected embryos were then sorted by fluorescence on day 0, raised to adulthood, and then screened for founders by outcrossing to wild type. Please refer to Table S2 for a full list of strains used in this study.

To create maternal zygotic tdgf1 mutants, tdgf1^tz57/+ parents were in-crossed, and all embryos were injected with tdgf1 mRNA so that homozygous embryos could survive. Genotyping was performed according to established protocols (Hashimoto et al., 2000; Pogoda et al., 2000).

Capped messenger RNA was synthesized using the mMESSAGE mMACHINE kit (Ambion). The following expression plasmids were used in this study: acvr1ba* in pCS2 (pCS2-acvr1ba*-tBFP), full-length zebrafish sox32 in pCS2 (pCS2-sox32; Chung and Stainier, 2008), ndr1 and ndr2 in pCS2 independently and GFP-UTRN in pCS2. The sox32 MO was designed to target the translation initiation site and was used at 2 ng (5′-CCTCCTCAGTGTTTATTTCGCTCAT-3′). ndr1 MO (5′-ATGTCAAATCAAGGTAATAATCCAC-3′) and ndr2 MO (5′-GCGACTCCCGAGCGTGTGCATGATG-3′) were used at 4 ng. MOs targeting aplnra (5′-CGGTGTATTCCGGCGTTGGCTCCAT-3′) and aplnrb (5′-CAGAGAAGTTGTTTGTCATGTGCTC-3′) were injected at the one-cell stage at 1 ng or 0.5 ng, respectively. Please refer to Table S2 for a full list of constructs used in this study.

mRNA, morpholino and dye injections were performed with a micromanipulator connected to Picospritzer III. Drop size was regulated by the duration and pressure of the pulse. mRNA of appropriate concentration for different genes was injected into the yolk of the embryo at the one-cell stage. To obtain induced endodermal cells, 0.5 pg acvr1ba* mRNA or 100 pg sox32 mRNA were injected into the embryo. To study the effect of Nodal ligands on ingression, 4 pg ndr1 and ndr2 mRNA were injected into the embryo. To visualize actin dynamics, 200 pg GFP-UTRN was injected into the embryo. Morpholinos were briefly incubated at 65°C to prevent precipitation and then injected into the yolk before the first cell division. To inhibit the translation of the corresponding genes, 4 ng ndr1, 2 ng ndr1 MO and 2 ng ndr2 MO (Feldman and Stemple, 2001; Karlen and Rebagliati, 2001), 2 ng sox32 MO (Sakaguchi et al., 2001), 1 ng aplnra MO or 0.5 ng aplnrb MO (Scott et al., 2007; Paskaradevan and Scott, 2012; Pauli et al., 2014) were injected into the embryo. Dyes including Dextran-FITC or Dextran-tetra-methyl-rhodamine-dextran (TMR-dextran) or Dextran-Alexa Fluor 680 (Life Technologies) were injected at 1 ng at the one-cell stage to label whole cells and 1 ng Histone H1-Alexa Fluor 488 conjugate (Thermo Fisher Scientific) was injected into the embryo at the one-cell stage to label the nucleus. Please refer to Table S2 for a full list of reagents used in this study.

For Nodal-activated conditions, wild-type embryos were injected at the one-cell stage with 2 pg acvr1ba* mRNA or mCherry mRNA as a control. Expression of sox17, a known Nodal target gene, as well as sox32, were used to confirm Nodal activation. Expression of cdh2 was measured under different Nodal-activated conditions. At shield stage, total RNA was extracted using the RNAqueous-Micro Kit, and 1 ng was used for reverse transcription with the SuperScript VILO cDNA Synthesis Kit (Invitrogen). The quantitative PCR reaction mixture contained 2 µl of 10-fold-diluted cDNA, 12.5 µl SYBR green PCR master mix (Applied Biosystems), 714 nM of each primer, and nuclease-free water for a total volume of 25 µl in 48-well plates (Ilumina). Reactions were performed in the Eco Real-Time PCR System (Illumina) as follows: initial activation at 95°C for 10 min followed by 40 cycles of 30 s at 95°C, 30 s at 60°C, and 30 s at 68°C. Once the PCR was completed, a melt-curve analysis was performed to determine reaction specificity. Samples were run in triplicate. The housekeeping gene ef1a was used as a reference. Refer to Table S1 for a list of oligonucleotides used in this study.

Donor and host embryos were dechorionated with forceps under a dissection stereomicroscope and transferred into a glass plate with 0.3× Ringer's Buffer. Approximately 25-50 cells were taken from a dechorionated donor embryo(s) at sphere stage (4 hpf) and transplanted into the animal pole of a dechorionated host at the same stage using a beveled borosilicate needle with a 35 μm inner diameter attached to a syringe system. In single donor transplantation experiments, the donor embryo was injected with mRNAs and/or morpholinos described in the main text and wild-type or MZ tdgf1 embryos were used as hosts. In double donor transplantation experiments, the endoderm donor embryo was injected with 2 pg acvr1ba* mRNA, control ectoderm donor embryo was injected with 2 ng sox32 morpholino, and wild-type embryos were used as the host. Dextran dyes were used to differentiate donor versus host cells. H1-Alexa Fluor 488 conjugate was used to label the nucleus for single-cell tracking. After transplantation, embryos were either immediately mounted for microscopy or maintained in 0.3× Ringer's Buffer at 28-31°C for further analysis.

For pharmacological treatment, embryos from one dish were removed at the desired stage and split into glass dishes containing the drug in 5 ml embryo medium, at a density of 25 embryos/dish. For SB505124, the lowest dose that produced the sqt; cyc phenotype ranged from 30 to 50 μM, depending on the age of the drug (Hagos and Dougan, 2007). Desired concentration is diluted from 10 mM stock. For transplants, drug treatment is initiated after the transplantation is finished at 4 hpf.

Dechorionated embryos, immediately after transplantation, were embedded in 1% low-melting agarose within glass-bottomed Petri dishes, with animal pole mounted towards the glass bottom. For tracking, transplanted embryos were imaged with a 20×/0.75 NA Plan Fluor multi-immersion objective with water as the immersion media. For actin dynamics visualization, a 40×/1.15 NA water immersion objective was used. A 10×/0.45 NA Plan Apo λ objective was used for imaging 24 hpf or 18-somite stage embryos. A high-speed widefield Nikon spinning disk confocal microscope was used for all imaging. This microscope is equipped with an Andor Borealis CSU-W1 unit, an Andor DU-888 EMCCD camera, and a stage-top incubator unit from OkoLab. Andor 4-line laser launch (100 mW at 405, 561 and 640 nm; 150 mW at 488 nm) was used for excitation. Micro-Manager Open Source Microscopy Software Version 2.0 Beta was used to control the microscope. Image stacks of 70-150 μm with 1-2 μm (1 μm for time lapse and 2 μm for end-point scanning) z stack were recorded in continuous mode, resulting in an image sampling rate of 2-4 min. Embryos were kept at 28.5°C throughout imaging.

Tracking with Gaussian Mixture Models (TGMM) software for automated large-scale segmentation and tracking of fluorescently labeled cell nuclei from the Keller Lab was adapted for single-cell tracking of the transplanted cells (Amat et al., 2014). Time-lapse datasets with z-stacks were rendered into 3D tracks and filtered by track length. A sphere was used for modeling the zebrafish embryo, and Cartesian coordinates were transformed into spherical coordinates to determine the radial distance traveled by the transplanted cells.

Quantification of the percentage of transplanted cells that localized to endodermal versus non-endodermal domains was performed by analyzing images with Fiji. z-stack images were converted to maximum intensity projections and thresholded by Renyi entropy. Particles were analyzed with Fiji and size of regions of interest were measured. For image re-slicing, z-stack images were re-sliced to achieve 1×1×1 voxel size, then converted to maximum intensity projections to generate an xz projection. Statistical data analyses were performed using Student's t-test in Matlab.

We thank Steve Harvey for plasmid (pmTol2-ef1a: Venus-Smad2), Lilianna Solnica-Krezel for providing the tdgf1 zebrafish line, Philipp Keller for providing guidance on TGMM tracking software and Alba Diz-Muñoz for feedback on the manuscript. We are grateful to Jeremy Reiter, Wallace Marshall, Sophie Dumont and the Weiner group for fruitful discussions, and the zebrafish facility at CVRI and NIC at UCSF for assistance.