A dual function of FGF signaling in Xenopus left-right axis formation

Left-right (LR) asymmetry of inner organs is present in all deuterostome lineages and depends on the left-sided activation of the evolutionarily highly conserved Nodal signaling cascade. In fish, amphibian and mammalian embryos, the symmetry-breaking event, which activates this asymmetric gene cascade in the lateral plate mesoderm (LPM), is represented by cilia-based extracellular fluid flow at the left-right organizer (LRO) of the early neurula embryo (Dasgupta and Amack, 2016; Yoshiba and Hamada, 2014). The flow-generating LRO epithelium consists of mono-ciliated cells of mesodermal fate, which are transiently embedded in the endodermal endothelium of the forming archenteron at the posterior end of the notochord (Blum et al., 2007). In amphibians, it is of triangular shape and termed the gastrocoel roof plate (GRP; Blum et al., 2009b). The Xenopus GRP is subdivided into a medial flow-generating part, with posteriorly localized motile cilia, and lateral cells with non-polarized and immotile cilia, which presumably sense flow. The former cells will later integrate into notochord and hypochord, whereas the latter will contribute to somites (Boskovski et al., 2013; Shook et al., 2004). Somitic GRP cells are characterized by co-expression of nodal and the Nodal inhibitor gene dand5. Flow is thought to downregulate dand5, de-repress Nodal and induce transfer of an asymmetric signal to the left LPM (Schweickert et al., 2010). The GRP derives from the superficial mesoderm (SM; Shook et al., 2004) of the early gastrula, which represents the outer-most layer of the Spemann organizer. The SM is marked by the expression of the transcription factor gene forkhead box J1 (foxj1; Blum et al., 2014; Stubbs et al., 2008), which induces motile ciliogenesis, and the Nodal-related growth factor gene nodal3 (nodal homolog 3; previously known as xnr3; Smith et al., 1995; Beyer et al., 2012).

A number of signaling pathways have been implicated in LR development, although conserved functions have only been described for a few, such as Nodal (Namigai et al., 2014). Conflicting results have been obtained for FGF signaling. In chick and rabbit embryos, the ligand Fgf8 seems to act as a right determinant, repressing asymmetric gene expression in the right LPM (Boettger et al., 1999; Feistel and Blum, 2008; Fischer et al., 2002). In the mouse, in contrast, Fgf8 acts as a left determinant. Here, FGF signaling is required both for nodal gene expression in the sensory crown cells of the LRO at about the 2-somite stage, and for asymmetric gene expression in the LPM later on (Meyers and Martin, 1999; Oki et al., 2010). In addition, FGF signaling has been implicated as a regulator of vesicle release at the LRO; these vesicles transfer through flow to the left side, where they activate Nodal signaling (Tanaka et al., 2005). In teleost fish, FGF signals were shown to be required for symmetry breakage upstream of leftward flow. Loss of FGF signaling resulted in reduction or loss of LRO cilia in specimens harboring five to ten pairs of somites, and, consequently, LR defects later on (Hong and Dawid, 2009; Neugebauer et al., 2009; Yamauchi et al., 2009). Reduced ciliary lengths were reported in the Xenopus GRP, but GRP morphogenesis, foxj1, flow and Nodal cascade genes have not been analyzed (Neugebauer et al., 2009). Thus, the role of FGF signaling seems to vary in the different model organisms analyzed to date.

Here, we studied the role of FGF signaling in LR axis formation in Xenopus. The frog is accessible to manipulation and analysis at all stages of LR development, from SM specification to GRP morphogenesis, leftward flow, Nodal cascade induction and organ morphogenesis (Blum et al., 2014, 2009a). Through pharmacological and molecular manipulation of Fgfr1 signaling, we identified a dual role of FGF signaling. During early gastrulation, FGF signaling was necessary for SM specification upstream of foxj1 but downstream of nodal3. During late gastrulation, a second, cilia-independent function was involved in morphogenesis of the lateral flow sensor of the LRO, independently of foxj1. Our data suggest that this late phenotype was caused by interfering with FGF/Ca²⁺ signaling, which additionally required input from the calcium channel Polycystin-2. Our work reconciles conflicting findings in model organisms and demonstrate a conserved role of FGF signaling in LR axis formation, in much the same dual way as has been shown for mesoderm induction and morphogenesis.

In order to determine the critical time frame of FGF signaling in Xenopus LR development, we treated embryos of defined stages with SU5402, an inhibitor of the tyrosine kinase activity of Fgfr (Fig. 1A). SU5402 incubation at late blastula or early gastrula stages (until stage 10.5) impaired gastrulation movements leading to blastopore closure defects, reduced somitic expression of myogenic differentiation 1 (myod1), and a shortened anterior-posterior axis, as previously described for inhibition of FGF signaling (Fig. 1B; Amaya et al., 1991). Concomitantly, FGF-dependent mesodermal marker genes were strongly (myogenic factor 5, myf5; Fig. S1A-C) or moderately (myf5; myod1; brachyury/T-box transcription factor T, tbxt; Fig. S1D-L) reduced at late gastrulation when treated from late blastula (stage 9) or early gastrula (stage 10) onwards, respectively. Mesodermal defects at tadpole stages could be minimized by treatment at mid gastrulation (stage 11/11.5), and circumvented when treated from late gastrula stages onwards (st.12; Fig. 1B; data not shown). In all cases, however, i.e. from early, mid or late gastrulation onwards, FGF inhibition resulted in absence of pitx2c expression in the left LPM (Fig. 1; Schweickert et al., 2000). Treatments at later stages did not impact on laterality (Fig. 1C). Inhibition of FGF signaling did also not impair the competence of LPM tissue to respond to Nodal, as injection of a nodal1 DNA construct into the LPM of SU5402-treated embryos rescued pitx2c expression (Fig. 1C). These experiments revealed a sensitive time window for FGF signaling in LR axis development from late blastula until late gastrulation, i.e. before establishment of leftward flow and, thus, upstream of nodal1 activation in the left LPM.

Loss of asymmetric pitx2c expression could be caused by lack of symmetry breaking due to impaired leftward flow. Gastrulation defects, however, prevented the analysis of leftward flow in embryos treated before stage 10.5 (Fig. 1B). We therefore decided to analyze the specification of the precursor tissue from which the LRO derives, namely the SM, using mRNA expression of the known marker genes nodal3, foxj1 and wnt11b. This reasoning was further supported by our recent demonstration that knockdown of the Fgfr1 ligand nodal3 in the SM inhibited foxj1 induction (Vick et al., 2018; Yokota et al., 2003). Early SM-specific expression of nodal3 was unaltered when treated from late blastula onwards, confirming its dependence on canonical Wnt but independence of FGF signaling pathways, as demonstrated previously (Fig. 2A,B,G; Fletcher and Harland, 2008; Smith et al., 1995). In agreement with this notion, embryos treated with SU5402 at early gastrula (stage 10.5) showed a significant reduction of foxj1 and wnt11b in the SM, confirming a role of Nodal3/FGF in setting-up a functional GRP (Fig. 2C-G). Endodermal expression of wnt11b was unaffected by SU5402 treatment, underscoring the FGF specificity of this effect (2E-F′). Injection of a dominant-negative version of Fgfr1 (dnfgfr1; Amaya et al., 1991) mimicked SU5402 treatment and resulted in a significant loss of foxj1 expression as well (Fig. S2A,B,D). Together, these experiments demonstrated that FGF signaling was required for induction of gastrulation movements and in parallel for induction of foxj1 and wnt11b expression in the SM. Because earlier nodal3 expression was unaffected by SU5402 treatments, the initial SM identity seemed not to be perturbed, but an early FGF signal was required at the beginning of gastrulation for further SM specification.

Interestingly, SU5402 incubation at late gastrula stages had no impact on foxj1, in agreement with the robust expression of foxj1 at these stages (Fig. 2H,I; cf. Beyer et al., 2012; Stubbs et al., 2008). Injection of low doses of dnfgfr1 did not cause a significant loss of foxj1 expression either, nor did it block gastrulation movements, although pitx2c expression in the left LPM was partially lost (Fig. S2C-E). At neurula stages, expression of the Foxj1 target gene tektin2 (tekt2) in the GRP was apparently unaffected by late-gastrula-stage SU5402 incubations. The tekt2 expression domain, however, appeared narrower than that of control embryos (Fig. 2J,K). We therefore analyzed cilia formation and functionality by scanning electron microscopy (SEM) and flow analysis, to test for effects downstream of ciliary gene activation. SEM analysis did not reveal significant differences in cilia length, GRP ciliation or posterior polarization of flow-generating motile cilia between treated and control specimens (Fig. S3A-E). Low dose dnfgfr1 injections did not alter ciliary length either (Fig. S3F). Cilia motility and leftward flow were also unaffected, as demonstrated by flow analysis of DMSO- and SU56402-treated GRP explants from stage 17/18 embryos (Fig. S3G-I; Schweickert et al., 2007). Together, these data hint at two different modes of action of FGF signaling in Xenopus LR axis development: a cilia- and foxj1-dependent function at early gastrulation, and a later, cilia-independent mode of action downstream of flow and upstream of nodal1 expression in the LPM.

Next, we investigated whether signaling downstream of flow was impaired in embryos treated at late gastrula stages, i.e. whether the left-asymmetric flow signal was perceived and transferred to the LPM. In wild-type (WT) embryos, polarized motile cilia at the center of the GRP generate flow, which is perceived by immotile and central cilia on lateral GRP cells (Boskovski et al., 2013; Schweickert et al., 2007; Shook et al., 2004). At the molecular level, the GRP in its entirety can be highlighted by expression of ciliary marker genes, such as tekt2 (Stubbs et al., 2008) or dynein axonemal heavy chain 9 (dnah9; Schweickert et al., 2007), or, indirectly, by the lack of endodermal marker gene expression, such as alpha-2-macroglobulin-like 1 (a2ml1, also known as panza; Pineda-Salgado et al., 2005; Fig. S4A-C). Sensory lateral GRP cells are characterized by the co-expression of both nodal1 and its inhibitor dand5 (Fig. S4D; Schweickert et al., 2010; Vonica and Brivanlou, 2007), whereas flow-generating central GRP cells express sonic hedgehog (shh; Fig. S4E). The combined analysis of shh (central GRP) and a2ml1 (endoderm) thus highlights the lateral, sensory part of the GRP by an absence of staining (Fig. S4F). SU5402 incubation at stage 12 resulted in lack of both nodal1 and dand5 expression (Fig. 3A,B,D,E,G), but effects were greatly reduced upon later incubations from stage 13 onwards (Fig. 3C,F,G). To confirm that lateral, sensory GRP cells were specifically absent upon late SU5402 treatment, we analyzed embryos from the same batches in parallel for expression of further marker genes. Whereas untreated embryos showed the signal-free area between the shh and a2ml1 expression domains, in addition to WT nodal1 expression (which together unequivocally mark the lateral GRP cells), about 50% of specimens treated at stage 12 lacked both nodal1-positive cells and the shh/a2ml1-negative region between central GRP and endoderm. Again, treatments from stage 13 onwards impacted only slightly (in 5%) on the presence of the sensory GRP (Fig. S4G-L). The lack of somitic GRP cells was not due to interfering with mesodermal maintenance during neurulation, as only early but not late treatments resulted in reduction of expression of mesodermal marker genes brachyury (tbxt) and myod1 (Fig. S4M-R). Confirming the specificity of this effect on the sensory GRP, low-dose dnfgfr1 injections in a sided manner resulted in a significant reduction or loss of nodal1 and dand5 on the injected side (Fig. S4S-X). In line with this notion, activation of FGF signaling by injection of the ligand fgf8b into the lateral GRP was sufficient to increase the expression of both nodal1 and dand5 in a highly significant proportion of specimens (Fig. 3H-L). Overexpression was achieved through injections of a DNA construct, which only gets transcribed after activation of the zygotic genome at late blastula stages, a procedure chosen to prevent earlier FGF-induced developmental defects. Together, these experiments pointed to a function of FGF signaling during late gastrulation to promote the lateral, somitic fate of the GRP.

As both nodal1 and dand5 expression were lost, we wondered if FGF inhibition prevented activation of these genes, or whether the late FGF signal was required for formation of sensory cells. These cells in due course detach from the gastrocoel roof and integrate into the presomitic mesoderm (Shook et al., 2004). Their somitic fate is marked already in the GRP by expression of myod1 (Fig. 4A,B,E; Schweickert et al., 2010). This contribution of myod1-positive cells to the GRP was essentially lost in specimens treated with SU5402 from stage 12 onwards (Fig. 4C,E), to a lesser degree when incubation started at stage 13 (Fig. 4D,E). This loss was quantified by assessing the widths of GRP subpopulations in transversal sections at the level of the posterior aspect of the GRP. The relative widths of the central GRP were not different between control and SU5402-treated specimens. The mean widths of the lateral, somitic part of the GRP, however, were reduced to less than 20% of the values of control specimens upon late FGF inhibition (Fig. 4E). Treatments from stage 13 onwards showed some reductions in widths as well (Fig. 4D′,E), although LR development was not affected. Low doses of dnfgfr1 injections likewise resulted in a lack of myod1-positive lateral GRP cells on the injected side (Fig. 4F), demonstrating specificity of effects.

Next, we wondered whether proliferation and/or apoptosis were involved in FGF-mediated GRP patterning, as the FGF pathway impacts on both processes (Schlessinger, 2000). The proliferation rate in the GRP was assessed in explants at two stages (13/14 and 16), using a phospho-histone3-specific antibody. No differences were recorded between untreated samples and specimens incubated with SU5402 from stage 11.5 onwards (Fig. S5A-C; Saka and Smith, 2001). Apoptosis was unaffected by FGF inhibition as well, as demonstrated by terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) staining of control and SU5402-treated embryos at stages 13/14 or stage 16. Here, we found apoptosis in the neuroectoderm, as previously reported (Yeo and Gautier, 2004), but never at the GRP (Fig. S5D,E). In agreement with the notion that apoptosis was not involved in FGF-mediated specification of lateral GRP cells, injection of the anti-apoptotic B-cell CLL/lymphoma (bcl-XL) DNA construct, which inhibits chemically induced apoptosis (Johnston et al., 2005), into the lineage of the lateral GRP was unable to rescue SU5402-induced loss of nodal1/dand5-positive GRP cells (Fig. S5F-K). Thus, neither apoptosis nor altered proliferation within the GRP are causative for the observed lack of somitic GRP cells upon late-gastrula-stage loss of FGF signaling.

In a last set of experiments, we analyzed the morphological distortions of GRP morphogenesis and patterning following FGF inhibition by SEM. Using GRP explants from DMSO- or SU5402-treated embryos (Fig. 5A), three traits were analyzed: cilia polarization, tissue morphology and vesicle abundance. Ciliated GRP cells are easily distinguishable from the much larger non-ciliated lateral endodermal cells (LECs; Fig. 5B). As previously reported (Schweickert et al., 2010), cilia at the central part of the GRP are posteriorly polarized, whereas lateral cells harbor central cilia (Fig. 5B′,B″). Cilia polarization was facilitated by viewing SEM preparations of GRP explants at an angle (cf. Fig. S6A). SU5402-treated GRPs lacked lateral GRP cells harboring non-polarized cilia (Fig. 5C,D; Fig. S6B), such that GRP cells with polarized cilia were directly bordered by LECs (Fig. 5C′,D). Quantification of cilia polarization revealed the same percentage of posteriorized cilia in central GRPs of DMSO-treated and whole GRPs of SU5402-treated specimens. The fraction of about 50% non-polarized cilia in lateral GRP cells of WT specimens (Fig. 5E) was lost following SU5402 treatment. With respect to overall tissue morphology, there is a clear separation between central and lateral parts of WT GRPs, enabling a direct assessment of tissue widths (Fig. 5B; Fig. S6C): cells align at the border between central and lateral GRP areas, giving rise to an almost straight line separating the two cell populations, as previously observed by Shook et al. (2004). SU5402-treated GRPs were found to be narrower, encompassing just the widths of WT central GRPs (Fig. 5F; Fig. S6B). Finally, higher magnification of SEM photographs revealed the presence of small vesicles across the apical surface of all GRP cells (Fig. S6C). Upon close inspection, we noticed that vesicles were enriched on lateral GRP cells, which offered itself for direct quantification. GRPs from SU5402-treated specimens showed lower vesicle abundance compared with untreated specimens (Fig. S6D), in line with a loss of lateral GRP cells. Of note, vesicle abundance did not differ between left and right somitic cells in WT GRP cells (Fig. S6E). SEM analysis of dnfgfr1-injected embryos revealed a strong reduction of lateral GRP cells on the injected side as well, confirming the specificity of SU5402 treatment (Fig. 5G). Finally, unilateral injection of fgf8b resulted in a marked increase of lateral GRP widths (Fig. 5H). Together, these analyses demonstrated that Fgfr-mediated signaling is required for the morphogenesis of lateral, flow-sensing GRP cells during late gastrulation.

A dual role of FGF signaling has been previously described in the context of mesoderm induction and subsequent morphogenesis (Nutt et al., 2001; Sivak et al., 2005). Specifically, MAPK-dependent FGF signaling is necessary during early/mid gastrulation for mesoderm induction and maintenance. At mid-gastrulation, the PKC-dependent FGF/Ca²⁺ branch is required for morphogenetic gastrulation movements, which, however, does not affect mesoderm induction. The FGF antagonist sprouty is expressed during early gastrulation; overexpression does not impact on MAPK-mediated mesoderm induction, but blocks FGF/Ca²⁺-activation during late gastrulation, i.e. specifically regulates FGF-mediated morphogenesis (Nutt et al., 2001; Sivak et al., 2005). The dual role of FGF signaling in LR axis formation made us wonder whether the FGF/Ca²⁺ branch was likewise involved in the FGF-mediated specification of the sensory GRP.

In order to test this possibility, we injected spry1 mRNA to target the SM and lateral GRP cells specifically. Sprouty overexpression did not affect foxj1 expression in the SM, nor subsequent activation of its target tekt2 in the GRP (Fig. 6A-F). Mesoderm induction and mesodermal marker gene expression (tbxt, myod1) at gastrula and neurula stages were unaffected, as previously described (Fig. S7A-F; Sivak et al., 2005). However, the width of the tekt2 expression domain was reduced in about 45% of injected specimens following spry1 injections (compare Fig. 6D and E). Concomitantly, expression of nodal1 and dand5 were lost on the injected side in about 50% of unilaterally injected specimens (Fig. 6G-L). The loss of the lateral GRP cells upon unilateral spry1 injection was confirmed by SEM analysis, which revealed about 75% of embryos with loss (Fig. S7G,I) or strong reduction of lateral GRP cell populations (Fig. S7H,I). Together, these results unveil an inhibitory effect of Sprouty proteins on the morphogenesis of the LRO flow sensor, i.e. somitic GRP cell formation during late gastrulation, suggesting a requirement of the FGF/Ca²⁺ pathway in this process.

Like FGF, pkd2 is required in a dual way during symmetry breakage in Xenopus. In a first stage, it cooperates with the FGF ligand nodal3 to induce foxj1 in the SM. Later, it is required for somitic GRP cells and initiation of nodal1 and dand5 (Vick et al., 2018). As pkd2 encodes the calcium channel Polycystin-2 (Busch et al., 2017; Koulen et al., 2002), we reasoned that it may interact synergistically with the FGF/Ca²⁺ pathway. To investigate this possibility, we injected a characterized pkd2 antisense morpholino oligomer (MO; Tran et al., 2010) to target the nodal1/dand5-expressing somitic GRP cells. The MO prevented expression of both genes specifically on the injected side, as described (Fig. S8A-D; cf. Vick et al., 2018). To test for genetic interaction, we lowered both pkd2-MO dose and concentration of spry1 mRNA such that nodal1 and dand5 expression were only mildly affected on the injected side (Fig. 7A-F). Co-injections of sub-phenotypic doses, however, resulted in a highly significant reduction of both genes (Fig. 7D-F). These experiments indicated that pkd2 loss- and sprouty gain-of-function impact on the same process, namely somitic GRP-induction. As Sprouty presumably blocks the FGF/Ca²⁺ pathway upstream of endoplasmic reticulum-mediated intracellular Ca²⁺-release (Akbulut et al., 2010; Nutt et al., 2001), pkd2 might be able to rescue the spry1-induced loss of sensory GRP cells. Indeed, co-injection of spry1 and pkd2 partially restored nodal1 expression in lateral GRP cells (Fig. 7G-J). These final experiments suggested a synergy of pkd2 and the FGF/Ca²⁺ branch of FGF signaling in inducing the somitic GRP, i.e. for specification of the LRO flow sensor.

Our analysis of FGF signaling during left-right axis formation in Xenopus demonstrated a dual role, namely for specification of the LRO precursor, the SM, during early gastrulation, and a later, independent role for sensory cell formation in the LRO itself. These findings reconcile conflicting data reported previously from zebrafish and mouse. LRO ciliogenesis and earlier foxj1 expression were reduced or lost in zebrafish between 90% epiboly and the 10-somite stage following SU5402 treatment or knockdown of fgfr1 or fgf8 (Caron et al., 2012; Hong and Dawid, 2009; Neugebauer et al., 2009). In contrast, data reported from a hypomorphic Fgf8 mouse line suggested that lateral, sensory LRO cells might be missing, based on the absence of Nodal expression at the LRO, which is an evolutionarily conserved hallmark of ciliated LROs (Blum et al., 2007, 2009b; Meyers and Martin, 1999). Strikingly, manipulation of FGF signaling during late gastrulation (late-bud to 3-somites) using FGF inhibitors also resulted in loss of Nodal expression at the mouse LRO (Oki et al., 2010). Whether or not this function was mediated through the FGF/Ca²⁺ pathway, as suggested here, was not addressed in either of these studies, nor a possible morphogenetic role in sensory cell formation, a concept that emerged only later (McGrath et al., 2003; Tabin and Vogan, 2003). For future analyses, it may be worthwhile to re-evaluate LRO specification (Foxj1 expression) and ciliation in mouse embryos with reduced FGF signaling (in the hypomorphic Fgf8 line or following early SU5402 treatment), in order to analyze whether the first phase of FGF-dependent laterality formation, i.e. LRO precursor formation, is conserved in mammalian embryos as well. Lateral, flow-sensing cells in zebrafish are less well defined (Ferreira et al., 2017). MO-mediated knockdown of fgf8 resulted in loss of dand5 at the LRO as well (Hong and Dawid, 2009). However, this loss-of-function scenario does not allow us to distinguish between an early inhibition of FGF-dependent LRO specification or a later FGF requirement for flow sensor morphogenesis.

Together, the data on FGF function in cilia-dependent LR axis formation in the vertebrates strongly suggest an evolutionarily conserved dual role during early and late gastrulation. Interestingly, this dual function parallels the known role of FGF signaling during Xenopus mesoderm induction (phase 1 during early gastrulation; Amaya et al., 1993; Schulte-Merker and Smith, 1995) and morphogenesis (phase 2 during late gastrulation; Dorey and Amaya, 2010; Pownall and Isaacs, 2010). Both processes are intimately linked through the timing of events. Morphogenetic gastrulation behavior in phase 2 requires calcium, and is mediated through the FGF/Ca²⁺ pathway as well (Sivak et al., 2005; Wallingford et al., 2001; Wang and Steinbeisser, 2009). In the lateral GRP of Xenopus, Polycystin-2 might cooperate with FGF/Ca²⁺ downstream of inositol trisphosphate receptor activation at the level of intracellular calcium release, a positive interaction that has been demonstrated in Xenopus oocytes (Berridge et al., 2000; Dorey and Amaya, 2010; Li et al., 2005). This notion is supported by thapsigargin-mediated interference with calcium fluxes, which also resulted in lack of nodal expression in mouse crown cells and Xenopus lateral LRO cells (Takao et al., 2013; Thastrup et al., 1990; Vick et al., 2018).

Although not considered mesodermal at the stage of emergence and function, cells at the Xenopus SM (i.e. superficial cells of the organizer) and LRO (embedded in the archenteron) are fated to become mesodermal: they integrate into the notochord (flow-generating central LRO cells) and somites (flow-sensing lateral LRO cells) once the LRO submerges under the endodermal cells, which migrate medially to close the gap in the posterior dorsal archenteron during neurulation (Shook et al., 2004). Thus, SM specification may be considered as one aspect of mesoderm induction, and LRO formation as a morphogenetic process during early mesoderm differentiation. In that sense, LRO cells take a unique detour to assume their final mesodermal fate, starting out as superficial cells overlying the organizer and dorsal mesoderm at the beginning of gastrulation, and lining the dorsal posterior archenteron later on (Blum et al., 2014). Interestingly, pkd2 is involved at both stages in Xenopus (Vick et al., 2018; data presented here). It remains to be seen to what extent this parallel holds, i.e. which other known molecules previously involved in either mesodermal process participate in early LR axis formation as well. Brachyury mutant mouse lines and zebrafish no tail morphants support this reasoning, as do our unpublished results on the role of brachyury during LR development in Xenopus (Sabrina Kurz and M.B., unpublished; Amack and Yost, 2004; King et al., 1998). In addition, mechanical strain has been recently shown to constitute a decisive physical force in LRO morphogenesis and function, namely cilia polarization and motility (Blum and Ott, 2018; Chien et al., 2018). Remarkably, strain is only effective when foxj1 is present; therefore, it would be interesting to analyze whether or not FGF signaling is a prerequisite of strain-mediated LRO morphogenesis and function.

There is but one presumed FGF function in LR axis formation that remains unresolved in an evolutionary context: FGF-mediated vesicle release at the mouse LRO (Tanaka et al., 2005). This is more than just a cursory inconsistency, as it impacts on our conceptional understanding and perception of cilia-mediated symmetry breaking in general. To date, two models of symmetry breaking, though not necessarily mutually exclusive, co-exist: (1) flow-sensing itself by immotile lateral sensory LRO cells, which through Polycystin-2 transmits a calcium signal into the cells (two-cilia model; McGrath et al., 2003; Tabin and Vogan, 2003), and (2) the morphogen model, in which a secreted factor (Nonaka et al., 1998; Okada et al., 1999), released from LRO cells, transfers through flow to the left side, where it initiates asymmetric signaling. In one variant of the morphogen model, the flow-transported factor is localized in vesicles (i.e. exosomes or exocytic vesicles), which are released in an FGF-dependent manner from the LRO. This variant has regained attention recently, following the proposal that sensory cilia do not qualify as calcium-responsive mechanosensors at the mouse LRO (Delling et al., 2016). This proposal contrasts with experimental and genetic evidence in the mouse, which demonstrated (1) that artificial flow was able to rescue lack of cilia motility as well as to revert laterality upon inversion of flow directionality (Nonaka et al., 2002); and (2) that Polycystin-2 is required on lateral LRO cells for flow sensing (McGrath et al., 2003; Yoshiba et al., 2012). The case, therefore, seems undecided on this point. In particular, it remains to be seen at what stage vesicles or a secreted morphogen arise at the LRO, and how they transfer and fuse with cells or cilia at the left margin of the LRO. In our experiments, application of SU5402 at stages following LRO formation, i.e. at time points when such vesicles could emerge for the first time, did not impact on LR axis formation. Our data thus suggest that the reported role of FGF in releasing vesicles at the mouse LRO is likely not conserved in the frog. In addition, the vesicles observed during our SEM analyses did not accumulate on the left side, i.e. at the left lateral GRP cells. Their distribution therefore seemed unaffected by flow (cf. Fig. S6).

In conclusion, our work re-defines the role of FGF signaling in vertebrate LR axis formation. In parallel with mesoderm induction, FGF specifies superficial organizer cells as LRO precursor during early gastrulation. This holds true despite apparent morphological differences between fish, amphibian and, perhaps, also mouse embryos. As mesodermal tissues undergo morphogenesis beginning at mid gastrulation, FGF signaling, specifically the FGF/Ca²⁺ branch, is instrumental for the generation of flow sensor cells at the lateral margins of the LRO. LR axis formation, thus, seems to be more highly conserved as previously anticipated. It remains to be seen whether this notion holds for flow-sensing mechanisms as well.

Frogs were purchased from Nasco (Fort Atkinson, WI, USA). Handling, care and experimental manipulations of animals was approved by the Regional Government Stuttgart, Germany (Vorhaben A379/12 ZO ‘Molekulare Embryologie’), according to German regulations and laws (§6, article 1, sentence 2, nr. 4 of the Animal Welfare Act). Animals were kept at the appropriate conditions (pH 7.7, 20°C) at a 12 h light cycle in the animal facility of the Institute of Zoology of the University of Hohenheim (Stuttgart, Germany). Female frogs (4-15 years old) were stimulated with 25-75 units of human chorionic gonadotropin (hCG; Sigma), depending on weight and age, which was injected subcutaneously 1 week prior to oviposition. On the day before ovulation, female frogs were injected with 300-700 units of hCG (10-12 h before). Eggs were collected into a Petri dish by careful squeezing of the females and in vitro fertilized. Sperm of male frogs was gained by dissection of the testes and was stored at 4°C in 1× MBSH (Modified Barth's saline with HEPES) solution. Embryos were staged according to Nieuwkoop and Faber (1994). Only clutches of embryos from healthy females were used for the experiments reported here, provided the early embryonic stages showed normal survival rates as well. Individual embryos from one batch were randomly picked and used either as control or tested specimens. If control groups displayed unusual developmental defects later in development, such clutches were excluded as well, based on empirical judgement.

Prior to in vitro synthesis of mRNA using the Ambion sp6 message machine kit, the plasmid was linearized with NotI. Embryos were injected at the 4- to 8-cell stage, using a Harvard Apparatus. Drop size was calibrated to 8 nl/injection. Amounts of injected mRNA or MOs are indicated in the main text. For specific lineage targeting of constructs/MOs, tracers used for injection control included fluorescein (70,000 MW) and rhodamine B dextran (10,000 MW; both Thermo Fisher). [For more detail on lineage-specific injections to target central or lateral GRP and ventro-lateral tissues (i.e. LPM) refer to Blum et al., 2009a; Vick et al., 2018; Vick et al., 2009.]

SU5402 (Calbiochem) was dissolved at 20 mM in DMSO and stock solutions were stored in aliquots at −20°C. Incubations were conducted in 24-well plates and protected from light at room temperature. Concentrations used ranged from 30 µM to 100 µM (diluted in 0.1× MBSH) as indicated. Incubation start points and duration are indicated by Nieuwkoop and Faber stages in the main text. Incubations were terminated by removal of the inhibitor and several rounds of washes using 0.1× MBSH buffer solution.

DMSO control or 60 µM SU5402-treated embryos from two independent experiments were fixed in 4% paraformaldehyde at stage 13/14 or 16 for 1 h at room temperature, followed by the preparation of dorsal explants. Immunofluorescence with anti-pH3 antibody [phospho-histone H3 (Ser10), 06-570, Millipore; diluted 1:140 and incubated at 4°C] was used to visualize proliferating cells. Cell borders were highlighted by cortical actin staining using Alexa Fluor 488 Phalloidin (Thermo Fisher; 1:40; 30 min incubation at room temperature) and nuclei were stained with Hoechst 33342 (Thermo Fisher; 1:10,000; 1 h at room temperature).

TUNEL analyses were performed using standard protocols (Hensey and Gautier, 1998). Dorsal explants of DMSO control specimens or 60 µM SU5402-treated embryos from two independent experiments were fixed at stage 13/14 or 16 and analyzed for apoptotic cells in the neuroectoderm and GRP. Briefly, fixed embryos were incubated at room temperature overnight with digoxygenin-dUTP (1 µM) and terminal deoxynucleotidyl transferase (150 U/ml). Following incubation with an anti-digoxygenin-AP (alkaline phosphatase) antibody (11093274910, Roche; 1:5000) overnight at 4°C, apoptotic cells were visualized using BM Purple AP substrate (Roche) at room temperature.

The apoptosis rescue experiment using Bcl-XL was performed by targeting 8 pg of Bcl-XL/CS2+ plasmid DNA (Johnston et al., 2005) into the left and right somitic GRP lineages (dorsolateral part of the 4-cell embryo). Embryos were treated with either DMSO or 60 µM SU5402 from stage 12-17, fixed and processed for in situ hybridization (ISH) using dand5- and nodal1-specific antisense probes.

For in situ mRNA detection, whole-mount ISH was performed. Embryos were fixed in MEMFA [3.7% formaldehyde, 100 mM 3-(N-morpholino)propanesulfonic acid, 2 mM EGTA, 1 mM MgSO₄] for 2-3 h at room temperature and processed following standard protocols (Sive et al., 2000). RNA in situ probes were transcribed using SP6 or T7 polymerases. Whole-mount ISH protocol was modified from Belo et al. (1997).

For SEM analysis, injected or treated specimens were fixed with 4% paraformaldehyde/2.5% glutaraldehyde in Sorenson's phosphate buffer overnight at 4°C. Dorsal explants were prepared for SEM and analyzed as previously described (Schweickert et al., 2007; Sulik et al., 1994). SEM photographs were analyzed for GRP ciliation, cilia polarization and apical cell surface area using ImageJ and evaluated as described (Beyer et al., 2012; Sbalzarini and Koumoutsakos, 2005).

For analysis of leftward flow, dorsal posterior GRP explants were dissected from control or SU5402-treated embryos at stage 17. Explants were placed in a Petri dish containing fluorescent microbeads (diameter 0.5 µm; diluted 1:2500 in 1× MBSH) and incubated for a few seconds. Explants were transferred to a microscope slide which was prepared with vacuum grease to create a small chamber that contained a fluorescent microbead solution; a cover slip was carefully pressed on to seal the chamber. Time-lapse movies of leftward flow were recorded using an AxioCam HSm video camera (Zeiss) at two frames per second using an Axioplan2 imaging microscope (Zeiss). Using the ImageJ ‘Particle-Tracker’ plug-in, leftward flow was analyzed and particle movement was assessed (Beyer et al., 2012; Sbalzarini and Koumoutsakos, 2005; Vick et al., 2009). For evaluation of flow quality, i.e. directionality of transported fluorescent particles, a custom-made analysis tool written in the statistical software R (R Core Team, 2013) by T. Thumberger and described by Vick et al. (2009) was used. Briefly, a mask covering the GRP was used to exclusively measure beads transported across the GPR. Only particles were evaluated that were detected in ten consecutive frames. To measure directionality of single beads and leftward flow, and to exclude Brownian movement, a Rayleigh's test of uniformity was performed, first on single beads, then on the mean angles of single particles moving in a directed manner. The parameter rho (ρ) describes the directionality of flow (i.e. the sum of all remaining particles), with P>0.6 defined as WT and P<0.6 considered as reduced or aberrant flow (cf. Vick et al., 2009). Diagrams of flow quality show mean direction of all particles and a P-value as the index of directionality (Fig. S3G-I).

For quantification of GRP widths, SEM photographs of GRP explants or vibratome sections of myod1-stained embryos were used. For evaluation of superficial myod1 expression in the GRP, transversal GRP sections containing the widest part of myod1-positive cells were selected. Widths were measured in pixels using ImageJ (Sbalzarini and Koumoutsakos, 2005). For each analysis, the mean value of DMSO-treated controls was set to 1.0 and the relative sizes of manipulated embryos was calculated to generate box plots. For measurements of myod1-positive, lateral somitic GRP widths, the left and right halves were added up.

Statistical calculations of marker gene expression patterns and cilia distribution were performed using Pearson's χ² test (Bonferroni corrected, if required). For statistical calculation of ciliation, cilia length, GRP widths, flow velocity and directionality Wilcoxon-Match-Pair test was used, for vesicle abundance per µm² Kruskal–Wallis test, and for left versus right vesicle abundance and pH3 analysis Mann–Whitney test. For calculations, the statistical software R was used (R Core Team, 2013; https://www.r-project.org).

In all figures, statistical results are represented as *P<0.05, **P<0.01, ***P<0.001.

We thank T. Thumberger for expert help with flow and GRP analysis, M. Tingler for help with pkd2 epistasis analysis, T. Beyer for technical advice, and E. Amaya, J. Heasman, D. Kessler, C. Kintner, R. Rupp, J. Smith, H. Steinbeisser and A. Vonica for constructs.