Divergent mechanisms specify chordate motoneurons: evidence from ascidians

One of the major mechanisms for generating diversity during evolution is thought to involve changes in the regulation of the genes controlling embryonic development (Duboule and Wilkins, 1998; Davidson and Erwin, 2006). According to this paradigm, conservation of the expression profile of developmental regulatory genes between species should pinpoint areas of evolutionary stability. Despite an evolutionary distance of over 500 million years (Shu et al., 2001), the tadpole larvae of ascidians has a body plan remarkably similar to that of vertebrate embryos, suggesting a degree of conservation in the developmental mechanisms used to generate this body plan. It is, thus, not surprising that some studies have revealed highly similar expression territories for developmental regulatory genes, acting in equivalent steps during embryogenesis [e.g. brachyury in notochord formation (Yasuo and Satoh, 1998) and MyoD for muscle formation (Meedel et al., 2007)]. There is increasing evidence, however, that equivalent morphological features can arise by distinct developmental routes, a phenomenon termed ‘developmental systems drift’ (True and Haag, 2001). For example, among nematodes, the formation of the vulva involves a surprisingly diverse array of developmental strategies to produce an equivalent structure (Kiontke et al., 2007). Similarly, interactions between equivalent cells, but using different signalling molecules, are involved in the fate determination of a particular muscle precursor in two different species of ascidian (Hudson et al., 2007; Hudson and Yasuo, 2008; Tokuoka et al., 2007). This means that during evolution, developmental mechanisms can be ‘rewired’ to produce a conserved phenotypic outcome.

Recently, a systematic large-scale comparison of whole-mount in situ hybridisation patterns of Ciona intestinalis and the zebrafish found an unexpectedly high level of divergence in the gene expression profiles between these two chordate species (Sobral et al., 2009), suggesting that the use of different developmental mechanisms to produce a similar morphological output may be surprisingly common. Interestingly, this comparison revealed that conservation of orthologous gene expression between the two species is highest in tissues involved in locomotion, including muscle, notochord and central nervous system. Here, we examine the mechanisms used to generate three out of the five pairs of somatic motoneurons in the ascidian Ciona intestinalis (hereafter referred to as Ciona) and compare them with the mechanisms used in vertebrates.

One of the hallmarks of the chordate body plan is the presence of a dorsal hollow central nervous system, which contains distinct types of neurons along the anteroposterior and dorsoventral axis. In vertebrates, the somatic motoneurons form at a precise position in the ventral spinal cord, close to the ventral midline (Fig. 1D) (Briscoe and Ericson, 2001).

A number of signalling pathways have been shown to be crucial for the formation of motoneurons in vertebrates. Sonic hedgehog (Shh) is expressed in the underlying notochord and in the ventral midline of the neural tube, the floor plate, and promotes motoneuron fate (Briscoe and Ericson, 2001; Ulloa and Briscoe, 2007; Wilson and Maden, 2005). Retinoids from the lateral somites are also required for motoneuron formation (Wilson and Maden, 2005). BMP and Wnt signals from the overlying ectoderm and roof plate, the dorsal-most structure of the spinal cord, are required for the formation of the dorsal spinal cord and repress formation of the ventral spinal cord, including motoneurons (Lewis and Eisen, 2003; Liu and Niswander, 2005). Antagonists of BMP signalling expressed in the axial and paraxial mesoderm, and in the dorsal spinal cord itself (roof plate) restrict the activity of BMP dorsally (McMahon et al., 1998; Mekki-Dauriac et al., 2002; Liem et al., 2000; Jessell, 2000; Liu and Niswander, 2005). Thus, BMP and Shh signals act broadly across the neural tube in a mutually restrictive fashion with BMP limiting the range of Shh activity and vice versa to establish the dorsal-ventral pattern of the vertebrate spinal cord.

Ascidian larvae have a chordate body plan of central notochord, dorsal neural tube and lateral muscles in miniature, consisting of only about 3000 cells (Lemaire et al., 2008). Ascidian development proceeds with a fixed cell cleavage pattern enabling the cell lineages to be described (Conklin, 1905; Nishida, 1987). In Ciona, the motoneurons arise from the A8.15 lineage of the early gastrula stage embryo or column 3 of the neural plate (Fig. 1A) (Cole and Meinertzhagen, 2004). By hatching stage, the five pairs of motoneurons along the anteroposterior axis are A12.239, A13.474, A11.118, A11.117 and A10.57. These motoneurons finally reside in a ventrolateral position of the larval trunk ganglion, considered to be the equivalent of the vertebrate spinal cord, and project axons that terminate in characteristic terminal endplates on the larval muscle (Fig. 1A-C) (Cole and Meinertzhagen, 2004; Dufour et al., 2006; Imai and Meinertzhagen, 2007; Horie et al., 2010). Thus, the position of the motoneurons within the neural tube and in relation to other embryonic tissues such as notochord and muscle in Ciona is equivalent to that in vertebrates (Fig. 1B-D). However, unlike in vertebrates, the tiny size of the ascidian trunk ganglion means that the motoneurons are in close proximity to both the dorsal and ventral cells of the neural tube. Moreover, motoneurons are initially formed in the lateral neural tube directly in between the dorsal and ventral cells, before shifting into their final ventrolateral position of the thickened trunk ganglion (Cole and Meinertzhagen, 2004).

In Ciona, hh-2, a homologue of vertebrate hedgehog, is expressed in the ventral-most cells of the neural tube, the positional equivalent of the vertebrate floor plate (Takatori et al., 2002), while BMP2/4 expression has been reported in the lateral neural plate (future dorsal and lateral neural tube) in another ascidian species [Halocynthia roretzi (Miya et al., 1997)]. The conserved expression pattern of these molecules strongly suggests that their crucial role in spinal cord patterning has been conserved during evolution of the chordate lineage. The aim of this study was to investigate this possibility.

Adult Ciona intestinalis were purchased from the Roscoff Marine Biological Station (France). Blastomere names, lineage and fate mapping are as described in previous studies (Conklin, 1905; Cole and Meinertzhagen, 2004; Nicol and Meinertzhagen, 1988a; Nicol and Meinertzhagen, 1988b; Nishida, 1987). Embryo culture, ablation, DiI labelling, microinjection, Nodal-morpholino, Smad1/5-morpholino (at 0.5 mM) and SB431542 treatment have been described previously (Hudson et al., 2003; Hudson and Yasuo, 2005; Pasini et al., 2006; Sardet et al., 2011). Xenopus chordin mRNA was injected into unfertilised eggs at a concentration of 0.6 μg/μl (Sasai et al., 1994). pSPE3-Chordin vector for RNA synthesis was made by taking the Ciona Chordin full-length construct (pENTR-Chordin, from the Unigene collection plates, purchased from Cogenics) (U. Rothbächer, M. J. Gilchrist and P. Lemaire et al., unpublished) and cloning into pSPE3 by Gateway LR reaction (Invitrogen) (Roure et al., 2007). Ciona Chordin RNA was injected at concentrations of 0.5 to 0.75 μg/μl.

Electroporation was carried out as described previously with modifications (Corbo et al., 1997; Bertrand et al., 2003). FOG::Nodal, FOG::Noggin and FOG::BMP2/4 and pvAChTP::EGFP constructs have been previously described (Pasini et al., 2006; Yoshida et al., 2004). FOG::Chordin was constructed by Gateway cloning pENTR-Chordin into pFOG::RfA (Pasini et al., 2006). A DNA solution (50 μg in 250 μl) was used in all experiments using FOG::Noggin, FOG::BMP2/4 and FOG::Nodal, except those in Fig. 5B and Fig. 7D where 25 μg of each plasmid was used to give a maximum total of 50 μg of DNA per electroporation. FOG::Chordin was toxic to embryos at high concentrations, so 15 to 25 μg of DNA per electroporation was used. Embryos were selected for normal development at the gastrula stage. Problems with development prior to and during gastrulation were not observed following injection of Ci-Chordin mRNA into eggs and therefore were most probably non-specific effects. For all experiments, data were pooled from at least two independent experiments.

Except for BMP2/4, all expression patterns have been described previously (Hudson et al., 2003; Hudson et al., 2007; Hudson and Yasuo, 2005; Ikuta and Saiga, 2007; Imai et al., 2004; Imai et al., 2006; Shi et al., 2009; Takamura et al., 2002) (http://ghost.zool.kyoto-u.ac.jp). In situ hybridisation was carried out as described previously (Hudson and Yasuo, 2006; Wada et al., 1995). For fluorescent in situ hybridisation, the first 2 days of the protocol was the same as above, after which the protocol described by Takatori et al. was used and adapted for double fluorescent in situ hybridisation (Takatori et al., 2010). Embryos were mounted in Vectashield DAPI (Vector Laboratories).

Confocal analysis was performed with a Leica SP5 microscope and images were processed using Image J software. For the comparative expression analyses of motoneuron markers (Fig. 1; see Fig. S1 in the supplementary material), all in situ hybridisations were carried out on the same batch of embryos. For ablation experiments in Fig. 3, one experiment gave no staining in notochord-ablated embryos. As this was observed only once in six independent notochord-ablation experiments, this entire data set was removed from the analysis. Sometimes embryos displayed more than three motoneurons on one side and fewer than three on the other side. In order to represent these cases and to distinguish, for example, embryos displaying six cells on one side and none on the other from those with three on each side, we decided to consider each half of the embryo independently for our analyses. The level of staining was not taken into consideration during cell counts, rather the number of nuclei with detectable levels of signal. For statistical analysis, Microsoft Excel, Excel Box-and-whiskers plots V2 (http://www.coventry.ac.uk/ec//~nhunt/boxplot.htm) and Mann-Whitney test V5 (Excel) (Siegel and Castellan, 1988) software were used.

In order to carry out an analysis of motoneuron formation, we chose three markers: Ciona vAChTP, which encodes a vesicular acetylcholine transporter; Ciona ChAT, which encodes choline acetlytransferase; and Ciona Mnx, an orthologue of vertebrate somatic motoneuron markers Hb9 and Mnr2 (Takamura et al., 2002; Wada et al., 2003; Tanabe et al., 1998; Thaler et al., 1999; Ikuta and Saiga, 2007). Comparing fluorescent in situ hybridisation analysis with neural maps previously published (Cole and Meinertzhagen, 2004), we confirmed expression of these markers in the motoneurons (Fig. 1B,C,E,F; see Fig. S1 in the supplementary material). At the late tail bud stage (13 hours of development), all three markers were co-expressed in three of the five pairs of motoneurons: A10.57, A11.117 and A11.118. Expression of Mnx was weak in A11.117 and this gene was also detected in cells of the tail neural tube, probably the descendents of A12.230 (not shown). Expression in the anterior two motoneurons A12.239 and A13.474, which form later than the three posterior motoneurons, probably starts later in development; however, at later stages of development non-specific staining of the tunic obscured the analysis.

Thus, vAChTP, ChAT and Mnx are expressed in three of the five pairs of motoneurons and these markers were used to investigate motoneuron formation in Ciona embryos. We will refer to these three pairs of motoneurons as vAChTP⁺ motoneurons.

In vertebrates, signals from the midline tissues, the notochord and the neural tube floorplate play a pivotal role in spinal cord patterning, a patterning process beginning as early as gastrula stages (Beattie et al., 1997; Placzek et al., 1991; Yamada et al., 1991; Yamada et al., 1993). These midline tissues express both Shh (which is required for formation of ventral spinal cord cell types, including motoneurons) and BMP antagonists (which act to protect the ventral neural tube from dorsalising BMP signals) (McMahon et al., 1998; Jessell, 2000; Lewis and Eisen, 2001). Correspondingly, hh-2 in Ciona is expressed in the ventralmost cells of the tail nerve cord (Takatori et al., 2002), while the BMP antagonist Chordin is expressed in the notochord (and lateral neural tube) from gastrula to tailbud stages (Imai et al., 2004). In order to test the requirement for midline signals in motoneuron formation in Ciona embryos, we performed cell ablations at the 64-cell stage, prior to gastrulation, to prevent formation of notochord and/or ‘floorplate’ tissues (Fig. 2A, Fig. 3). We monitored motoneuron specification using ChAT probe for conventional in situ hybridisation and vAChTP for fluorescent in situ hybridisation. We first prevented formation of the ascidian ‘floor plate’, the ventral row of cells in the tail nerve cord and trunk ganglion, by ablating the A7.4 pair of cells (removing all the yellow cells in Fig. 1A) (Cole and Meinertzhagen, 2004; Nishida, 1987; Taniguchi and Nishida, 2004). Ablation of these cells had no effect on ChAT expression. We noticed, however, that ventral neural tissue was not entirely absent in these embryos, as assayed by Ciona FGF3 expression (Shi et al., 2009), with a small domain of expression persisting in the anterior part of the central nervous system (Fig. 2A). This remaining ventral tissue of the neural tube derives from the a7.10 cells of the 64-cell stage embryo (Cole and Meinertzhagen, 2004). Removal of a7.10 together with A7.4 (‘FP-ablated’) resulted in a complete loss of FGF3 expression, but ChAT expression still persisted. We quantified the effect of ablations by counting the number of vAChTP⁺ cells following fluorescent in situ hybridisation (Fig. 3), analysing each half of the embryo independently. In 65% of cases the normal number of three motoneurons could be detected per embryo half. The remaining embryo halves had 1 to 2 motoneurons, to give an average of 2.5 motoneurons each side. Thus, we conclude that the ascidian ‘floor plate’ cells are not essential for the formation of vAChTP⁺ motoneurons.

We next addressed whether the axial mesoderm plays a role in formation of motoneurons. ChAT expression was not affected following ablation of the four anterior notochord precursors (‘N-ablation’) (Fig. 2A). Confocal analysis revealed an average of 2.3 vAChTP⁺ cells each side, with 39% of embryo halves displaying the correct number of motoneurons (Fig. 3). When we ablated both primary (A-line) and secondary (B-line) notochord lineages (‘allN ablation’), which also prevents development of some mesenchymal cell fates, embryo halves had an average of 2.4 motoneurons and 71% displayed the correct number of three (Fig. 3). These results show that the majority of motoneurons form following notochord ablation and suggest that the notochord is not crucial for vAChTP⁺ motoneuron formation in Ciona embryos.

Similar results were found following FP+N-ablation, with an average of 2.4 cells expressing vAChTP each side and 28% of embryos displaying the correct number of motoneurons (Fig. 3). Furthermore, expression of all three motoneuron markers was detected in N+FP ablated embryos by conventional in situ hybridisation (Fig. 2A). The reduction and increase in variability in the number of motoneurons formed on each side of the embryo following ablations may indicate some role for midline structures, particularly the notochord, in controlling the precise number of motoneurons formed. Alternatively, it might be a non-specific consequence of the general morphological disorganisation caused by the ablation of these structures. The fact that motoneurons do form in most cases indicates strongly, however, that there is no absolute requirement for either the floorplate, the notochord or a combination of these tissues for the specification of vAChTP⁺ motoneurons in Ciona embryos. This is clearly different to the situation in vertebrates where these tissues are essential for the formation of somatic motoneurons.

In vertebrates, dorsally derived TGFβ family signals play a prominent role in specifying cell fates in the dorsal part of the neural tube, in particular, members of the BMP family (Lee and Jessell, 1999). In Ciona embryos, a TGFβ ligand from a distinct class, Nodal, is expressed lateral to the neural plate precursors and is required for expression of the motoneuron markers Mnx and ChAT (Hudson and Yasuo, 2005). Consistently, ablation of b5.3, the mother cell of the Nodal-expressing b6.5 cell, resulted in loss of motoneuron markers on the ablated side (Fig. 2B). We hypothesised that Nodal may act via BMPs in ascidians as BMP2/4 expression has been reported in the lateral neural plate precursors of Halocynthia embryos and promotes motoneuron fate (Miya et al., 1997; Katsuyama et al., 2005). We confirmed that the one BMP2/4 orthologue in the Ciona genome (Hino et al., 2003) was also expressed laterally in the neural plate (Fig. 4A). In gastrula embryos, expression was detected in the lateral A-line neural precursors A8.15 (progenitor of all the larval motoneurons) and A8.16, as well as b-line (b8.19) neural precursors and some B-line muscle precursors (Fig. 4A). By neural plate stages, BMP2/4 expression was detected in lateral b-line cells bordering the neural plate and a9.49 (Fig. 4A). Expression of BMP2/4, initially variable and often patchy (see Fig. 4B, control), became robust by this stage. By the time the neural plate consists of six rows of cells, BMP2/4 expression was restricted to b9.37, b9.38 and the ventral epidermis precursors (Fig. 4A and data not shown). During neurula stages, expression was restricted to the ventral epidermis (Imai et al., 2004).

We confirmed that BMP2/4 expression in the lateral neural plate precursors is downstream of Nodal by using a pharmacological Nodal type I receptor (ALK4/5/7) inhibitor (SB431542), by injection of Nodal-morpholino and by electroporating the FOG::Nodal overexpression construct, which drives expression throughout the animal cells from the 16-cell stage (Inman et al., 2002; Hudson and Yasuo, 2005; Pasini et al., 2006). Both methods used to inhibit Nodal signalling resulted in loss of BMP2/4 expression in the A- and b-line cells (Fig. 4B). Overexpression of Nodal resulted in ectopic expression of BMP2/4 throughout the A-line neural precursors, as well as ectopic expression of its antagonist Chordin (Fig. 4B, Fig. 5A). The expression of Chordin in a broad lateral domain encompassing the lateral neural plate and the epidermal cells bordering the neural plate has previously been shown to be a target of Nodal signals, although Chordin expression in the presumptive notochord is Nodal independent (Hudson and Yasuo, 2005; Imai et al., 2006).

We conclude that, in Ciona, both BMP2/4 and its antagonist Chordin are expressed in the lateral neural plate, directly or indirectly downstream of Nodal, and so are well placed to participate in the patterning of the neural tube.

As BMP2/4 expression begins at the early gastrula stage, we investigated its role in patterning the A-line neural precursors at the gastrula and neural plate stages. We found that overexpression of BMP2/4 had effects very similar to overexpression of Nodal, with expression of lateral markers expanded into medial neural plate cells at the expense of medial marker expression (Fig. 5; see Fig. S2 in the supplementary material). Thus, as previously shown for Nodal, BMP2/4 appears capable of transforming medial neural plate into lateral neural plate fates (Hudson and Yasuo, 2005; Imai et al., 2006; Hudson et al., 2007). We ruled out the possibility that the lateralising effect of BMP2/4 was indirect via the ectopic activation of the Nodal signalling pathway by comparing the effects of different combinations of treatments on the expression of Snail, an early target of Nodal signalling in Ciona (Hudson and Yasuo, 2005). Snail expression was totally abolished when Nodal signalling was inhibited using the ALK4/5/7 inhibitor SB431542, even in the presence of excess Nodal ligand (Fig. 5Bi-iv). However, overexpression of BMP2/4 in the presence of SB431542 was able to restore Snail expression to significant levels (Fig. 5Bvii).

Given that BMP2/4 was clearly able to lateralise the neural plate independently of Nodal signals, we were surprised to find that the antagonists Noggin and Chordin had little effect on any of the neural plate marker genes analysed (Fig. 5A; see Fig. S2 in the supplementary material). We confirmed that Noggin overexpression effectively inhibited BMP2/4 but not Nodal in Ciona embryos, as it had no effect on Snail expression induced ectopically by Nodal, but reduced to one-fifth its ectopic expression induced by BMP2/4 (compare Fig. 5iii,v,vi,viii).

We conclude that although BMP2/4 has the ability to lateralise the neural plate independently of Nodal signals, BMP2/4 signals do not play a major role in neural plate patterning during normal development. These findings are not contradictory but suggest a redundant role for BMP2/4 in this process, possibly with Nodal itself.

It has previously been shown that modification of BMP signalling in ascidians caused a number of characteristic morphological defects, including rounding and fusion of the palps and a gain or loss of epidermal sensory neurons (Miya et al., 1997; Pasini et al., 2006; Darras and Nishida, 2001). We observed the same effects following introduction of the FOG::Noggin, FOG::Chordin and FOG::BMP2/4 constructs, and by injection of Ciona and Xenopus Chordin mRNA and Smad1/5-morpholino (Fig. 6; see Fig. S3 in the supplementary material). In addition, following BMP signalling the tail bent up dorsally, sometimes forming a corkscrew curl (Fig. 6; see Fig. S3 in the supplementary material). Inhibition of BMP signalling, however, had no effect on motoneuron gene expression (Fig. 6; see Fig. S3 in the supplementary material), with no difference in the number of vAChTP⁺ cells formed per half embryo following Chordin or Noggin overexpression (Fig. 7A,B). By contrast, overexpression of BMP2/4 resulted in a marked increase in the expression domains of the three motoneuron markers (Fig. 6) as well as the previously reported upregulation in epidermal sensory neurons (Fig. 6) (Pasini et al., 2006). The average number of vAChTP⁺ cells per whole embryos increased from 7.2 to 19 following BMP2/4 overexpression (Fig. 7C). Nodal overexpression had similar effects on the expression of motoneuron markers, but resulted in loss or severe reduction of all epidermal sensory neurons (Fig. 6).

The formation of excess motoneurons following Nodal or BMP2/4 overexpression is probably a result of the transformation of medial into lateral neural plate, as observed at gastrula and neural plate stages (Fig. 5; see Fig. S2 in the supplementary material). Indeed, DiI labelling of the medial pair of blastomeres at the 64-cell stage (A7.4 pair) demonstrated that medial cells produced motoneurons following BMP2/4 or Nodal overexpression, but never in unmanipulated embryos (Fig. 7D). Consistently, markers of tissues derived from the medial neural plate, 08C09 (which marks the posterior sensory vesicle) and Fgf3 (which marks the ventral neural tube) were lost following BMP2/4 or Nodal overexpression (see Fig. S4 in the supplementary material).

We conclude that BMP2/4 is capable, like Nodal, of transforming medial into lateral neural plate fates, thereby inducing excess vAChTP⁺ motoneuron formation in Ciona embryos, but that during normal development BMP2/4 does not play an essential role in this process.

The evolutionary split between the vertebrate and urochordate lineages took place more than 500 million years ago. Although the morphology of their common ancestor is not certain, it appears likely that in the ascidian lineage evolutionary pressures favouring rapid development have driven a remarkable acceleration of the early developmental programme. Fate specification of key chordate features such as notochord and neural tube therefore occur in a very different cellular context, typically with signalling occurring between one or two defined cells rather than across fields of hundreds of cells (Lemaire, 2009). Comparisons of the developmental programmes, cellular interactions and regulatory gene expression, between vertebrates and ascidians, can thus be informative both about the ancestral developmental strategy and the degree of evolutionary flexibility, allowing mechanistic rewiring of developmental processes (Lemaire et al., 2008). There is now ample evidence that changes in gene expression drive changes in development and the final body plan (Davidson and Erwin, 2006; Erwin and Davidson, 2009). Here, we highlight an example where the gene expression patterns of key regulatory genes and the final overall organisation of the resulting structure appear to be well conserved; a Hh gene is expressed in the ventral midline of the neural tube, BMP2/4 is expressed in the lateral/dorsal neural plate and the final position of the motoneurons in a ventrolateral position all appear to be conserved. However, we nonetheless find evidence for divergence in the mechanisms that control the specification of this important cell type. Future studies should aim to identify the changes in the cis-regulatory control sequences within the gene regulatory network of motoneuron specification that underlie this signal deployment change.

In vertebrate embryos, the formation of somatic motoneurons in ventrolateral regions of the neural tube depends on Shh signals emanating from the notochord and floor plate (Beattie et al., 1997; Placzek et al., 1991; Weinstein et al., 1994; Yamada et al., 1991; Lewis and Eisen, 2001; Wilson and Maden, 2005). Similarly, expression of a Ciona homologue of hedgehog, hh-2, is restricted to the ventral midline of the neural tube during tailbud stages (Takatori et al., 2002). Although this expression occurs later than that observed in vertebrates, in which Shh begins to pattern the neural tube at gastrula and neural plate stages (Beattie et al., 1997; Yamada et al., 1993), it is nonetheless well placed for a role in motoneuron development. However, our experiments involving ablation of the hh-2-expressing ‘floor plate’ argue against a crucial role for Hedgehog in Ciona vAChTP⁺ motoneuron formation. Although we cannot rule out that some Hedgehog activity remains in these embryos, it is likely to be significantly reduced. The fact that the majority of Ciona vAChTP⁺ motoneurons form following ablation of the hh-2 expression territory strongly suggests that the floor plate and Hedgehog signalling are not acting in the same way during vertebrate and Ciona motoneuron formation. Consistently, injection of hh-2 morpholino or a dominant-negative form of mouse Patched (mPtcΔloop) (Briscoe et al., 2001) in Ciona did not disrupt motoneuron formation (Imai et al., 2009) (H.Y., unpublished). Similarly, in Halocynthia embryos, attempts to activate or inhibit the Shh signalling pathway had no effects on the formation of motoneurons (Katsuyama et al., 2005). Furthermore, ablations causing the entire loss of both notochord and floorplate failed to prevent the formation of vAChTP⁺ motoneurons (Figs 3, 4). Although we cannot rule out some role for midline structures and hh-2 in controlling precisely the number and position of motoneurons, our data indicate that these midline tissues are not playing the same role during the induction of motoneuron fate in the ascidian and vertebrate spinal cord.

On the other hand, despite Shh having been assigned a pivotal role in vertebrate motoneuron specification, there is some evidence for parallel Shh-independent pathways (Eggenschwiler et al., 2001; Litingtung and Chiang, 2000; Ruiz i Altaba et al., 2003; Wijgerde et al., 2002; Wilson and Maden, 2005). This evidence comes from embryos carrying double mutations, in Shh itself (or its receptor) as well as in a negative regulator of the Shh signalling pathway such as Rab23 or Gli3 (Eggenschwiler et al., 2001; Litingtung and Chiang, 2000; Wijgerde et al., 2002). In these double mutant embryos, motoneurons still form, as do some other types of ventral neurons that are normally dependent on Shh signalling, although the positioning of these neurons within the spinal cord is perturbed. This suggests that a Shh-independent patterning mechanism exists that is sufficient for the formation of ventral neuronal cell types. It is possible that the patterning system that is operating in ascidian embryos is the same as the Shh-independent system uncovered in vertebrate double mutant embryos, and that Shh was recruited to refine the neural tube patterning process during evolution of the vertebrate lineage. Alternatively, a more crucial role for Hedgehog signalling may have been lost or reduced in the ascidian lineage. This may have been a consequence of the drastic reduction in cell number during evolution of the ascidian neural plate/tube, which brought the motoneuron precursors into closer proximity to and therefore into the influence of dorsal signals (Fig. 1). This in turn may have allowed dorsal signals to take on a more prominent and positive role during motoneuron induction and ventral signals to be dispensed with. Functional data from Amphioxus, which belongs to the cephalochordates, the basal group of the chordate phylum and in which Hedgehog is expressed in both the floor plate and notochord from neural plate stages (Shimeld, 1999), would shed light on which of these scenarios is likely to be the most plausible.

It remains possible that Hedgehog signalling is involved in other aspects of ascidian neural tube patterning, including the specification of particular neuronal cell types. Recently, putative cholinergic and glycinergic cells have been identified in the tail nerve cord of Ciona, which was previously thought to be devoid of neurons (Imai and Meinertzhagen, 2007; Horie et al., 2010). It will be interesting in future studies to test the role of the ventral midline and Hedgehog in the specification of these cell types.

Despite the differences in the initial induction mechanisms of motoneurons between ascidian and vertebrates, a shared set of transcription factors are expressed in the trunk ganglion of ascidians, including Lhx, Mnx and Islet (Katsuyama et al., 2005; Ikuta and Saiga, 2007). Thus, distinct inductive mechanisms may converge on a common gene regulatory cassette, although thorough functional studies would be required in ascidians to support this hypothesis.

In vertebrates, two other types of signal act in parallel with Shh in spinal cord patterning: retinoids and BMPs (Litingtung and Chiang, 2000; Ruiz i Altaba et al., 2003; Wilson and Maden, 2005). Retinoids from lateral somites are vital for the formation of vertebrate motoneurons (Novitch et al., 2003). Although not yet fully investigated in ascidians, treatment of embryos with retinoic acid does not seem to affect the expression of motoneuron markers (Katsuyama and Saiga, 1998).

In vertebrates, BMPs appear to act over the entire neural plate, promoting dorsal and repressing ventral cell types, including motoneurons (Helms and Johnson, 2003; Wilson and Maden, 2005; Barth et al., 1999; Lewis and Eisen, 2001). Indeed, BMPs are involved in dorsoventral neural patterning in many species (Denes et al., 2007; Katsuyama et al., 2005; Lowe et al., 2006; Mizutani et al., 2006; Rusten et al., 2002). In both Halocynthia and Ciona embryos, BMP2/4 expression is detected in the lateral/dorsal neural plate, as in vertebrates (this study) (Miya et al., 1997). However, our experiments revealed no role for BMP signalling during formation of Ciona vAChTP⁺ motoneurons, a conclusion supported by BMP2/4 morpholino knockdown experiments (Imai et al., 2009). This was somewhat surprising, given that, in Halocynthia, inhibition of BMP signalling resulted in loss of motoneurons, suggesting that BMPs may promote motoneuron formation in ascidians rather than inhibiting their formation as in vertebrates (Katsuyama et al., 2005). Correspondingly, overexpression of BMP2/4 in Ciona (this study) and Halocynthia (Katsuyama et al., 2005) resulted in an increase in motoneurons. This may point to a recent loss of a requirement for BMP signalling in motoneuron formation in Ciona embryos, with Nodal a good candidate to have replaced BMP as the principal motoneuron inducer. The role of Nodal itself is also different from that in vertebrates, where Nodal is required for formation of the ventral most part of the spinal cord: the floor plate (Strähle et al., 2004). Intriguingly, Ciona Nodal is expressed in a bilateral pair of cells, b6.5, ablation of which leads to loss of early lateral neural plate markers and of motoneurons (Fig. 2B) (Hudson and Yasuo, 2005). This cell will give rise to the dorsal epidermis and dorsal neural tube. Thus, the embryonic origin of the Nodal signal in Ciona is equivalent to that of the dorsal signalling centre that patterns the spinal cord of vertebrates (epidermis and roof plate). This may be an example where geometric conservation of cellular interactions is accompanied by divergence in the signalling molecules deployed. We recently highlighted another example during the induction of a specific muscle precursor in Ciona and Halocynthia. Although the geometric identity of the interacting cells during the induction of this cell type in these two ascidian species is remarkably similar, the signalling molecules involved are not the same (Hudson and Yasuo, 2008). Surprisingly, it may be a common theme that, during evolution, the use of cellular interactions can be apparently conserved, while the underlying molecular mechanisms have been modified. A parallel conclusion from this study with important implications for evo-devo studies is that conservation of gene expression patterns does not necessarily point to an equivalent role for these molecules during embryonic patterning.

We thank N. Takatori (Osaka University) for generously sharing in situ hybridisation protocols prior to publication, and U. Rothbächer (IBDML), M. Gilchrist (Wellcome Trust/Cancer Research UK Gurdon Institute), P. Lemaire (IBDML) and collaborators for releasing the Unigene Collection Plates prior to publication. We are very grateful to the following people who helped this study: E. Houliston (UMR7009) for her constructive input and insightful discussions; J. Briscoe (MRC, London) for mPtcΔloop2; T. Kusakabe (Konan University) for pvAChTP::EGFP; T. Momose (UMR7009) for help with confocal microscopy and Image J; A. Roure, S. Darras and P. Lemaire (IBDML) for help with Gateway cloning, FOG constructs and Smad1/5-MO; H. Saiga and T. Ikuta (Tokyo Metropolitan University) for vAChTP cDNA; N. Satoh and colleagues (Kyoto University) for the Gene Collection Plates; and staff of the UMR7009 for all their help and support. All authors are CNRS staff. This work was supported by the Centre National de la Recherche Scientifique, the Université Pierre et Marie Curie, the Association Française contre les Myopathies (number 11734) and the Agence Nationale de la Recherche (ANR-09-BLAN-0013-01).