Antagonizing Retinoic Acid and FGF/MAPK Pathways Control Posterior Body Patterning in the Invertebrate Chordate Ciona intestinalis

Vertebrate embryos exploit the mutual inhibition between the RA and FGF signalling pathways to coordinate the proliferative elongation of the main body axis with the progressive patterning and differentiation of its neuroectodermal and paraxial mesodermal structures. The evolutionary history of this patterning system is still poorly understood. Here, we investigate the role played by the RA and FGF/MAPK signals during the development of the tail structures in the tunicate Ciona intestinalis, an invertebrate chordate belonging to the sister clade of vertebrates, in which the prototypical chordate body plan is established through very derived morphogenetic processes. Ciona embryos are constituted of few cells and develop according to a fixed lineage; elongation of the tail occurs largely by rearrangement of postmitotic cells; mesoderm segmentation and somitogenesis are absent. We show that in the Ciona embryo, the antagonism of the RA and FGF/MAPK signals is required to control the anteroposterior patterning of the tail epidermis. We also demonstrate that the RA, FGF/MAPK and canonical Wnt pathways control the anteroposterior patterning of the tail peripheral nervous system, and reveal the existence of distinct subpopulations of caudal epidermal neurons with different responsiveness to the RA, FGF/MAPK and canonical Wnt signals. Our data provide the first demonstration that the use of the antagonism between the RA and FGF signals to pattern the main body axis predates the emergence of vertebrates and highlight the evolutionary plasticity of this patterning strategy, showing that in different chordates it can be used to pattern different tissues within the same homologous body region.


Introduction
Antagonism between the Retinoic Acid (RA) and FGF signalling pathways controls the development of the posterior vertebrate body [1,2]. The tailbud provides a posterior source of FGF8 that keeps the caudal presomitic mesoderm (PSM) cells in an immature state and maintains the proliferative status of the spinal cord precursors [1,[3][4][5][6][7]. RA produced in the segmented somites and the anteriormost PSM by the enzyme Retinaldehyde dehydroge-nase2 (Raldh2) promotes PSM segmentation and triggers spinal cord patterning and differentiation [1,8]. The signalling cascades initiated by RA and FGF inhibit each other at multiple levels. In chicken, FGF8 prevents Raldh2 and the RA receptor RARb from being transcribed in the intermediate and posterior PSM, while RA downregulates FGF8 in the anterior PSM and the overlying neuroepithelium [1,9]. In Xenopus, RA induces the expression of MKP3, a phosphatase that inhibits the MAPK branch of the FGF signalling pathway, while FGF controls the expression of Cyp26, a P450-family cytochrome that hydrolyses RA [8]. The posterior displacement of the FGF8-producing tailbud during embryo elongation results in the formation of a caudalward-travelling wave of RA signal. This determination front allows paraxial mesoderm segmentation and spinal cord differentiation to proceed coordinately with the elongation of the embryo [1,2,8,10]. The opposing RA and FGF signals also control the collinear activation of Hox gene transcription: FGF activates genes located at the 59 end of the Hox cluster, while RA induces expression of the 39 end Hox cluster members [2,11,12]. Besides the RA/FGF system, a posterior-to-anterior gradient of canonical WNT activity has been described, with a role in coordinating the PSM maturation and segmentation [13]. Canonical WNT signalling also stimulates the expression of Raldh2 in the low FGF signalling environment of the anterior PSM, thus mediating the transition between FGF-and RA-sensitivity [9].
The strategy relying on opposing FGF/WNT and RA gradients to coordinate embryo elongation and patterning has been identified only in vertebrates, and it is unclear whether it is exploited by other chordates. Some support for the existence of a two opposing gradient system in the last common chordate ancestor comes from amphioxus, currently seen as the most basal extant chordate [14]. The amphioxus tailbud expresses several Wnt genes and at least one Fgf8-orthologue, Fgf8/17/18 [15,16]; RA controls amphioxus Hox gene expression, as well as AP patterning of the spinal cord, the epidermis and epidermal neurons [17][18][19]. On the other hand, recent work only shows a role of FGF signalling in the formation of the anterior paraxial mesoderm in amphioxus, but not in the segmentation and patterning of the more posterior somites [20].
Ascidians belong to the subphylum Tunicata (or Urochordata), the sister group to vertebrates [21]. Although their larvae have a tadpole-like chordate body plan, they show a number of derived features: their embryos are made of few cells and develop according to a fixed lineage [22]; elongation of the embryo takes place by spatial rearrangement of postmitotic cells without posterior growth [23,24]; mesoderm segmentation is absent [24]; the Hox gene cluster is disorganized and dispersed across two chromosomes; the temporal collinearity of Hox gene expression is lost and the spatial collinearity is only partially retained [25].
If the last common chordate ancestor already exploited opposing FGF/WNT and RA signals to coordinate AP elongation and patterning, then such a mechanism is likely to have undergone profound modifications in tunicates: it may have been altogether lost for lack of selective constraints, simplified, adapted to different constraints such as a determinative mode of development or recycled to fulfil novel functions. In any case, a better understanding of the molecular and functional interactions between FGF, WNT and RA pathways in the development of the posterior structures of ascidian larvae will not only contribute to clarifying the existence of a common minimum blueprint for generating the chordate body plan, but also provide insights into how the evolution of gene regulatory networks and other complex patterning mechanisms accompanies and/or underlies morphological and morphogenetic simplification in an extremely derived organism that retained a prototypical chordate body plan [22,26,27].
Here we show that in the elongating embryonic tail of Ciona intestinalis, the RA pathway is specifically active within the anterior tail epidermis, while the activities of the FGF/MAPK and WNT canonical pathways are restricted to the tailtip epidermis, a situation reminiscent of the two opposing anterior RA and posterior FGF/WNT gradients described in vertebrates. The RA and FGF/MAPK signals counteract each other, and this antagonism is required for the proper AP patterning of the tail epidermis and PNS. Thus, our data provide the first functional evidence that a strategy relying on the mutual antagonism between RA and FGF acts to pattern the posterior body of an invertebrate chordate, although in the very derived Ciona this may be used only to control the patterning of epidermal structures. This could reflect either a shared ancestral chordate usage of this system to control the anteroposterior patterning of all posterior body structures, followed by taxon-specific losses, or a case of repeated cooption of a particularly efficient developmental regulatory network during the course of chordate evolution.

Results
Regionalized activity of the RA, FGF/MAPK and canonical WNT pathways in the Ciona tail epidermis To address the potential role of the RA, FGF/MAPK and WNT signals in the patterning of Ciona embryonic posterior structures, we first defined the territories where these pathways are active during the processes of tail patterning and elongation.
We found that in the Ciona embryo tail, immunoreactivity against diphosphorylated Erk, a hallmark of the activation of the RTK-dependent MAPK pathway, is detected in the epidermal cells surrounding the tailtip from neurula [28] to tailbud stages (Fig. 1A, 2K, L). The dpErk signal is expanded by treatment with bFGF ( Fig. S1A) and is suppressed or decreased following the targeted epidermal expression of a dominant-negative form of the only Ciona FGF receptor, Ci-FGFR (Fig. 1A9). Two FGF factors are expressed in the tailtip region: Ci-fgf8/17/18, restricted to a very small population (2 to 4 cells) of epidermal cells at the tip of the extending tail from the gastrula stage onwards (Fig. 1B-F and [29]) and Ci-fgf9/16/20, expressed by some tailtip muscle cells ( [29] and Fig. S1B). Likewise, only two WNT genes are expressed at the end of the Ciona tail: Ci-wnt5 is expressed in tailtip epidermal cells and their progenitors from the late gastrula stage and throughout tailbud stages (Fig. 1H-L); Ci-wnt11-1/Ci-orphan wnt e is expressed by a few tailtip muscle cells ( [30] and http://aniseed-ibdm.univmrs.fr/gene-card.php?clusterid = cluster9820). Although WNT5 and WNT11 have often been considered as members of a ''noncanonical'' signalling subgroup, it has been recently shown that whether the canonical or the non-canonical pathways are activated depends more on the receptor/co-receptor context than on strictly defined WNT functional classes [31]. Interestingly, we found that Ci-LRP5/6, the Ciona homologue of LRP5/6, the WNT co-receptor required for canonical pathway activation, is expressed in the ventral midline epidermis (Fig. S2A). Consistently, WNT canonical activity, mapped with a reporter construct in which expression of b-galactosidase is under the control of multiple binding sites for the canonical WNT pathway effector, TCF [32], is restricted to two ventral territories in the Ciona tail, the endodermal strand and the posterior ventral midline epidermis (Fig. 1G). While the endodermal activity is the outcome of the pregastrula b-catenin requirement for endoderm specification [33], still visible due to the long half-life of b-galactosidase, the posterior epidermis activity likely reflects the activation of the WNT canonical pathway during tail elongation (Fig. S2B). Co-expression of an epidermally-targeted dominant-active form of b-catenin or embryo incubation with the GSK3-b inhibitor LiCl both result in ectopic reporter activity ( Fig. S2C and data not shown). Finally, expression of Ci-aldh1a1/2/3a, the Ciona homologue of the RAsynthesizing enzyme Raldh2 [34], was first detected in the anterior muscle precursors at late gastrula stage, then confined to the anteriormost 3-5 muscle cell pairs throughout the process of tail extension ( Fig. 1N-P and [35]). The RA-responsive territories, identified by using the reporter construct pCi-Hox1(intron2)/lacZ, in which the RA-responsive cis-regulatory sequences of the Ci-Hox1 gene control the expression of b-galactosidase ( [36] and Fig.  S3A), are localized in the trunk mesenchyme and epidermis, the muscle and the anterior tail epidermis (Fig. 1M). These structures coincide with, or are adjacent to, the territories expressing Ci-aldh1a1/2/3a.
In conclusion, our data reveal that, despite the major morphogenetic differences in the posterior elongation process existing between Ciona and the vertebrates, the regions of activity of the RA, FGF/MAPK and canonical WNT pathways in the posterior body bear striking analogies: RA is active in the anterior territories, FGF/MAPK and canonical WNT in posterior ones (Fig. 1Q).

Reciprocal control of FGF/MAPK and RA pathways
In vertebrates, the signalling cascades triggered by RA and FGF have been found to inhibit each other at multiple levels [1,2,8,9]. To address whether the RA and FGF/MAPK pathways also antagonize each other in the Ciona tail epidermis, we first analysed the expression of the RA-synthesising enzyme, Ci-aldh1a1/2/3a, and the RA-catabolising enzyme, Ci-cyp26, following activation or blockade of the RA and FGF/MAPK pathways between early gastrula and early tailbud stages. The expression of Ci-aldh1a1/2/ 3a is independent of RA signalling itself, as it is not modified by either exogenous RA, nor the known Raldh2 inhibitor DEAB (Diethylaminobenzaldehyde at any stage analysed (Fig. 2B, C and data not shown). As shown in Fig. 2D and data not shown, the number of muscle cells expressing Ci-aldh1a1/2/3a does not appear to increase in response to treatment of embryos with recombinant bFGF. On the other hand, the expression of Ci-aldh1a1/2/3a in the anterior tail muscle cells was lost or severely downregulated in embryos treated with the MAPK inhibitor U0126 between early gastrula and neurula stages ( Fig. 2E and data not shown).
Ci-cyp26, the Ciona homologue of the cytochrome enzyme responsible for degrading RA, has a complex pattern of expression in the tail epidermis, being strongly expressed in the anterior third of the tail, more weakly in the posterior third, and absent from the central third (Fig. 2F, and [35]). A feedback control exists between RA and Ci-cyp26, since its expression is strongly upregulated throughout the tail epidermis by treatment with RA at all stages analysed (Fig. 2G, data not shown and [35]), and suppressed in anterior tail region by 150 mM DEAB between early gastrula and neurula stages ( Fig. 2H and data not shown). At these stages, Ci-cyp26 expression in the tail epidermis is also expanded by treatment with bFGF, albeit to a lesser extent than by RA and mostly within the anterior and middle third of the tail ( Fig. 2I and data not shown). On the other hand, treatment with U0126 results in a complete loss of the anterior expression of Ci-cyp26 (Fig. 2J), a likely secondary effect of the suppression of Ci-aldh1a1/2/3a expression, as this loss is rescued in embryos treated with both U0126 and RA (Fig. S3B), while epidermally-targeted expression of a dominant-negative form of Ciona FGFR results in the partial loss of Ci-cyp26 expression (Fig. 2K).
Thus, our data show that the levels of RA in the Ciona embryo tail are regulated by both the RA pathway itself and the FGF/ MAPK pathway, with MAPK controlling the production of RA within the anterior tail muscle cells and FGF promoting its degradation in the tail epidermis. To explore whether the FGF/MAPK control of the RA pathway is mirrored by a symmetrical control exerted by RA on the FGF/MAPK activity, we analysed the phosphorylation status of Erk in the tail epidermis of Ciona embryos treated with 1.5 mM RA at gastrula stage. As shown in Fig. 2L-O, RA treatment leads to a decrease of Erk diphosphorylation in the epidermis, distinct from the global extinction of the signal following treatment with the MAPK inhibitor U0126 (Fig. S1A).
To summarize, multiple levels of interaction exist between the FGF/MAPK and RA pathway (Fig. 2P).

FGF/MAPK and RA pathways control the AP patterning of Ciona tail epidermis
The timing of our pharmacological treatment experiments shows that the antagonism between the RA and FGF/MAPK pathways in the Ciona tail epidermis is set up between the midgastrula and late neurula stages, before the initiation of tail elongation and concomitantly with the establishment of the epidermal AP patterning [25]. We therefore explored whether the AP patterning of Ciona tail epidermis might be controlled by the RA and the FGF/MAPK signals.
The tail epidermis shows only a limited degree of AP patterning: in addition to the above described expression of Ci-fgf8/17/18 and Ci-wnt5 by cells at or around the tailtip, Ci-hox12 is expressed by epidermal cells in the posteriormost third of the tail, and Ci-hox1 by those of the anteriormost third (Fig. 3A, B and [25]). As shown in Fig. 3E-H and data not shown, treatment of embryos with 1.5 mM RA from early gastrula to neurula stages leads to an expansion of Ci-hox1 expression throughout the entire tail epidermis and to a complete loss or a severe reduction of both Ci-hox12 and Ci-wnt5 expression, with only a residual signal left in cells at the very tip of the tail, while the expression of Ci-fgf8/17/18 is not affected (Fig. 3H). When the RA signal is blocked by treatment with 150 mM DEAB at the same stages, the expression of Ci-hox1 is completely lost from both epidermis and neural tube (Fig. 3I), whereas Ci-hox12, Ci-wnt5 or Ci-fgf8/17/18 are not significantly affected (Fig. 3J-L and data not shown).
On the other hand, treatment with recombinant bFGF from early gastrula to neurula results in an anterior expansion of the Ci-hox12 and of the Ci-wnt5 expression domains (Fig. 3N, O), again without affecting the expression of Ci-fgf8/17/18 (Fig. 3P). Interestingly, bFGF treatment leads to the loss of Ci-hox1 expression from the anterior tail epidermis, but not from the anterior neural tube, in about 30% of the embryos (Fig. 3M) and to a similar, albeit weaker, reduction of the pCi-Hox1(intron2)/ lacZ reporter epidermal activity (Fig. S3C). The upregulation of Ci-hox12 and Ci-wnt5 following treatment with bFGF is at odds with a previous report in which SU5402, a pharmacological inhibitor of the FGF receptor, was used to rule out a role of the FGF pathway in controlling the expression of these markers [37]. To better understand the role of the FGF pathway in patterning the tail epidermis, we analysed embryos in which the reception of the FGF signal was specifically disrupted in the epidermis by means of the dominant-negative form of Ci-FGFR. As shown in Fig. 3Q-T, blocking the FGF pathway in the epidermis results in the ectopic expression of Ci-hox1, while leading to the decrease or loss of the expression of Ci-hox12, Ci-wnt5 and Ci-fgf8/17/18.
Inhibition of the MAPK pathway with 10 mM U0126 leads to the complete suppression of Ci-hox1 (Fig. 3U), revealing an action of the MAPK pathway that does not depend on the reception of the FGF signal in epidermis, and is likely related to the MAPK requirement for Ci-aldh1a1/2/3a expression (Fig. 2C). U0126 treatment also leads to the downregulation of Ci-hox12, Ci-wnt5 and Ci-fgf8/17/18 (Fig. 3V-X).
Overall, our data show that the RA and FGF/MAPK pathways exert partially opposing effects on the AP patterning of Ciona tail epidermis (Fig. 3Y): RA is required to establish and maintain the anterior identity of the tail epidermis and is sufficient to inhibit its posterior identity, while FGF is required to establish the posterior tail epidermis character, but also negatively controls the anterior epidermis identity. Therefore, the antagonism between the RA and FGF/MAPK signals in the Ciona tail is no mere atavism, but a functioning patterning system.

Distinct anterior and posterior subpopulations of caudal epidermal neurons (CENs)
We next searched for differentiated Ciona tail structures that could be under the control of the patterning system relying on the antagonism between RA and FGF. In vertebrates, this antagonism has a role in the control of paraxial mesoderm segmentation and neuroectoderm differentiation [1,2,8]. Although no segmentation has been identified in Ciona embryos and larvae, pairs of caudal epidermal neurons (CENs) constituting the posterior larval peripheral nervous system (PNS) are distributed with some periodicity along the dorsal and ventral midlines of the tail (Fig. 4A and [38]) and, together with the acellular tail fin, are the only morphologically differentiated tail structure. We tested whether the CENs are regionalized along the AP axis by analyzing their birthdate and morphology. Examination of embryos hybridized with the neuronal-specific marker Ci-etr at initial-to early-tailbud stages revealed that the posteriormost ventral CENs (vCENs) are consistently born earlier than the anterior ones (Fig. 4C, C9; see also [38]). We next drove GFP expression in CENs using the Ci-Vesicular Glutamate Transporter (Ci-VeGluT) promoter [39] and analysed the morphology of the labeled CENs cell bodies and axons at the hatching larva stage. We found that all the dorsal CENs as well as the one or two posteriormost vCENs emit long axonal projections (Fig. 4B, D-E9), while the anteriormost vCENs have only short, stumpy axons or no axon at all (Fig. 4F, F9). Moreover, the posteriormost vCENs establish contact with the axonal projections of other GFP-positive cells, such as the dorsal CENs and the dorsal subepidermal bipolar cells (Fig. 4E, E9) [40], while the anterior vCENs do not appear to establish contact with any other GFP-positive cell. To quantify the morphological differences between anteriormost and posteriormost vCENs, we measured the axonal length in fluorescence microscopy pictures of 42 randomly chosen pCi-VeGluT::GFPexpressing vCENs from 14 different larvae and plotted it against the cell body position along the larval tail (Fig. 4G). K-means parameter cluster analysis of the scatterplot revealed the existence of two distinct clusters of 21 points each, whose centroid coordinates are axonal length = 15.9 mm, distance from trunk/tail junction = 233.8 mm, and axonal length = 173,8 mm, distance from trunk/tail junction = 565.4 mm respectively.
Thus, our data reveal unexpected morphological differences among Ciona CENs and support the existence of distinct anterior and posterior subpopulations of vCENs. Consistent with this hypothesis, posterior and anterior vCENs derive from different epidermis precursor blastomeres at the onset of gastrula and thus have a partially distinct lineage history [38]. These two neuronal subpopulations may form in response to distinct developmental programs, as it is also suggested by the observation that the anterior vCENs are absent in the distantly related ascidian Halocynthia roretzi [41,42].

FGF/MAPK, RA and canonical Wnt pathways control the number and distribution of vCENs
We finally asked whether the FGF/MAPK and RA pathways, in addition to controlling the AP patterning of the tail epidermis, also affect the vCENs formation and distribution. The number and distribution of Ci-etr-positive vCENs at late tailbud stage (stages 23 and 24 according to [43]) were thus analyzed in embryos treated at different time points with bFGF, RA or with the pharmacological inhibitors U0126 and DEAB. In addition, embryos treated with the GSK3-b inhibitors LiCl and BIO, as well as embryos expressing a dominant-active form of b-catenin, were also analyzed to explore a possible role of the canonical Wnt pathway. As shown in Fig. 5A, B and data not shown, all the treatments tested have an effect on the number of vCENs. Treatment with RA starting at early gastrula to late neurula stages (stages 11 to 16 according to [43]) leads to a significant increase in vCEN number, mirrored by a comparable decrease in embryos treated with the RA pathway inhibitor DEAB (Fig. 5A, B and data not shown). Interestingly, posterior and anterior vCENs were differentially affected. RA stimulation leads to an increased number of anterior vCENs and to a loss of posterior ones (Fig. 5A, C). The supernumerary anterior vCENs in RA-treated embryos arise within the neurogenic midline region and are always intercalated with Ci-etr negative epidermal cells, thus placing RA downstream of the BMP signal which induces the neurogenic ventral midline [38] and upstream of the Delta/Notch-dependent lateral inhibition process which controls the number of CENs and epidermal cells within the midline [38]. Conversely, DEAB treatment had little effect on the posterior vCENs, but strongly affected the anterior ones (Fig. 5A, C). Consistent with the opposing activities of RA and FGF pathways during gastrula and early neurula stages, ectopic application of bFGF from the early gastrula stage repressed the formation of anterior vCENs, without effect on the posterior population (Fig. 5A, B, C). Treatment with U0126 at early gastrula stage (stage 11) led to the loss of the posteriormost CENs, and also had a weak effect on the anterior ones (Fig. 5C), likely reflecting the control of RA synthesis by MAPK. On the other hand, U0126 treatment at later stages did not affect significantly the number of vCENs (Fig. 5A, B, C). Finally, activation of the canonical Wnt pathway by treatment with either LiCl (Fig. 5A, B, C) or BIO (data not shown) or by epidermally targeted overexpression of activated b-catenin (Fig. 5A) from the mid-neurula stage represses the formation of the anterior vCENs, without affecting the posterior ones.
To correlate these observations with the morphological differences between anterior and posterior vCENs described in Fig. 4, we measured the axonal length and the cell body position along the larval tail in pCi-VeGluT::GFP-expressing vCENs from embryos treated with RA or bFGF at (Fig. 5D). In both cases, cluster analysis revealed the presence of only one cell population: in RA-treated larvae the coordinates of the cluster centroid are axonal length = 8.4 mm, distance from trunk/tail junction = 276.1 mm, while in bFGF-treated larvae they are axonal length = 198.7 mm, distance from trunk/tail junction = 575.1 mm. This shows that RA treatment results in a depletion of the posterior vCENs population, while treatment with bFGF has the opposite effect. Figure 5E summarizes the effect of the interference with the RA, FGF/MAPK and Wnt canonical signalling pathways on vCENs formation.

Conservation and adaptation of a signalling and transcriptional patterning network
Although Ciona larvae and tailbud-stage vertebrate embryos have a remarkably similar body plan, the morphogenetic processes used to build their shared tadpole shape differ significantly. In vertebrates, embryo elongation is brought about by a combination of continuous cell proliferation in the tailbud and convergent-extension movements. On the other hand, Ciona embryos lack a posterior growth zone and tail elongation is mostly obtained via changes in the position and shape of postmitotic cells [23,24]. Moreover, the AP patterning of the Ciona epidermis is not progressively laid down throughout the course of tail extension, but initially established between the late gastrula and early tailbud stages, then maintained during the subsequent elongation step [25]. Despite the different morphogenetic strategies exploited by Ciona and the vertebrates to shape their posterior body and the divergence in the differentiated structures that constitute it, we found extensive conservation of a system of opposing RA and FGF/MAPK signals to control spatial Hox genes expression and posterior body patterning. This is in contrast with what described in insects, where two divergent morphogenetic strategies (shortand long-germband) use different segmentation and patterning mechanisms [44], and is also surprising in view of the global divergence in orthologous gene expression patterns between ascidians and vertebrates [45].
Our data show that not only small cell-autonomous kernels based on interactions among transcription factors [27], but also a longer-range patterning mechanism relying on diffusible molecules may be conserved in its basic principles and some of its transcriptional targets over at least 500My of independent evolution. However, in the simplified context of Ciona embryos, the details of interaction between the RA and FGF signals differ from the model described in vertebrates. Accordingly, the mutual inhibition between RA and FGF in Ciona seems to take place only at the level of signal transduction or degradation rather than synthesis, as is the case in mouse or chicken [1,2]. Just as the regulatory interactions which structure the antagonism between the two pathways differ from the vertebrate paradigm, the links between this patterning system and its final output are modified. In vertebrates, the caudal-related homeobox transcription factors of the Cdx family mediate the FGF transcriptional activation of posterior Hox genes [12]. In Ciona, the unique Cdx factor Ci-cdx is surprisingly expressed throughout the tail epidermis with the exception of the posteriormost cells, where FGF/MAPK signalling is active [37]. Despite these differences our data reveal an unexpected level of conservation in Hox gene regulation between Ciona and vertebrates: as in vertebrates, RA promotes the expression of the anterior gene Ci-hox1 and represses the posterior Ci-hox12, while FGF has the opposite effect [11,12]. Thus, although in Ciona the loss of temporal collinearity has allowed for (or was allowed by) the disintegration of Hox clustering, the maintenance of spatial collinearity is accompanied by the conservation of upstream regulatory factors. To what extent this holds true in other metazoans with various degrees of Hox gene clustering and temporal/spatial collinearity [46] remains an open question.
Our work thus provides a further addition to a growing list of cases where the precise relationships among regulatory network components vary from one metazoan species to the next, but the ultimate outcome of the network is comparable [44]. In particular, recent work has revealed differences in the precise interactions of the RA, FGF and Wnt signalling cascades during posterior axis elongation and segmentation in different vertebrates [47]. Network wiring differences between Ciona and the vertebrates may be related to the different dynamics of the antagonistic RA and FGF activities: rather than generating a moving determination front that progressively shifts posteriorly during elongation, in Ciona embryos RA and FGF define two stable determination fields in the epidermis before the onset of the tail extension process or at its very beginning.

Plasticity of an ancestral chordate strategy?
In vertebrates the antagonism between RA and FGF has been shown to control the patterning of the paraxial mesoderm and spinal cord [1,2], whereas our data show that in Ciona the opposition between these two signals is required to pattern the posterior epidermis. This raises the possibility that in ancestral chordates, the RA/FGF antagonistic coupling was exploited to regulate the patterning of a broad range of embryonic derivatives, including mesoderm, neurectoderm and non neural epidermis, and that in the course of evolution, different chordate taxa have undergone germ layer-specific losses of this patterning mechanism. Such a situation could reflect taxon-specific differences in the degree of patterning complexity of each germ layer, exemplified by the complex patterning of epidermis and the absence of paraxial mesoderm patterning in Ciona. Alternatively, it is possible that in some cases the requirement for the antagonism between RA and FGF has been overlooked. As an example, it has been shown in amphioxus that RA affects the AP patterning of the epidermis and the distribution of the epidermal neurones [18], but it is currently not known whether and how FGF also affects these processes; FGF controls the segmentation and patterning of the anteriormost Amphioxus somites [20], but the role of RA in the segmentation and patterning of the paraxial mesoderm is still unclear. Similarly, almost no data is currently available concerning the AP patterning of trunk epidermis in vertebrates and the role that the antagonism between RA and FGF signals might exert on this process.
Morphological heterogeneity of the Ciona CENs: a driving force for AP tail patterning?
Our work reveals the existence of subpopulations of Ciona CENs with distinct morphologies and different susceptibility to the RA, FGF and Wnt signals. The morphological heterogeneity of the tail PNS neurons is most likely imposed by their function during the Ciona life cycle, which is still unknown. Whatever its function is, it is tempting to speculate that in Ciona, the presence of a strongly regionalized tail PNS represents the selective force requiring the maintenance of a complex signalling system responsible for controlling the AP pattern of the posterior body. In this respect, it is worth noting that in the larvacean Oikopleura dioica, which lacks a tail PNS analogous to that of Ciona (H. Nishida, personal communication), most of the RA signalling machinery is missing [48] and the Hox cluster is completely disorganized [49]. Confirmation of this hypothesis will require further investigation of the tail AP patterning, as well as of the PNS anatomy, development and function in a wider, evolutionarily significant range of ascidians and tunicates.

Conclusions
We show that, regardless of the different morphogenetic strategies they adopt to build the chordate body plan, the vertebrates and the tunicate Ciona intestinalis exploit a very similar mechanism, based on the antagonism between the RA and FGF/ MAPK signals, to pattern the posterior body of their embryos and larvae. Thus this mechanism, until now described only in vertebrates, likely represents an ancestral trait common to all olfactores. Remarkably, while in the vertebrates the opposition between RA and FGF/MAPK coordinates the patterning of the posterior neural tube and paraxial mesoderm, in Ciona it controls the AP patterning of the posterior epidermis and the associated PNS, suggesting a high degree of functional plasticity throughout evolution. We also propose that the presence of a relatively complex posterior PNS might have been the selective constraint responsible for retaining the RA vs FGF/MAPK antagonism in Ciona. Altogether, our work supports the existence of a common minimum blueprint for generating the chordate body plan but also shows how this has been modified and restricted or redeployed to only a subset of tissues in an extremely derived organism that retained a prototypical chordate body plan.

Materials and Methods
Embryo culture and pharmacological treatments C. intestinalis adults were collected by the Biological Sample Collection Service of the Station Biologique de Roscoff, France. Fertilization and embryo culture were performed as previously described [50]. Embryos were treated with soluble recombinant human bFGF (10 ng/ml; Sigma), UO126 (10 mM; Calbiochem), or LiCl (150 mM or 300 mM), as described in [32,50]. Treatment with RA (1.5 mM; Sigma) or DEAB (150 mM; Sigma), were performed according to [36].

Anti-dpErk immunostaining
Detection of diphosphorylated Erk was performed with a modified version of the protocol described in [50]. Briefly, embryos were fixed for 309 in 4% paraformaldehyde, 0.2% glutaraldehyde in artificial sea water, supplemented with 50 mM NaF and 100 mM Na 3 VO 4 to inhibit endogenous phosphatases. After EtOH dehydration, embryos were re-hydrated in PBTri (PBS+ Triton 0.1%), treated for 109 with H 2 O 2 0.6% in PBTri, blocked in PBTri, 1% Roche Blocking Reagent, 10% goat serum, then incubated overnight with an anti-dpErk antibody (1/1200, Sigma). After extensive washing in PbTri, embryos were incubated overnight with a HRP-conjugated anti-mouse antibody (1/200, Molecular Probes), followed by revelation with the TSA Plus Fluorescein Kit (Perkin Elmer). After the revelation, the embryos were fixed for 309 in 4% paraformaldehyde, extensively washed in PBTri, then blocked and incubated overnight with an APconjugated anti-fluorescein antibody (Roche), followed by NBT/ BCIP staining.
Quantification of vCENs number, position and morphology vCENs were stained by ISH against Ci-etr, and their number was counted in control and treated embryos. To quantify their distribution along the tail, embryos were photographed and analysed with the ImageJ software. The following parameters were analysed: total embryo length (from palps to tailtip); total tail length (from the ventral junction between trunk and tail to the tail tip); anteriormost vCEN position (distance between the trunk/tail junction and the anteriormost Ci-etr+ cell); posteriormost vCEN position (distance between the tailtip and the posteriormost Ci-etr+ cell). All the statistical analyses on neuron number and position were performed using ANOVA with Tukey post hoc test in R environment. To quantify the morphological differences between anterior and posterior vCENs, fertilized eggs were electroporated with the pVGluT::EGFP construct, then let develop until hatching larva stage in controls condition or in the presence of RA or bFGF from late gastrula stage. The larvae were processed for anti-GFP immunofluorescence as described above, then individual vCENs were photographed and analysed with the ImageJ software. The following parameters were measured: cell body surface, axonal length and distance of the cell body from the trunk-tail junction. Cluster analysis was performed with the TANAGRA software (http://chirouble.univ-lyon2.fr/,ricco/tanagra/en/tanagra.html).