A novel role for retinoids in patterning the avian forebrain during presomite stages

The precise nature of signals and mechanisms underlying the process of AP patterning is not entirely understood. At present, two main models have been put forward to explain how these events work at a molecular level in the early vertebrate embryo. Whereas Spemann and collaborators proposed separate organizer centres for inducing different regions of the central nervous system(CNS) (Spemann, 1931; Spemann, 1938), Nieuwkoop postulated that during gastrulation, induction and patterning of the nervous system occur in two steps (Nieuwkoop et al., 1952; Nieuwkoop and Nigtevecht, 1954). During the first step or `activation', neural fate is induced and the forebrain is specified, whereas later in development,some cells receive other signals that cause them to acquire a more caudal character (`transformation or caudalization step'). Recently, Stern and collaborators have proposed a modification of Nieuwkoop's model that consists of three successive steps (Foley et al.,2000; Stern,2001). The first step occurs before gastrulation in which a labile, pre-neural and pre-forebrain state is induced in the overlying epiblast (`activation step'). This is mainly based on the observation that, in chick, the hypoblast, an early endodermal layer present before gastrulation,can transiently induce expression of early neural markers when transplanted in contact with competent epiblast (Foley et al., 2000). However, to maintain a forebrain or even a neural phenotype, cells must receive stabilizing signals (`stabilization step'). This maintenance function might be provided by prechordal plate mesendoderm and/or anterior head process, both of which emerge from the organizer region during gastrulation (Ang and Rossant,1993; Foley et al.,1997; Pera and Kessel,1997; Schier et al.,1997; Dale et al.,1999; Rowan et al.,1999; Camus et al.,2000). Finally, the `transformation' step caudalizes the nervous system through transforming signals emanating from more posterior structures(reviewed by Wilson and Rubenstein,2000). What are the signalling molecules involved in patterning the early CNS? Candidate signals include fibroblast growth factors (FGFs), and components of the Nodal and Wnt pathway and/or their antagonists. Most of these molecules are expressed transiently before gastrulation in the chick hypoblast or in its mouse equivalent, the anterior visceral endoderm (AVE),and during gastrulation in node-derived structures(Varlet et al., 1997; Foley et al., 2000; Streit et al., 2000; Lawson et al., 2001; Chapman et al., 2002),suggesting conserved molecular mechanisms during the various steps of early AP patterning.

Like FGF and Wnt proteins, retinoic acid (RA) is mainly considered to be a posteriorizing factor with a high concentration posteriorly and a decreasing gradient toward the anterior end of the embryo(Chen et al., 1994; Maden et al., 1998; Maden, 1999). Retinoids comprise various isoforms, including all-trans-RA and 9-cis-RA, which are generated from vitamin A in a two-step oxidation process involving a group of enzymes belonging to the family of alcohol dehydrogenases(Duester, 2000; Ross et al., 2000). Retinoids act via specific nuclear receptors belonging to the steroid superfamily of ligand-activated transcription factors, which play fundamental roles in embryological morphogenetic processes (reviewed by Zile, 2001). The effects of retinoid signalling on the CNS have been analysed in gain- and loss-of-function experiments, which have generally revealed an involvement of RA in AP patterning. By adding exogenous RA or constitutively activating retinoic acid receptors (RAR) at early stages of development, the forebrain becomes dramatically reduced, whereas the hindbrain is expanded(Durston et al., 1989; Sive et al., 1990; Simeone et al., 1995; Avantaggiato et al., 1996; Zhang et al., 1996). To reproduce a loss-of-function of retinoid signalling, various strategies have been employed: vitamin A has been depleted in pregnant mothers of quails or rodents (Maden et al., 1996; Gale et al., 1999; White et al., 2000), dominant negative RARs have been injected into Xenopus embryos(Blumberg et al., 1997; Kolm et al., 1997; van der Wees et al., 1998) or,more recently, antagonists to RARs have been added in a temporally and spatially controlled fashion (Wendling et al., 2000; Dupé et al., 2001). These, and the inactivation of the retinaldehyde dehydrogenase Raldh2 in mouse and zebrafish(Niederreither et al., 1999; Begemann et al., 2001; Grandel et al., 2002) have pointed to defects restricted mainly to the hindbrain and spinal cord regions. Thus, it remains unclear whether retinoids have an endogenous role in forebrain patterning.

Two sites of localized retinoid activity have been described in the developing forebrain at later stages. One site is localized in the rostrolateral head mesenchyme and has been visualized with RA-reporter mice;at this site, activity overlaps with a transient expression of Raldh2(Rossant et al., 1991; LaMantia et al., 1993; Wagner et al., 2000). A second source of RA synthesis is represented by the expression of Raldh3 in the surface ectoderm and in a large part of the ventral rostral head at later stages (Li et al., 2000; Mic et al., 2000; Smith et al., 2001). It has been suggested that the ventral ectoderm RA source, which is also present in the chick embryo, might be involved in sustaining the outgrowth of the frontonasal process and forebrain(Schneider et al., 2001). This and other reports have therefore confirmed an essential role for retinoids in the maintenance of a proper craniofacial structure, mainly through the survival and proliferation of head mesenchyme(Smith et al., 2001). However,a direct role for RA in the early events of anterior neural patterning remains unclear.

In this report, we show for the first time that, in avian embryos, RA is synthesized in the hypoblast before gastrulation, and in anterior definitive endoderm and prechordal mesendoderm during gastrulation. As a first attempt to challenge the role of RA in anterior patterning, we used the well-established vitamin A-deficient quail model (VAD), which has been considered to be similar to an `RA knockout' (Zile,2001). We show that VAD embryos have altered molecular properties in the anteriormost endoderm, which comprises hypoblast and definitive endoderm, during the process of axial mesoderm specification. Furthermore,prechordal mesendodermal and prospective telencephalic markers are expanded posteriorly, and in early somite embryos increased apoptosis is detected in ventral neuroectoderm and foregut endoderm. At later stages, VAD quail embryos show a single undivided telencephalic vesicle, increased cell death in ventral mesenchymal and neuroectodermal cells and abnormal dorsoventral patterning in the diencephalon. Thus, we hypothesize that, in the forebrain, retinoids have a biphasic activity that reflects differential functional contexts of RA signalling in anterior regions: an early phase at gastrula stages, during which RA refines the rostrocaudal identity of the anterior neural plate by modulating signalling involved in axial mesoderm specification (this study);and a late phase at neural pore closure, during which RA is involved in the survival of the frontonasal mesenchyme mass and forebrain neuroectoderm (this study) (Schneider et al.,2001). We propose that during the early phase, RA might be one of the signalling molecules involved in the `stabilization step' of the model proposed by Stern and colleagues (Foley et al., 2000; Stern,2001).

Fertilized chick and quail eggs were incubated at 38°C until the required developmental stages. Incubation times and staging were as per Hamburger and Hamilton (Hamburger and Hamilton, 1951). Normal and VAD embryos were staged based on as many parameters as possible; however, in some cases, VAD embryos looked more differentiated than their normal counterparts despite having the same stage or somite number (differences in stage matching are indicated in the legends). VAD quail embryos were obtained from Japanese quail (Coturnix coturnix japonica) hens fed on a semi-purified diet with 13-cis-retinoic acid as the only source of vitamin A, as described in detail by Dersch and Zile(Dersch and Zile, 1993),whereas normal quail embryos were obtained from the same quail hens fed on a normal diet. Embryos for whole mount in situ hybridization and immunohistochemistry were fixed in 4% PFA/PBS at 4°C, whereas embryos for histology were fixed in Bouin's solution (0.9% picric acid, 9%formaldehyde).

Whole-mount in situ hybridization with digoxigenin-labelled riboprobes was performed as previously described (Grove et al., 1998). Chick riboprobes were used in quail embryos after careful comparison with their equivalent published chick expression profiles. The Gsc riboprobe differs from the original version(Izpisua-Belmonte et al.,1993) by covering most of the 5′ part of the cDNA sequence(50 bp of 5′ UTR and 738 bp of coding sequence; see also GenBank Accession Number X70471). Embryos for sectioning were embedded in 20% gelatine and fixed in 4% PFA/PBS for at least 2 days. Sections were cut at 30 μm on a vibratome and mounted in 90% glycerol before being photographed using a digital camera. For histology, embryos were embedded in Fibrowax and sectioned at 10 μm on a microtome. After the dewaxing and rehydration procedures,sections were stained with Haematoxylin and Eosin for 2 minutes.

Free-floating 30 μm gelatine-embedded sections were washed in PBS followed by three 1 hour washes in a blocking solution consisting of 5% goat serum and 1% Triton X-100. Sections were then incubated overnight with anti-phospho-Histone H3 (Upstate, USA) or anti-HNK1 (Zymed Laboratories, USA)at a concentration of 1:500. After three 1 hour PBS washes, sections were incubated overnight with Cy3 (Jackson IRL, USA) at a concentration of 1:100. For the TUNEL method we used the In Situ Cell Death Detection, POD kit(Roche), with some modifications. Free-floating sections were first washed in 1% Triton X-100 in PBS and, subsequently, incubated in the TUNEL-reaction mix provided in the kit for 3 hours at 37°C. Sections were then washed in 1%Triton X-100/PBS and incubated at 4°C overnight in the converter POD provided in the kit. All sections were mounted in 90% glycerol and viewed under a confocal microscope (BioRad, MRC-600).

To establish whether there are sources of retinoids in anterior regions, we looked at the expression of RA-synthesizing enzymes in the early chick embryo. We found that Raldh2 starts to be expressed at stage XIV-2 in the posterior region where the primitive streak is formed (data not shown). By stage 3, Raldh2 is expressed posteriorly in the elongating primitive streak and more anteriorly in the migrating primary hypoblast(Fig. 1A). By stage 4–the hypoblast expression is displaced further anteriorly into the forming germinal crescent, which extends bilaterally and caudally(Fig. 1B). In addition, Raldh2 is expressed posteriorly in most of the primitive streak and in non-axial mesoderm as previously described(Berggren et al., 1999). At the definitive primitive streak stage, Raldh2 expression is detected rostral to the node in the anteriormost endoderm, which is composed of hypoblast and anterior definitive endoderm(Fig. 1C; inset)(Vesque et al., 2000). At head-process stages, anterior Raldh2 is expressed in the characteristic fan-shaped structure corresponding to prechordal mesendoderm and at lower levels in anterior endoderm(Fig. 1D; see Fig. 3A)(Seifert et al., 1993). Transverse sections along the rostrocaudal axis show strong Raldh2expression in prechordal mesendoderm cells and in non-axial mesoderm, whereas expression in the node is weak (Fig. 1D′-D′′). At early somite stages, anterior Raldh2 expression is restricted to prechordal mesoderm and foregut endoderm, with a sharp posterior boundary at the level of the notochord(arrows in Fig. 1E,E′). Interestingly, we also found that Raldh3 (previously known as Aldh6) is transiently expressed at stage 4+/5– in the node and in migrating mesodermal cells, as clearly shown in mid-sagittal sections(Fig. 1F,F′). This expression is quickly downregulated and, as already described, Raldh3expression becomes confined to the oral ectoderm at later stages (data not shown) (Suzuki et al., 2000). No expression of Raldh1 has been detected at these developmental stages (data not shown) (Godbout,1992).

To determine the sites of RA action in the early embryo, we analyzed the distribution of various isoforms of retinoic acid and retinoid X receptors(RAR and RXR, respectively), when both Raldh2 and Raldh3show localized anterior expression. We found that most RAR isoforms are either ubiquitously expressed in the embryo or localized at the anterior and lateral marginal zone, whereas RXR isoforms show a more restricted expression pattern(data not shown). RXRγ expression is restricted to mesodermal precursors located in the node region (Fig. 1G,G′), suggesting that RXRγ could be the ideal partner in the specification and migration of mesodermal cells during gastrulation.

In summary, we found that two RA-synthesizing enzymes are expressed in anterior structures involved in early forebrain induction and patterning. Whereas Raldh2 expression is located in the hypoblast in pre-gastrula embryos and in prechordal mesendoderm during gastrulation, Raldh3 is transiently expressed in the node and in axial mesodermal precursors emerging from the node.

The striking expression of Raldh2 in anterior endoderm at a stage when this structure provides signals necessary for axial mesoderm specification (Vesque et al.,2000) suggests an endogenous role of RA signalling in patterning early endodermal and mesodermal properties. To functionally test this hypothesis, we took advantage of the VAD quail model, which lacks biologically active RA throughout development. We first assessed whether TGFβ family members were altered in anterior endoderm, which at stage 4/5 is composed of hypoblast and anterior definitive endoderm(Fig. 2A-A′′,C-C′′)(Vesque et al., 2000). Surprisingly, in VAD embryos Bmp2 and Bmp7 have stronger levels of expression in hypoblast cells when compared with normal embryos(n=6; asterisks in Fig. 2B,B′,D,D′), which suggests an upregulation of BMP signalling in the hypoblast. Furthermore, the anterior definitive endoderm domain of Bmp2, and to a lesser extent of Bmp7, expression is laterally expanded, although levels of transcripts do not change(arrowheads in Fig. 2B,D,B′′,D′′).

Next, we tested whether the homeobox gene Hex, a necessary component for anteriormost axial patterning(Martinez-Barbera et al.,2000) and expressed in hypoblast and anterior definitive endoderm(Fig. 2E,E′,E′′)(Yatskievych et al., 1999; Chapman et al., 2002), was affected in VAD embryos. At stage 5, VAD embryos show a dramatic downregulation of Hex expression in anterior regions (arrow in Fig. 2F). Transverse sections indicate that the decreased expression is mainly localized in anterior definitive endoderm (n=5; Fig. 2F′′), whereas the hypoblast layer has only slightly reduced expression levels (Fig. 2F′). Furthermore, expression of the Wnt inhibitor crescent (Crs) (Pfeffer et al., 1997) is also dramatically downregulated in anterior and lateral definitive endoderm of stage 6– VAD embryos (n=3; Fig. 3E-F′′),whereas the prechordal mesodermal expression is expanded posteriorly (arrows in Fig. 3F; see also next section).

Altogether, these data suggest an endogenous role for retinoids in modulating anterior endodermal signalling, which has been shown to be required in forebrain development. However, whereas BMP genes have altered expression levels mainly in primitive endoderm of VAD embryos, the expression of Hex and Crs is affected more in anterior definitive endoderm, suggesting distinct roles for RA in patterning anterior endodermal properties.

Considering the fundamental role attributed to anterior endoderm (hypoblast and anterior definitive endoderm) in the specification of axial mesoderm(Vesque et al., 2000), we were interested in establishing whether prechordal mesoderm and notochordal markers were perturbed in VAD quail embryos.

As described above, anterior Raldh2 expression at stage 5/6 is specifically upregulated in prechordal mesendoderm but is absent from notochord cells (Fig. 1; gap of expression in Fig. 3A and inset). Surprisingly, we found that stage 6 VAD embryos have patches of ectopic Raldh2 expression along the whole anterior axial mesoderm(n=2; Fig. 3B,arrowheads in inset). Transverse sections at the level of presumptive rostral notochord indicate the lack of Raldh2 expression in normal embryos and ectopic mesodermal expression in VAD embryos(Fig. 3A′,B′). Next, we looked at the expression of the transcription factor gene goosecoid (Gsc) and at the late expression of Crs,Bmp2 and Bmp7 in prechordal mesodermal cells(Fig. 3C,E,G; data not shown). Transverse sections reveal no Gsc expression in the notochord of normal embryos (Fig. 3C′), whereas abnormal ectopic expression is detected in VAD embryos (n=5; Fig. 3D′), and Crs and Bmp2 midline expression is expanded towards the node when compared with normal embryos (n=7;arrows in Fig. 3F,H). Next, to ascertain whether the extension of the anterior notochord was altered in VAD embryos, we used chordin as a molecular marker for notochord cells in stage 7 embryos (Fig. 3I,I′) (Vesque et al.,2000). In VAD embryos the chordin-free region is posteriorly expanded, and at presumptive anterior notochord levels, expression is mostly undetectable (Fig. 3J, inset,J′). In summary, these data indicate abnormal patterning of axial mesoderm, i.e. prechordal mesodermal markers are expressed more posteriorly,in presumptive anterior notochord.

Furthermore, and surprisingly, transverse sections at all rostrocaudal levels reveal ectopic Bmp2 and Bmp7 expression in the ventral neural midline of VAD embryos (n=8; Fig. 3H′; data not shown), which suggests that BMP signalling is also altered in the ventral neural tube. It has been reported that the coordinate action of TGFβfamily members and SHH is required for the correct induction of anterior properties of ventral midline cells at prospective forebrain levels(Dale et al., 1997; Dale et al., 1999; Ericson et al., 1995). Thus,to ascertain whether the abnormal distribution of BMP genes accompanied altered Shh expression, we looked at Shh distribution at stage 6 of normal and VAD embryos (Fig. 3K-L′). Fig. 3L shows that, in VAD embryos, overall levels of Shhexpression are slightly decreased along the rostrocaudal axis; however, lower levels are more evident in anterior regions (n=4; inset in Fig. 3L).

In summary, our molecular analysis suggests an endogenous role for retinoids in the AP specification of axial mesoderm into prechordal mesoderm and anterior notochord.

In addition, abnormal BMP gene expression in the ventral neural tube and lower levels of Shh expression in prechordal mesoderm suggest an altered balance of signalling molecules involved in anterior ventral neural patterning.

Heterotopic grafting and tissue recombination experiments have demonstrated that prechordal mesendoderm is a potent anterior neural inducer (reviewed by Kiecker and Niehrs, 2001). Therefore, the posterior expansion of prechordal mesendoderm markers observed in VAD embryos could result in the induction of neural tissue with a more anterior character. To test this hypothesis, we looked at the expression of various anterior neural markers in stage 6-11 VAD embryos. The transcription factor Otx2 is a marker for presumptive fore- and midbrain regions in all vertebrate species examined, and its expression is reduced in presence of excess RA (Bally-Cuif et al.,1995; Simeone et al.,1995; Swindell et al.,1999). Concordantly, we observed an AP enlargement of the Otx2 expression domain in the absence of RA (n=3; Fig. 4B), which suggests an expansion of prosencephalic and mesencephalic tissue. To assess the degree of expansion at the level of the prosencephalon, we used Pax6 as a presumptive diencephalic marker (Bell et al., 2001). At stage 7+, Pax6 is expressed in the caudal half of the prosencephalon but is excluded from the most anterior region of the embryo (Fig. 4C). In VAD embryos the posterior Pax6-positive region is not affected, whereas the most anterior Pax6-negative region has an increased AP length(n=3; Fig. 4D). To verify that the affected region corresponds to the future telencephalon, we used Nkx2.1 as a ventral regional marker(Crossley et al., 2001) and Bf1 (qin) as a prospective ventral telencephalic marker(Bell et al., 2001). At stage 9, the Nkx2.1 expression domain encompasses mainly the hypothalamic primordium (Fig. 4E; inset). In VAD embryos the posterior expression boundary is clearly shifted caudally,whereas the anterior boundary has not changed (n=5; Fig. 4F; arrowhead in F′). Furthermore, the caudomedian expression boundary of Bf1(red line in Fig. 4G) has shifted towards the prospective diencephalon at stage 11+ (n=4; Fig. 4H; inset). Thus, our analysis with regional forebrain-specific markers shows an expansion of telencephalic tissue in the absence of retinoids, and suggests a role for RA in maintaining a proper boundary between the future telencephalon and diencephalon.

Next, we investigated whether the above-described molecular changes in presomite and early somite VAD embryos precede putative cellular and morphological abnormalities at later stages. We first assessed the presence of apoptotic cells using the TUNEL technique(Gavrieli et al., 1992). In anterior regions, stage 10 normal embryos show localized cell death at the anterior neural pore (Fig. 5A)(Maden et al., 1997) and a few apoptotic cells in ventral regions (Fig. 5A; inset). In VAD embryos of similar stages we noticed an increase in the rate of apoptotic cells in ventral neuroectoderm and in foregut endoderm (n=4; Fig. 5F; inset), whereas the extent of apoptotic cells at the anterior neural pore shows no obvious change. To ascertain whether the rate of dying cells was maintained at later stages, frontal and horizontal sections of stage 12-20 VAD quail embryos were examined for the presence of apoptotic cells(Fig. 5B,C,G,H; data not shown). At stage 13/14, a dramatic increase of dying cells is evident in the ventral neural tube and in head mesenchymal cells, which in many cases invade the forebrain vesicle (n=5; Fig. 5G). Furthermore, high levels of programmed cell death are retained in head mesenchyme and in ventral neuroectoderm until stage 19/20(n=2; Fig. 5H). To establish whether the dying mesenchymal cells are of neural crest origin, the HNK1 antibody (Luider et al.,1992) was used in adjacent para-sagittal sections(Fig. 5D,E,I,J). Whereas at stage 10, HNK1-positive crest cells are normally distributed in VAD embryos(n=3; Fig. 5I), their presence decreases dramatically at later stages, consistent with an increased incidence of apoptotic cells (n=6; Fig. 5E,J). In summary, our data indicate that the absence of retinoids induces a first wave of programmed cell death in the ventral neuroectoderm and foregut endoderm, whereas the migration of forebrain neural crest cells is not affected. However, at later stages a second and more dramatic wave of cell death occurs within the neural crest-derived head mesenchyme, whereas the high rate of apoptotic cells is maintained in ventral neuroectoderm.

It has been previously reported that VAD embryos die at ∼4 days of development because of cardiovascular defects(Twal et al., 1995; Chen et al., 1996; Maden et al., 1996; Gale et al., 1999). Up to stage 14, VAD and normal embryos are morphologically indistinguishable;however, after 72 hours of incubation (stage 19/20), VAD embryos are about three quarters the head-to-tail length of normal embryos, they develop no body torsion but at a morphological level are comparable in stages with normal embryos (Maden et al., 1996; Zile, 1998). At the level of the rostral brain, we found that 100% of stage 20 VAD embryos lack a cephalic flexure, and have a single and reduced telencephalic vesicle, which is variable in size and placed either centrally or slightly laterally(Table 1; Fig. 6D,E). However, all VAD embryos develop two eyes, although they can be different sizes (see Table 1). We classified VAD embryos as having a moderate (12%), severe (77%) and very severe (11%)phenotype according to the degree of telencephalic reduction(Table 1). Moreover, we found that 70% of VAD embryos have an enlarged diencephalon and/or develop ectopic transverse ridges within the diencephalon(Fig. 6B), whereas 100% of VAD embryos lack a morphological zona limitans intrathalamica (zli) as seen by histological and molecular analysis (A.H. and M.S., unpublished). Coronal sections at anterior levels confirm the presence of a smaller and single vesicle with no interhemispheric fissure and an abnormally thick neuroectoderm(Fig. 6G); at more posterior levels the neuroectoderm is dramatically reduced (see also Fig. 5J; data not shown). Thus,at the level of the forebrain, VAD embryos have a single telencephalic vesicle but retain two eye fields and have an abnormally patterned diencephalon.

It is well established that the primitive (or extra-embryonic) and definitive (or embryonic) endoderm are required for forebrain development;however, little is known about the molecular mechanisms and signalling pathways involved in this process.

In the mouse, the AVE is necessary for the development of anterior structures (Thomas and Beddington,1996); however, its transplantation does not generate an ectopic forebrain (Tam and Steiner,1999; Kimura et al.,2000). In the chick, the hypoblast can only transiently induce early anterior and neural markers(Knötgen et al., 1999; Foley et al., 2000). Therefore, the role of the AVE and the hypoblast remain unclear. One hypothesis is that the hypoblast might pre-pattern the naïve epiblast,before it becomes responsive to additional stabilizing signals secreted by the node. Alternatively, both the AVE and the hypoblast could be involved in regulating cell movements, as suggested by fate mapping in mouse and transplantation studies in chick (Foley et al., 2000; Kimura et al.,2000). Finally, removal of the hypoblast in chick causes the formation of ectopic primitive streaks, which suggests that the hypoblast could position the primitive streak by emitting an inhibitor of axis formation(Bertocchini and Stern, 2002). Interestingly, it has been shown that local application of RA can cause a pronounced curving of the primitive streak, away from the retinoid source(Knezevic and Mackem, 2001),in a manner reminiscent to that observed with hypoblast rotation(Foley et al., 2000); however,no presence of retinoids has been documented in pre-gastrula stage embryos so far. We found that Raldh2 is highly expressed in the expanding hypoblast and thus it is conceivable that retinoids participate in the early events in which the hypoblast is implicated. We propose that one of the roles of RA in the hypoblast might be to provide a proper developmental environment for migrating cells, a role that has already been suggested in branchial arch patterning (Wendling et al.,2000; Zile et al.,2000).

During early gastrulation the anteriormost endoderm, comprising the hypoblast and anterior definitive endoderm, is required for proper forebrain development. In particular, it has been shown that BMP2 and BMP7 can partially mimic the action of anterior endodermal cells in suppressing notochord and in inducing prospective prechordal mesoderm characteristics in the axial mesoderm(Vesque et al., 2000). Our study shows that during axial mesodermal specification, retinoid synthesis is present in anterior endoderm, as seen by the expression profile of Raldh2. Therefore, RA could freely diffuse from the endoderm into the overlying anterior neural plate and towards the emerging axial mesoderm. However, the presence of the RA-degrading enzyme CYP26 in anterior neuroectoderm from stage 3+/4– onwards (data not shown)(Swindell et al., 1999)excludes a vertical action of RA directly into the neural plate at this stage,and favours instead a direct or indirect role of RA in axial mesoderm specification (see model in Fig. 7). Accordingly, our functional data demonstrate that in the absence of retinoids, Bmp2 and Bmp7 expression is upregulated in the hypoblast, and axial mesodermal cells acquire altered prechordal mesoderm properties. Thus, we propose that retinoids cooperate in refining the extent of prechordal mesododerm characteristics by modulating adequate levels of BMP signalling in the hypoblast(Vesque et al., 2000). The need to maintain precise levels of both RA and BMP signalling has been already described in other systems (Schultheiss et al., 1997; Yatskievych et al.,1997; Andree et al.,1998; Ladd et al.,1998; Ghatpande et al.,2000; Schlange et al.,2000). In the head, low levels of BMP proteins and higher levels of RA are necessary to pattern the frontonasal mesenchyme in chick embryos(Lee et al., 2001). Our data validate that precise levels of RA and BMP proteins are also a requisite for early anterior patterning, and add further insight into the regulation of BMP proteins by retinoids.

Finally, it has been demonstrated that removal of anterior definitive endoderm during stages 4+ and 5 leads to dramatic forebrain reductions and concomitant gene expression alterations, such as abolishment of the homeobox gene Hex and the Wnt antagonist Crs(Withington et al., 2001). Furthermore, Hex deficiency(Martinez-Barbera et al.,2000) and overactivation of Wnt signalling produced by abolishing Wnt antagonism (reviewed in Yamaguchi,2001) will lead to different degrees of forebrain malformations,which suggests an essential role for these genes in anterior patterning. Accordingly, our data show that Hex and Crs anterior definitive endoderm expression is severely reduced in the absence of RA,whereas expression of Bmp2, and to a lesser extent Bmp7, is laterally expanded, which suggests abnormal characteristics of the anterior definitive endoderm cells in the absence of RA. Our data are in agreement with a recent study in zebrafish, which demonstrates that RA signalling plays an important early role in regionalizing the zebrafish endoderm(Stafford and Prince, 2002),and therefore we suggest that retinoids might act independently on anterior endodermal cells thereby modulating instructive signals necessary for proper head patterning.

In summary, we propose that among the various signalling molecules involved in endodermal and early anterior CNS patterning, RA is an essential component that might act at the level of primitive and definitive anterior endoderm thereby modulating levels of expression of various target genes. More studies need to be performed in order to characterize the precise role of retinoids in the genetic cascade of endodermal signalling.

Various reports have shown that prechordal mesoderm is a potent anterior inducer when transplanted ectopically, and is absolutely necessary for maintaining and/or reinforcing anterior neural plate specification and, thus,head structures (reviewed in Kiecker and Niehrs, 2001). Moreover, in chick, signals provided by axial mesendoderm refine the rostrocaudal character of the overlying neuroectoderm(Rowan et al., 1999). Therefore, accurate AP specification of prechordal mesoderm is a prerequisite for correct regionalization of the overlying anterior neuroectoderm. Our data are in accordance with these findings and show that an extension of prechordal mesendoderm caused by lack of retinoids induces a reciprocal expansion of the anterior neural plate. In Xenopus, injection of a dominant-negative retinoid acid receptor that can inhibit RA signalling leads to an expansion of anterior structures, as observed by the increased expression of the forebrain/midbrain marker Otx2(Blumberg et al., 1997). Conversely, in mouse and chick, RA exposure abolishes Otx2 expression in midline mesendoderm, and induces various degrees of forebrain malformations in mouse embryos (Bally-Cuif et al.,1995; Simeone et al.,1995). Our data not only confirm the endogenous requirement of retinoids in restricting Otx2 expression in avian embryos, but also identify the ventral telencephalon as the main target subregion within the developing prosencephalon that is affected by the absence of retinoids. Therefore, the early endogenous role of RA might be to restrict the AP extent of the ventral telencephalon. However, our data cannot distinguish between a posterior to anterior fate conversion or an expansion of the future telencephalon. The maintenance of a normal diencephalic Pax6 domain prompts us to opt for the latter possibility; however, studies involving tissue explants with the use of regionalized molecular markers in the presence or absence of RA would help solve this issue.

Finally, the expansion of the posterior expression boundary of Nkx2.1, which normally marks the border between prechordal and epichordal neural plate (Shimamura et al.,1995), could imply a specific role of retinoids in maintaining the correct position of morphological boundaries within the prosencephalon. Various observations are in agreement with this hypothesis. First, VAD embryos fail to develop a morphological zli boundary, as seen in histological preparations and in the absence of Shh expression in the zli (A.H. and M.S., unpublished). Second, VAD embryos present ectopic constrictions and/or ridges in the dorsal diencephalon(Fig. 6B). Finally, in the hindbrain, local application of RA causes the disappearance of posterior hindbrain boundaries (Nittenberg et al.,1997) and, conversely, treatment with a retinoid receptor antagonist fails to specify gene expression borders involved in boundary formation (Dupé et al., 2001). Thus, as in the hindbrain, the endogenous role of RA in anterior regions might be to restrict the expression domains of regulatory genes involved in regionalizing the developing prosencephalon during a crucial time-window.

Although VAD embryos show an expansion of anterior neural tissue at early stages, the telencephalic vesicles are not separated and are dramatically reduced at later stages. How are these events correlated? We propose various hypotheses to explain this discrepancy. First, we found a reduction of Shh expression in prechordal mesoderm before the appearance of an increased rate of cell death in ventral neuroectoderm and foregut endoderm. SHH has been shown to act as an anti-apoptotic factor and might therefore be responsible for the increased ventral cell death seen in VAD embryos (see also Ahlgren and Bronner-Fraser,1999; Britto et al.,2002; Charrier et al.,2001). Second, VAD embryos show abnormal expression of Bmp2 and Bmp7 in ventral neural midline, which might induce,together with reduced Shh expression, altered rostrocaudal properties in the overlying neuroectoderm. Therefore, mis-specification of axial identity in the mesenchyme and neuroepithelium could lead to programmed cell death, a phenomenon referred to as positional apoptosis and already described in the hindbrain (Maden et al.,1997). Third, decrease of retinoid signalling in the frontonasal mass, either through inactivation of RAR at the gene level or through RAR antagonists, generates various degrees of forebrain malformations and increased death of head mesenchymal cells(Lohnes et al., 1994; Schneider et al., 2001). The study by Schneider et al. (Schneider et al., 2001) has shown that, in stage 10 chick embryos, retinoids might be involved in maintaining a proper growth rate of head mesenchymal cells, and thus an adequate outgrowth of the frontonasal process and forebrain structures. Our study demonstrates that, at stage 10, VAD embryos have already increased cell death in ventral neuroectoderm and endoderm but migration of cranial neural crest cells is normal. Subsequently, from stage 12 onwards, VAD embryos show a dramatic increase in programmed cell death of mesenchymal and neuroectodermal cells. Thus, in contrast to the findings of Schneider and collaborators, we identified apoptotic cells in ventral neuroectoderm before the appearance of dying cells in neural-crest derived cells, which suggests that retinoids might have different roles at different stages in patterning the developing forebrain. We propose, therefore, that the overall increase of cell death observed in VAD embryos at stage 20, is derived from an additive effect between an early event restricted mainly to the ventral neuroectoderm and foregut endoderm, and a later event confined to the head mesenchyme.

In summary, this study has unravelled a novel role for RA signalling during gastrulation in anterior patterning that has not been reported previously (see model in Fig. 7). We speculate that RA synthesized in hypoblast cells might be involved in providing a proper developmental environment for cell motility and/or regulating primitive streak formation. Subsequently, during the process of axial mesoderm specification into prechordal mesoderm and notochord, RA is synthesized from two sources,one located in the node and another in anterior endoderm, which is composed of remnants of hypoblast and anterior definitive endoderm. Whereas the node source might be involved in the differentiation of axial mesodermal precursors, the anteriormost endodermal source might be required in modulating regulatory genes and signalling molecules during the specification of axial mesodermal cells (Fig. 7A). In particular, RA would restrict the AP extent of prechordal mesoderm, thereby maintaining and stabilizing anterior neural plate characteristics in conjunction with other signalling pathways, in accordance with the`stabilization step' proposed by Stern and collaborators(Stern, 2001)(Fig. 7B). Therefore, altered vertical signals that are due to the absence of retinoids and to abnormal AP specification of prechordal mesoderm would fail to maintain a proper forebrain anlage (Fig. 7C). Previous data obtained from Xenopus and mouse embryos with exogenous doses of RA had suggested a direct action of RA on axial mesodermal cells(Ruiz i Altaba and Jessell,1991; Bally-Cuif et al.,1995; Simeone et al.,1995); however, no endogenous sources of RA synthesis had been described at that time to justify such a hypothesis.

Finally, a late source of retinoids localized in the head mesenchyme and involved in sustaining proper forebrain outgrowth would be responsible for the late phase of cell death observed in VAD and antagonist-treated embryos (this study) (Schneider et al., 2000). Thus, although presumptive telencephalic structures are more widely induced in early VAD embryos, these cannot be maintained at later stages.

We thank V. Heller, M. Maden and E. Gale for providing us with normal and VAD quail embryos; M. Maden for Raldh2; E. Swindell for CYP26; R. Godbout for Raldh1; F. Grun for Raldh3;J. J. Michaille for RARα2-1; P. Brickell for RARβ, RXRα, RXRγ, Bmp2and Bmp4; A. Graham for Bmp7; S. Chapman for Nkx2.1; C. Tabin for Shh and L-fng; M. Gulisano for Emx1 and Otx2; M. Goulding for Pax6; A. McMahon for Wnt3A; I. Mason for Fgf8; P. Vogt and J. Li for Bf1; J. C. Izpisua-Belmonte for Gsc; P. Pfeffer for Crs; and G. Goodwin for Hex. We also thank D. Echevarria, A. Lumsden, J. P. Martinez-Barbera, C. Stern, C. Vesque and L. Zeltser for helpful comments on the manuscript. A. H. was an MRC student fellow, M.H.Z. is supported by an NIH grant (RO1 HL61982) and by the Michigan Agricultural Experiment Station, and M.S. was an MRC Career Development Fellow.