Retinoic acid-binding protein, rhombomeres and the neural crest

The hypothesis that retinoic acid (RA) is a morphogen in vertebrate development is an amalgamation of two lines of evidence: the effects of administration of excess RA to the developing embryo and the identification of RA as an endogenous component of the embryo. The two developing systems that have so far been investigated in this regard are the limb and the nervous system.

In the developing chick limb bud, when RA is applied to the anterior margin, a 6-digit, double posterior limb develops instead of the normal 3-digit one (Tickle et al. 1982; Summerbell, 1983). RA is present endogenously in the normal chick limb bud and is concentrated 2.5 – fold on the posterior side compared to the anterior side (Thaller and Eichele, 1987). This is precisely the behaviour we would expect if RA was the morphogen released by the zone of polarizing activity, the region on the posterior margin of the limb bud that controls the specification of pattern across the anteroposterior axis of the limb bud (Tickle et al. 1975). In the regenerating amphibian limb, RA causes respecification not only in the anteroposterior axis (Kim and Stocum, 1986; Maden, 1983), but also in the proximodistal axis (Maden, 1982) and the dorsoventral axis (Ludolph et al. 1990). RA is also present endogenously in the blastema of the regenerating limb (M.Maden; N.Waterson and D.Summerbell, unpublished data). Thus RA may be the morphogen responsible for specifying axial information in the developing and regenerating limb.

In the early nervous system of the Xenopus embryo, RA administration causes an anteroposterior transformation whereby the forebrain and midbrain are reduced or absent and the hindbrain and spinal cord are correspondingly exaggerated (Durston et al. 1989). RA is also present endogenously in the whole Xenopus embryo at these stages thus fulfilling the two criteria listed in the first paragraph that suggest a role for RA in the specification of pattern in the early nervous system.

Inside the cell, RA interacts with two classes of protein: (1) cellular retinoic acid-binding protein (CRABP) which is located in the cytoplasm, (2) two famihes of retinoic acid receptors which are located in the nucleus. CRABP is a 15.6 ×10³M_r protein present in a variety of RA-responsive tissues in the adult mammal (Ong et al. 1982). Of the naturally occurring retinoids this protein specifically binds RA: there are other proteins that specifically bind other retinoids such as retinol. The role that CRABP plays in the mechanism of action of RA is still unknown as is its relationship to the retinoic acid receptors (RARs), RAR α, RAR β, RARγ (Zelent et al. 1989) and RAR<5 (Ragsdale et al. 1989) and the new family of receptors termed RXR (Mangelsdorf et al. 1990). Several suggestions have been made as to the function of CRABP. It may be involved in the metabolism of RA, it may transport RA from its point of entry into the cell to the nucleus where the RARs are located thereby defining populations of cells that are RA-responsive (Maden et al. 1989a) or it may act as a buffer regulating the amount of RA available to the RARs and generating concentration gradients of RA (Maden et al. 1988; Smith et al. 1989).

One obvious way to begin investigations into the role of CRABP is to examine its distribution during development. This may also provide us with a more precise indication than that available by HPLC as to which tissues in the embryo utilise RA. Such an analysis with the RARymRNA, for example, has shown that it is present in the branchial arches, limb buds, differentiating cartilages and keratinising epithelia of mouse embryos, suggesting a role for this receptor in cartilage and epithelial differentiation (Ruberte et al. 1990). Our previous experiments on the distribution of CRABP by immunocytochemistry in the chick embryo have identified two systems with interesting patterns of immunoreactivity, the limb bud and the nervous system i.e. the very systems that are RA responsive.

In the chick limb bud, CRABP is concentrated in the progress zone at the distal tip of the bud. Across the anteroposterior axis of the progress zone, CRABP is concentrated at the anterior margin, producing a gradient of opposite polarity to its ligand (Maden et al. 1988). In the mouse limb bud, an analysis of CRABP mRNA distribution by in situ hybridisation produced a similar result (Dollé et al. 1989). Dollé et al. also examined the distribution of RAR transcripts and found that RAR α and RAR γ transcripts were uniform and there were no RARβ transcripts in the early limb bud thereby emphasising the importance of CRABP in the morphogenesis of pattern in the limb.

In the nervous system of the chick embryo, we have previously shown that CRABP immunoreactivity appears in the neuroblasts of the mantle layer, which subsequently become the neurons of the commissural system (Maden et al. 1989b). Outside the neural tube, individual CRABP-immunoreactive cells were seen spreading ventrally through the sclerotome to accumulate beside the dorsal aorta. They also appeared in the walls of the developing gut and as a population of dorsal root ganglion cells. We suggested that these were neural crest cells. A similar spectrum of CRABP-containing cells have been observed in the mouse embryonic nervous system both by in situ hybridisation (Perez-Castro et al. 1989) and immunocytochemistry (Maden et al. 1990).

In the work reported here, we examine in detail the contention that neural crest cells contain high levels of CRABP by comparing CRABP and HNK-1 patterns of immunoreactivity in the early chick embryo. HNK-1 is a monoclonal antibody that reacts with migrating neural crest cells (Vincent et al. 1983; Vincent and Thiery, 1984) and its distribution in the trunk of the chick embryo has been well characterised (Rickmann et al. 1985; Bronner-Fraser, 1986). We show that CRABP is indeed present in neural crest cells and their derivatives such as the frontonasal mass, branchial arches and cephalic ganglia. We have also investigated the time of first appearance of CRABP immunoreactivity in the neural tube because of the effects that RA has on the early nervous system (see above). We find a fascinating pattern of labelling, which begins in the hindbrain just after neural tube closure and produces a distinct striping array in the rhombomeres. The rhombomeres are currently arousing great interest as the sites of discrete homeobox gene localisations (Wilkinson et al. 1989) and homeobox gene expression, at least in cultured cells, is responsive to RA. Therefore we have also compared this pattern of CRABP immunoreactivity with that of the chicken Hox-2.9 gene using in situ hybridisation. We report that Hox-2.9 has a similar pattern of labelling in the rhombomeres. These results suggest that there may be a relationship between RA, CRABP and homeobox genes in the establishment of pattern in the nervous system, notably the hindbrain.

Fertile chicken eggs were obtained from a mixed local flock (Needle Farm), incubated, and staged according to Lillie (1952).

Embryos at stages 4 – 23 were fixed for 3 h in Perfix (Fisher Scientific, New Jersey) for immunocytochemistry, dehydrated, wax embedded and sectioned at 7μm.

CRABP was immunolocalised with an affinity-purified rabbit polyclonal antibody raised against a synthetic peptide corresponding to a sequence from bovine CRABP (Eriksson et al. 1987). The characteristics and specificity of this antipeptide antibody has been established and it has been used as an immunocytochemical reagent to localise CRABP in rat testis (Eriksson et al. 1987). We first confirmed that the antibody was appropriate for use with chicks by performing a Western blot. Whole chick embryos were homogenised in solubilisation buffer for SDS-polyacrylamide gel electrophoresis according to Laemmli (1970) and centrifuged to remove debris. Aliquots of the extract were run on a 12.5% SDS-polyacrylamide gel and the proteins transferred to nitrocellulose paper. The immunoreactivity was visualised using the avidin-biotinylated peroxidase technique with a kit from Vector Laboratories (Peterborough, UK). The Western blot revealed that the antibody reacted with a single band of approximately 16 × 10³ (Fig. 1), the expected molecular weight of CRABP.

Both HNK-1 (a gift of Dr Claudio Stem, University of Oxford) and CRABP antibodies were used on sections at a dilution of 5 μg ml⁻¹ and immunoreactivity detected with the Vector Laboratories ABC kit as for the Western blot.

For in situ hybridisation a 1 kb Pstl-Xhol fragment containing part of the chicken Hox-2.9 gene cloned into the pSP65 vector (Promega) was used to generate an anti-sense S³⁵-RNA-labelled probe by transcription with SP6 polymerase. The hybridisation protocol used was essentially that of Wilkinson et al. (1987) with some modifications. Briefly, chick embryos were fixed overnight in 4% paraformaldehyde in PBS at 4°C and then embedded in paraffin wax. Sections (6 μm) were dried onto TESPA-coated slides, dewaxed in xylene, subjected to proteinase K treatment and further treated with acetic anhydride. After dehydration, RNA probe, reduced in size by controlled cleavage at alkaline pH, was added at a final activity of l–3 × 10⁻¹ disints min⁻¹kb⁻¹ μl⁻¹ in hybridisation buffer (50% formamide, 0.3M NaCl, 20mM Tris, 5ITIM EDTA, 10mM sodium phosphate, 10% dextran, 1 × Denhardt’s solution, 0.5μg μl⁻¹ tRNA). Hybridisation was overnight at 55°C. Sections were washed at 65°C in 50% formamide, 2xSSC, 100mM DTT for 40min followed by incubation with RNAse A at 20 μ gml⁻¹ in 0.5 M NaCl, 10mM Tris, 5mM EDTA for 30min followed by 2 × l5min washes in 2 × SSC and 01 × SSC. Sections were dehydrated through alcohol solutions containing 0.3 M ammonium acetate. Slides were dipped in a mixture of Ilford K5 nuclear emulsion and glycerol water (6 ml emulsion in 8.82 ml water and 0.18 ml glycerol) and stored at 4 °C until developed. Sections were stained in 0.02% toluidine blue for 1 min and mounted in permount.

This is the earliest stage at which CRABP was detected. Serial sections of several embryos revealed that fewer than 10 cells per embryo were immunoreactive and these were all located in the neural tube at the level of somites 1-3 (Figs2A, 3). Since the neural folds had only just fused at the level of the mesencephalon/ anterior rhombencephalon at stage 8, about 3h earlier (Lillie, 1952), the appearance of CRABP-positive neuroblasts is a very early event in the development of the neural tube. The first 5 somites are located beside the posterior half of the rhombencephalon, thus conveniently marking the location of these early staining cells.

No other CRABP-positive cells were detected in the embryo either in the primitive streak, epiblast or mesoderm.

HNK-l-positive cells in the embryo were detected only in a small patch dorsal to the mesencephalon. This is precisely the stage and location (junction of prosencephalon and mesencephalon) of the first migrating neural crest cells (Le Douarin, 1982).

Within the space of 3 – 4 h, the numbers of cells in the neural tube that had CRABP immunoreactivity increased markedly. The area over which they were localised had now spread caudally to include the neural tube adjacent to the first 5 somites (Fig. 2B) thus confirming that the first neuroblasts to show this phenotype were in the future rhombencephalon. Outside the neural tube, there were no immunoreactive cells either in the somites, primitive streak, notochord, invaginating mesoderm etc.

The location of these CRABP neuroblasts within the neural tube is of interest. In cross section, they were excluded from the extreme dorsal or ventral poles of the neural tube and were always located on the pial surface rather than the ventricular surface (can be seen in Fig. 3), although occasionally immunoreactivity was present in a cell that extended across the whole width of the neural tube. In the anteroposterior dimension, the first cells to become labelled were located adjacent to somites 6–8 behind the last-formed somite. These principles were apparent throughout this study as the embryos elongated and added somites at their posterior ends.

HNK-1-positive cells could be seen in the cephalic mesenchyme adjacent to the developing mesencephalon migrating out from the neural tube in two distinct streams of cells.

This was the first stage at which recognisable rhombomeres, the series of swellings in the rhombencephalon were seen. There are 8 rhombomeres in all, but at this early stage only 5 (numbers 2–6) could be delineated as they develop over a period of time from stages 9–12 (Vaage, 1969; Fraser et al. 1990; Lumsden, 1990). The numbering of individual rhombomeres can be assigned by reference to the otic vesicle. What was immediately obvious was that when sections were scored for the presence of CRABP immunoreactivity only certain rhombomeres contained positive neuroblasts (Figs 2C, 4). Thus rhombomeres 2, 3 and 5 were CRABP negative and rhombomeres 4, 6 and continuing into the anterior spinal cord were CRABP positive (Fig. 4). As can be seen in Fig. 4, not every cell in the rhombomeres or neural tube was immunoreactive, only those towards the pial surface and this peripheral location remained as differentiation proceeded (see Figs 8, 10, 12). In Fig. 4 only 7 cells are labelled in rhombomere 4, but when all serial sections are added together and plotted on a representative section, as in Fig. 2, a large number of cells are immunoreactive. An additional area of immunoreactivity in the neural tube appeared for the first time in these embryos, in the anterior mesencephalon (Fig. 2C).

HNK-l-positive cells continued to invade the cephalic mesenchyme by migrating in streams and became packed in the space between the developing optic lobes and the ectoderm. Further posterior, neural crest cells could now be seen leaving the spinal cord by the ventral route, between the neural tube and the somite as has been described before for trunk neural crest cells (Rickmann et al. 1985; Bronner-Fraser, 1986).

By now the pattern of CRABP immunoreactivity in the neural tube had become well-established in three locations (Fig. 2D): (1) in the developing neural tube from the level of 6 – 8 somites behind the last-formed somite through the posterior rhombencephalon up to rhombomere 6; (2) in a single stripe in rhombomere 4 and (3) in the anterior mesencephalon both dorsally and ventrally (Fig. 2D).

For the first time, CRABP-positive cells in the mesenchyme were observed at this stage. These were located in the cephalic mesenchyme at the anterior end of the embryo in between the optic vesicle and the ectoderm and dorsally above the prosencephalon (Fig. 2D). What was most noticeable about these cells is that they were indistinguishable in terms of location from the early neural crest cells that populated the same area around the developing forebrain at stage 12. It is likely therefore that these CRABP-positive mesenchymal cells are a subpopulation of neural crest cells that have begun the synthesis of the retinoic acid-binding protein.

HNK-l-positive cells were by this stage widely distributed throughout the embryo in well-characterised locations, namely: mesenchyme surrounding the fore- and midbrain; the region of the future facial primordia and branchial arches, although discrete arches were not yet evident; the facial and glossopharyngeal ganglia extending ventrally from the rhombencephalon; the developing heart and gut; the posterior half of the otic vesicle; and at the level of the somites, in the intersomitic clefts and in the anterior half of each somite.

During these stages there were several changes in CRABP immunoreactivity both in the neural tube and the mesenchyme. In the anterior neural tube, the area of labelled cells expanded posteriorly from the mesencephalon to encompass rhombomeres 1 and 2, thus changing the pattern from one labelled stripe (Fig. 2C, D) to two unlabelled stripes in rhombomeres 3 and 5 (Fig. 2E). The prosencephalon remained unlabelled.

In the mesenchyme, labelling of the most anterior cells surrounding the developing eye and telencephalon intensified. These were in the same location as the first HNK-1 cells to be labelled and, since the HNK-1 immunoreactivity in this area had now ceased, it seemed that CRABP was switched on in these cells when HNK-1 was switched off. Immunoreactivity also appeared for the first time in the branchial arches (Figs 2E, 5), again in cells that had switched off HNK-1. This branchial arch mesenchyme was clearly located peripherally (Figs 6, 8) being displaced by the cranial nerves and aortic arches. CRABP-positive staining also appeared in the otic vesicle (Fig. 5), as HNK-1 reactivity had done one stage earlier. In the region of the somites at stage 14, CRABP-positive cells could be seen in the anterior half of each somite and the intensity of this labelling increased over subsequent stages (see Fig. 7). These cells were thus likely to be neural crest migrating through the somites exactly as had been seen at earlier stages for HNK-l-positive cells. Lastly, occasional labelled cells began to appear in the cephalic ganglia (see Fig. 9).

The patterns of HNK-1 staining in cephalic regions over these stages showed an interesting complementarity to the CRABP staining. For example, in the branchial arch mesenchyme, the initial HNK-1 immunoreactivity that was noted at stage 13 had by now disappeared to be replaced by CRABP labelling. Thus HNK-1 cells could be seen approaching the arches, but within the arches themselves there was no labelling whereas those same arches labelled intensely with CRABP (Figs 6, 8). Also, the cells surrounding the optic vesicle and telencephalon had now ceased to be HNK-l-positive and had become CRABP-positive instead.

During the final period of development that we studied, the patterns of immunoreactivity described above intensified rather than changed dramatically. Fig. 12 shows an overall picture of CRABP immunoreactivity in the cephalic region and Fig. 2F a summary diagram.

The two gaps in CRABP immunoreactivity in the rhombencephalon gradually filled in so that labelling of these neuroblasts was continuous from the mesencephalon, through the rhombomencephalon all the way to the end of the neural tube (Figs 8, 12). The prosencephalon remained unlabelled as it had done throughout development (see Fig. 12). In a cross section of the rhombencephalon, the lateral and ventral neurons of the mantle layer were the immunoreactive ones (Fig. 10). In the spinal cord, these neurons became those of the commissural system as we have previously described (Maden et al. 1989b).

The mesenchyme of the frontonasal mass, maxillary prominence and branchial arches remained intensely CRABP positive. Fig. 11 shows the immunoreactive frontonasal mass mesenchyme contrasting with the unstained diencephalon and the epidermis. Two other points are worth emphasising about the staining patterns of the arches. First it was specific to arch and facial mesenchyme, that is, the mesenchyme outside the developing arches or the mesenchyme not immediately underneath the facial epithelium was unstained. Second there was a considerable difference in the intensity of staining between different arches. The maxillary prominence and arch 2 were the faintest in CRABP-staining intensity, the frontonasal mesenchyme and arches 3 and 4 had intermediate intensity and arch 1 the strongest (Fig.9).

The three other areas of the embryo that first showed CRABP immunoreactivity at earlier stages changed little during this period. The otic vesicle remained labelled, occasional labelled cells were seen in the cranial nerves (Fig. 9) and the immunoreactive cells migrating through the anterior halves of the somites became clearer (Fig. 7).

The distribution of CRABP immunoreactivity in the rhombomeres immediately brings to mind the expression patterns of Hox-2 genes in the same structures in the mouse (Wilkinson et al. 1989). To examine similarities in the chick we have, as an initial experiment, looked at the distribution of the chick Hox-2.9 mRNA by in situ hybridisation, in the absence of an antibody to the Hox-2.9 protein.

In stage 9 – 10 embryos, the Hox-2.9 gene is expressed in a continuous domain along the neural tube extending into the hindbrain in a region coincident with that of CRABP. The anterior boundary of this domain is rhombomere 4. At stage 14 – 15 expression of Hox-2.9 is clearly seen in rhombomere 4, but there is now a gap in the domain of expression posterior to rhombomere 4 as shown in Fig. 13. Expression in this posterior domain extends through the anterior neural tube up to the region between rhombomeres 6 and 7 and appears to correspond to the boundary between these rhombomeres. However, as seen in Fig. 13 B – C on the right side of the neural tube some expression does appear to extend into rhombomere 6. The relevance of the association between CRABP and Hox genes is considered further below.

We have described here the distribution of cellular retinoic acid-binding protein (CRABP), the protein to which retinoic acid (RA) specifically binds in the cytoplasm of cells, and compared it with the distribution of HNK-1, a monoclonal antibody that recognises migrating neural crest cells, at a full range of stages of early chick development. We also report the expression pattern of the chicken Hox-2.9 gene.

We are interested in CRABP because it may indicate those systems in development that use RA as a morphogen. The developing chick limb bud is one such case (Thaller and Eichele, 1987) and on the basis of the results reported here we propose that the generation of pattern in the early nervous system is another case. We used HNK-1 staining because our previous studies had suggested that CRABP was present in neural crest cells (Maden et al. 1989b) and we wished to see if CRABP and HNK-1 stained the same populations of cells.

After examining serial sections of embryos up to stage 23, it was apparent that CRABP was present in only two areas of the embryo, the neural tube and a subpopulation of neural crest and its derivatives (the additional area is the limb bud – see Maden et al. 1988).

From the comparative patterns of immunoreactivity of CRABP and HNK-1, it was clear that the neural crest does indeed contain high levels of CRABP as we had previously suggested (Maden et al. 1989). However, CRABP immunoreactivity appeared in the crest some time after HNK-1 immunoreactivity. In the trunk, for example, HNK-l-positive cells were present in the anterior halves of the somites at stage 13, but CRABP-positive cells in the same location were not seen until stages 14 – 16. This unique behaviour of trunk neural crest cells in migrating through the anterior half of each somite and the same behaviour being elicited by CRABP-positive cells confirms that migrating neural crest does contain CRABP.

In cephalic regions, however, the behaviour of these two antibodies was slightly different. CRABP did not appear until after HNK-1 staining had disappeared, the two antibodies thus showing a complementary pattern. This was particularly apparent in the branchial arch mesenchyme in which HNK-1 staining disappeared as the neural crest entered the arches to be replaced by intense CRABP staining. Thus rather than playing a role in migration, RA is more likely to be involved in the differentiation of the neural crest upon arrival at its destination. Our previous observations in the trunk of the chick embryo (Maden et al. 1989) showing CRABP immunoreactivity in a population of dorsal root ganglion cells, in the sympathetic ganglia and in the walls of the developing gut, support this contention.

The first neuroblasts to become CRABP positive appeared at stage 9 in the rhombencephalon, very soon after closure of the neural folds (at stage 8). These neuroblasts were situated within the mantle layer of the neural tube although an occasional positive cell could be seen stretching from the pial to the ventricular surface. This location at the pial surface is the same as that seen with N-CAM and the 68 × 10³M_r neurofilament protein labelling at later stages in the rhombencephalon (Lumsden and Keynes, 1989).

As the neural tube continued to develop and somites were added in a posterior direction, CRABP immunoreactivity also spread posteriorly within the neural tube to remain about 6 – 8 somites behind the last formed somite. By stage 11, a remarkable pattern of labelling had appeared in the newly formed rhombomeres such that a single stripe was created. Immunoreactivity was present up to and including rhombomere 6, it was absent in rhombomere 5, present in rhombomere 4 and absent again in the rest of the hindbrain (Fig. 2C, D). This pattern remained until stages 14 – 16 by which time rhombomeres 1 and 2 had become labelled thus creating two unlabelled stripes (Fig. 2E). Following this the entire hindbrain became labelled and the striping pattern was lost (Fig. 2F). Over this period, the mesencephalon was also CRABP positive, especially the roof, whereas the prosencephalon remained unlabelled.

CRABP immunoreactivity is thus one of the earliest markers of differentiation yet described within the rhombomeres, the homeobox genes being the other (see below). It preceeds the appearance of neurofilament immunoreactivity of reticular neurons which appears in the neurones of even-numbered rhombomeres at stage 11/12 (Lumsden, 1990). CRABP appears during the period that the rhombomeric ‘compartments’ are established (Fraser et al. 1990) and its sequence of appearance seems to be the same as the sequence of differentiation, that is, even numbered rhombomeres before odd numbered ones (Lumsden and Keynes, 1989).

As described in the Introduction, one reason for examining the distribution of CRABP in the embryo is to provide an indication of which structures use RA in their development. Since CRABP specifically binds RA, it may be valid to assume that cells containing higher levels of CRABP contain higher levels of RA than cells that do not contain CRABP. Because of the impossibility of collecting large amounts of embryonic material, this information would not be available by current HPLC techniques. But is this assumption valid?

Evidence that this is valid has recently been obtained by injection of pregnant mice with ³H-RA (Dencker et al. 1987) or ¹⁴C-Ro 13 –6298, a RA analogue (Dencker et al. 1990). Radioactivity localises to specific regions of the mouse embryo. These regions are: early neural crest migrating from the neural folds, the midbrain roof, the hindbrain, the frontonasal mass, the branchial arches (particularly the periphery rather than the core region), dorsal root ganglia and the mantle layer of the neural tube. These are exactly the regions that contain high levels of CRABP both in the chick (Maden et al. 19896; these results) and the mouse (Perez-Castro et al. 1989; Vaessen et al. 1989; Dencker et al. 1990; Maden et al. 1990) embryo. Thus the coincidence between high RA and high CRABP in discrete regions of the embryo is experimentally verifiable.

It would be a valuable exercise to compare the pattern of CRABP expression with the effects of teratogenic doses of RA on the embryo to see if the two are related. Compared to the mammalian embryo, however, few teratogenicity studies have been conducted on the chick embryo. Nevertheless it seems clear that neural tube development is arrested and cardiovascular, limb and beak defects appear (Jelinek and Kistler, 1981). It may be no coincidence that the cells of each of these systems contain high levels of CRABP. Interestingly, during the development of the face only the upper beak is affected by RA, the lower beak being immune from its teratogenic action (Tamarin et al. 1984) and this effect is mediated via the mesenchyme (Wedden, 1987). As we have shown here, the components that go to make up the face and beak show differing intensities of CRABP staining, which may be related to their differing susceptibility to RA. For example, only the mesenchyme of the frontonasal mass and branchial arches is CRABP-positive, the epidermis is negative and it is only the mesenchyme that is affected by RA. However, the upper beak/lower beak discrepancy is more difficult to explain in these terms. The upper beak develops from the frontonasal mass and maxillary process whereas the lower beak develops from the first arch. But the first arch shows the most intense CRABP staining which, according to the above scheme, should make it more susceptible to teratogenic doses of RA. So there must be other components such as the RARs and RXRs involved in the teratogenicity of RA.

The dynamic spatial and temporal patterning of CRABP that we have described in the rhombomeres is reminiscent of the patterns of activity of the Hox-2 genes, which show segment-restricted expression in the hindbrain (Wilkinson et al. 1989). Recently we have found that the Hox-2 genes are also expressed in cranial neural crest cells, displaying sharp boundaries of expression in crest from different arches (Hunt, Wilkinson and Krumlauf, in preparation).

One obvious question to ask therefore is whether any of these Hox and CRABP patterns are coincident. Previous in situ hybridisations in the hindbrain with homeobox probes have been performed on the mouse and as part of a similar analysis in the chick with the homologous chick probes we examined here the distribution of Hox-2.9. It is important to note that, in the absence of a Hox-2.9 antibody, we were comparing mRNA accumulation with protein accumulation, the two not necessarily being comparable. Nevertheless, the patterns of Hox-2.9 and CRABP were very similar between stages 9 and 15. Initially, a uniform domain of expression extending into the hindbrain up to the anterior boundary of rhombomere 4 was observed for both genes. This was followed by a regression or drop in expression in rhombomere 5. These temporal changes in the establishment of segment-restricted patterns are of particuar interest in view of the fact that they are displayed by two different kinds of genes, and may reflect events in the ordering and specification of different rhombomeres. The only difference between CRABP and Hox-2.9 in the neural tube at stage 14-15 occurred in rhombomere 6 where CRABP was expressed but Hox-2.9 was only faintly detected in part of rhombomere 6 on one side of the neural tube (Fig. 13 B, C). At later stages the pattern of CRABP continued to change while that of Hox-2.9 remained the same at least until stage 19.

CRABP and Hox-2 mRNA also co-localise in the mantle layer of the mouse spinal cord in the region forming the commissural and association neurons (Graham et al. 1990; Maden et al. 1990). The only other analysis of the codistribution of Hox genes and retinoidbinding proteins concerns Hox-2.5 and cellular retinolbinding protein (CRBP) in the transverse axis of the mouse spinal cord (Graham et al. 1990). In this case we found that these two occupied reciprocal and adjacent domains with Hox-2.5 being dorsal and CRBP ventral. Understanding the significance of these observations must await a more complete analysis of the role of the retinoid binding proteins and the role of the retinoic acid receptors (see Introduction) in relation to Hox gene activity. Nevertheless they suggest that other such coincidence should be sought, for example, between CRABP, RAR, RXR and Hox genes.

When several pieces of evidence are brought together it appears likely that RA may play a role in the development of pattern in the hindbrain. First, RA causes anteroposterior transformations in the Xenopus neural tube whereby the forebrain and midbrain are reduced and the hindbrain and spinal cord are correspondingly exaggerated (Durston et al. 1989). In fact, alterations in specific cranial ganglia associated with the rhombomeres can be detected in Xenopus embryos at certain concentrations of exogenously administered RA (N.Papalopoulou and R.Krumlauf, unpublished results). Second, in mouse F9 teratocarcinoma cells all of the genes in the Hox-2 complex are rapidly inducible by RA. The relative responsiveness of any particular gene varies with RA concentration and with time and there is a linear correlation between their ability to respond to RA and their organisation in the Hox-2 cluster (Simeone et al. 1990; Papalopoulou et al. 1990). This correlation between homeobox genes and RA responsiveness was also observed for several Xenopus Hox genes when whole embryos were exposed to brief treatment with RA (Papalopoulou et al. 1990, in preparation). Finally from the results of this study the distribution of CRABP and Hox-2.9 are coincident in the hindbrain. Thus it is possible that RA via CRABP and the RARs and RXRs could also be used to establish gradients of homeobox gene expression, thereby imparting identity upon previously uniform segments of the developing central and peripheral nervous system.

M.M. gratefully acknowledges support from Hoffmann-La Roche & Co., Basel.

We thank Ian Muchamore for producing Fig. 13.