Retinoic acid receptors and cellular retinoid binding proteins : II. Their differential pattern of transcription during early morphogenesis in mouse embryos

Retinol (vitamin A) and its active derivative retinoic acid (RA) play important roles in vertebrate morpho-genesis. In mammals retinoids are well known to be potent teratogenic agents, in both deficiency and excess (Wilson et al. 1953; Shenefelt, 1972; Morriss, 1972; Rosa, 1983; Lammer et al. 1985). Exposure to high levels of retinoids during early stages of development interferes with the normal proliferation and differen-tiation of the neural epithelium leading to brain and spinal cord deformities (Langman and Welch, 1967; Lammer et al. 1985). Retinoid excess has also been reported to alter the normal migration of neural crest cells into the pharyngeal arches resulting in the abnormal development of the craniofacial skeletal elements (Morriss and Thorogood, 1978; Webster et al. 1986). The ability of RA to induce pattern duplications in the developing chick limb and regenerating amphib-ian limb has led to the idea that it could be a natural morphogen conferring positional values on the limb bud mesenchymal cells (reviewed by Brockes, 1989; for additional refs, see Dollé et al. 1990).

The mechanism by which retinoids exert their physiological role is not well understood. RA-induced differentiation of F9 embryonal carcinoma cells is associated with the expression of specific sets of genes (Silver et al. 1983; La Rosa and Gudas, 1988; and refs therein), suggesting that retinoic acid mediates its effect through the nucleus. The cloning of three receptors that specifically bind retinoic acid with high affinity, RAR-α (Petkovich et al. 1987; Giguère et al. 1987), RAR-β (Brand et al. 1988; Benbrook et al. 1988) and RAR-γ (Krust et al. 1989; Zelent et al. 1989), and that belong to the family of the steroid/thyroid hormone nuclear receptors, strongly supports the idea that RA exerts its effects by modifying the transcriptional activity of specific target genes. Previous studies on the distri-bution of RARs in limb development, and of RAR-γ in whole embryos, indicate that RARs may mediate region-specific and tissue-specific effects of RA during development (Dollé et al. 1989; Ruberte et al. 1990). Possible target genes include homeogenes and genes encoding proteins and proteases of the extracellular matrix (Colberg-Poley et al. 1985a,b;Deschamps et al. 1987; Kessel et al. 1987; Murphy et al. 1988; Mavilio et al. 1988; La Rosa and Gudas, 1988; Vasios et al. 1989; Simeone et al. 1990, and refs therein); homeobox gene expression and extracellular matrix molecules play important developmental roles during early morpho-genesis stages of mammalian development (see Dis-cussion). .

In addition, there exists a family of proteins in the cytoplasm, the cellular retinoic acid and retinol binding proteins (CRABP I and II, and CRBP I and II, called hereafter CRABP and CRBP) (Ong and Chytil, 1978a,b; for additional refs see Dollé et al. 1990). Their role is unknown, but it has been suggested that CRABP could act as a shuttle between the cytoplasm and the nucleus (Takase et al. 1979 and, 1986). However, the relatively high RA content in the chick limb bud, and the fact that some cells lacking CRABP are responsive to RA, has led to the idea that CRABP could control the concentration of RA, reducing the amount of RA available for binding to the nuclear receptors (Robert-son, 1987; Maden et al. 1988; Smith et al. 1989). In a related study on the distribution of RAR and binding protein mRNA transcripts during organogenesis, we made some correlations between CRBP distribution and the vulnerability of specific tissues to malformation in conditions of vitamin A deficiency (Dollé et al. 1990). We suggested that CRBP I plays an essential role in generating RA for cells whose requirement for RA is high. Conversely, CRABP expression may be corre-lated with low RA requirement, so that cells expressing CRABP would be those that are most vulnerable to RA excess. This second hypothesis would be most appropri-ately tested by looking for correlations between CRABP expression and retinoid-induced teratogenesis during preorganogenesis stages of development, since detailed morphogenetic studies on the cell and tissue effects of retinoid excess have been carried out in embryos exposed during the period of gastrulation, neurulation and neural crest cell migration (Morriss, 1973; Morriss and Thorogood, 1978; Webster et al. 1986).

In this study, we describe the distribution of RAR and retinoid binding protein RNA transcripts in mouse embryos from day 7.5 to day 9.0 of development. The results are discussed in terms of RA-related effects on cells involved in specific morphogenetic events in normal development and teratogenesis.

In situ hybridizations (see Dollé et al. 1990) using ³⁵S-labelled anti-sense RNA probes synthesized from cDNAs of mouse RAR-α, RAR-β, RAR-γ, CRABP I and CRBP I were performed on consecutive sections to study the distribution of their transcripts at early morphogenetic stages of mouse development. Exposure times were 7 days for CRABP I and 15 days for the other four probes. The stages analysed range from the presomitic stage (7.5 days p.c.) to the 20-somite stage (9.0 days p.c.). Since the correlation between the copulation age and the developmental age is variable at these early stages, we have used somite number as the criterion of developmental stage.

All sections were examined by both brightfield and darkfield light microscopy at low and high magnifications. The textual description of the transcript distribution patterns is an account of all of our observations. The photographs have been selected to provide illustrations as clear as possible of the points made in the text. It is not valid to compare background labelling levels between different probes. Intensity of grain depends on exposure times of the sections and of the photographs. It should also be noted that some sections contain appreciable numbers of maternal red blood cells in the trophoblastic lacunae (e.g. Fig. 6B, RAR-β), or embry-onic blood cells in the mesodermal layer of the yolk sac (e.g. Fig. 8A, RAR-α). The brightness in these sites is due to refraction, not silver grains, and is not referred to in the text.

The signal obtained with the RAR-α and RAR-γ probes in embryos at the presomitic (late gastrulation, 7.5 to 8 days) stage was diffuse, indicating either very low levels of expression or background of hybridization (data not shown). In contrast, well-defined labelling was seen in these embryos with RAR-β, CRABP and CRBP probes (Fig. 1), in specific but overlapping regions. CRBP expression was highest in both epiblast (ep) and primary mesenchyme (mesoderm) of the primitive streak (ps), and in the most recently formed adjacent mesenchyme. There was also some CRBP labelling of the basal and apical surfaces of the more rostral neural epithelium (Fig. 1B). CRABP expression was observed throughout the primary mesenchyme including the allantois (al), overlapping with CRBP expression just lateral to the primitive streak, but absent from the streak itself (Fig. 1B). RAR-β ex-pression was lower in midline than in lateral regions of the embryo, and was not clearly tissue-related except in the most rostral area, where it was present in the mesenchyme, but not in the overlying neural epithelium (nf).

Embryos with 4 – 7 pairs of somites (day 8) have well-developed cranial neural folds, which are convex in shape except for the caudal hindbrain, where they are V-shaped (Fig. 2B). Cranial mesenchyme is abundant. During this period, the two heart tubes come together and fuse in the ventral midline as the foregut forms, and the heart begins to beat. Neural tube closure is initiated in the upper cervical region at the 7-somite stage. The embryo has not turned yet, so the relationship between the three germ layers is unchanged from that of the previous stage. Embryos at this stage hybridized with the RAR-α probe showed a slightly higher signal in epiblast and mesoderm of the primitive streak (ps) than in the rest of the embryo (not illustrated); however, this difference may be due to the higher cell density in this region of the embryo rather than to differential levels of expression.

In embryos with 4 – 6 pairs of somites, RAR-β transcripts within the neural epithelium were restricted to caudal hindbrain levels (hb), (Fig. 2B). As at the previous stage, mesodermal RAR-β labelling was regionally restricted. From the 6-somite stage onwards, RAR-β transcripts were also found in the foregut endoderm (fg), (Fig. 2A). As previously described, the earliest RAR-γ transcripts were found in the primitive streak (Fig. 3 and Ruberte et al. 1990).

In 4-to 6-somite-stage embryos, CRABP transcripts were present in the neural epithelium, where they were restricted to a small area of the hindbrain, caudal to the preotic sulcus (po) (Figs 2B, 3). They were also present in the primary mesenchyme (me), immediately sub-jacent to the epithelium of both midbrain and hindbrain (Fig. 2). At the 7-somite stage, labelling had extended to the region of hindbrain just rostral to the preotic sulcus, again with the basal region of the epithelium most intensely labelled (Fig. 4A). The region caudal to the preotic sulcus, in which the signal was present earliest, showed the highest level of labelling in the neural epithelium (Fig. 4B). At this stage CRABP transcripts were also found in the lateral plate mesoderm dorsal to the heart (he) (Fig. 4A) and in the caudal primary mesenchyme as at earlier stages (Figs 1, 2). We cannot distinguish unequivocally between neural crest and primary mesenchyme cells in the cranial region of these sections, but information from previous studies (Nichols, 1981; Tan and Morriss-Kay, 1985, 1986), indicates that the strong CRABP signal seen at the 4-somite stage is in cranial primary mesenchyme (me), (Fig. 2A); the signal in this tissue had diminished considerably by the 7-somite stage (me), (Fig. 4A).

As at the presomite stage, CRBP labelling was present in the epiblast and mesenchyme of the primitive streak (Figs 2, 3). In the neural epithelium CRBP labelling was strongest in the caudal hindbrain (Fig. 2B); at rostral hindbrain and midbrain levels, it was present at the basal and apical neuroepithelial cell surfaces (Figs 2A, 3).

Embryos of this age group (approximately 8.5 days) are beginning to turn, and neural tube closure is extending from the upper cervical region into the hindbrain, cranially, and to lower cervical regions caudally. RAR-α transcripts were observed within the neural epi-thelium of the forebrain (fb), (Fig. 5A), and at the level caudal to the preotic sulcus (Fig. 5C), where they were strongest in the lateral neural epithelium from which neural crest cells were emigrating, and in the crest cells themselves. The RAR-αgene was also expressed in the neural crest cells of the frontonasal and first pharyngeal arch regions (Fig. 5A,B), and less strongly in migrating crest cells (upper nc labels) in the same sections.

RAR-β transcripts were observed, as at the previous stage, in the foregut endoderm (fg) (Fig. 5B – D), in the lateral mesoderm and in the sinus venosus region of the heart (he), (Fig. 5C,D). The somites were unlabelled with the RAR-β probe. RAR-β transcripts in the neural epithelium showed a limit of expression just rostral to the first somite (Fig. 5D, compare with Fig. 5A–C); the expression extended caudally as far as the limit of neural tube closure, the open trunk neural fold epithelium (nf) being unlabelled (Fig. 5D). In contrast, RAR-γ transcripts were specifically present in the open neural epithelium of the caudal neuropore, but not rostral to it (Fig. 5D, and data not shown).

CRABP was strongly expressed in specific regions of the neural epithelium. The limits of expression were from the midbrain-forebrain junction, rostrally, to the first occipital somite level of the hindbrain caudally (Fig. 5A–D, and data not shown). The intensity of the signal was greater in the hindbrain than in the midbrain (not shown). At a higher magnification (Fig. 4C) it could be seen that in the most rostral hindbrain the CRABP signal was strongest on the basal side of the epithelium. At more caudal levels there was strong labelling both apically and basally (Fig. 4D). There were also differences in the intensity of the signal within the transverse plane: the midline supranotochordal cells (future floor plate of the closed neural tube) were unlabelled (Fig. 4, all sections); the signal was also reduced or absent from the lateral edges of neural epithelium from which crest cells had emigrated (Fig. 4D). Cranial mesenchyme cells in positions corresponding to expected positions of neural crest cells, e.g. in the first pharyngeal arch (ph), (Fig. 5B) also showed high levels of CRABP expression (Figs 4, 5A–C), while caudally the strongest signal was in mesoderm lateral to the primitive streak (Fig. 5C) and in the presomitic and lateral mesoderm (me) cranial to the primitive streak (Fig. 5D), extending with diminish-ing intensity into the most recently formed two or three pairs of somites. Mesoderm-derived cranial mesen-chyme appears unlabelled with the CRABP probe (Figs 4, 5); however, it is not clear whether the line of labelled subectodermal cells in Fig. 5D originates from the neural crest or from the mesoderm.

The CRBP gene was expressed within the hindbrain neuroepithelium, the signal being strongest apically and basally (Fig. 5D); the rostral limit of expression was not a clear boundary, but a gradual diminution from caudal to rostral levels (not illustrated), the signal being very low in the forebrain (Fig. 5A). In the cranial mesen-chyme, the pattern of labelling was similar to but weaker than that of CRABP, being strongest in the areas assumed to be occupied by neural-crest-derived cells (Fig. 5A,B). More caudally, CRBP transcripts were detected in the undifferentiated mesenchyme (Fig. 5C,D), and in the youngest somites (data not shown). Unlike CRABP, the CRBP probe gave a strong signal in the primitive streak (ps) itself, where epithelium is being converted to mesenchyme (Fig. 5C), and in the neural folds (including the midline supranotochondal cells) just rostral to the streak (Fig. 5D). At this stage of development the CRBP gene is also expressed in the endodermal layer of the yolk sac (Fig– 5).

In late 8.5 day embryos, RAR-α transcripts were almost ubiquitously expressed. As at the previous stage, presumed neural crest cells were most strongly labelled for RAR-α in hindbrain levels from which neural crest cells were emigrating (compare Figs 6B and 5C), with a lower intensity of labelling in the frontonasal (fn) and pharyngeal arch (ph) mesenchyme (Fig. 6A,B), and little or no expression in the cranial primary mesenchyme (me) or in the occipital somites (oc) (Fig. 6C).

RAR-β transcripts were detected in the caudal hindbrain, extending caudally into the trunk neural tube as far as the caudal neuropore (i.e. labelled in the closed trunk region shown in Fig. 6C, but not in the open neural folds in Fig. 6B). In these embryos, RAR-β transcripts were detected in the frontonasal mesen-chyme (fn) (Fig. 6A) and in the gut endoderm (Fig. 6B,C), extending into the medial part of the ectoderm-derived epithelial covering of the mandibular arch (ph) (Fig. 6B). Transcripts were also present in the splanchnopleuric mesoderm (sp) of the upper trunk (Fig. 6C), but were absent from the somites (so).

As at the previous stage, RAR-γ transcripts were found in the caudal region, in the open neural folds and in the closing region of the neural tube where the RAR-β gene was not yet expressed (Fig. 6A, B, but not C). RAR-γ transcripts were also found in the presomitic mesenchyme of the trunk, and the frontonasal and mandibular arch mesenchyme of the head (Fig. 6A,B).

Embryos at this developmental stage hybridized with the CRABP probe showed no labelling in the forebrain neural epithelium (fb); transcripts were present in the midbrain, extending caudally into the hindbrain and for a short distance into the cervical region (Fig. 6). The highest intensity of the signal in the neural epithelium was in the hindbrain just rostral to the otic placodes (this region is illustrated in Fig. 6B, which is in fact more strongly labelled than the section illustrated in Fig. 6C). The signal was high throughout the hindbrain, except for a small group of midline supranotochordal cells, which were unlabelled (this shows as a rounded instead of pointed shape to this region in Fig. 6B,C). Rostral and caudal to this region, where the signal was less strong (Fig. 6A), the medial transcript-free area was more extensive and, where the signal was present, it was strongest on the basal aspect of the epithelium (Fig. 6A). A group of cells containing high levels of CRABP transcripts, presumably neural crest cells, was seen at the occipital hindbrain level close to the heart, with smaller groups of labelled cells close to the neural tube (Figs 6C, 7A,B). Higher levels of transcripts were detected in the neural-crest-cell-derived mesenchyme of the head and pharyngeal arches than in primary mesenchyme (Fig. 6A,B). Caudally, a lower signal than at the previous stage was observed in the lateral mesoderm. No labelling with the CRABP probe was present in the ectoderm or endodermal layers, or in the somites (Fig. 7C).

CRBP transcripts were present in the medial half of the most rostral hindbrain neural epithelium, with the exception of the midline supranotochondal cells (Fig. 6A). In the hindbrain level with the mandibular arch (ph), they were even more restricted to a small band just dorsal to the supranotochordal cells (Fig. 6B), similar to the band of CRBP protein detected in the trunk neural tube at day 10 by Maden et al. (1990). Caudal to this, as far as the posterior neuropore, only the basal and apical surface regions of the neural epithelium were labelled with the CRBP probe (Fig. 6B, caudal part, and Fig. 6C); the same pattern of apical and basal epithelial labelling was found in the surface ectoderm-derived otic placodes (not shown). CRBP transcripts were detected in the lateral and intermediate mesoderm, in the mesenchyme adjacent to the heart and in the frontonasal and pharyngeal arch mesenchyme. The tail bud mesoderm and epiblast (Fig. 6A) as well as the youngest somites (Fig. 6B,C) showed high levels of CRBP transcripts. At this stage transcripts were also found in the gut endoderm (fg), (Fig. 6C).

In 9 day embryos, the pattern of expression was not very different from that described above, except that in the cranial mesenchyme, the distinction between the labelling intensities of neural-crest-cell-derived and primary mesoderm-derived mesenchyme was clearer for all of the probes, with higher levels of transcripts in the crest-derived cells (Fig. 8). In near-sagittal sections (Fig. 8A), this can clearly be seen in the ring of frontonasal mesenchyme around the forebrain vesicle, and (with the exception of RAR-β) in the mandibular mesenchyme (ph). At this stage, the anterior limit of expression of RAR-α transcripts appears to be the rhombomere 3 –4 boundary, with higher levels of transcripts detected in the outer cells of the neural tube (Fig. 8A); this is the newly forming mantle layer, containing neuroblasts. RAR-α transcripts were found in all mesodermal components of the trunk and limb (lb), but were absent from the heart (he) (Fig. 8A and data not shown).

The rostral limit of expression of RAR-β in the hindbrain neuro-epithelium corresponds to the rhom-bomere 6 –7 boundary, just caudal to the otic vesicles (which were not labelled with the RAR-β probe, not shown), extending caudally along the whole length of the closed region of the neural tube (Fig. 8A). RAR-β transcripts were detected in the frontonasal mesen-chyme, but not in the mandibular arch mesenchyme (ph) (Fig. 8A). RAR-β transcripts were also observed in the lateral and intermediate mesoderm and in the most proximal mesenchyme of the developing limbs. As at the previous stage, RAR-β transcripts were found in the gut endoderm (fg), but were absent from the somites (so) (Fig. 8A).

RAR-γ transcripts in these embryos were found as at the previous stage in the presomitic mesoderm and in the overlying neural folds of the caudal neuropore (not shown). Labelling was also present in the frontonasal and pharyngeal arch mesenchyme as well as in the lateral and limb mesoderm (Fig. 8A, and see Ruberte et al. 1990).

The rostral limit of neuroepithelial expression of CRABP transcripts was at the forebrain-midbrain boundary, with labelling extending caudally to the most rostral levels of the trunk neural tube (Fig. 8A,B). Labelling was particularly strong in the marginal layer of the hindbrain neural tube, which at this stage forms the basal plate (future motor region) ventrally (Fig. 8B). High levels of CRABP transcripts were present in the cranial ganglia (V, VII/VIII, IX and X) and in the otic pit (Fig. 8B). CRABP transcripts were expressed in the frontonasal and pharyngeal arch mesenchyme of the developing face (Fig. 8A,B), in the lateral mesoderm of the trunk and in the limbs. As at the previously described stages, CRABP labelling was present in the mesenchyme associated with the heart, ventral to the dorsal aorta (Fig. 8A and data not shown). No transcripts were found in the gut endoderm or in the somites (Fig. 8A,B).

The strongest CRBP signal at this stage was on the basal side of the otic pit epithelium (Fig. 8B). Else-where in the head, labelling was present in the frontonasal and pharyngeal arch mesenchyme, in the first seven rhombomeres of the hindbrain, and in the foregut endoderm (Fig. 8). In the trunk, CRBP transcripts were observed in the most recently formed somites, and in the gut endoderm. For the first time, the CRBP gene was strongly expressed in the yolk sac endoderm (Fig. 8A,B); at this stage the vitelline circulation is well established, so CRBP in this position is well placed to bind maternal retinol which has traversed Reichert’s membrane, and to provide reti-noids for transport to the embryonic tissues via the vitelline veins and embryonic circulation.

We report here the distribution of the transcripts coding for the three RARs (α, β and γ), CRABP and CRBP during mouse embryogenesis from day 7 to day 9. This period begins at the late gastrulation stage, with differentiation of the cranial neural plate and migration of the newly formed primary mesenchyme. It includes major events in craniofacial and trunk development, including neural crest cell migration and early ganglion development. The central nervous system is established as the neural tube forms and begins histogenesis to form neuroblasts. Each of the five genes studied shows a differential pattern of expression in the embryo, suggesting that they play important and specific roles from the earliest stages of morphogenesis.

The CRBPI probe used here is unlikely to cross-hybridize with CRBP II RNA, but it is probable that the CRABP I probe is detecting both CRABP I and CRABP II transcripts (see Dollé et al. 1990, for further explanation). The RAR-α, RAR-β and RAR-γ probes detect all mRNA isoforms of the corresponding genes (Kastner el al. 1990, and unpublished data from our laboratory for RAR-α and RAR-β isoforms).

There was a clear pattern of differential gene ex-pression for CRBP and CRABP in relation to mesoderm formation: CRBP was expressed in the epiblast and mesenchyme of the primitive streak, while CRABP was expressed in the mesenchyme immediately lateral to the streak and under the newly differentiated neural plate. This pattern persisted in the primitive streak region throughout the period studied. It suggests a role for CRBP in the conversion of epithelium to mesenchyme in the primitive streak, and for CRABP in migration of the primary mesenchyme cells. No RAR probe gave a clear signal in the streak itself before the 4-to 6-somite stage, when RAR-γ was present in the whole caudal region of the embryo (Ruberte et al. 1990). Expression of RAR-β in the newly formed mesenchyme lateral to the streak at early stages was reflected at later stages as expression in the lateral plate mesoderm, including that associated with the heart. Its presence during gastrulation is therefore unlikely to be related to any role in cell migration. Failure to observe RAR-α- or RAR-γ labelling in presomitic stage embryos does not necessarily mean that these genes are not expressed. In fact, either one of the two receptors is probably expressed at this stage, but below the level of sensitivity of our techniques. The expression of RAR-β is indeed known to be induced by RA (see for instance Zelent et al. 1989) and this induction is presumedly mediated by either RAR-α or RAR-γ.

Many other genes are also known to be expressed in the primitive streak region. From the 5-somite stage onwards, the patterns described for Hox-2.6, Hox-2.7, Hox-2.8 and Hox-2.9 (Wilkinson et al. 1989b) are similar to those described here for RAR-α, RAR-γ and CRBP, being in both epiblast and primary mesen-chyme. CRABP transcript distribution resembled that of the FGF-like proto-oncogene int-2 (Wilkinson et al. 1988) in being specifically mesenchymal in the primitive streak, except that it was not expressed in the newly formed midline cells.

Migration of primary mesenchyme cells is inhibited by retinoid excess: retinol has been observed to decrease locomotory activity of explanted primary mesenchyme cells (Morriss, 1975); whole rat embryos exposed in vitro to both retinol and retinoic acid during late gastrulation stages have decreased numbers of cranial mesenchyme cells (Morriss and Steele, 1974, 1977); similar results are obtained by exposing embryos to excess retinoids in utero (Morriss, 1972). The cellular binding proteins may therefore act to control the levels of retinoic acid and retinol that are available to primary mesenchyme cells during their formation in the primitive streak and during migration to their ultimate destinations.

Conversion of epithelium to mesenchyme also occurs during neural crest cell emigration from the neural epithelium. In mouse embryos, crest cell migration begins at the 4-to 6-somite stage, from the midbrain-rostral hindbrain region (Nichols, 1981). Although CRBP expression was observed in the neural epi-thelium and in the migrating cranial neural crest cells, it did not show the clear association with mesenchyme formation seen in the primitive streak. CRABP expression was much more significant in this context: as the cranial neural crest cells began to emigrate, there was a concomitant intensification of the signal in the neural epithelium, and the crest cells retained a high level of CRABP expression during their migration. Only in the preotic hindbrain was a strong signal present in the neural epithelium before the onset of crest cell emigration. This region, which continued to label more strongly with the CRABP probe than elsewhere throughout development, is particularly susceptible to retinoid excess. It is shortened in embryos exposed to excess retinoids in vivo and in vitro, so that the otocyst is shifted rostrally with respect to the pharyngeal arches (Morriss and Thorogood, 1978; Webster et al. 1986). CRABP was not expressed in the forebrain, which, in contrast to that of avian embryos (Le Douarin, 1982), makes no contribution to the neural crest in mammals (Tan and Morriss-Kay, 1986; O’Rahilly and Müller, 1987) except for a small emigration from the optic epithelium (Bartelmez, 1960). In the trunk, where neural crest cells emigrate after neural tube closure (Martins-Green and Erickson, 1986), CRABP transcripts appeared in the neural tube as it closed, with concomitant labelling in adjacent groups of cells which appeared to be neural crest. These observations suggest that the presence of CRABP transcripts is in some way related to the formation and migration of neural crest cells, and may be a useful crest cell marker during the migratory phase. Maden et al. (1990) have shown by immunohistochemistry the selective presence of CRABP in neural crest cells. Our results indicate that this selective expression corre-sponds to a control at the mRNA level.

The sites of CRABP expression in the cranial neural epithelium and neural crest are also specific sites of uptake of RA: Dencker et al. (1987) administered RA-15-¹⁴C to day 8.5 pregnant mice, and found after 4h that uptake was greatest in the hindbrain neural epithelium and neural crest, and absent from the forebrain epithelium; on day 9.5, after cranial neural tube closure and crest cell migration, crest-derived facial and pharyngeal arch mesenchyme was labelled. These tissues are specific targets of retinoid excess-induced teratogenesis (Morriss and Thorogood, 1978; Webster et al. 1986). The pattern of CRABP expression therefore provides important evidence in support of the hypothesis that CRABP is involved in controlling RA levels in cells whose requirement for RA is low, and which are therefore specifically vulnerable to RA excess, when the binding capacity of CRABP is saturated (see Dollé et al. 1990).

The distribution of RARs in relation to neural crest cells are less clearly suggestive of possible roles. RAR-α was the only nuclear receptor expressed at higher levels in the migrating neural crest than in the surrounding mesodermal mesenchyme. It was expressed at high levels in sites of neuroepithelium to neural crest mesenchyme conversion.

After crest cell migration, differences in the intensity of labelling were observed between the neural crest cells and the primary mesenchyme of the head with RAR-α and RAR-γ, as well as with the two retinoid binding proteins, all of them being expressed at higher levels in the neural crest-derived mesenchyme. At this stage, regional differences began to be established, with RAR-β transcripts present within the frontonasal mesenchyme, but absent from the pharyngeal arch mesenchyme. At subsequent stages, each transcript became progressively more restricted in its distribution in the developing craniofacial region, suggesting that they play a role in pattern formation through regulation of expression of different sets of genes (see Dollé et al. 1990).

The roles of retinol and retinoic acid in craniofacial development are clearly complex, as indicated both by the patterns of distribution of the five transcripts reported here and in the related study of Dollé et al. (1990) and by observations on the teratogenic proper-ties of retinoid excess, which are particularly important in the craniofacial region, thymus and heart (Morriss and Thorogood, 1978; Geelen, 1979; Rosa, 1983; Lammer et al. 1985; Webster et al. 1986). These authors have attributed an important role in the genesis of the malformation to an inhibition of neural crest cell migration. Our results support this hypothesis by correlating the distribution of RAR-α- and retinoid binding protein gene expression with neural crest cell migration, so that retinoid excess could induce abnor-mal gene expression in the crest cells, resulting either in a direct effect on their migratory capacity (e.g. by an effect on the cytoskeleton or cell surface receptors) or in a modification of the molecular composition of the extracellular matrix which provides the substratum for migration.

The effects of excess retinoic acid on heart develop-ment are clearly distinct from those of vitamin A deficiency. In retinol deficiency, failure of closure of the interventricular septum was common (Wilson and Warkany, 1949), whereas retinoic acid excess is associated with septation defects of the outflow tract (aorticopulmonary septation failure) (Lammer et al. 1985). Closure of the interventricular septum depends on proliferation and fusion of the cardiac jelly-rich endocardial cushions and upper interventricular septum tissue, in which CRBP is expressed (Dollé et al. 1990). In contrast, septation of the cardiac outflow tract has been shown in avian embryos to depend on the immigration of a specific population of neural crest cells from the occipital region of the hindbrain (Kirby et al. 1983,1985; Kirby, 1989). A crest cell migration pathway from the occipital hindbrain to the outflow tract has now been identified in rat embryos (Fukiishi and Morriss-Kay, 1991), adding further weight to the hypothesis that RA-induced failure of aorticopulmon-ary septation is due to neural crest cell migration failure. The CRABP-expressing cells that we observed close to the outflow region of the heart in 12-to 14- and 18-to 20-somite-stage embryos almost certainly rep-resent this occipital/cardiac neural crest cell popu-lation.

The expression of specific genes has been correlated with specific segmental levels in the central nervous system of mouse embryos. This is of particular interest in the hindbrain, which is clearly segmental in nature, being divided into eight rhombomeres by a series of sulci and gyri (Orr, 1887; Adelmann, 1925; Tuckett and Morriss-Kay, 1985). Since the initial observation of an expression boundary between rhombomeres of a homeobox gene within the developing hindbrain (Gaunt et al. 1986), segment-specific gene expression in the hindbrain has been demonstrated for several members of the Hox-1 (Gaunt, 1987; Duboule and Dollé, 1989) and the Hox-2 complex (Murphy et al. 1989; Wilkinson et al. 19896) of murine homeobox genes, as well as for the zinc-finger-encoding gene Krox-20 (Wilkinson et al. 1989a) and the int-2 proto-oncogene (Wilkinson et al. 1988). Retinoic acid is known to be able to control the expression of a number of developmental genes including homeobox genes in vitro (Colberg-Poley et al. 1985a,6; Deschamps et al. 1987; Kessel et al. 1987; Murphy et al. 1988; La Rosa and Gudas, 1988; Mavilio el al. 1988; Simeone et al. 1990). Thus RARs may mediate RA-regulated ex-pression of genes associated with specific rhombo-meres, as part of the mechanism of rhombomere-specific development at these and subsequent stages. RAR-α is expressed only caudal to rhombomere 3. Hence the RAR-α expression boundary in the rhom-bencephalon appears quite similar to that of Hox-1.6, the extreme 3’-located gene of the Hox-l complex, which has been shown to display the most rostral expression boundary among genes of its complex (Duboule and Dolió, 1989). The RAR-α expression boundary in the hindbrain may also coincide with the anterior boundary of expression of Hox-2.9, which is confined to rhombomere 4 (Murphy et al. 1989). RAR-γ is expressed within and caudal to rhombomere 7, which coincides with the rostral limit of expression of Hox2.6 (Wilkinson et al. 19896), and the caudal limit of expression of int-2 (Wilkinson, 1990). These correspon-dences of position suggest that segment-specific gene expression could be differentially controlled by RARs, possibly with the help of CRABP whose expression is particularly high in the hindbrain. In fact, an altered pattern of expression of Hox 2.9 and Krox-20 has recently been observed in RA-exposed mouse embryos (Morriss-Kay et al. unpublished data). RA also affects hindbrain development in amphibian embryos (Dur-ston et al. 1989).

The expression patterns of RAR-βand RAR-γ in the neural epithelium showed an interesting correlation. RAR-β was expressed specifically in the closed neural tube, whereas RAR-γ was expressed in the open neural epithelium of the caudal neuropore (see also Ruberte et al. 1990). As the closing region of the neural tube moved caudally, this boundary of expression moved caudally with it. This change in RAR gene expression may be correlated with maturational change within the trunk neural epithelium, which begins to produce neuroblasts only after closure (Langman et al. 1966). Retinoid-related gene expression in the region of the caudal neuropore is also of interest in the context of neurulation, since exposure to high RA levels at early stages of development can cause spina bifida (Shene-felt, 1972; Alles and Sulik, 1990). Expression of CRABP (as well as RAR-γ) in the primary mesen-chyme adds weight to the hypothesis that failure of neural tube closure may be due to mesenchymal deficiency, through one or more mechanisms that may involve increased cell death (Alles and Sulik, 1990), decreased migration efficiency (see earlier discussion), or failure of a tissue interaction with the neural epithelium involving synthesis of extra-cellular matrix molecules (e.g. heparan sulphate proteoglycans, which are known to be essential for the generation of neuroepithelial curvature, Tuckett and Morriss-Kay, 1989).

The floor plate of the neural tube consists of the ventral midline supranotochordal cells, which become wedge-shaped through an inductive interaction with the notochord (Staaten et al. 1988; Schoenwolf and Smith, 1990). This tissue has recently been shown to have polarizing activity when grafted under the apical ectodermal ridge of the chick wing bud (Wagner et al. 1990); the activity was restricted to the closed region of the neural tube, the midline cells of the open region being inactive. The authors suggested that, in situ, the floor plate might impose polarity along the dorsoventral axis, and that this role might involve retinoids, since CRBP (CRBP I) is present in the floor plate at later stages (Maden et al. 1989). Our results, however, show no expression of CRBP (CRBP I) in the midline neuroepithelial cells except just rostral to the primitive streak, where they were present throughout the neuroepithelium (Fig. 5D). No transcripts of CRBP I or CRABP could be detected in the midline cells in the open region rostral to this position, or in the floor plate of the closed neural tube during the period in which dorsoventral axis polarity is established. Similarly, Maden et al. (1990) observed absence of both CRBP I and CRABP proteins from the floor plate of the recently closed neural tube of rat and mouse embryos. CRBP I-derived retinoic acid generated from the floor plate is therefore unlikely to play a role in establishing dorsoventral polarity in the neural tube of mammalian embryos, but CRBP II, which our CRBP I RNA probe is unlikely to detect, may be involved.

In conclusion, we have shown here that the pattern of gene expression of RARs and retinoid binding proteins in mouse embryos correlates well with some specific morphogenetic events, particularly in relation to gastrulation and primary mesenchyme cell migration, cranial neural crest cell migration, hindbrain segmen-tation and neural tube closure. These observations provide a basis for further molecular analysis of the roles of retinoids in normal and abnormal mammalian development. In particular, the CRABP transcript distribution patterns correlated well with known target tissues of excess retinoic acid-induced teratogenesis, providing further evidence in support of our hypothesis that cells expressing CRABP are those which cannot tolerate high levels of RA for their normal developmen-tal function.

We are grateful to Drs P. Kastner, C. Stoner, L. Gudas, A. Krust, P. Leroy and A. Zelent for generous gifts of probes. We thank all the members of the retinoic acid receptor group for useful discussions, Martin Barker for technical assistance, C. Werlé, A. Landmann, B. Boulay and J-M. Lafontaine for illustrations, and the secretarial staff. This work was supported by the INSERM (grant CNAMTS), the CNRS, the Fondation pour la Recherche Médicale and the Association pour la Recherche sur le Cancer. G.M-K is supported by a grant from Hoffmann-La Roche, and E.R. by a fellowship from the Université Louis Pasteur.