Function of the retinoic acid receptors (RARs) during development: (I) Craniofacial and skeletal abnormalities in RAR double mutants

It has long been known that retinol (vitamin A) is crucial for normal growth, vision, reproduction, maintenance of numerous tissues and overall survival (Wolbach and Howe, 1925; see Sporn et al., 1994 and Blomhoff, 1994, for reviews and references). Retinol is also essential for normal development, as shown by the appearance of multiple congenital abnormalities in fetuses from dams fed a vitamin A-deficient (VAD) diet (the fetal VAD syndrome, see Wilson et al., 1953 and references therein). Interestingly, with the exception of vision (Wald, 1968), retinoic acid (RA) appears to be the active derivative of vitamin A, since its administration can prevent or reverse most of the defects induced by postnatal VAD (Thompson et al., 1964). Furthermore, RA excess is much more teratogenic than retinol excess, causing many developmental abnormalities, the precise malformation depending largely on the time of administration (reviewed in Morriss-Kay, 1993; Nau et al., 1994; Hofman and Eichele, 1994). The spectacular effects of topical RA application on limb development and regeneration popularized the belief that RA could in fact be a morphogen (for reviews, see Tabin, 1991; Hofman and Eichele, 1994).

The discovery of a nuclear receptor for RA, acting as a ligand-inducible transcriptional regulator (RAR; Petkovich et al., 1987; Giguère et al., 1987), greatly advanced our understanding of the molecular mechanisms underlying the pleiotropic effects of retinoids (synthetic and natural derivatives of RA; reviewed in Leid et al., 1992; Kastner et al., 1994; Mangelsdorf et al., 1994; Linney and LaMantia, 1994). Since this initial finding, it has been shown that the RA signal can be trans-duced in cultured cells through two families of retinoid receptors. The RAR family (RARα, β and γ and their isoforms) are activated by both all-trans RA and 9-cis RA, whereas the RXR family (RXR α, β and γ) are activated only by 9-cis RA. The DNA-binding and ligand-binding regions (regions C and E, respectively) of the three RAR types are highly similar, whereas the C-terminal F region and the central D region exhibit little, if any, conservation. The three RAR types also diverge in their N-terminal B regions and further diversification is generated for each receptor type by variant isoforms differing in their N-terminal-most A regions (RARα1 and α2, β1 to β4, and γ1 and γ2), which originate from alternate splicing and differential promoter usage (reviewed in Leid et al., 1992). Amino acid sequence comparisons have revealed that the interspecies conservation of a given RAR type and of each of its isoforms is greater than the similarity found between the three RARs within a given species (see Kastner et al., 1994 for review). Furthermore, the various RAR isoforms contain two transcriptional activation functions (AFs), located in the N-terminal A/B region (AF-1) and C-terminal E region (AF-2) which act synergistically, and sometimes differentially, to activate various RA-responsive promoters. Taken together with the distinct spatiotemporal transcript distribution observed for each RAR and isoforms during mouse embryogenesis and in adult tissues, the above interspecies sequence conservation and transcriptional activation specificities suggested that each RAR isoform may perform unique functions (for refs see Kastner et al., 1994; Chambon, 1994). Furthermore, the finding that RA-responsive promoters are likely controlled in cultured cells through RAR-RXR heterodimers (reviewed in Kastner et al., 1994; Mangelsdorf et al., 1994; Chambon, 1994) suggested that the diverse effects of retinoids may also reside in the control of various subsets of retinoid-responsive promoters by different combinations of RAR-RXR types (and isoforms).

To evaluate the function of the various RARs (types and isoforms) in vivo, we have created mice lacking several of these receptors. Surprisingly, mice deficient for RARα1 (Li et al., 1993; Lufkin et al., 1993), RARβ2 (Mendelsohn et al., 1994a) or RARγ2 (Lohnes et al., 1993) isoforms were apparently unaffected. In contrast, mice deficient for RARα or RARγ receptors (all α or γ isoforms disrupted) exhibited postpartum lethality and growth deficiency (Lohnes et al., 1993; Lufkin et al., 1993). Furthermore, RARα null mice presented a degeneration of the testicular germinal epithelium, which was similar to that observed in male rats maintained on a VAD diet (Howell et al., 1963). Male RARγ null mutants also exhibited VAD-like abnormalities, namely squamous metaplasia of the seminal vesicles and prostate gland. Agenesis of the ocular Harderian gland and homeotic transformations of the axial skeleton were also observed in RARγ null mutants, although both of these defects occurred with incomplete penetrance and expressivity. In no case, however, have these RAR null mutants displayed any of the congenital malformations observed in fetal VAD studies (Wilson et al., 1953). Furthermore, the phenotype of RARα and RARγ null mice was confined only to a small subset of tissues expressing these receptor types. Thus, contrary to our expectations, the RAR types and isoforms disrupted to date do not apparently possess the unique functions that were predicted on the basis of their evolutionary conservation, expression pattern and in vitro transcriptional regulatory characteristics.

Taken together, the above findings suggest either that there is a high degree of functional redundancy among members of the RAR family, or that the RARs are not essential transducers of the retinoid signal in vivo. To discriminate between these possibilities, we have generated and analyzed RAR compound null mutants. The defects displayed by various double mutants recapitulate essentially all of the congenital malformations found in fetal VAD. These double mutants also exhibit a number of abnormalities not previously described in VAD experiments (see also the accompanying study). We report here a detailed analysis of the craniofacial and skeletal defects found in RAR double null mutants.

The generation of RARα1, α, β2 and γ single null mutants has been described (Lohnes et al., 1993; Lufkin et al., 1993; Mendelsohn et al., 1994a). Initial intercrosses of these single mutants were performed to derive double heterozygotes. With the exception of RARα^+/–/γ^+/– offspring, second generation animals were obtained by mating double heterozygotes with the appropriate RAR heterozygous or homozygous single mutants to test for viability and fertility of compound mutants. This was performed in order to optimize subsequent generation of double null mutants. The RAR double mutants and the crosses used to generate them were as follows: RARα1^–/–/β2^–/– were derived from RARα1^+/–/β2^–/– intercrosses or crosses between RARα1^+/–/ β2^–/– and RARα1^+/–/β2^+/– animals; RARα^–/–/β2^–/– and RARα^–/–/ β2^+/– mutants were derived from crosses between RARα^+/–/β2^–/– males and RARα^+/–/β2^+/– or RARα^+/–/β2^–/– females; RARα1^–/–/γ^–/– mutants were derived from crossing RARα1^–/–/γ^+/– males and RARα1^+/–/γ^+/– females; RARα1^–/–/γ^–/–/α2^+/– mutants were derived from crosses between RARα1^–/–/γ^+/– and RARα^+/–/γ^+/– animals; RARα^–/–/γ^–/– mutants were obtained from intercrosses between RARα^+/–/γ^+/– animals; RARβ2^–/–/γ^–/– mutants were obtained from crosses between RARβ2^–/–/γ^+/– males and RARβ2^+/–/γ^+/– or RARβ2^–/–/γ^+/– females.

Animals were mated overnight and females examined for a vaginal plug the following morning. Noon of the day of evidence for a vaginal plug was considered 0.5 dpc. Embryos (10.5-14.5 dpc) or 18.5 dpc fetuses were obtained by Cesarean section and genotypes determined by genomic Southern blotting using DNA isolated from the yolk sac or placenta. Probes, digests and other conditions for southern blotting have been detailed elsewhere (Lohnes et al., 1993; Lufkin et al., 1993; Mendelsohn et al., 1994a).

For whole-mount skeletal analysis, fetuses were collected at 18.5 dpc and stored at –20°C. Skeletons were prepared as described (Lufkin et al., 1992). For histological analysis, embryos or skinned fetuses were fixed in Bouin’s solution. Paraffin sections, 7 μm thick, were stained with hematoxylin and eosin or Groat’s hematoxylin and Mallory’s trichrome (Mark et al., 1993).

Animals lacking the RARα1 or β2 isoforms were apparently normal (Table 1A; see refs also Lufkin et al., 1993; Mendelsohn et al., 1994a). Although exhibiting a high degree of neonatal lethality, animals lacking all isoforms of either RARα or RARγ survived in isolation for at least 24 hours when delivered by Cesarean section at 18.5 day postcoitum (dpc), being in this respect as viable as their control littermates (Table 1A; see Lohnes et al., 1993; Lufkin et al., 1993). With the exception of αγ double mutants, the distribution of RARα1β2, αβ2, α1γ, α1γα2^+/– and β2γ double mutant offspring at 18.5 dpc indicated that loss of these receptors did not result in embryonic lethality [Table 1A; for the sake of simplicity mutants null for RARα1 and RARβ2, RARα (all isoforms) and RARβ2, etc, are called hereafter and in the accompanying study α1β2, αβ2, etc, mutants]. However, in contrast to the RAR single mutants, these double mutants invariably died within at most 12 hours following Cesarean delivery at 18.5 dpc.

The frequency of RARαγ mutants found at 18.5 dpc was lower than predicted from Mendelian distribution (Table 1B), indicating partial embryonic lethality. Although analysis of earlier stages of development (10.5-13.5 dpc) yielded the expected frequency, a large fraction were either dead or partially resorbed (Table 1B). The time of death appeared variable, as judged by the size and relative proportion of resorbed and dead mutant embryos (Table 1B, data not shown). The embryonic and postpartum lethality may be due to one or more malformations affecting the heart, aortic arches, kidney, lung or trachea which were observed in RAR double mutant fetuses (see the accompanying study).

Upon external inspection at 18.5 dpc, α1β2, αβ2 and β2γ fetuses could not be distinguished from their littermates. In contrast, αγ mutant fetuses could be readily identified by their reduced size (compare Fig. 1a and c), the small size or apparent absence of the eyes (asterisk, compare Fig. 1a with c and g-i) and the aspect of their mid-facial region (compare Fig. 1d and f). Their snout was markedly foreshortened and divided by a sagittal median cleft (large arrow, Fig. 1f). The prolabium (median third of the upper lip, dashed box in Fig. 1d) was absent (compare Fig. 1d and f). The maxillary processes which bear whiskers (double arrow, Fig. 1f) were located farther apart than in WT animals. Paramedian swellings likely corresponding to the nasomedial processes (single small arrow, Fig. 1f and see below) were fused to the maxillary processes ventral to the nostrils, which opened dorsally instead of rostrally (compare arrowheads, Fig. 1d and f). Additional external defects occasionally found in these mutants included: exteriorized brain (exencephaly, Fig. 1h and i, and Table 1B), bilateral agenesis of the auricle (open arrows, compare Fig. 1a with c and g), umbilical hernia (Fig. 1i, large black arrow), and abnormal limbs (e.g. Fig. 1c,h and i, and see below).

A persistent opening of the rhombencephalic neural tube was observed in nine out of twenty living αγ mutant embryos between 10.5 and 11.5 dpc (e.g. compare Fig. 2b and e) and in one of seven living αγ embryos at 13.5 dpc (Table 1B). In these embryos, the telencephalic hemispheres appeared markedly underdeveloped (TE, compare Fig. 2a,b with Fig. 2d, and Fig. 2c with Fig. 2f). Some αγ mutant embryos also showed a severe degree of lateral deviation of the vertebral axis (scoliosis, arrows in Fig. 2d,e). However, the only pathognomonic external feature of αγ mutant embryos between 11.5 and 13.5 dpc was the aspect of the frontonasal segment of the face: both the nasolateral (NL) and the nasomedial (NM) processes, located on either side of the ofactory pit (OP), were present (compare Fig. 2c and f). The nasomedial processes were normally fused with ipsilateral maxillary processes, but were never fused at the midline resulting in a median facial cleft (FC; compare Fig. 2c with f, and Fig. 2g with h). This lack of fusion may result from cell deficiency caused by excessive cell death in the frontonasal mesenchyme, which was observed on serial histological sections in 10.5 dpc αγ embryos (dashed box, Fig. 2i).

18.5 dpc α1γα2^+/– fetuses were brachycephalic (compare Fig. 1a and b) and occasionally showed a median cleft of the upper lip (compare Fig. 1d and e). In contrast, α1γ mutants did not exhibit any abnormal external features.

Craniofacial skeletal deficiencies were not observed in 18.5 dpc αβ2, α1β2, or β2γ mutant fetuses. In contrast, most of the neural crest cell (NCC)-derived craniofacial skeletal elements were altered in αγ mutants. The most severe skeletal defects were observed in the midfacial region and rostral cranial base, consistent with the loss of midfacial structures described above. The skeletal elements normally derived from the frontonasal mesectoderm are the frontal (F) and nasal (N) bones (Fig. 3a), the cartilaginous template of the ethmoid bone [comprising the nasal septum (NS, Figs 3k and 4a), the nasal capsule (NC, Figs 3a,k, 4a) and the lamina cribriform (LC, Fig. 4a)], and the incisive (or premaxillar, PX, Fig. 3a,k) and vomer (not shown) bones (De Myer, 1975). These elements were grossly deficient or absent in αγ mutant fetuses. The medial portions of the frontal (F) and nasal (N) bones were lacking (compare Fig. 3a with c,d). The nasal capsule was apparently reduced to laterocaudal rudiments (NC, compare Fig. 3a with c,d, and Fig. 3k with m). The rest of the nasal capsule, the nasal septum, the lamina cribriform and the vomer and incisive bones could not be identified (NS and PX, compare Fig. 3a with c,d, and Fig. 3k with m), and were apparently replaced by aggregates of cartilaginous and bony nodules or rods (asterisks in Figs 3m, 4c, 5c,d). In the cranial base, caudal to the ethmoid bone, similar aggregates replaced the presphenoid bone (data not shown). The upper incisors, which are largely derived from the nasomedial processes (Lumsden and Buchanan, 1986), were lacking (not shown). The hypophyseal foramen of the basisphenoid was never closed (arrowhead in Fig. 3m), thus the pituitary gland (HY, Fig. 4g) remained in contact with the pharynx.

Many of the first pharyngeal arch-derived skeletal elements (Noden, 1988; Le Douarin et al., 1993 and references therein) were also malformed (e.g. maxillary and palatine bones) or hypoplastic (e.g. alisphenoid) (compare X, AL and PL in Fig. 3k,m). Surprisingly, however, the mandibular (dentary) bone (compare D in Fig. 3a,c,d), the temporomandibular joint (not shown), the malleus middle ear ossicle (not shown) and the tympanic bone (T, compare Fig. 3k with m) appeared normal, as did the patterning of the lower dentition and the shape (cuspal pattern) of the first and second upper and lower molars and of the lower incisors (compare UM and LM in Fig. 4a,c, and data not shown). Second and third pharyngeal arch-derived skeletal elements were either absent (stapes, not shown) or highly malformed (styloid and hyoid bones, see the accompanying study and data not shown).

In non-exencephalic αγ mutants, the skull vault caudal to the frontal region was complete, although markedly underossified [compare the size of the parietal (P) and interparietal (IP) bones in Fig. 3a,d; also note the absence of interparietal (IP) and supraoccipital (S) ossification centers in Fig. 3c]. In contrast, the entire cranial vault was absent in exencephalic mutants (Fig. 7k).

The cartilaginous otic capsule was always small and incomplete in αγ mutants (O, compare Fig. 3a,c,d), resulting in a cystic protrusion of the epithelial inner ear within the braincase (not shown). These malformations are likely to be secondary to defects of the otocyst (discussed in Mark et al., 1993), which was consistently hypoplastic in 10.5 dpc αγ mutants (not shown). That the epithelial inner ear is a RA-target organ affected early during embryogenesis is supported by the absence of some of its derivatives (i.e. the spiral organ of Corti and the spiral ganglion, data not shown) in 18.5 dpc αγ fetuses.

Although less affected than αγ mutants, α1γα2^+/– mutants (but not α1γ mutants) also exhibited several defects of the cranial skeleton. These included shortening of the frontal bone (F, compare Fig. 3a and b), duplication of the cartilaginous nasal septum (NS, compare Fig. 3k and l), cleft palate, aplasia of the presphenoid (PS, compare Fig. 3e and f), persistence of the hypophyseal foramen (arrowhead Fig. 3f) and absence of the incisive foramen (IF, compare Fig. 3k and l). The latter malformation was closely correlated with the absence or dysplasia of the upper incisors and may be secondary to this defect. Ectopic cartilaginous and bony nodules, formed from the meninges, were also found in 18.5 dpc αγ and α1γα2^+/– mutant fetuses. In particular, in αγ mutants, the falx cerebri was completely chondrified (FX, Fig. 4c).

With the exception of the ectopic cartilaginous and bony deposits, the above skeletal abnormalities correspond to deficiencies, including the duplicated nasal septum which arises by failure of coalescence of the nasomedial processes (De Myer, 1975). Aside from these deficiencies and ectopias, two supernumerary skeletal elements were frequently detected.

In placental mammals, two cartilaginous pillars, the pila prooptica and pila metoptica, connect the orbital (optic) region of the fetal skull to the floor of the braincase (De Beer, 1985). In normal mice at 18.5 dpc, these pilae (PP and PM in Fig. 3e) are located on either side of the optic foramen (OF) and are fused ventrally to the presphenoid bone (PS). All 18.5 dpc αγ, α1γα2^+/– and α1γ mutant fetuses possessed a third, more caudal, cartilaginous pillar which was fused ventrally to the basisphenoid bone to form a cartilaginous medial wall to the cavum epiptericum [PA, compare Fig. 3e with f, and Fig. 4e-g with 4d; the size of this additional pillar was greater in αγ and α1γα2^+/– (not shown) mutants than in α1γ mutants]. The cavum epiptericum (bracketed in Fig. 4d), which corresponds to a normal extracranial space (i.e. located outside of the dura mater), is so called as it is limited laterally by the alisphenoid bone (i.e. the mammalian homologue of the reptilian epipterygoid bone) (AL, Figs 3e,k, 4d). This cavum lodges the trigeminal ganglion (G5, Fig. 4d-f) and is crossed by cranial nerves III, IV, V (N5, Fig. 4g) and VI, and the internal jugular vein (JV, Fig. 4g).

In mammals, the alisphenoid bone contributes to the base and lateral walls of the skull between the optic and otic regions where it forms the lateral limit of the cavum epiptericum, whereas the incus (Fig. 3g) represents one of the three middle ear ossicles. In the middle ears of α1β2, αβ2^+/–, αβ2, α1γα2^+/– and αγ, but not α1γ or β2γ mutants, the medial aspect of the body of the incus was continuous with a rostrally oriented cartilaginous or osseous rod (Q, Fig. 3h-j) which was frequently fused to the alisphenoid bone (Table 2). In a number of these mutants, the short process of the incus (SI) was conspicuously larger than its wild-type homologue (compare Fig. 3g and i).

The brains of 18.5 dpc α1β2, αβ2 and β2γ mutants appeared normal. In contrast, in 10.5 dpc αγ mutants, a wide persistent opening of the rhombencephalic neural tube was frequently observed (see above and Fig. 2e). The midbrain and forebrain of these exencephalic embryos were invariably closed and covered by ectoderm (Fig. 2d,e). However, the neurectoderm of the telencephalic vesicles (TE) was abnormally folded and the lateral ventricles (LV) were collapsed (compare Fig. 2g,h). Histological analysis of two 18.5 dpc exencephalic fetuses showed a lack of hindbrain and cerebellar structures; furthermore the cerebral hemispheres were small and displayed hemorragic foci (not shown). Taken together with the finding that persistent opening of the rhombencephalon at 10.5-11.5 dpc and exencephaly at 18.5 dpc occurred with similar frequencies (Table 1B), these results suggest that failure of closure of the rhombencephalic neural tube is the primary defect leading to exencephaly. This failure, which leaves the rhombencephalic neurectoderm exposed to the amniotic fluid, may result in rhombencephalon degeneration and impair the accumulation of cerebrospinal fluid in the ventricular system. The absence of hydrostatic pressure would then lead to the abnormal folding of the cerebral hemispheres (Pexieder and Jelinek, 1970; Jacobson, 1981), thus altering the relationship between the neurectoderm and overlying osteogenic cranial mesectoderm and subsequently impairing the epithelial-mesenchymal skeletogenic interactions required for the formation of the bones of the skull vault (Hall, 1991). The improper interactions between the folded neuroepithelium and presumptive dermis might also lead to the absence of skin covering the brain in 18.5 dpc exencephalic mutants.

The brain of 18.5 dpc non-exencephalic αγ mutant fetuses appeared considerably distorted (BR, Fig. 4c and data not shown), probably secondary to increased intracranial pressure caused by the shortening of the braincase and compression by intracranial ectopic cartilaginous and bony nodules (e.g. FX in Fig. 4c). In addition, failure of the rostral interhemispheric commissures (i.e. corpus callosum, hippocampal commissure and anterior commissure) to cross the midline was consistently observed, a condition that is also frequently encountered in humans with median cleft face syndrome (De Myer, 1975; Cohen and Sulik, 1992). In the mutant hindbrain, the motor nucleus of the abducens nerve (derived from rhombomeres 5 and 6, Lumsden et al., 1991) was not identifiable (not shown); whether this reflects a primary effect on the hindbrain or is secondary to the abnormalities present in the ocular region (see below) is unclear, since the target organ of the abducens nerve is the external rectus muscle of the eye.

Distortions of the brain and absence of the interhemispheric commissures was also found in two out of five 18.5 dpc α1γα2^+/– mutants (BR, Fig. 4b). In these mutants, which display nearly normal eyes, the abducens nucleus was always present. The brains of α1γ mutants, which did not exhibit gross cranial skeletal malformations, appeared normal.

The eye of a normal 18.5 dpc fetus is covered by fused eyelids (Y, Fig. 5a,e). A wide conjunctival sac (J, Fig. 5a,e) is present between the lids and the cornea (C). There are two cell-free spaces, one between the cornea and the lens (the anterior chamber; A, Fig. 5a,e), and the other between the lens and the retina (the vitreous body; V, Fig. 5a,e). The retina, comprising the internal (neural) and external (pigmented) epithelia (IR and OR, Fig. 5a,e), encloses the lens and the vitreous body and is continuous with the iris (I, Fig. 5a,e) laterally.

RARαγ, α1γα2^+/–, α1β2, αβ2 and β2γ mutants exhibited a number of ocular defects (Table 3). Microphtalmia, coloboma of the retina and abnormalities of the cornea, eyelids and conjunctiva, were constant features of 18.5 dpc αγ fetuses. As shown in Fig. 5c, the eye was conspicuously smaller than normal (compare Fig. 5a and c). The dorsal margin of the retina extended well past the equator of the lens (L, Fig. 5c), where it formed a rudimentary iris (I, Fig. 5c). Ventrally, the retina barely reached the equator (possibly secondary to the coloboma), so that the ventral lens region was in direct contact with periocular mesenchyme (PO, Fig. 5c). Medially, a second ventral gap permitted communication between periocular mesenchyme and persistent retrolenticular mesenchyme (F, Fig. 5c), which occupied the space normally taken by the vitreous body (V, Fig. 5a). The two ventral gaps were joined caudally. In this plane of section, however, they were separated by a portion of the retina with a duplicated internal leaf. This aspect is characteristic of an eversion of the retina (ER, Fig. 5c), which is thought to arise by metaplastic transformation of pigmented epithelium into neural epithelium (Coulombre and Coulombre, 1977). The cleft in the ventral portion of the retina, the penetration of the optic cup by mesenchymal tissue and the eversion of the retina in the cleft region are characteristic of the typical complete coloboma of the retina (Mann, 1937). The developmental fault underlying this defect is a complete persistence of the optic fissure (also called fetal or choroid fissure, which normally closes completely by 14.0 dpc) through its entire length from the region of the optic disc (the optic nerve exit point) to the iris. Additional defects in this eye included: absence of fusion of the eyelids (compare Y, in Fig. 5a and c), agenesis of the upper (dorsal) conjunctival sac (compare J, in Fig. 5a and c), persistent corneal-lenticular stalk (see below), absence of differentiation of the corneal stroma (C, in Fig. 5a,e) and absence of the anterior chamber of the eye (A, in Fig. 5a,e). Another αγ mutant eye is displayed in Fig. 5d showing a cleft restricted to the medial portion of the ventral retina (i.e. typical ‘partial’ coloboma of the retina). The corneal stroma was present (C, in Fig. 5d,f), but was fused with that of the iris (I, Fig. 5e,f), resulting in an absence of the anterior chamber; the corneal epithelium was locally hyperplastic and keratinized (EP, compare Fig. 5e and f). Additional ocular abnormalities in two 18.5 dpc exencephalic αγ mutants included agenesis of the conjunctiva and cornea, and abnormal lens fibers (data not shown). Primary aphakia (i.e. failure of lens formation) was observed bilaterally in one non-exencephalic 12.5 dpc αγ mutant (data not shown; see also Table 3).

Interestingly, the only ocular malformations found in two of five α1γα2^+/– mutants were unfused eyelids (not shown) and a corneal-lenticular stalk. This latter abnormality was characterized by a persistent continuity of the corneal and lens epithelia at the center of the cornea (large arrow, Fig. 5g), which most probably results from the failure of the lens cup to pinch off from its parental surface ectoderm (Coulombre and Coulombre, 1977). The eyes of α1γ mutants were unaffected (Fig. 6b; Table 3).

A partially chondrified fibrous retrolenticular membrane (F and arrow, Fig. 5h,i) was a specific feature of β2γ mutants. It was always continuous with a thick strand of mesenchyme, which was embedded in the optic nerve (ON, Fig. 5h) and was carried along with it into the eye through a coloboma of the optic nerve. The retina appeared normal, with the exception of a unilateral coloboma of the ventral retina in one out of three mutants (see ER, Fig. 5h). All β2γ eyes also exhibited a small conjunctival sac (J, compare Fig. 5a and h, and Fig. 5e and i), poorly differentiated corneal stroma and absence of the anterior chamber of the eye (Fig. 5h,i; see also Table 3).

A fibrous retrolenticular membrane (F, Fig. 5b) was the only abnormality present in the eyes of all αβ2 and of two (out of three) α1β2 18.5 dpc fetuses (Table 3). It is noteworthy that, with the exception of the fibrous retrolenticular membrane, the ocular defects were confined to animals disrupted for RARγ plus either α or β2 (Table 3).

The intraorbital, submandibular and sublingual glands and their ducts were normal in 18.5 dpc α1β2 and αβ2 mutants (Table 4). In contrast, in all 18.5 dpc α1γ (Fig. 6b) and β2γ (not illustrated) mutant fetuses, the epithelial rudiments of all intraorbital glands [which include the lachrymal and Harderian glands (HG, Fig. 6a)] were missing bilaterally (Table 4). However, the neural crest-derived, melanocyte-containing stroma of the Harderian glands was always present (data not shown; see Lohnes et al., 1993). The nasolachrymal duct, which was consistently missing in α1γ fetuses (compare NLD in Fig. 6c and d), was sometimes present in β2γ fetuses, indicating that its development was independent from that of the lachrymal glandular epithelium.

The parenchyma of the submandibular and sublingual salivary glands develop from downgrowth of the buccal ectoderm and from the adjacent mesenchyme, whereas the main ducts of these two glands are formed by closure, in a rostral direction, of gutter-like grooves in the oral ectoderm (Hamilton et al., 1945). In 18.5 dpc wild-type fetuses, the two ducts (MAD and LID, Fig. 6e) open in the mouth cavity at the sublingual caruncle, a median mucosal fold located at the level of the rostral third of the lower incisors (LI, Fig. 6e). Both ducts were shortened in all α1γ fetuses (Table 4): the submandibular duct (MAD, Fig. 6f) opened at the caudal end of the lower incisor (LI, Fig. 6f), whereas the sublingual duct opened even more caudally at the level of the 2nd lower molar (not shown). The sublingual caruncle was always absent. In contrast, only the sublingual duct was shortened in β2γ fetuses and the sublingual caruncle was always present (Table 4). Dysplasia of the sublingual gland, consisting of cystic epithelial formations within the parenchyma (compare LIG, Fig. 6g and h), was frequently observed bilaterally in α1γ fetuses (Table 4), whereas the submandibular glands appeared normal (compare MAG, Fig. 6g with h; Table 4).

As expected, glandular defects similar those described above in α1γ mutants were found in all α1γα2^+/– and αγ 18.5 dpc fetuses (Table 4), with agenesis of both sublingual and submandibular glands being occasionally observed (Table 4).

RARγ null mice exhibit various vertebral abnormalities which include homeotic transformations and affect primarily the cervical region (Lohnes et al., 1993). These abnormalities occurred with variable penetrance and bilateral expressivity, although the majority (89%) of RARγ null offspring exhibited one or more vertebral malformations. Although vertebral defects were not initially found in RARα null mutants (Lufkin et al. 1993), analysis of RARα null offspring inbred in a 129 SV background (third generation of inbreeding) revealed a low frequency of malformations affecting the second (C2) and third (C3) cervical vertebrae (Table 5).

With the exception of fusion of the basioccipital bone with the anterior arch of the atlas, which likely represents a posterior homeotic transformation (Lohnes et al., 1993), α1γ and α1γα2^+/– mutants showed increases in the frequency of all homeotic transformations previously observed in RARγ null fetuses (Table 5). Anterior transformation of C2 to a first cervical identity, evidenced by an increase in the thickness of the neural arches and the appearance of an ectopic anterior arch appeared two to four times more frequently in α1γ and α1γα2^+/– mutants, respectively, when compared to γ null fetuses (this transformation was, however, incomplete since the axis dens was unaffected). Although an ectopic anterior arch of atlas was occasionally observed in αγ skeletons (e.g. AAA* in Fig. 7i and l), the severe malformations of the cervical vertebrae in these mutants usually precluded evaluation of homeosis.

Anterior transformation of C6 or C7 to a C5 or C6 identity, respectively, was found twice as frequently in α1γ null mutants and approximately five times more frequently in α1γα2^+/– mutants than in RARγ null fetuses (Table 5 and data not shown). C7 to C6 transformation was evidenced by the appearance of tuberculi anterior (TA) on the ventral aspect and of foramina transversaria on the lateral processes of C7 (see Lohnes et al., 1993 and Fig. 7d). C6 to C5 transformation was inferred from loss of the C6-specific tuberculi anterior. Interestingly, in 50% of the affected α1γα2^+/– mutants, the C6 to C5 and C7 to C6 transformations were bilateral (data not shown), whereas these transformations were essentially unilateral in RARγ and α1γ null mutants. Additional anterior transformations of cervical vertebrae may exist in α1γ and α1γα2^+/– offspring that cannot be identified due to the lack of morphological landmarks (see Kessel and Gruss, 1991). Since the C2 to C1, C6 to C5 and C7 to C6 anterior transformations were found simultaneously in two α1γ and in six α1γα2^+/– mutants (data not shown), the entire cervical region may have undergone an anterior transformation in these fetuses.

RARα1γ, α1γα2^+/– and αγ (but not γ) mutants also exhibited a posterior homeotic transformation characterized by an extensive rib anlage on C7, which in some cases fused ventrally with the first thoracic rib (data not shown; Table 5). This transformation was usually unilateral and the ectopic C7 rib never contacted the sternum. Ectopic cervical ribs (CR) were found unilaterally on both C7 and C6 in two αγ mutants, with the C6 rib joining the cervical rib projecting from C7 (compare Fig. 7i with d; Table 5). These additional ribs did not alter the total number of presacral vertebrae, thus this likely represents homeotic transformations of C6 and C7 to thoracic vertebral identities. (‘T1’ and ‘T2’ in Fig. 7i).

Relative to RARγ null offspring, αβ2 mutant skeletons exhibited an increase only in the frequency of anterior transformations of C6 to C5, and C7 to C6, which were bilateral transformations in 50% of the cases (Table 5 and data not shown), whereas RARβ2γ mutants did not exhibit any increase in the frequency of the homeotic transformations described for RARγ null mutants (Table 5). No skeletal abnormalities were found in α1β2 mutant fetuses.

In RARα1γ, α1γα2^+/– and β2γ mutants, bifidus of C1 and/or fusion with C2 occurred eight to ten times more frequently than in RARγ null mutants (Table 5 and data not shown). A similar increase was also observed for fusion of the neural arches of C2 and C3 (compare Fig. 7d and e, white arrow; Table 5 and data not shown). RARα1γ, α1γα2^+/– and β2γ mutants also exhibited malformations not previously observed in RARγ null offspring, including dyssymphysis of the neural arch of C1 (asterisk in Fig. 7e compare to 7d) or C2 (not shown), and fusion of C3 with C4 (white arrow in Fig. 7e compare to 7d).

RARαγ mutant fetuses had severe defects of all cervical vertebrae, usually making detailed analysis impossible, save for cervical ribs on C6 or C7 (discussed above). Bifidus (not shown) and dyssymphysis of C1 (asterisks in Fig. 7f,i,l, compare to Fig. 7d,j) and agenesis or fusions of the neural arches of C2 to C5 (large arrow in Fig. 7f, compare to 7d; note the absence of the neural arch of C2 in Figs. 7f,i,l) were observed in all specimens. In two cases an ectopic ossified structure between C1 and C4 or C5 was observed dorsal to the cervical region (EC in Fig. 7k).

Most interestingly, with the exception of rib fusions, which occurred in α1γ, α1γα2^+/– and αγ mutants with a frequency similar to that in RARγ null mutants (arrow in Fig. 7l; Table 5), vertebrae caudal to the cervical region appeared unaffected in all double mutants analyzed (e.g. compare thoracic (T) and lumber (L) vertebrae in Fig. 7j and k; note the complete loss of the cranial vault of the specimen in Fig. 7k). The only additional malformations found in these animals involved occipital bones (see below) and the sternum. In some 18.5 dpc α1γ, α1γα2^+/– and αγ mutant fetuses, the sternum was distorted and, in one case, appeared incompletely closed (arrowhead, Fig. 7l) and possessed four instead of five sternebrae.

The cervical region of αβ2 mutants was also malformed, with a high frequency of dyssymphysis and bifidus of C1, and fusions between C1 and C2 or C2 and C3 (Fig.7a compare to b and c; arrow in Fig. 7b denotes C1 bifidus and arrowhead indicates C2-C3 fusion; asterisk in Fig. 7c denotes dyssymphysis of C1). Furthermore, the majority of αβ2 and 1 (of 6) αγ mutant fetuses exhibited an osseous fusion of the basioccipital (BO) and exoccipital (E) bones (asterisk in Fig. 7h, compare to g; Table 5), a malformation not observed in RARγ mutants.

The limbs of all double mutants were essentially normal with the exception of RARαγ mutants.

Although forelimbs were affected in all 18.5 dpc αγ mutants examined, the nature of the limb defects showed considerable variation. Syndactyly was frequently observed (compare Fig. 8e and f). The precise digits fused varied among animals and fusion was at the level of soft tissues only (see below). Digits frequently looked abnormal (compare Fig. 8c and d, e and f) and polydactyly (with 6 digits; compare Fig. 8a and b), or ectrodactyly (3 or 4 digits; not illustrated) was often apparent.

RARαγ mutant hindlimbs exhibited a consistent aberrant external aspect with the hindfoot twisted such that the footplate faced the lateral abdominal wall (compare Fig. 1a and c). However, no cases of syndactyly or other digit malformations were found in the hindlimbs upon external examination.

Whole-mount skeletal preparations from six 18.5 dpc αγ mutants were analyzed in detail. All six exhibited a number of malformations affecting diverse skeletal elements of the forelimb. However, the severity and frequency of most of these defects varied both among animals and between contralateral forelimbs from the same animal (Table 6).

The entire forelimb was shorter when compared to 18.5 dpc control littermates (Fig. 8i-l). This may be related to the general growth deficiency of these mutants (e.g. compare Fig. 1a and c). The scapula was always malformed in αγ mutants. In one fetus, the two scapulae were partially agenic with a greatly reduced shaft diameter and vertebral (medial) region (asterisk in Fig. 8l). In another fetus, the median portion of one scapula appeared bifurcated (Fig. 8k). In a third fetus, one scapula was partially agenic, whereas the central portion of the contralateral scapula was bifurcated (data not shown). The other scapula defects corresponded to mild aplasia of the superior or inferior margins of the vertebral region (e.g. Fig. 8i and j, asterisks).

The humerus, although smaller than controls, appeared otherwise normal (e.g. Fig. 8i-l). In contrast, the radius and ulna, which were also reduced in size, were abnormal in all αγ mutants. In four of six fetuses, the radius of either the left or right forelimb was missing unilaterally (e.g. Fig. 8l; Table 6). The ulna was always formed, but exhibited an abnormal curvature as did the radius when present (e.g. Fig. 8i-l).

With the exception of the pyramidal (PY) and pisiform (PI) bones [the nomenclature of Milaire (1978) is used for the carpal bones], which were well defined (e.g. compare Fig. 9e with f and g with h), all other carpal bones were malformed. The central bone (C) was missing in all but one case (compare Fig 9e with f, g with h; asterisks denote the missing central bone). The scapholunatum (SL), which was always present regardless of the absence or presence of the radius (e.g. see SL in Fig. 9m), was always misshapen and small (compare SL in Fig 9e and f, g and h), and in many cases was partially cleft (arrow in Fig. 9f). All distal carpals (D1-D4/5) were consis-tently hypoplastic (compare Fig. 9e and f; g and h). The prepollex (PX) was frequently rudimentary or absent (compare Fig. 9i and j).

In 6 (of 12) mutant forelimbs, only 4 digits were present, but this defect was bilateral in only one instance (Table 6 and see below). In five cases, the missing digit was presumably digit 1 (e.g. compare Fig. 9c and d), as determined by phalangeal count (digit one has two phalanges, whereas digits 2 to 5 have three) and the position of the remaining digits with respect to the distal carpal bones (using D4/5 as a landmark). The loss of digit 1 was correlated with the absence of the carpal D1 in four cases (Table 6 and data not shown). In one other specimen, the first digit was present, but markedly smaller (Table 6 and data not shown). In one fetus, which was missing the first digit on the right forelimb, the second digit was absent on the contralateral forelimb, with only a small cartilaginous anlage found in its place (arrow II in Fig. 9h). Interestingly, the first digit in this forelimb resembled a normal second digit rather than a thumb, as judged by its length and the advanced ossification of the metacarpal bone, yet only two phalanges were present (not shown). Unilateral ectrodactyly with three digits was observed in one animal, where digits 1 and 2 were missing and carpal D1 was agenic (Table 6 and data not shown).

Polydactyly with six digits was found in one mutant (compare Fig. 9a and b, e and f); the additional preaxial element was associated with an ectopic distal carpal bone, whereas the prepollex was lacking (data not shown). The diameter of the putative metacarpal bone was reduced and only one phalange was apparent. Interestingly, in this specimen the normal first digit may have undergone a transformation to a second digit, as supported by the increase in its length and diameter, advanced ossification of the metacarpal bone and larger size of the phalanges (compare digit I in Fig. 9a and b, e and f; note that the third phalange was broken during preparation of this specimen). Remarkably, the contralateral limb of this mutant possessed only four digits (Fig. 9d; Table 6).

Additional malformations observed in αγ mutant forelimbs included a delay in the ossification of the metacarpals and/or phalanges (e.g. compare Fig. 9c and d). Although this may result from the general growth deficiency of these animals, comparison to 18.5 dpc controls of only slightly larger size (e.g. compare Fig. 9a and b) indicates that this delay may be a direct effect of the mutations. In this respect, note a complete lack of metacarpal ossification centers in the mutant forelimb shown in Fig. 9m (although hypertrophic chondroblasts are evident), even though ossification of metacarpals 3 and 4 usually commences as early as 15.5 dpc (Dollé et al., 1993; Kaufman, 1992). The phalanges of αγ mutants were also abnormal, appearing bulbous with the outline of the synovial joints between the phalanges difficult to discern (compare Fig. 9a to b, c to d, k to l). This latter malformation occurred irrespective of other forelimb malformations.

Examination of the hindlimbs revealed (in five of six cases) a dramatic bilateral reduction in the length and increase in the diameter of the ossification of the fibula (F), with a concomitant bending of the tibia (T; compare Fig. 8g and h; note that a connection exists between the head of the tibia and the proximal end of the fibula in the specimen in Fig. 8h). The disproportionate growth between the fibula and tibia is the likely cause of the bending of the tibia and of the abnormal external aspect of the hindlimbs. In one αγ mutant fetus, the ossification of the fibula was intermediate between controls and severely affected mutants (data not shown). The only additional defects found in the hindlimbs of RARαγ skeletons were a slight retardation in ossification of the metatarsals and phalanges, and bulbous phalanges with indistinct synovial joints as described above for the forelimbs (compare Fig. 8h and g). No loss or gain of hindlimb digits or tarsal elements was observed.

The forelimbs of α1γα2^+/– fetuses were unaffected. However, of 11 RARα1γα2^+/– fetuses examined, 1 exhibited a defect of the fibula identical to that shown in Fig. 8h and two other specimens exhibited a similar, but milder malformation (data not shown).

The present results show that RARs are essential for normal development of many structures in the mouse. However, with the exception of malformations of the axial skeleton and agenesis of the Harderian glands (which have been previously described in RARγ null offspring; Lohnes et al., 1993), congenital defects are observed only in RAR double mutants. Offspring from dams fed a vitamin A-deficient (VAD) diet also exhibit a number of developmental abnormalities (Wilson et al., 1953). The recapitulation of most of these VAD-associated congenital malformations (see also Mendelsohn et al., 1994b) indicate that retinoic acids, the known ligands of the RARs, most probably represent the active retinoid developmental signal. With the exceptions of eye defects (Warkany and Schraffenberger, 1946) and cleft palate (Hale, 1933), the malformations described here in RAR double mutants have not been reported in VAD studies, probably because extreme deprivation of vitamin A results in embryonic lethality (Mason, 1935; Wilson and Barch, 1949). Complete inactivation (by full dietary deprivation of RA) of all RARs is likely to result in early embryonic death and resorption, as reflected here by the embryonic death of approximately 50% of the RARαγ double mutants. These observations indicate that some RA-dependent developmental processes are more sensitive than others to RA deficiency (e.g. eye development; see below) and that only abnormalities related to these events are observed in VAD studies.

RARβ isoform transcripts and amount of RARβ2 protein were not affected in 13.5 or 14.5 dpc αγ double mutant embryos, respectively, nor was the pattern of expression of RARβ transcripts apparently altered in 10.5, 11.5 or 13.5 dpc αγ mutant embryos as judged by in situ hybridization data (our unpublished results). Thus the extreme malformations specific to αγ double mutants are unlikely to reflect an additional defect in RARβ expression. Moreover, since RARβ2 is believed to be transcriptionally modulated by RA in vivo, its expression in RARαγ double null mutants may be controlled by the other RARβ isoforms (RARβ1, β3 and β4), through an autoregulatory mechanism (i.e. by RARβ2 itself), or by other nuclear receptors, such as the RXRs.

The eye is the most sensitive organ to retinol deprivation and in less severely affected VAD fetuses, it is often the only site of malformation (Warkany and Schraffenberger, 1946). The spectrum of VAD-induced ocular malformations is largely recapitulated in αγ (with the exception of a shorter ventral retina) and to a lesser extent in β2γ, double mutants (see Table 3).

Eye formation involves the coordinated development of forebrain neuroectoderm (which gives rise to the retina, optic nerve and epithelial portion of the iris), surface ectoderm (which gives rise to the lens and epithelia of the cornea and conjunctiva) and cranial neural crest-derived mesenchyme (which forms the choroid, sclera, stroma of the cornea and iris, anterior chamber, and vitreous body; Pei and Rhodin, 1970; Johnston et al., 1979; Le Douarin et al., 1993; and references therein). In WT mice at 9.5 dpc, the optic vesicle, which has evaginated from the forebrain neuroectoderm, comes into contact with the lens placode, which immediately invaginates and, at 11.5 dpc, pinches off from its parental surface ectoderm to become a hollow lens vesicle, leaving behind the presumptive corneal epithelium at the surface. At 10.5 dpc, the optic vesicle invaginates from its ventral side to form the optic cup (i.e. the anlage of the retina and of the epithelial portion of the iris; Kaufman, 1992; Pei and Rhodin, 1970). The area of invagination represents the optic fissure. Mesoderm extends from this fissure into the cup forming the primary vitreous body. The two lips of the optic fissure come into contact and fuse at 12.5 dpc, except in the regions of the iris and of the optic disc, where closure is delayed until 14.0 dpc. By 14.5 dpc, the mesenchymal cells of the vitreous body (i.e. the retrolenticular mesenchyme) disappear.

The ocular defects in RAR double mutants correspond mostly to structures arrested in ontogenesis and partly to abnormal embryonic formations. Absence of the lens, cornea, conjunctiva or anterior chamber, persistence of a corneal lenticular stalk or a fibrous retrolenticular membrane, coloboma of the retina or optic nerve, poor differentiation of the corneal stroma or lens fibres, hypoplasia of the conjunctival sac and lack of fusion of the eyelids are developmental arrests. However, the chondrification of the persistent retrolenticular tissue in RARβ2γ mice and the keratinization of the corneal epithelium have no equivalent at earlier developmental stages.

RARs are clearly required for morphogenetic processes at distinct stages of eye development. The complete absence of lens tissue observed in one 12.5 dpc RARαγ mutant embryo likely results from a fault in the process of lens induction occurring before 9.5 dpc (reviewed in Grainger et al., 1992). The presence of a corneal lenticular stalk implies an arrest at 11.5 dpc in the separation of the lens from the parental ectoderm. These processes depend on interactions between the lens placode and the optic vesicle. Interestingly, the embryonic retina is capable of synthesizing RA (McCaffery et al., 1992) and RA-reporter mice suggest that the eye contains RA as early as 9.5 dpc, with later synthesis (12.5 dpc) in the neural retina (Balkan et al., 1992; Rossant et al., 1991). These data suggest that the lack of a RA-dependent inductive signal from the optic vesicle may be the underlying basis for these lens defects.

The fusion of the two lips of the optic fissure, which starts with an inversion of the retinal pigmented epithelium, occurs first between undifferentiated cells at the junction of the neural retina and the retinal pigmented epithelium, and involves the disintegration of the basement membrane between the retina and the mesodermal tissue (Geeraets, 1976; Suzuki et al., 1988; Hero, 1989, 1990). Partial or complete persistence of the optic fissure (coloboma) in 18.5 dpc RARαγ and β2γ mutants might then be caused by: (i) overgrowth of the inner layer of the optic cup relative to its external counterpart, thus preventing the normal inversion of the latter along the line of the optic fissure, (ii) precocious differentiation of the retinal cells located at the junction between the two layers or, (iii) maintenance of the basement membrane preventing the fusion of the two lips of the optic fissure. Further studies at the electron microscopic level are required to distinguish between these possibilities. Although the RA-dependent events leading to coloboma are unknown, this defect appears to be determined shortly before the begining of the closure of the optic fissure (12.5 dpc), as administration of vitamin A to VAD rat embryos can reduce the incidence of coloboma if given before the equivalent of 11.5 dpc of mouse gestation (Wilson et al., 1953). The persistence of the retrolenticular mesenchyme appears to correspond to a later arrest, since its occurrence could be likewise prevented in VAD embryos by vitamin A administration before 13.5 dpc (Wilson et al., 1953). The additional malformations in the eyes of both VAD offspring and RAR double mutants (including isolated persistent retrolenticular mesenchyme and malformations of the eyelids, conjunctival sac, cornea and anterior chamber) may be due to defects in mesenchymal NCC, as all of these structures are derived from NCC originating from the forebrain and/or midbrain of the developing embryo (see below; Serbedzija et al., 1992; Le Douarin et al., 1993 and references therein).

The major ocular defects were found in RARαγ and RARβ2γ double null mutants, suggesting that RARγ is essential (in the absence of RARα1 and RARα2 or RARβ2) for normal eye development. Furthermore, in contrast to RARαγ mutants, RARα1γα2^+/– double mutants had near normal eyes, suggesting that (in the absence of RARγ) one copy of RARα2 suffices for most of the events needed for eye development. The eye defects observed in RAR mutants are in good agreement with the presence of RA in the eye (see above) and the expression pattern of the RARs during the period of eye development known to be sensitive to vitamin A deprivation (8.5 to 13.5 dpc in the mouse; Dollé et al., 1990; Ruberte et al., 1990, 1991). RARα is expressed ubiquitously, whereas RARγ and RARβ are expressed in the periocular mesenchyme throughout this period of development. Furthermore, RARβ2 (but not RARγ) transcripts are also detected at 12.5 to 13.5 dpc in the retrolenticular mesenchyme (our unpublished results), the abnormal persistence of which is the only ocular abnormality found in RARα1β2 and αβ2 double mutants. Thus, in these mutants, this abnormality may be a primary defect whereas, in RARαγ and β2γ mutants, it may be secondary to the coloboma.

The homeotic transformations and other axial malformations that occur in RAR double mutants are strikingly confined to cervical vertebrae. The present results indicate that RARs may be functionally redundant for specification of these vertebrae, as the penetrance and expressivity (bilateral versus unilateral defects) of cervical anterior transformations previously observed in RARγ null offspring (Lohnes et al., 1993) increased in a graded manner with subsequent loss of RARα1 and RARα2 isoforms from the RARγ^–/– background. Furthermore, RARβ2 (in the absence of RARα) also appears to play a role in axial specification, as RARαβ2 double mutants (but not RARα mutants) displayed a high frequency of anterior homeotic transformations, particularly of the sixth and seventh cervical vertebrae. That the posteriorization of the sixth and/or seventh cervical vertebrae was observed only in α1γ, α1γα2^+/– and αγ double null mutants suggests that RARα and RARγ are redundant with regard to events leading to these posteriorizations and that RARβ2 has no role in this particular event.

Gain-of-function (Kessel et al., 1990: Lufkin et al., 1992) and loss-of-function (Le Mouellic et al., 1992; Ramirez-Solis et al., 1993; Jeannotte et al., 1993; Condie and Capecchi, 1993) studies have shown that some Hox genes specify the identity of somites. Although there are notable exceptions (Pollock et al., 1992; Jegalian and De Robertis, 1992), Hox gain-of-function mutations usually lead to posteriorization, while Hox loss-of-function mutations lead to anteriorization of vertebral identities. RAR double mutants and some Hox null mutants exhibit similar cervical vertebral transformations, suggesting that RA may affect vertebral patterning by controlling Hox gene expression. The most notable similarities are with Hoxb-4 null mice, which exhibit anterior transformation of C2 to a C1 identitiy (Ramirez-Solis et al., 1993), and with Hoxa-5 null mice, which exhibit anterior transformation of C6 to a C5 identity and posterior transformation of the C7 to a T1 identity (Jeannotte et al., 1993). That Hox gene expression may be controlled by RA during development has been suggested by the observation that some Hox gene transcripts accumulate in cultured embryonal carcinoma (EC) cells exposed to RA (Simeone et al., 1990, 1991; Mavilio, 1993 and refs therein) and by the presence of functional RAREs in the promoter region of some of these RA-responsive genes (e.g. Hoxa-1, Langston and Gudas, 1992; and Hoxd-4, Pöpperl and Featherstone, 1993). That only cervical vertebrae were transformed in RAR double mutants may reflect the greater sensitivity (at least in EC cells) of 3′ Hox paralogues to RA (Mavilio, 1993). Some of these 3′ genes may be preferentially affected in RAR double mutants, leading to selective vertebral transformation in the cervical region. The severe malformations of cervical vertebrae observed in αγ double mutants may occur when the expression of several of these Hox genes are concommitantly altered.

In situ hybridization studies have shown that RARγ (and α) are expressed posterior to the caudal neuropore in the late gastrulating mouse embryo in all three germ layers prior to somite formation. RARγ expression then apparently disappears concomitant with the appearance of somites (Ruberte et al., 1990), suggesting that RARγ and α may control Hox gene expression during somite formation and specification. This also coincides with the embryonic stages when RA excess affects both Hox gene expression and vertebral identities (Kessel and Gruss, 1991). However, there is also evidence indicating that vertebral transformations can occur at later stages (10.5-11.5 dpc) through mechanisms not involving Hox genes (Kessel, 1992). At these later stages, RARα and γ transcripts are found in sclerotomes (Ruberte et al., 1990, 1991; Dollé et al., 1990), suggesting that RA could also be involved in the maintenance of vertebral identities through a Hox gene-independent mechanism. Thus, the vertebral malformations observed here could reflect a RA requirement at these later stages. RARβ transcripts were not detected in presomitic mesoderm, somites or sclerotomes in mouse embryos, while present in the neural tube (Ruberte et al., 1991; Dollé et al., 1990; and our unpublished results). This suggests that the defects seen in RARαβ2 and RARβ2γ mutants may reflect an indirect effect of RARβ in vertebral morphogenesis, involving RA-dependent signals emanating from the neural tube, since perturbations of this structure can result in vertebral malformations (Hall, 1977). However, isolated transformations of C6 and C7 and a restriction of malformations to the cervical region have not been reported in such studies.

We previously speculated that the lack of limb malformations in RARα or γ null mutants may be due to functional redundancy between these receptors (Lohnes et al., 1993; Lufkin et al., 1993), since each is uniformly expressed throughout the mesenchyme of the limb bud at 9.5-11.5 dpc (Dollé et al., 1989). This is clearly the case, as the limbs from RARαγ double mutants consistently exhibited malformations. Strikingly, forelimbs from RARα1γα2^+/– mutants were normal, showing that (as for the eye) a single copy of RARα2 suffices for limb morphogenesis in the absence of RARγ and RARα1.

The anteroposterior axis of the limb is patterned by the zone of polarizing activity (ZPA), whereas limb outgrowth requires a functional apical ectodermal ridge (AER; see Tabin, 1991 for review). The effect of the ZPA on the specification of the anteroposterior axis of the limb can be mimicked by topical application of RA, and both RA and the ZPA can trigger the expression of some Hox genes believed to be critical for limb specification (for reviews, see Duboule, 1992; Dollé and Duboule, 1993). A role for RA in maintaining AER activity is suggested by the finding that RA, in combination with FGF-4, can fulfill most AER functions (Niswander et al., 1993). However, the AER of RARαγ mutants appeared histologically normal, although its formation may be slightly delayed (our unpublished results). It is currently believed that RA may indirectly influence limb patterning by generating (or maintaining) a functional ZPA (Wanek et al., 1993), possibly through the regulation of a secreted protein, sonic hedgehog (Riddle et al., 1993). The malformations in RARαγ mutants do not appear to result from an early ZPA defect, since the limbs displayed a clear anteroposterior asymmetry. This does not exclude a role for RA in anteroposterior limb patterning, since RARβ transcripts (which appear unaffected in the limbs of αγ mutants) are expressed in the flanking mesenchyme and proximal regions of the early limb bud in a region that overlaps with the ZPA (Dollé et al., 1989; Mendelsohn et al., 1991; our unpublished results). In addition, the expression of some molecular markers of ZPA and AER activity, including Hoxd-9 and Hoxd-13 (Dollé and Duboule, 1993), MSX-1 (Robert et al., 1989; Hill et al., 1989), BMP-2 (Lyons et al., 1990), FGF-4 (Niswander et al., 1993) and sonic hedgehog appeared unaffected as judged from in situ hybridization studies from 10.5 and 11.5 dpc RARαγ mutants (P. Dollé and D. Décimo, unpublished data).

Although two RARαγ mutants exhibited putative digit transformations, firm conclusions cannot be drawn concerning the role of RA in specification of limb axes. It is clear, however, that RA is essential for the generation of some skeletal elements of the forelimb. In tetrapods, the limb skeletal elements are established through a conserved series of branching and segmentation events from prechondrogenic blastemas arising during limb outgrowth (reviewed in Shubin and Alberch, 1986; Shubin, 1991). Branching of the humeral blastema gives rise to those of the radius and ulna. The proximal and central carpals, as well as the digital arch, arise from segmentation and branching events initiating from the ulnar condensation, while distal carpal bones and subsequent digit formation generally proceed in a posterior-to-anterior direction in the mouse. All of the elements consistently affected in RARαγ mutants are derived from the preaxial (anterior) portion of the limb bud (i.e. radius, central and D1 carpals, digits 1 and 2 and the prepollex) and arise from the last branching event occurring in a given region of the limb. Loss of radius, first digit, central carpal bone and prepollex may be due to the absence of the final specific branching events giving rise to these structures (i.e. branching of the radial condensation from the humeral condensation, branching of the central carpal condensation from the pyramidal condensation and final branching of the digital arch to yield the prepollex or first digit). These defects may reflect a requirement of RA to generate the proper amount of limb mesenchyme, since a deficit in this mesenchyme leads to a preferential loss of anterior skeletal elements (Alberch and Gale, 1983). Such a deficit is also suggested by the generalized size reduction of the carpals in RARαγ double mutants (note also in this respect that the entire forelimb of RARαγ mutants was often shorter). Smaller mesenchymal condensations may also provide the extra space needed for generating supernumerary condensations, occasionally resulting in polydactyly.

The phalanges of both the forelimbs and hindlimbs of 18.5 dpc RARαγ double mutants were consistently malformed, appearing bulbous with poorly defined boundaries. This malformation does not appear to correspond to a developmental delay, since the phalanges of WT 15.5 to 17.5 dpc fetuses are always well defined (our unpublished observation; Kaufman, 1992). The perichondrial cells, surrounding the emerging blastemae, are believed to be important in directional growth of the blastemae and in determining their final shape (Shubin and Alberch, 1986; and refs therein). Since RARα and RARγ are both expressed in the perichondrial region of blastemal condensations (Dollé et al., 1989), the loss of these receptors could conceivably alter the functional integrity of the phalangeal perichondrium, thus leading to their altered morphology.

It is noteworthy that preaxial malformations of the limbs exhibited by RARαγ double mutants are restricted to the forelimbs (with the exception of bending of the tibia, which is likely secondary to the severe malformation of the fibula). These observations suggest that either RA plays different roles in forelimbs and hindlimb development, or that events related to differences in time of development of the two limbs allow phenotypic rescue of the preaxial derivatives of the hindlimb. Note in this respect that several mouse mutants, such as Po (postaxial; Nakamura et al., 1963) and Px (postaxial hemimelia; Searle, 1964) also exhibit defects confined largely to the forelimbs. However malformations affecting the anteroposterior axis of the limb usually affect homologous structures in both forelimbs and hindlimbs (e.g. Xt, Batchelor et al., 1966).

Malformations of most of the structures derived from cranial and cardiac mesenchymal neural crest cells (NCC) were observed in RAR double mutants (this and the accompanying study of Mendelsohn et al., 1994b). These structures originate from osteogenic NCC (i.e. membrane bones of the skull and face), from chondrogenic NCC [i.e. endochondral bones of the skull and face and second (hyoid) and third arch skeletal elements], from odontogenic NCC (i.e. upper incisors), from smooth muscle NCC precursors (i.e. the tunica media of the aortic arches and the aorticopulmonary septum), from dermal NCC precursors (contributing to the pinna, eyelids and prolabium) and from periocular NCC (i.e. corneal stroma and retrolenticular mesenchyme). Mesenchymal NCC also contributes to the stroma of the glands whose development and/or migration are altered in RAR double mutants (i.e. Harderian, lachrymal, thyroid, thymus and parathyroid glands) (for review and references see Noden, 1988; Le Douarin et al., 1993). Since NCC appeared with the emergence of vertebrates (Gans and Northcutt, 1983) and RARs have been found only in vertebrates (Kastner et al., 1994; Linney and LaMantia, 1994; and references therein), our observations raise the interesting possibility that these receptors may have evolved to fulfill functions necessary for the development of mesenchymal NCC-derived structures.

In αγ (and to a lesser extent in α1γα2^+/–) mutant mice, all the structures derived from mesenchymal NCC originating from the forebrain and the rostral midbrain (i.e. prolabium, frontal, nasal, premaxillary, ethmoid, presphenoid and sphenoid bones, and upper incisors; Le Douarin et al., 1993 and references therein) were either agenic or severely malformed. Ocular structures derived from fore- and rostral midbrain mesenchymal NCC were also affected in αγ and other RAR mutants (see above). All elements derived from more caudal mesenchymal NCC emanating from the level of rhombomeres 4 and 6 (R4 and R6) and from the unsegmented caudal portion of the rhombencephalon, and populating the second, third, fourth and sixth arches, (Lumsden et al., 1991; Noden, 1988; Le Douarin et al., 1993; Serbedzija et al., 1992; Sechrist et al., 1993), were also likely malformed or ectopi-cally localized in RARαγ and other double mutants. These structures included the hyoid and styloid bones and stapes (i.e. second and third pharyngeal arch derivatives), aortic arches, aorticopulmonary septum and thymus, thyroid and parathyroid glands (Mendelsohn et al., 1994b). In contrast, the first pharyngeal arch skeletal elements, which are derived from caudal midbrain and rostral hindbrain (i.e. R1 and R2) levels, were all identifiable in RARαγ double mutant fetuses, although some were abnormal. Most of the misshapen first pharyngeal archderived skeletal elements (e.g. maxillary and palatine bones) are derived from the maxillary (rostral) process of this arch, which is populated mainly by mesenchymal NCC from the caudal mesencephalon. Note that the alisphenoid and incus bones, which are also derived from the maxillary process, were fused, but otherwise not grossly affected (see below). In contrast, most of the unaffected skeletal elements (e.g. dentary bone, Meckel’s cartilage, malleus and tympanic bone) are derived from the mandibular (caudal) process of the first pharyngeal arch, which is essentially populated by mesenchymal NCC originating from R1 and R2 (Lumsden et al., 1991).

This lack of effect of RARαγ mutation on first pharyngeal arch mesectoderm is unlikely due to a compensation by RARβ, since RARβ transcripts are expressed at a much lower level in the first arch than in the frontonasal region, nor can it be explained by the absence of expression of RARα and γ, since their transcripts are equally abundant in the first arch and frontonasal region (Dollé et al., 1990; Ruberte et al., 1990, 1991). Interestingly, first arch NCC appear to be embodied with a ground state morphogenetic program, which is also present in the second pharyngeal arch mesectoderm where it is respecified by expression of Hoxa-2 (Rijli et al., 1993). Frontonasal mesectodermal cells are also embodied with a similar program, since they can generate first arch skeletal elements when grafted to the level of either the presumptive first or second arch (Noden, 1983). Thus, our data suggest that the realization of at least part of the ground state program present in the first arch does not require RA, whereas modification of this program both in the frontonasal and second arch mesectoderm involves RA-dependent processes. Since the lack of expression of Hoxa-2 results in homeotic transformation of second arch to first arch skeletal elements (Rijli et al.,1993), the second arch defects seen here cannot simply reflect a requirement of RA for Hoxa-2 expression.

Substantial evidence suggests that RA excess can affect premigratory NCC (Morriss-Kay, 1993 and references therein). However, our results suggest a RA requirement for events occurring during and/or after NCC migration, since RARγ has not been detected in the presumptive forebrain or midbrain, nor in presumptive NCC progenitors in the rhombencephalon (Ruberte et al., 1990). In contrast, all three RARs are highly expressed in the frontonasal process, whereas RARα and RARγ, and to a lesser extent RARβ2, are expressed in the pharyngeal arches after (or possibly during) NCC migration into these structures (Dollé et al., 1990; Ruberte et al., 1990, 1991; our unpublished results). Furthermore, VAD-induced aortic arch and aorticopulmonary septal defects can be prevented by vitamin A administration up to mouse 9.5 or 10.5 dpc, respectively, whereas all NCC contributing to these structures have migrated by 9.0 dpc (Wilson et al., 1953; Le Douarin et al., 1993; and references therein; see also Mendelsohn et al., 1994b). Although the molecular defects underlying the effect of RAR inactivation on mesenchymal NCC are unknown, two phenomena have been observed: (i) abnormal cell death in the mesectoderm of the frontonasal process, which precedes the aplasia of midfacial structures in RARαγ mutants, and (ii) abnormal specification of the fate of some NCC populations, as evidenced by chondrification of the meninges and the persistence and occasional chondrification of the retrolenticular mesenchyme. Additional cartilaginous ectopias found in the diaphragm, the peritoneum and the semilunar cusps of the heart (see Mendelsohn et al., 1994b), may also be of NCC origin, here reflecting either an abnormal specification of trunk NCC (which normally have no chondrogenic potential; Hall, 1991 and references therein) or abnormal migration or specification of cranial NCC.

It is remarkable that, with the exception of the lack of the spiral ganglion (acoustic portion of the VIIIth cranial nerve) in RARαγ mutants (which, however, may be secondary to abnormalities of the otocyst; see results section), we have not observed primary malformations of neurogenic NCC-derived structures. Thus, whether the effect of RA excess on many of these structures (reviewed in Armstrong et al., 1994) is a pharmacological phenomenon awaits examination of RARβ (all isoforms) null mice. In this respect, RA has been proposed to regulate directly transcription of the Hoxa-1 gene (Langston et al., 1992; Boylan et al., 1993). However, with the exception of the lack of structures derived from the otocyst and lack of the abducens nerve, RAR double mutants do not exhibit any of the defects found in Hoxa-1 null mice (Lufkin et al., 1991; Chisaka and Capecchi, 1991; Mark et al., 1993).

Finally, the phenotype of αγ mutants (see also Mendelsohn et al., 1994b) present some striking similarities with a human neurocristopathy called the CHARGE syndrome (Pagon et al., 1981; Siebert et al., 1985): Coloboma of the retina, Heart disease (consisting of aorticopulmonary septal defects and abnormalities of aortic arch-derived great arteries), Atresia of the choana (likely resulting from a defect in the formation of the ethmoid bone), Retardation of physical and mental development, Genital hypoplasia in males and Ear abnormalities (i.e. malformation of the pinna) and/or deafness. Additional defects in these patients include hypoplasia or agenesis of the thymus and parathyroid glands. Although it is unlikely that the CHARGE association is due to inactivation of both RARα and RARγ, it could result from the mutation of a RA-dependent gene that is mis-expressed in αγ mutant mice (potential candidates are discussed in the conclusion section of Mendelsohn et al., 1994b).

Two types of supernumerary skeletal structures were often detected in the skulls of RAR double mutant fetuses, the first forming a cartilaginous medial wall to the cavum epiptericum, the second linking the incus and alisphenoid. Comparative anatomical data strongly suggest that these two skeletal elements correspond to atavistic structures (Allin, 1975; Presley, 1989; De Beer, 1985).

In reptiles and monotremes (lower mammals), the orbital and the basal regions of the skull are connected by three pillars, the pilae prooptica, metoptica and antotica (cranial to caudal order) all of which arise by local chondrification of the primitive cranial wall (i.e. the dura mater). The caudalmost reptilian pillar (the pila antotica) forms a solid medial wall to the cavum epiptericum, thus physically separating it from the braincase. The ability to form a pila antoptica from the dura mater has been lost in placental mammals, thus permitting the expansion of the braincase. The resulting gap caused by this loss became filled in a more lateral plane by a first pharyngeal arch-derived skeletal element, the alisphenoid bone. In the RARγ^–/– genetic background, disruption of the RARα1 isoform resulted in the bilateral appearance of a small supernumerary chondrification center at the base of the medial wall of the cavum epiptericum. With subsequent loss of only one copy of the α2 isoform, a complete supernumerary pillar was formed which, based on its anatomical relationships, likely corresponds to the reptilian pila antotica. Thus RA appears to modify the ancestral reptilian program, suppressing the formation of this pillar in mammals.

In reptiles, the quadrate bone (the dorsal element of the jawjoint) and the epipterygoid bone (which forms the lateral limit of the cavum epiptericum) develop from a single element, the pterygoquadrate (upper jaw) cartilage. During the emergence of mammal-like reptiles (therapsids), the quadrate bone, while losing its connection with the epipterygoid bone, underwent size reduction, with its otic process becoming the short process of the incus middle ear bone [the long or stapedial process of the incus likely represents a mammalian neomorph (Presley, 1989)]. It is also widely held that the mammalian pterygoid and alisphenoid bones evolved from the reptilian epipterygoid bone which develops from the rostral portion of the pterygoquadrate cartilage. A number of RAR double mutant fetuses exhibited abnormal skeletal structures in which the body of the incus was continuous with a rostrally oriented cartilaginous or osseous rod, which was often fused to the alisphenoid (Table 2), while the short process of the incus was larger than in normal fetuses. It is likely that this is an atavistic structure corresponding to the therapsid evolutionary state in which the incus had appeared, but was still linked to the newly derived alisphenoid bone through quadrate remnants. Interestingly, the present day ground state morphogenetic program of the first arch appears to encode a similar atavistic structure (Rijli et al., 1993). It is tempting to conclude that an RA-dependent process has been recruited during the course of evolution to modify this program.

The re-emergence of ancestral skeletal features in RAR double mutants not only indicates that the underlying mesectodermal developmental programs are still present in mammals, but also that RA-dependent mechanisms have been recruited during the reptilian-mammalian transition to modify some features of the reptilian skull.

We would like to thank Dr T. Pexieder for a critical reading of the manuscript, Dr M. LeMeur for her collaboration and the members of the retinoid group for useful discussions; B. Weber, C. Fischer and V. Giroult and the technical staff of the animal facility for excellent help; B. Boulay, J. M. Lafontaine and C. Werlé for the illustrations and the secretarial staff for assembling the manuscript. D. L. was a recipient of a fellowship from the MRC Canada and C. M. was supported by a fellowship from the NIH (5F32 GM13597-03) and from the ARC. This work was supported by funds from the Institut National de la Santé et de la Recherche Médicale, the Centre National de la Recherche Scientifique, the Centre Hospitalier Universitaire Régional, the Association pour la Recherche sur le Cancer, the Human Frontier Science Program and the Fondation pour la Recherche Médicale.