ABSTRACT
The eye is the organ whose development is the most frequently altered in response to maternal vitamin A deficiency [VAD; Warkany, J. and Schraffenberger, S. (1946). Archs Ophthalmol. 35, 150-169]. With the exception of prenatal retinal dysplasia, all the ocular abnormalities of the fetal VAD syndrome are recapitulated in mouse mutants lacking either RARα and RARβ2, RARα and RARγ? RARγ and RARβ2, or RXRα [Lohnes, D., Mark, M., Mendelsohn, C., Dollé, P., Dierich, A., Gorry, P., Gansmuller, A. and Chambon, P. (1994) Development 120, 27232748; Mendelsohn, C., Lohnes, D. Décimo, D., Lufkin, T., LeMeur, M., Chambon, P. and Mark, M. (1994) Development 120, 2749-2771; Kastner, P., Grondona, J. Mark, M., Gansmuller, A., LeMeur, M., Décimo, D., Vonesch, J.L., Dollé, P. and Chambon, P. (1994) Cell 78, 987-1003], thus demonstrating that retinoic acid (RA) is the active vitamin A metabolite during prenatal eye morphogenesis. Whether retinoids are also involved in postnatal eye development could not be investigated, as VAD newborns are not viable and the above RAR double null mutants and RXRα null mutants died in utero or at birth.
We report here the generation of viable RARβ2/RARγ2 double null mutant mice, which exhibit several eye defects. The neural retina of newborn RARβ2γ2 mutants is thinner than normal due to a reduced rate of cell proliferation, and from day 4 shows multiple foci of disorganization of its layers. These RARβ2γ2 mutants represent the first genetically characterized model of retinal dysplasia and their phenotype demonstrates that RARs, and therefore RA, are required for retinal histogenesis. The RARβ2γ2 retinal pigment epithelium (RPE) cells display histological and/or ultrastructural alterations and/or fail to express cellular retinol binding protein I (CRBPI). Taken altogether, the early onset of the RPE histological defects and their striking colocalisation with areas of the neural retina displaying a faulty laminar organization, a reduced neuroblastic proliferation, and a lack of photoreceptor differentiation and/or increased apoptosis, make the RPE a likely target tissue of the RARβ2γ2 double null mutation. A degeneration of the adult neural retina, which may similarly be secondary to a defective RPE, is also observed in these mutants, thus demonstrating an essential role of RA in the survival of retinal cells. Moreover, all RARβ2γ2 mutants display defects in structures derived from the periocular mesenchyme including local agenesis of the choroid and of the sclera, small eyelids, and a persistence of the primary mesenchymal vitreous body. A majority of the RARβ2 single null mutants also exhibit this latter defect, thus demonstrating that the RARβ2 isoform plays a unique role in the formation of the definitive vitreous body.
INTRODUCTION
Vitamin A deficiency (VAD) studies have shown that vitamin A (retinol) is required during prenatal and postnatal development, and during adult life. After birth, retinol is indispensable for survival, growth, reproduction and vision and also for the maintenance of numerous tissues. Widespread squamous metaplasia of the conjunctival, corneal, respiratory, urinary and various glandular epithelia (i.e. olfactory, salivary, Harderian, seminal vesicle and prostate glands and/or their associated excretory ducts), together with degeneration of the seminiferous tubules and of the neural retina, are hallmarks of the postnatal VAD syndrome (Wolbach and Howe, 1925; Johnson, 1939). Interestingly, retinoic acid (RA) can prevent or reverse the deleterious effects of a postnatal VAD diet, with the exception of night blindness and photoreceptor degeneration (Dowling and Gibbons, 1961; Dowling, 1964; Howell et al., 1963; Thompson et al., 1964; Van Pelt and de Rooij, 1991). Furthermore, conceptuses of VAD dams exhibit a large number of congenital malformations (i.e., the fetal VAD syndrome) affecting the eye, the kidney and genitourinary tract, the heart and aortic arch-derived great arteries, the lung and the diaphragm. Retinol can prevent these malformations provided that it is supplied to the dams at specific times of gestation, thus demonstrating that it is required at several stages during ontogenesis (reviewed in Wilson et al., 1953).
Two families of nuclear receptors for retinoids have been characterized. Members of the RAR family (types α, β and γ? and their isoforms α1, α2, β1 to β4, and γ1 and γ2) are activated by most physiologically occurring retinoids (all-trans RA, 9-cis RA, 4-oxo RA and 3,4 dihydro RA). In contrast, members of the RXR family (types α, β and γ, and their isoforms) are activated by 9-cis RA only. In addition to the multiplicity of receptors, the complexity of retinoid signalling is further increased by the fact that, at least in vitro, RARs bind to their cognate response elements as heterodimers with RXRs. Moreover, RXRs can also bind in vitro to some DNA elements as homodimers and are heterodimeric partners for a number of nuclear receptors other than RARs (reviewed in Chambon, 1994; Giguère, 1994; Mangelsdorf and Evans, 1995).
Null mutations of the RAR genes (either α, β or γ), as well as isoform-specific knock-outs for RARα1, RARβ2/β4 and RARγ2 have been generated (reviewed in Kastner et al., 1995). RARβ single null mutant were apparently normal (Luo et al., 1995), whereas RARα, and RARγ single null mutants were viable and displayed abnormalities which, however, were confined to a small subset of the tissues that express these receptors (Lufkin et al., 1993; Lohnes et al., 1993). These findings suggested that there could be some functional redundancy in the RAR family. To test this hypothesis, double null mutants lacking either RARα1 and RARβ2, RARα and RARβ2, RARα1 and RARγ? RARα and RARγ? or RARβ2 and RARγ? were generated (Lohnes et al., 1994; Mendelsohn et al., 1994a). In contrast to RAR single mutants, these double mutants exhibited a dramatically reduced viability, as about half of the RARαγ mutants died in utero, and the remaining half as well as the other RAR double null mutants survived for 12 hours at most following delivery by Caesarean section at full term (Lohnes et al., 1994). Furthermore, almost all of the malformations of the fetal VAD syndrome were recapitulated in the different RAR double mutants, with the exception of a shortening of the ventral retina, which was, however, found in RXRα single null mutants (Kastner et al., 1994) and of a prenatal retinal dysplasia (Warkany and Schraffenberger, 1946). These findings demonstrated that RA is the vitamin A derivative that is active during ontogenesis, and that its effects are mediated by the RARs.
We have recently generated double null mutant mice lacking the RARβ2 and RARγ2 isoforms. These RARβ2−/−/RARγ2−/− mutants (hereafter referred to as RARβ2γ2 mutants) were normally viable and fertile. However, they displayed severe ocular defects. Their analysis demonstrates that RARs, and therefore RA, play a crucial role in histogenesis and maintenance of the neural retina, and that the retinal pigment epithelium (RPE) most probably represents the primary target tissue of the RARβ2γ2 compound mutation.
MATERIALS AND METHODS
Generation of RARβ2γ2 double null mutants
RARγ2 (Lohnes et al., 1993) and RARβ2 (Mendelsohn et al., 1994b) mutant mice were bred to generate double heterozygote mice, which were intercrossed to produce RARβ2−/−/RARγ2−/−, RARβ2−/−/RARγ2+/− and RARβ2+/−/RARγ2−/− mice, from which most of the mutants used in this study were generated.
Histology and immunohistochemistry
Mice were killed by cervical dislocation. The eyes were enucleated and fixed by immersion in Bouin’s fluid for 2 days, transferred to 70% ethanol overnight, and then bisected with a razor blade along a plane defined by the superior-inferior axis and the optic nerve. Lenses were removed and the eyes embedded in paraffin. Occasionally, skinned skulls from newborn, 4 days and 1-week-old mice were fixed in toto in Bouin’s fluid to prevent any damage that might have occurred during enucleation. 7 μm serial sections were mounted on slides coated with 0.01% poly-L-lysine (Mr 350,000; Sigma). The sections were stained with Groat’s hematoxylin and Mallory’s trichrome (Mark et al., 1993) or employed for immunohistochemistry. The following antibodies were used : a mouse monoclonal IgG directed against rod-specific opsin (4D2, in the form of a hybridoma supernatant) (a gift from R. Molday, University British Columbia, Vancouver and D. Hicks, University Louis Pasteur, Strasbourg; Hicks and Barnstable 1987); a rabbit polyclonal antibody directed against CRBPI (in the form of an antiserum; a gift of U. Eriksson, Ludwig Institute for Cancer Research, Stockholm; Gustafson et al., 1993); a polyclonal antibody against glial fibrillary acidic protein (GFAP; Sigma); and an anti-BrdU monoclonal antibody (Boehringer, Mannheim). These antibodies were used at dilutions of 1:40, 1:100, 1:400 and 1:100, respectively. Biotinylated anti-mouse IgG, biotinylated anti-rabbit IgG (Vectastain Elite ABC Kit, Vector) were used according to the manufacturer’s instructions. The reagents of the Vectastain ABC Elite system and diaminobenzidine (Sigma) were employed for immunoperoxidase labelling. Cy3conjugated streptavidin (Jackson ImmunoResearch) was used for immunofluorescence labelling. All primary antibodies, anti-IgG antibodies and components of the ABC system were diluted in PBS pH 7.3, containing 0.05% Tween 20 and 0.5% normal goat serum. Control sections were incubated with either mouse or rabbit preimmune serum instead of the primary antibody.
Fixation by perfusion was employed in one histological experiment (illustrated in Fig. 2) in order to visualize better the capillary network of the periocular mesenchyme. Under general anaesthesia, adult mice were perfused with 1% glutaraldehyde–4% paraformaldehyde in PBS (pH 7.2) at an outflow rate of 5 ml/minute for 5 minutes. The eyes were enucleated, then postfixed in Bouin’s fluid for 24 hours.
Electron microscopy
Eyes from 7-day-old and 1-month-old animals were fixed by immersion in 2% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.4) for 16 hours at 4°C, and the lenses removed. The eyes were then rinsed in cacodylate buffer, postfixed in 1% osmium tetroxide in the same buffer for 2 hours at 4°C, dehydrated with graded alcohols series and embedded in Epon. 1 μm sections were stained with toluidine blue. Ultrathin sections from selected areas were contrasted with uranyl acetate and lead citrate and examined with a Philips 208 electron microscope operating at 80 kV.
Labeling of S-phase nuclei
Bromodeoxyuridine (BrdU) (Sigma) dissolved in PBS was injected intraperitoneally at a dose of 50 mg per kg of body weight. The mice were killed 2 hours later, the eyes fixed in Bouin’s fluid for 2 days, then embedded in paraffin. BrdU incorporation was detected by using an anti-BrdU monoclonal antibody (Boehringer) and immunoperoxidase labelling.
End-labeling of DNA nicks in tissue sections (TUNEL staining)
In situ detection of fragmented DNA was performed as described by Gavrieli et al. (1992) with some modifications. 7 μm sections from Bouin-fixed, paraffin-embedded eyes were collected on poly-L-lysine coated slides. The sections were dewaxed and then hydrated. After rinsing in distilled water (3× 5 minutes) the sections were digested for 15 minutes at 24°C with 20 μg/ml proteinase K in 50 mM Tris-HCl pH 7.5 containing 50 μM EDTA, rinsed in distilled water (3×5 minutes), and then incubated for 1 hour at 37°C with biotinylated dUTP in terminal transferase buffer (all from Boehringer). The reaction was terminated by transferring the slides to distilled water. The sections were permeabilized with 0.05% Tween 20 in PBS (3×5 minutes) and biotin incorporation was revealed with Cy3-conjugated streptavidin (Jackson ImmunoResearch) diluted 1:400 (30 minutes at 24°C). Negative controls were obtained by omitting terminal transferase in the incubation mixture.
Detection of RARs by in situ hybridization
The RARβ and γ probes used for in situ hybridization were synthesised from cDNA fragments covering the entire open reading frame (Dollé et al., 1990). In situ hybridization was performed on 10 μm frozen sections from albino mice, as described (Décimo et al., 1995).
RESULTS
RARβ2γ2 double null mutant mice are viable
In contrast to the RAR compound mutants analysed previously (Lohnes et al., 1994; Mendelsohn et al., 1994a), which all died within 12 hours after birth, RARβ2γ2 homozygotes were normally viable and fertile. From external inspection, however, their eyes appeared abnormal (e.g. compare Fig. 1a and b).
Comparison of the palpebral fissure, eyeball and retina between WT and RARβ2γ2 (β2γ2) mutant adults, at one year (a,b,e-m) and 1 month (c,d). h, i and j correspond to histological sections of e, f and g, respectively; note that the lenses were removed prior to paraffin embedding. k, l and m correspond to high magnification of regions boxed in h, i and j respectively. The open arrow in j points to a focus of degeneration. The arrowheads in l and m indicate the absence of retinal pigmented epithelium (RPE). The detachment of the neural retina from the RPE (asterisk in l) is artifactual. Abbreviations: CH, choroid; F, retinal fold; GCL, ganglion cell layer; HG, Harderian gland; I, iris; IC, inner canthus; INL, inner nuclear layer; IPL, inner plexiform layer; L, lens; OC, outer canthus; ONL, outer nuclear layer; OPL, outer plexiform layer; PS, photoreceptor segments; R, retina; RL, retrolenticular membrane; Ro, rosettes; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Mallory’s trichrome-hematoxylin (h-m). Magnifications: ×13 (c and d); ×9 (e-j); ×155 (k-m).
Comparison of the palpebral fissure, eyeball and retina between WT and RARβ2γ2 (β2γ2) mutant adults, at one year (a,b,e-m) and 1 month (c,d). h, i and j correspond to histological sections of e, f and g, respectively; note that the lenses were removed prior to paraffin embedding. k, l and m correspond to high magnification of regions boxed in h, i and j respectively. The open arrow in j points to a focus of degeneration. The arrowheads in l and m indicate the absence of retinal pigmented epithelium (RPE). The detachment of the neural retina from the RPE (asterisk in l) is artifactual. Abbreviations: CH, choroid; F, retinal fold; GCL, ganglion cell layer; HG, Harderian gland; I, iris; IC, inner canthus; INL, inner nuclear layer; IPL, inner plexiform layer; L, lens; OC, outer canthus; ONL, outer nuclear layer; OPL, outer plexiform layer; PS, photoreceptor segments; R, retina; RL, retrolenticular membrane; Ro, rosettes; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Mallory’s trichrome-hematoxylin (h-m). Magnifications: ×13 (c and d); ×9 (e-j); ×155 (k-m).
Scleral and choroidal defects in RARβ2γ2 mutant eyes. Histological sections from perfusion-fixed, 6 months old, WT (a) and RARβ2γ2 eyes (left and right eyes from the same mutant, as indicated in b and c). Note the paucity (c) or the absence (b) of the choroidal melanocytes (CH) and the thinning of the sclera (SC). Also note that b and c represent typical aspect of the left and the right neural retinas in this mutant. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; ONL, outer nuclear layer; OPL, outer plexiform layer; RPE, retinal pigment epithelium. The arrowheads point to periocular capillaries. Magnification: ×380 (a-c).
Scleral and choroidal defects in RARβ2γ2 mutant eyes. Histological sections from perfusion-fixed, 6 months old, WT (a) and RARβ2γ2 eyes (left and right eyes from the same mutant, as indicated in b and c). Note the paucity (c) or the absence (b) of the choroidal melanocytes (CH) and the thinning of the sclera (SC). Also note that b and c represent typical aspect of the left and the right neural retinas in this mutant. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; ONL, outer nuclear layer; OPL, outer plexiform layer; RPE, retinal pigment epithelium. The arrowheads point to periocular capillaries. Magnification: ×380 (a-c).
Retinal degeneration and dysplasia in adult RARβ2γ2 mutants
In the neural retina of the normal (wild-type, WT) adult mouse (R, Fig. 1c,e,h), there are 8 major cell types located in characteristic positions in the outer nuclear layer (ONL: rod photoreceptors), the inner nuclear layer (INL: horizontal, bipolar, amacrine and Müller cells) and the ganglion cell layer (GCL: ganglion cells and astrocytes). These layers of cell bodies are separated by two layers of synaptic interplay, namely the outer plexiform layer (OPL) between the ONL and INL and the inner plexiform layer (IPL) between the INL and GCL (Fawcett,1986; Huxlin et al., 1992; Sarthy et al., 1991 and refs therein; Fig. 1k).
The neural retina of adult RARβ2γ2 mutants (i.e. 1-month old or older) exhibited two types of abnormal phenotypes, namely a marked atrophy (degeneration) and a disorganisation (dysplasia) of the retinal layers
Retinal degeneration
Macroscopic examination of 1-month-old eyes revealed a marked reduction in the thickness of some portions of the neural retina (compare R, Fig. 1c and d). In older mutants (6 to 12 months), this reduction in thickness was observed over larger patches (compare R, Fig. 1e and f). In the most severely affected eyes (Fig. 1f and i), only 2 to 3 rows of nuclei persisted in the central retina, whereas in the peripheral retina, close to the iris, all nuclear and the synaptic retinal layers remained identifiable although the ONL, INL, OPL and IPL were conspicuously thinner and the GCL contained fewer cells, as compared to their wild-type counterparts (compare Fig. 1h and k with i and l). The patches of neural retinal atrophy always involved the ONL (Table 1, results not shown), whereas the INL and GCL were affected only in place where the ONL had almost disappeared (e.g. Figs 1l and 2c).
In WT mice, the glial fibrillary acidic protein (GFAP) gene is specifically expressed in the astrocytes present in the GCL (Sarthy et al., 1991; Huxlin et al., 1992). In contrast, accumulation of GFAP in Müller cells represents an unspecific response to eye injuries and retinal degeneration or detachment (references in Sarthy et al., 1991). Immunostaining with antiGFAP of the mutant retina showed, in the thin areas, numerous positive cellular processes, consistent with a degenerative process (data not shown).
Even though the penetrance of the retinal degeneration was complete (i.e. every eye displayed the defect; Table 1), its extent was highly variable among mutants from the same age, particularly among young adults (i.e. 1-month-old); note in this respect that Fig. 1d shows an example of one of the most affected eyes at 1-month. Moreover, in a given mutant, the two eyes could be affected to very different degrees. For instance, Fig. 2 shows the two eyes from the same mutant: the left neural retina is essentially normal (e.g. compare PS, ONL, OPL and INL in Fig. 2a and b), exhibiting only rare foci of degeneration, whereas the right retina (Fig. 2c) is essentially degenerated.
Retinal dysplasia
The second abnormal phenotype displayed by about one third of the adult mutant eyes (Table 1) corresponds to a disorganisation of the layers of the neural retina manifested by the presence of retinal folds and rosettes (F and Ro in Fig. 1g,j,m). Note that each of these dysplastic neural retinas also exhibited occasional foci of degeneration (open arrow in Fig. 1j, and data not shown) and large areas with a normal laminar pattern (e.g., R in Fig. 1j).
It is noteworthy that these retinal defects were never observed in the RARβ2 and in RARγ2 single null mutants.
Developmental abnormalities of RARβ2γ2 mutant neural retinas
In order to gain insights into the pathogenesis of the above retinal abnormalities in adult RARβ2γ2 mutants, we studied the processes of lamination, rod functional differentiation, cell proliferation and apoptosis in the neural retina.
Abnormal retinal lamination
During normal mouse retinal development, the IPL begins to form by embryonic day 17.5 (E17.5) and separates the future GCL from the outer neuroblastic layer (Pei and Rhodin, 1970). Likewise, by postnatal day 4 (P4), the OPL appears in the central retina, thus dividing the outer neuroblastic layer into a ONL and a INL (Young, 1984). By P7, the retinal layers (i.e. ONL, INL and GCL) are distinctly present in the entire neural retina (Young, 1984).
In ∼70% of the RARβ2γ2 newborn mutant eyes (Table 1), the entire outer neuroblastic layer appeared thinner than its WT counterpart (compare NBL, in Figs 3a, 6a and 6c with 3b, 6b and 6d; the overall thickness of these mutant retinas was 25% lower than normal). Additionally, in a few mutant eyes (3 out of 22; Table 1), focal areas of RPE agenesis were observed in the central retina (arrowheads in Figs 3b, 6b and d, and see below). However, even in these areas devoid of RPE, the lamination of the mutant neural retinas was normal (e.g. NBL, IPL and GCL, in Figs 3b and 6d).
Aspects of retinal histogenesis in WT and RARβ2γ2 mutants at birth (a,b), P4 (c-e) and P14 (f,g). c-e correspond to three successive sections through the same portion of a mutant retina. The arrowheads point to areas of missing RPE and the open arrows to scattered RPE cells within the periocular mesenchyme. The arrows in g indicate an area where the ONL is atrophic. Asterisk indicate artifactual detachment of the RPE and neural retina during tissue processing. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; NBL, outer neuroblastic layer; OPL, outer plexiform layer; ONL, outer nuclear layer; PS, photoreceptor segments; Ro, Ro1 and Ro2, rosettes; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Mallory’s trichrome-hematoxylin. Magnifications: ×155 (a,b,f,g) and ×195 (c-e).
Aspects of retinal histogenesis in WT and RARβ2γ2 mutants at birth (a,b), P4 (c-e) and P14 (f,g). c-e correspond to three successive sections through the same portion of a mutant retina. The arrowheads point to areas of missing RPE and the open arrows to scattered RPE cells within the periocular mesenchyme. The arrows in g indicate an area where the ONL is atrophic. Asterisk indicate artifactual detachment of the RPE and neural retina during tissue processing. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; NBL, outer neuroblastic layer; OPL, outer plexiform layer; ONL, outer nuclear layer; PS, photoreceptor segments; Ro, Ro1 and Ro2, rosettes; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Mallory’s trichrome-hematoxylin. Magnifications: ×155 (a,b,f,g) and ×195 (c-e).
In P4, P7 and P14 mutant eyes, the generalized thinning of the neural retina, already noticed at birth, was conspicuous. It affected the photoreceptor segments (PS), the ONL and the INL. In contrast, at the same developmental stages, the mutant GCL was indistinguishable from its WT counterpart (Table 1 and compare PS, ONL and INL in Fig. 3f and g). Additionally, multiple foci of dysplasia were observed in 80% of the mutant neural retina that displayed rosettes (Ro1 and Ro2 in Fig. 3c-e; Ro in Figs 3g, 4b, d), folds (F, Fig. 4c) and/or local retinal detachments (RD, Fig. 4e). The rosettes comprised tubular arrangements of photoreceptors displaying well defined segments (e.g. Ro, Fig. 3g). Interestingly, the earliest-formed (i.e. P4) rosettes were often connected with the periocular mesenchyme by bridges of cells ‘escaping’ from the ONL (Fig. 3ce). Folds are probably generated at sites of retinal detachment (RD, Fig. 4e) and involve all the layers of the neural retina (F, Fig. 4c). Extreme thinning of the ONL (unlabelled arrows in Fig. 3g) and of the INL (not shown) were frequently observed in these dysplastic areas at P14. At P4, P7 and P14 the dysplastic areas of neural retina were always contiguous with patches of abnormal or absent RPE, and always commenced abruptly at the point where the morphologically normal RPE terminated (open arrows in Figs 3c,d,g and 4b-e; arrowheads in Figs 3g, 4b,d, 6f; see below for a description of RPE defects). Thus, on the basis of histological criteria, the RARβ2γ2 mutant neural retinas display two distinct abnormal developmental phenotypes: (1) a congenital thinning of the neural retina, which is ubiquitous and thus does not correlate with the patchy distribution of the RPE defects, and (2) multiple foci of retinal dysplasia, which are spatially correlated with defects in the RPE.
Expression of the rod-specific opsin evaluated by immunohistochemistry in P4 WT (a) and RARβ2γ2 (b-e) mutants. The open arrows point to scattered RPE cells in the periocular mesenchyme, the arrowheads to portions lacking RPE, and the small arrows to areas of the ONL lacking expression of opsin. Note that these latter areas are always juxtaposed to an abnormal or absent RPE. The detachments between the RPE and neural retinal such as the one illustrated in e likely occur in vivo; they have regular outlines which are maintained on more than 20 serial sections; their size is small and the tissues bordering them are not damaged in any fashion. These features distinguish them from artifactual retinal detachments (asterisk in b and c) generated during tissue processing. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer; Ro, rosettes; RD, retinal detachment; RPE, retinal pigment epithelium; V, vitreous body. Immunoperoxidase with hematoxylin counterstain. Magnifications: ×77 (a-c), ×155 (d) and ×90 (e).
Expression of the rod-specific opsin evaluated by immunohistochemistry in P4 WT (a) and RARβ2γ2 (b-e) mutants. The open arrows point to scattered RPE cells in the periocular mesenchyme, the arrowheads to portions lacking RPE, and the small arrows to areas of the ONL lacking expression of opsin. Note that these latter areas are always juxtaposed to an abnormal or absent RPE. The detachments between the RPE and neural retinal such as the one illustrated in e likely occur in vivo; they have regular outlines which are maintained on more than 20 serial sections; their size is small and the tissues bordering them are not damaged in any fashion. These features distinguish them from artifactual retinal detachments (asterisk in b and c) generated during tissue processing. Abbreviations: CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer; Ro, rosettes; RD, retinal detachment; RPE, retinal pigment epithelium; V, vitreous body. Immunoperoxidase with hematoxylin counterstain. Magnifications: ×77 (a-c), ×155 (d) and ×90 (e).
Abnormal differentiation of mutant photoreceptors
Functional differentiation of the rods, the major photoreceptor type in the mouse retina, occurs during the first days after birth and can be followed by the expression of the rod-specific opsin in the presumptive ONL (Hicks and Barnstable, 1987). Rod differentiation is achieved at P14 which corresponds to the completion of segment formation (Obata and Usubura, 1992; Theiler, 1972).
The 4D2 monoclonal antibody, which specifically recognises opsin in rod photoreceptors (Hicks and Barnstable, 1987), was employed as a marker of the differentiated state of the ONL in newborn and P4 retinas (Fig. 4). In both WT mice and RARβ2γ2 mutants, the first opsin-immunoreactive cells were detected in the central retina at the time of birth (not shown). In P4 WT eyes, almost all cells of the ONL expressed opsin (Fig. 4a), and the P4 RARβ2γ2 mutant ONL was indistinguishable from its WT counterpart in areas facing a normal RPE (Fig. 4b,c,e). In contrast, the ONL cells adjacent to areas lacking an histologically normal RPE did not express opsin (unlabelled small arrows in Fig. 4b-e), with the notable exception of some rosettes (Ro in Fig. 4d), nor did they form segments (PS in Fig. 5, compare a and c with b and d). These results strengthen the conclusion that the presence of RPE is essential for proper photoreceptor differentiation (Stiemke et al., 1994, and references therein), and further suggest that the abnormalities in the RARβ2γ2 RPE could be, at least in part, responsible for the defects observed in the neural retina (see below for further analysis of the mutant RPE).
Semithin sections and electron microscopy of P7 RPE from WT and RARβ2γ2 mutants. Note the fibroblastic aspect of the mutant RPE cells in b and d. Abbreviations: C, capillaries; CH, choroid; ONL, outer nuclear layer; N, RPE cell nuclei; PS, photoreceptor segments; RPE, retinal pigment epithelium; SC, sclera. Magnifications: ×380 (a,b); ×2250 (c,d).
Semithin sections and electron microscopy of P7 RPE from WT and RARβ2γ2 mutants. Note the fibroblastic aspect of the mutant RPE cells in b and d. Abbreviations: C, capillaries; CH, choroid; ONL, outer nuclear layer; N, RPE cell nuclei; PS, photoreceptor segments; RPE, retinal pigment epithelium; SC, sclera. Magnifications: ×380 (a,b); ×2250 (c,d).
Reduced cell proliferation in RARβ2γ2 mutant neural retinas
Cell proliferation in the retina was examined by bromodeoxyuridine (BrdU) incorporation (Fig. 6). The fraction of cells incorporating BrdU during a short exposure reflects the fraction of cells in S phase during this period, therefore permitting an assessment of the proliferation rate of a population of cells.
Comparison of cell proliferation evidenced by BrdU incorporation in newborn (a-d) and P4 (e,f) neural retinas from WT (a,c,e) and RARβ2γ2 (b,d,f) mutants. The arrowheads indicate the absence of the RPE. Note that, in the neural retinas of newborn mutants, the cell proliferation is unaffected in areas of RPE agenesis; however, in P4 neural retinas the areas lacking RPE show a lower proportion of labelled cells. Abbreviations: GCL, ganglion cell layer; L, lens; NBL, outer neuroblastic layer; R, neural retina; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Immunoperoxidase with hematoxylin counterstain. Magnifications: ×40 (a,b); ×77 (c,d); ×155 (e,f).
Comparison of cell proliferation evidenced by BrdU incorporation in newborn (a-d) and P4 (e,f) neural retinas from WT (a,c,e) and RARβ2γ2 (b,d,f) mutants. The arrowheads indicate the absence of the RPE. Note that, in the neural retinas of newborn mutants, the cell proliferation is unaffected in areas of RPE agenesis; however, in P4 neural retinas the areas lacking RPE show a lower proportion of labelled cells. Abbreviations: GCL, ganglion cell layer; L, lens; NBL, outer neuroblastic layer; R, neural retina; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Immunoperoxidase with hematoxylin counterstain. Magnifications: ×40 (a,b); ×77 (c,d); ×155 (e,f).
In RARβ2γ2 newborns, the labelling index in the outer neuroblastic layer was markedly reduced (between 18 and 28% of the total cell number, depending of the animal) when compared with WT (40% of the total cell number) (compare Fig. 6a and c with 6b and d). This reduced labelling index affected both the central and peripheral retinas (Fig. 6b), and was apparently identical whether the RPE was missing (arrowheads in Fig. 6b,d) or present. At P4, the rate of cell proliferation had decreased in both WT and mutant neural retinas, which displayed similar labelling indices in places where the RPE was present (data not shown). In contrast, in portions of the mutant neural retina facing large areas devoid of RPE, the percentage of proliferating cells was markedly reduced (12% in peripheral retina) compared to WT (26% in peripheral retina) (compare Fig. 6e and f).
Thus, a reduced cell proliferation rate during the perinatal period of retinal development may account for the generalized thinning of the RARβ2γ2 neural retina. However, this decrease in cell proliferation cannot explain the loss of the retinal cells in adult mutants, since cell proliferation already ceases by P6 and by P11 in the central and peripheral portions of the WT retina, respectively (Young, 1985a,b).
Increased apoptosis in RARβ2γ2 mutant neural retinas
In the normal mouse retina, programmed cell death occurs primarily during the first 2 weeks after birth and is essentially completed by the end of the third week (Young, 1984). The in situ end-labelling method of DNA nicks (TUNEL method; Gavrieli et al., 1992) was employed to compare cell death between normal and mutant mouse retinas (Fig. 7).
Comparison of apoptosis evidenced by the TUNEL method in P4 (a,b) and P14 (d,e) retinas from WT (a,d) and RARβ2γ2 mutant mice (b,e). c and f are brightfield views of b and e, respectively. The open arrows in b and c indicate the absence of apoptosis in the pigmented cells scattered within the periocular mesenchyme. The small arrows and the arrowheads in e point to apoptotic bodies in the RPE and in the ONL respectively. Abbreviations: GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer; Ro, rosettes; V, vitreous body. Magnifications: ×77 (a-f).
Comparison of apoptosis evidenced by the TUNEL method in P4 (a,b) and P14 (d,e) retinas from WT (a,d) and RARβ2γ2 mutant mice (b,e). c and f are brightfield views of b and e, respectively. The open arrows in b and c indicate the absence of apoptosis in the pigmented cells scattered within the periocular mesenchyme. The small arrows and the arrowheads in e point to apoptotic bodies in the RPE and in the ONL respectively. Abbreviations: GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer; Ro, rosettes; V, vitreous body. Magnifications: ×77 (a-f).
At P4 and P14, the majority of apoptotic cells of WT retinas were located in the INL and the GCL in agreement with Young’s data (Young, 1984); apoptotic cells were scarce in WT ONL (Fig. 7a,d). In dysplastic areas of mutant retinas at the same ages (i.e. P4 and P14), the number of apoptotic cells in the INL was markedly increased (compare Fig. 7a and d with b and e), and dying cells were also often observed in the ONL (arrowheads in Fig. 7e). This suggests that, in addition to the reduced rate of cell proliferation mentioned above, an increase of apoptosis also contributes to the extreme thinning of the INL and/or ONL which is frequently observed in the dysplastic areas of P14 mutants (e.g. unlabelled arrows in Fig. 3g).
In WT animals older than 1 month, retinal cell death was no longer detectable (Young, 1984, and data not shown). In contrast, retinas from adult mutant mice displayed an average of 4–6 apoptotic cells per section that were preferentially located in areas of thin or folded retina (data not shown). Thus, apoptosis occurring after the period of physiological retinal cell death is likely to account for the progression of degeneration in the retina of adult mutants.
Retinal pigment epithelium defects in RARβ2γ2 mutants
As mentioned above, foci of dysplastic neural retina were always located adjacent to an absent or morphologically abnormal RPE. Moreover, the RPE defects preceded the neural retina dysplasia in mutant eyes. Since the RPE is known to play an important trophic influence in the development and maintenance of the neural retina (Campochiaro, 1993; Bok, 1993), these observations suggested to us that a defective RPE could be instrumental in the generation of the RARβ2γ2 retinal defects. Therefore, the morphology of the mutant RPE was analysed by electron microscopy, and its functional state was investigated by examining the distribution of cellular retinol binding protein I (CRBPI) which is normally expressed uniformly in RPE cells (Fig. 9a) and is believed to play a crucial role in the delivery of retinol to the neural retina (reviewed in Saari, 1994). From an analysis during the period of retinal histogenesis (i.e. at P4 and P7) and in adult retina (i.e. at 1 month), mutant RPEs could be classified within three different categories: disorganised RPE, abnormal RPE without loss of epithelial organisation and apparently normal RPE.
Disorganised RPE
During the period of retinal histogenesis, scattered pigmented cells were often detected within the periocular mesenchyme adjacent to the areas of dysplastic neural retina (open arrows in Figs 3c,d,g, 4b-e, 7c). Semithin sections and electron microscopy of P7 retina (Fig. 5b,d) showed flattened RPE cells that had lost their epithelial arrangement and, in some cases, their contact with the neural retina (compare Fig. 5a and c with b and d). Note that these flattened pigmented cells represent altered RPE cells, not choroidal melanocytes, since they were usually lying within large portions of the periocular mesenchyme that were totally devoid of choroid (see Figs 4a-e and 5b).
Abnormal ‘epithelial’ RPE
Electron microscopic analysis of mutant RPE at P7 (results not shown) and 1 month (Fig. 8b,d) revealed abnormalities in areas where its epithelial organization was preserved: the cytoplasm of these RPE cells was highly vacuolated (V in Fig. 8d) and their apical microvilli (Mi, Fig. 8c), which normally interdigitate with the photoreceptor outer segments, were absent. The photoreceptor outer segments facing these altered RPE cells were reduced in number and/or disorganized, showing improper piling of their disks (compare POS in Fig. 8a and c with b and d). Immunostaining on P7 (not shown) and 1-monthold mutant RPEs, revealed weak (with respect to wild type) or absent CRBPI expression in several areas where this tissue appeared histologically normal (compare Fig. 9a with b and c).
Electron microscopy of 1 month-old retinas from WT (a,c) and RARβ2γ2 (b,d) mutant mice. d corresponds to a higher magnification of the area boxed in b. The mutant RPE cells lack microvilli (Mi) and their cytoplasm is highly vacuolated. Abbreviations: BM, Bruch’s basement membrane; M, mitochondria; Mi, microvilli; N, nuclei of RPE cells; PH, phagosomes; POS, photoreceptor outer segments; V, vacuoles. Magnifications: ×4500 (a,b); ×7200 (c,d).
Electron microscopy of 1 month-old retinas from WT (a,c) and RARβ2γ2 (b,d) mutant mice. d corresponds to a higher magnification of the area boxed in b. The mutant RPE cells lack microvilli (Mi) and their cytoplasm is highly vacuolated. Abbreviations: BM, Bruch’s basement membrane; M, mitochondria; Mi, microvilli; N, nuclei of RPE cells; PH, phagosomes; POS, photoreceptor outer segments; V, vacuoles. Magnifications: ×4500 (a,b); ×7200 (c,d).
Immunodetection of CRBPI in 1 month old retinas from WT and RARβ2γ2 mutant. Note the near absence of CRBPI expression in the lower right portion of the mutant’s RPE in panel b. Abbreviations: R, neural retina; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Magnifications: ×40 (a) and ×45 (b).
Immunodetection of CRBPI in 1 month old retinas from WT and RARβ2γ2 mutant. Note the near absence of CRBPI expression in the lower right portion of the mutant’s RPE in panel b. Abbreviations: R, neural retina; RPE, retinal pigment epithelium; SC, sclera; V, vitreous body. Magnifications: ×40 (a) and ×45 (b).
Apparently normal RPE
In some areas of the mutant retina, the RPE was indistinguishable from its WT counterpart, based on ultrastructural characters or levels of CRBPI expression.
It is noteworthy that, during the period of histogenesis, the neural retina in contact with unaffected RPE always appeared normal. However, in the neural retina of adult mutants, degenerating, dysplastic and normal areas could be observed adjacent to both normal or abnormal RPE (based on the same ultrastructural and immunohistochemical criteria; data not shown). In any event, these data suggest that (i) large portions of the mutant RPE are morphologically and/or functionally abnormal, (ii) during retinal differentiation, but not in the adult retina, there is a tight spatial correspondence between the morphological defects in the RPE and the abnormalities of the neural retina (i.e. rosettes, folds, foci of degeneration, reduced cell proliferation, absence of opsin expression by photoreceptors and/or lack of the photoreceptor segment formation).
Apoptotic bodies were occasionally observed in RPE of both differentiating (small arrows in Fig. 7e) and mature mutant retinas (results not shown), but never in the RPE-derived cells scattered in the periocular mesenchyme (open arrows in Fig. 7b and c). These observations suggest that the loss of mutant RPE cells occurs either by migration away from the neural retina or by cell death.
Distribution of RARs in ocular tissues
The distribution of the transcripts of the 3 RARs was investigated by in situ hybridization at several prenatal and postnatal stages of eye ontogenesis as well as in the adult eye (Table 2 and Fig. 10). Both RARβ and RARγ were strongly expressed in the periocular mesenchyme from E15.5 to P7 (e.g. POM, Fig. 10b-d). Both RARγ and RARβ transcripts were also detected in the RPE at the same developmental stages (Fig. 10d,e). Previous studies in E9.5 to E14.5 mouse embryos and fetuses (Dollé et al., 1990; Ruberte et al., 1990) showed expression of RARγ in periocular mesenchyme and expression of RARβ in the periocular mesenchyme, the primary vitreous body and the RPE. Altogether, these observations indicate that the periocular mesenchyme and the RPE express RARγ and/or RARβ throughout the period of eye development. RARβ and RARγ transcripts were occasionally present in the INL and GCL, whereas RARα transcripts were detected in all nuclear layers; RARα appears to be the main RAR expressed in the prenatal and the postnatal neural retina (Table 2).
Detection RARβ and RARγ transcripts in a E17.5 WT eye by in situ hybridization. a is a bright field of b. The large arrows in d and e point to RPE cell nuclei. Note that the silver grains located over the neural retina (R) correspond to background labelling. Abbreviations: C, cornea; GCL, ganglion cell layer; EY, eyelids; HG, Harderian gland; I, iris; L, lens; POM, periocular mesenchyme; R, neural retina; RPE, retinal pigment epithelium; V, vitreous body. Magnifications: ×20 (a-c); ×310 (d,e).
Detection RARβ and RARγ transcripts in a E17.5 WT eye by in situ hybridization. a is a bright field of b. The large arrows in d and e point to RPE cell nuclei. Note that the silver grains located over the neural retina (R) correspond to background labelling. Abbreviations: C, cornea; GCL, ganglion cell layer; EY, eyelids; HG, Harderian gland; I, iris; L, lens; POM, periocular mesenchyme; R, neural retina; RPE, retinal pigment epithelium; V, vitreous body. Magnifications: ×20 (a-c); ×310 (d,e).
Thus, even though our probes did not specifically identify RARβ2 or RARγ2 transcripts, the in situ hybridization data strongly suggest that the periocular mesenchyme and/or the RPE are the target tissues of the double mutation.
Non-retinal abnormalities in RARβ2γ2 mutants
RARβ2γ2 adult mutant eyes exhibited several abnormalities in addition to retinal dysplasia and degeneration (Table 1): (i) a reduction of the palpebral aperture (i.e. blepharophimosis) was observed in all RARβ2γ2 mutants (Table 1). Its degree, evaluated by measuring the intercanthal distance (e.g. compare the distance between IC and OC in Fig. 1a and b), ranged from small decreases to near absence of the palpebral fissure. (ii) a posterior persistent hyperplastic primary vitreous (retrolenticular membrane: RL in Fig. 1d,f,g) was observed in all RARβ2γ2 mutants. However, this defect was not specific to these double mutants, since it was also seen in ∼70% (11 out of 16) of the RARβ2 single null mutant eyes examined, in ∼90% (16 out of 18 eyes) of RARβ2−/−/RARγ2+/− compound mutants (data not shown) and in ∼90% (62 out of 72 eyes) of the RARβ single null mutant eyes (N. Ghyselinck, M. M. and P. C., unpublished results). (iii) colobomas of the sclera and/or of the choroid (i.e., large portions of the eyeball completely lacking the sclera and/or the choroid) were present in all RARβ2γ2 eyes; these defects were often readily visible in the form of distended and translucent regions at external inspection of enucleated eyes (Fig. 1d,f). Where present, the sclera (SC) and the choroid (CH) usually showed a marked thinning (compare Fig. 2a with b and c). It is noteworthy that the choroid is made up of two main cell types, which have distinct embryological origins, namely neural crestderived melanocytes and capillary endothelial cells originating from the head mesoderm (Johnston et al., 1979). At the histological level, fewer capillaries were observed adjacent to the retina in areas completely lacking melanocytes (arrowheads in Fig. 2b, compare with 2a). (4) Unilateral or bilateral absence of the Harderian gland (i.e. the main periocular gland in the mouse) was observed in approximately one fourth of the mutant eyes. Similar agenesis of the Harderian gland occurs occasionally in RARγ single null mutants and is fully penetrant in RARα1γ and RARβ2γ double null mutants (Lohnes et al., 1993, 1994). (5) Lens cataracts were observed in about one third of the RARβ2γ2 eyes (Table 1), and were characterized macroscopically by the presence of large vacuoles between the lens fibers and/or by the presence of a posterior lenticonus (which consists of a bowing of the posterior pole of the lens capsule; data not shown).
The retrolenticular membrane, the local agenesis of the sclera and choroid and the agenesis of the Harderian glands were also observed in newborn RARβ2γ2 mutants, and thus correspond to congenital defects (Table 1). The local absence of the RARβ2γ2 choroid was conspicuous as early as P4, when ocular melanocytes start to differentiate. In contrast, the lens abnormalities were only detected in adult mutants and their penetrance appeared to increase with aging (Table 1). Since, in the mouse, the palpebral fissure starts to form only at P12-P14 (Theiler, 1972), the blepharophimosis could only be diagnosed in adult RARβ2γ2 mutants; note, however, that in all E14.5 RARβ2γ2 mutants the palpebral fissure was conspicuously smaller than in wild-type fetuses of the same age and weight (Fig. 11a,b, and N. Ghyselinck, M. M. and P. C., unpublished results) implying that eyelid hypoplasia, which is the underlying cause of the blepharophimosis, is determined early, in any event before eyelid closure at E15-E16 (Theiler, 1972; Juriloff and Harris, 1993 and references therein). It is also noteworthy that the defects observed in the periocular connective tissues were not spatially correlated with the neural retina defects in newborn, young or adult mutants.
DISCUSSION
The phenotypic analysis of the RARβ2γ2 mutant mice provides the first evidence of the indispensability of retinoic acid (RA), both for the postnatal stages of neural retina histogenesis and the survival of differentiated neural retina cells, in vivo. At least some of the effects of RA on the neural retina which are revealed by the RARβ2γ2 compound mutation are likely to be mediated by the RPE, as discussed below.
Pathogenesis of the retinal dysplasia in RARβ2γ2 null mutants
The RARβ2γ2 null mutant mice analysed in this report represent the first genetically characterized model of retinal dysplasia. Retinal dysplasia usually refers to a disorganization of the laminar pattern of the developing neural retina and is defined histologically by foldings of the neural retina or by the presence of rosettes composed of neurons and glial cells (Silverstein et al., 1971; Lahav and Albert, 1973). Retinal dysplasia in humans (Lahav and Albert, 1973; Godel et al., 1981; Potter and Traboulsi, 1993) and animals (Randall et al., 1983; Fite et al., 1982; Cook et al., 1991; Whiteley, 1991; Caffé et al., 1993; Toole, 1983) can be caused by a variety of genetic and environmental factors and are frequently associated with multiple eye defects.
The RARβ2γ2 dysplastic retina is associated with a persistent primary vitreous body, an absent or abnormal RPE, a partial agenesis of the choroid, and sclera and lens degeneration, raising the question as to which ocular tissue(s) is (are) primarily affected by the compound mutation. The earliest ocular defect observed in RARβ2γ2 mutants at E14.5 (data not shown) is the persistence of the primary mesenchymal vitreous body, resulting in the presence of a retrolenticular membrane in all postnatal mutants. A persistent hyperplastic vitreous body is often associated with retinal dysplasia in human patients (Lahav and Albert, 1973; Godel et al., 1981; Potter and Traboulsi 1993; and references therein). However, the two defects appear to be unrelated in the present mouse model, since ∼70% of the RARβ2 single null mutants display a retrolenticular membrane which, in this case, coexists with a normal retina.
Both embryonic and adult neural retinas contain high levels of RA as well as the enzymatic machinery required for its synthesis (McCaffery et al., 1993, and references therein). RA was shown to promote the differentiation and survival of isolated embryonic retina photoreceptors (Kelley et al., 1994; Stenkamp et al., 1993). However, the neural retina is unlikely to represent a primary target tissue of the compound mutation since RARβ and RARγ do not appear to be expressed in the outer neuroblastic layer (which expresses RARα) during the period of retinal lamination. Moreover, the differentiation of the ganglion cells, in which RARβ transcripts are detected from E17.5 onwards, seems to occur normally in RARβ2γ2 mutants as, in newborn mutants, these cells have withdrawn from the mitotic cycle and possess well-developed axons which have reached the diencephalon, and both the thickness of the ganglion cell layer and calibre of the optic nerve appear normal at this stage.
In contrast, the wild-type RPE expresses both RARβ and RARγ transcripts before and during the period of retinal lamination. In RARβ2γ2 mutants, focal histological abnormalities of the RPE precede the retinal dysplasia and their spatial distribution strikingly corresponds to that of the dysplastic areas. A spatial correlation between RPE structural defects and retinal dysplastic areas has also been reported in heritable forms of retinal dysplasia in mice (Cook et al., 1991) and chicks (Randall et al., 1983; Fite et al., 1983), as well as in a variety of human retinal dysplasia (reviewed in Silverstein et al., 1971). Contact between the RPE and the neural retina is required for morphological and functional photoreceptor differentiation in cultures of Rana pipiens ocular rudiments (Hollyfield and Witkowsky, 1974) and coculture of RPE cells is necessary for laminar organisation within the neural retina of chick (Vollmer and Layers, 1986). Fetal RPE cells secrete a protein that induces a neuronal phenotype in cultured retinoblastoma cells, suggesting that RPE-derived paracrine factors play a role in the differentiation of the neural retina (Steele et al., 1990). The importance of the RPE in the maintenance of the structural integrity of the retina, in vivo, has been elegantly demonstrated by Raymond and Jackson (1995) in a transgenic mouse line expressing the attenuated diptheria toxin A gene under the control of the tyrosinase-related protein1 promoter which is specifically active in RPE cells and melanocytes. Perinatal toxicogenic ablation of the RPE results in a retinal dysplasia strikingly resembling that of the RARβ2γ2 mutants, notably with respect to the possible origin of rosette formation: the rosettes observed in both Raymond and Jackson’s RPEdeficient mice and in our mutants appear to arise from the migration of ONL cells between the neural retina and the locally disrupted RPE. Since, in the adjacent areas, the RPE is attached to the neural retina (note that the detachment indicated by asterisks in Fig. 3c-e are typical artifacts due to tissue processing), mechanical constraints are generated, which cause an inward bending of the ONL. This bending, in turn, might provoke an inversion of the polarity of the photoreceptors located in the hinge region. Interestingly, this pathological cell rearrangement permits the photoreceptor to express differentiated features (i.e. segment formation and expression of opsin) in the absence of contact with the RPE (Figs 1m, 4d). Taken together, these data strongly suggest that a defective RPE might be responsible for the retinal dysplasia in the RARβ2γ2 mutants, either through a local disruption of the blood-retinal barrier (whose major component is the RPE) subjecting the neural retina to systemic disrupting influences, or through the lack of a RPE-derived positive signal, normally inducing and maintaining the neural retina.
The periocular mesenchyme is another candidate as a primary target tissue of the double null mutation, as it is the ocular tissue that expresses the highest levels of RARβ and RARγ: it is conceivable that, under the influence of RA, some periocular mesenchymal cells might synthesize paracrine factors required for the laminar organisation of the retina. Such factors might act either indirectly, e.g. on differentiation, proliferation or survival of RPE cells, or directly after having crossed the blood-ocular barrier. Recent transplantation experiments demonstrating the absence of specificity of the peri-ocular mesenchyme in supporting the differentiation of the RPE cells (Buse et al., 1993) argue against the first of these two possibilities.
Interestingly, Johnson (1939, 1943) also reported the occurrence of rosettes in the retinas of VAD rats. However, contrary to the rosettes seen in our RARβ2γ2 mutants, these VADinduced rosettes were determined after the period of retinal lamination and were interpreted as being the result of the reorganisation of the remaining photoreceptors in the most severely degenerated areas (Johnson, 1939).
Pathogenesis of the retinal degeneration in RARβ2γ2 null mutants
Retinal degeneration is characterized histologically by an atrophy of one or more layers of the neural retina. Several lines of evidence suggest that retinal degeneration might be a direct consequence of retinal dysplasia. Firstly, massive apoptosis is observed within the rosettes and folds at P4, suggesting that these sites will evolve toward degeneration. Accordingly, the frequency of the mutant retinas displaying signs of dysplasia apparently decreases as degeneration proceeds (Table 1). Secondly, the first histological signs of retinal degeneration that are manifested by focal atrophy of the ONL, are observed at the dysplastic sites at P14. Thirdly, several similar cases of retinal atrophy accompanying or preceded by retinal dysplasia have been described in mice and chicks (Randall et al., 1983; Cook et al., 1991; Caffé et al., 1992). Therefore retinal degeneration in RARβ2γ2 mutants might be initiated at sites of retinal dysplasia, and then spread toward the retinal periphery either because the cells of the dysplastic areas release diffusible ‘death’ factors or because they fail to produce a trophic factor normally required for the survival of neural retinal cells (see Huang et al., 1993 for further discussion).
Alternatively, or additionally, degeneration might occur de novo, at least in some portions of the mutant retina. In contrast to the ‘dysplastic phenotype’, the ‘degenerative phenotype’ is fully penetrant, suggesting that degeneration may not be solely a consequence of dysplasia. Moreover, areas with a normal laminar pattern, but displaying both a marked atrophy of all 3 nuclear layers (ONL, INL and GCL) and apoptotic cells in the ONL and INL, were observed in adult mutant retinas. Such areas of retinal degeneration were not always contiguous to areas of absent or histologically abnormal RPE. However, it is noteworthy that many more RPE cells were found to be altered on the basis of ultrastructural and immunocytochemical criteria than on histological criteria alone. The adult RPE has a critical role in the homeostasis of the neural retina through the regeneration of the 11-cis retinaldehyde (the chromophore of the visual pigment) in the visual cycle, the selective transport of retinoid and nutrients to the photoreceptors (forming part of the blood-retinal barrier), the phagocytosis of the distal portion of the rod outer segments, the production of interphotoreceptor matrix material and of trophic factors for the neural retina (Bok, 1993; Campochiaro, 1993; Saari, 1994). Culture media conditioned by normal RPE cells promote photoreceptor survival, indicating that cytokines and other growth factors produced by RPE cells may exert a trophic influence on the maintenance of the neural retina (reviewed in Campochiaro et al., 1993 and Sheedlo et al., 1993). A defective RPE appears to be the direct cause of neural retina degeneration in the Royal College of Surgeons strain of rats (RCS rats; Bok and Hall, 1971; Malecaze et al., 1993; and references therein) and in human choroideremia (reviewed in Bird and Jay, 1994). Recent immunohistochemical studies indicate that RPE cells synthesize bFGF, which is an important photoreceptor survival factor as it can rescue photoreceptor degeneration in RCS rats and in light-damaged rats (Steinberg, 1994, and references therein). Moreover, the RPE appears to be a target tissue of RA action since (i) RA can prevent dedifferentiation and loss of density-dependent growth control of human RPE cells in culture (Campochiaro et al., 1991), and (ii) the VAD-induced RPE defects in adult rats (i.e. flattening and degeneration of RPE cells; Johnson, 1939, 1943; Dowling and Wald, 1958) can apparently be prevented by supplementing their diet with RA (Dowling, 1964; Dowling and Gibbons, 1961; Carter-Dawson et al., 1979). Taken together, all of these data raise the possibility that a defective RPE could also be instrumental in the genesis of the RARβ2γ2 retinal degeneration through events that are not secondary to the retinal dysplasia.
The choroid is thought to represent the main source of blood supply for the ONL (reviewed in Bernstein, 1961) and large portions of this tissue are lacking in RARβ2γ2 mutants. However, we did not find any correlation between the regions displaying choroidal agenesis and the presence of lesions in the neural retina, e.g. RARβ2γ2 mutants often show portions lacking choroid juxtaposed with areas of normal retina (see Fig. 2b). Moreover, mutant mice deficient in melanocyte precursors, such as Dominant white spotting (w) and Steel (Sl), do not develop retinal defects (Jackson, 1994). These data suggest that the choroidal defects are not the cause of retinal degeneration.
In any event, the phenotype of the RARβ2γ2 mutants strongly suggests that RA is most likely required for the survival of the rod photoreceptors, the bipolar neurons, (which represent the major INL cell type), and the ganglion cells. Experiments aimed at rescuing the RARβ2γ2 mutant phenotype through specific reexpression of RARβ or RARγ in the RPE should demonstrate whether these trophic effects of RA on the neural retina are mediated by the RPE, as proposed above.
Pleiotropic role of RARs in retinal maintenance and eye development
Retinoids have trophic effects on the neural retina as first demonstrated by Johnson (1939, 1943) and later by Dowling and colleagues (Dowling and Wald, 1958; Dowling and Gibbons, 1961; Dowling, 1964) in studies of degenerative changes in retina of rats deprived of vitamin A. According to Johnson’s data (1939, 1943), the degeneration of the neural retina induced by avitaminosis A is progressive. It begins with the loss of the photoreceptor outer segment, then involves successively ONL and the INL, and is always more pronounced in the central than in the peripheral region of the retina. The identity of the retinoids exerting these trophic effects is unclear (discussed in Stenkamp et al., 1993). Systemic administration of RA to VAD rats apparently prevents the death of the cells of the INL, but not the night blindness, the deterioration of the photoreceptor outer segment and death of the photoreceptor cells, whereas administration of retinol can prevent the appearance of all of these defects, (Dowling and Gibbons 1961; Dowling, 1964). However, as it is the case for the blood-testis barrier (Van Pelt and de Rooij, 1991), RA is probably not transported across the blood-retinal barrier (Bridges et al., 1983) and RA synthesized within the neural retina (or the RPE) through metabolic conversion of retinol could be involved in the trophic effect of vitamin A.
In this context, it is noteworthy that there are at least two points of convergence between the neural retina degeneration in VAD animals and RARβ2γ2 mutants. The first is purely morphological: the neural retina degeneration in RARβ2γ2 mutants progresses from the center towards the periphery of the retina and from the ONL towards the internal retinal layers, thus resembling the progression observed in VAD animals (see above). The second deals with the physiopathology of the lesions: in RARβ2γ2 mutants, CRBPI expression by RPE cells is impaired and, since this protein likely plays an important role in the delivery of retinoids to the photoreceptors (Saari et al., 1994), this condition could create a state of VAD in the neural retina. Interestingly, CRBPI mRNA levels are increased by RA treatment of whole animals (Haq and Chytil, 1988) and the CRBPI gene contains a RA response element in its promoter (Smith et al., 1991) which further supports the proposal that the RPE could be a primary target of the double null mutation (see above).
Warkany and Schraffenberger reported half a century ago (1946) that the developing rat eye is the organ that is most sensitive to vitamin A deprivation since, in less severely affected VAD fetuses, it represented the only site of malformations. We show here that more than two thirds of the RARβ2 single null mutants display a retrolenticular membrane that actually corresponds to the commonest abnormality of the fetal VAD syndrome. It arises by persistence and hyperplasia of the primary vitreous body, a structure that starts to develop at E10.5 from periocular mesectodermal cells that enter the optic cup, and has regressed at E14.0 by mechanisms that are still unknown. In adult RARβ2 or RARβ2γ2 mutants, the persistence of the primary vitreous is manifested by the presence of a plaque of pigmented fibrovascular tissue connecting the posterior pole of the lens with the optic papilla (the optic nerve exit point and point of entry of retinal blood vessels). This abnormality, which must result in poor vision, was previously overlooked in RARβ2 mutants (Mendelsohn et al., 1994b) due to its lack of behavioural manifestation in the laboratory (note, in this respect, that blindness does not overtly affect the behaviour of the laboratory mouse; Grüneberg, 1952). RARβ2+/−/RARα−/− mice (Lohnes et al., 1994) and RARβ2+/−/RARγ2−/− mice (our present data) never displayed a retrolenticular membrane. Thus, one functional copy of the RARβ2 gene is sufficient to ensure the involution of the primary mesenchymal vitreous. The penetrance of the persistent retrolenticular membrane phenotype increased in a graded manner upon removal of one, and then of both alleles of the RARγ2 gene from the RARβ2 null genetic background and, in RARβ2γ2 double null mutants, it was fully penetrant. These results suggest that RARγ2 can functionally compensate for the lack of RARβ2 in some RARβ2γ2 mutants.
Aside from the neural retina, the RPE and the vitreous body, the sclera, the choroid, the eyelids, and the lens were affected in RARβ2γ2 mutants. Eyelids start to develop at E13.5 as mesenchymal outgrowths of the neural crest-derived periocular mesenchyme covered by the ectoderm, whereas the sclera arises from the compaction of the peripheral layers of the periocular mesenchyme at E16.5 This compaction event also individualizes the choroid, a loose, highly vascularized mesenchymal tissue located between the sclera and the RPE. The blepharophimosis and the local agenesis or thinning of the choroid and sclera in RARβ2γ2 mutants might reflect a direct effect of the double mutation in the periocular mesenchyme which normally expresses high levels of both RARβ and RARγ in embryos, fetuses and young mice. However, the cataracts could be secondary to vascular invasion of the lens by blood vessels coming from the retrolenticular membrane (discussed in Traboulsi, 1993).
Our previous analysis of RARαβ2, RARαγ, RARβ2γ and RXRα mutants (Lohnes et al., 1994; Kastner et al., 1994), together with the classical VAD studies of Warkany and Schraffenberger (1946), have implicated retinoid signaling at almost every step of prenatal eye morphogenesis. These include lens formation, separation of the lens from the ectoderm, development of the outer layer of the optic cup as RPE (demonstrated by the development of neural retina in the place of RPE on the dorsal aspect of RARαγ mutant eyes; Lohnes et al., 1994; P. Gorry, M. M. and P. C., unpublished data), development of the ventral retina, closure of the optic fissure, involution of the primary vitreous body, development of the eye’s anterior segment (cornea, conjunctival sac, anterior chamber), development of periocular structures (sclera, choroid and Harderian gland) and formation and fusion of the eyelids). In addition, RA may be implicated in the formation of the optic cup, as the development of the eye anlage is arrested at the optic cup stage in cultured mouse embryos deprived of RA by inhibition of yolk-sac retinol binding protein (RBP) synthesis (Båvik et al., 1996). Thus, our present demonstration that RARs are also required for retinal histogenesis and survival of retinal cells, further establishes the pleiotropic role of RA in eye development. Interestingly, in the retina of VAD rat fetuses, the formation of the inner neuroblastic layer (the future GCL) and of the IPL apparently fails to occur (Warkany and Schraffenberger, 1946). This absence of retinal laminar organisation is the only eye defect of the fetal VAD syndrome which was not recapitulated in the RAR and RXR single and double null mutants studied so far (Lohnes et al., 1994; Kastner et al., 1994). It remains therefore to be seen whether RA is also involved in fetal retinal histogenesis.
The problems of penetrance and expressivity of the retinal defects
The retinas (i.e. neural retina and RPE) of young RARβ2γ2 mutants usually displayed only focal lesions (dysplasia and abnormal RPE) coexisting with large, apparently intact areas. Thus, removal of RARβ2 and RARγ2 does not completely abolish retinoid responsiveness in target cells, but rather seems to bring this responsiveness close to a threshold level below which the realization of RA-dependent cellular events is impaired. Stochastic variation of this residual retinoid responsiveness among the target cells may account for the observation that defects are confined to limited portions of the neural retina and of the RPE, and also for the incomplete penetrance and expressivity of the retinal dysplastic phenotype (Table 1). This possibility is further supported by the observations that additional RAR inactivations in the RARβ2γ2 mutant background (e.g. removal of one allele of RARγ1 or RARα1) result in a marked increase in the number of dysplastic foci within a given retina (our unpublished data). These observations also suggest that the incomplete penetrance and expressivity of the RARβ2γ2 retinal dysplastic phenotype does not result from functional redundancy with RA-independent regulators.
The stochastic variation of retinoid responsiveness within retinal cells may occur both at the spatial (i.e. between different cells at a given time) and temporal (i.e. in a given cell at different times) levels. For instance in a given RPE cell, this responsiveness may be adequate at the time of birth and fall below the critical threshold only after the completion of retinal histogenesis, which would then impair the function of retinoids in retinal maintenance. This may account for the fact that, in old animals, retinal degeneration is fully penetrant and affects extensive portions of the retina. Thus, the degenerative phenotype may be the consequence of either the retinal dysplasia and/or a further impairment of RA function in retinas (or regions of the retina) that had escaped defects during histogenesis. Even though we cannot exclude that differences in genetic background may account for some of the phenotypic variations seen among different animals, the considerable variations in expressivity often observed between the two eyes of a given animal cannot be explained on that basis, and thus must be related to the stochastic processes mentioned above.
CONCLUSION
The present study demonstrates an essential role for RARs, and therefore for retinoic acid, in retinal histogenesis and survival of retinal cells. The present RARβ2γ2 mutant mice represent the first genetically characterized animal model for retinal dysplasia. Even though homozygous null compound mutations for both RARβ2 and RARγ2 are unlikely to occur at a significant rate in humans, our data raise the possibility that genetic lesions affecting the retinoid signalling pathway could underlie some cases of human retinal dysplasia and/or degenerations, for which the genetic basis is currently unknown. RARand/or RXR-deficient mice may also provide interesting models to investigate the mechanism underlying the therapeutic effects of vitamin A in some retinal degenerations (Jacobson et al. 1995; Acott and Weleber, 1995).
ACKNOWLEDGEMENTS
We are grateful to Drs R. Molday and D. Hicks for the gift of antiopsin antibody, Dr Eriksson for the gift of anti-CRBPI antibody and C. Mendelsohn for the RARβ2 mutant mice. We thank B. Weber, C. Fisher, V. Giroult and S. Heyberger for excellent technical assistance. We also thank Drs D. Hicks and J. Sahel for advice, B. Boulay, J. M. Lafontaine and the secretariat staff for their help in the preparation of this manuscript, and S. Ward for critically reading the manuscript. This work was supported by funds from the Centre National de la Recherche Scientifique, the Institut National de la Santé et de la Recherche Médicale, the Centre Hospitalier Universitaire Régional, the Association pour la Recherche sur le Cancer, the Human Frontier Science Program, the Collège de France and the Bristol-Myers Squibb Pharmaceutical Research Institute. J. M. G. was supported by a longterm fellowship from the European Communities (Human Capital and Mobility).