The small lipid retinoic acid is known to promote neuronal differentiation in vitro and to act as a teratogen in the embryonic brain, but very little is known about the natural role of endogenously synthesized retinoic acid in forebrain development. Retinoic acid is synthesized mainly by three retinaldehyde dehydrogenases. We show here where the retinaldehyde dehydrogenases for the developing telencephalon are expressed and how their expression patterns change over developmental time. Retinoic acid diffusing from the retinaldehyde dehydrogenase sites is likely to influence the early telencephalon before the beginning of neurogenesis, as well as differentiation and radial migration of neurons into the cerebral cortex. Because of its diffusible character, retinoic acid represents a unique tool for the coordination of growth processes over an intermediate distance range in the developing telencephalon.
Retinoic acid (RA), the active form of vitamin A, is essential for adult survival and embryonic development. It functions through activation of a family of transcription factors, the nuclear RA receptors (Mangelsdorf et al., 1995). Expression of a large fraction of all proteins involved in brain development can be influenced by RA (Ross et al., 2000). Its actions at a specific anatomical site depend on the cellular and developmental context. It tends to activate complex programs; in particular, at many sites it promotes differentiation processes. RA is a small diffusible lipid which is synthesized locally in different organs in the vicinity of its sites of action. In addition, variable fractions of the total tissue RA content derive from the circulation, as the plasma contains low RA levels bound to albumin (Kurlandsky et al., 1995). Since animals cannot synthesize vitamin A de novo, all of it originates from carotenes made by green plants and vertebrates have evolved a very large storage capacity for retinol in the liver. Adult animals are relatively insensitive to fluctuations in daily vitamin A intake, but for the developing embryo an appropriate supply is critical, as both vitamin A deficiency and excess can cause permanent malformations (Wilson et al., 1953; Cohlan, 1954). The most teratogenic retinoid is RA itself, as is known in humans from malformations caused by maternal intake of acne medications (Acutane = isotretinoin, 13-cis-RA) during pregnancy (Rosa et al., 1986). Even rather low isotretinoin doses that produce no symptoms in the mother can have deleterious effects on the embryo, depending on the developmental time of exposure.
This leads to the question of why RA is teratogenic, although it is required for embryonic development and is synthesized at a high rate in the embryo itself. The explanation must lie in an interference with its normal distribution patterns. It has not been possible to visualize this distribution at histological resolution, because RA cannot be fixed in the tissue. We here provide an indirect lead to the RA distribution in the embryonic telencephalon of the mouse through localization of the enzymes synthesizing it. The last step in RA synthesis is mediated mainly by retinaldehyde dehydrogenases (RALDHs), which catalyze the irreversible oxidation of retinaldehyde to RA (Lee et al., 1991). Measurements in embryonic tissues show that the RALDHs represent the main determinants of local RA concentrations: when embryonic tissues are dissected into small pieces and cultured separately the levels of RALDH activities in each sample predict the RA levels in the supernatants (McCaffery and Dräger, 1994b).
Most RA in the embryo is synthesized by three RALDHs, which were first identified by their most characteristic expression sites: RALDH1 in dorsal retina, RALDH2 in the embryonic trunk and RALDH3 in ventral retina (McCaffery et al., 1992, 1993; Zhao et al., 1996; Li et al., 2000). Here we describe where the RALDHs for the developing telencephalon are located and how their expression changes over developmental time. Whereas RA and RALDH levels in the embryonic eye and trunk are extremely high, overall RA levels in the brain are very much lower and RALDHs are restricted to a few sites. RA diffusing from these sites is likely to set up differential concentration patterns. Such tissue RA gradients must be of morphogenetic significance, because the specific RA concentrations required for activation of different genes are known to vary over several orders of magnitude (Simeone et al., 1990). Exogenous RA probably exerts its teratogenic actions through transient leveling of the normal differentials and through spatially and temporally inappropriate activation of genes.
The locations and severity of malformations caused by exogenous RA depend on developmental age. The most severe defects are elicited during the early stages of embryogenesis when the anlagen of the nervous system and major organs are formed (Shenefelt, 1972). In the mouse this is around embryonic days 7–8.5 (E7–8.5), when even relatively low RA doses often cause lethal malformations. In the developing forebrain region these early RA effects include shortening or complete absence of the rostral part of the body and the brain, exencephaly, severe microcephaly or hydrocephaly (Avantaggiato et al., 1996; Clotman et al., 1998). Almost all experimental studies on RA teratology have focused on the early stages and much less is known about later effects. Most available information about less severe RA effects are clinical observations on children that survived mild, variable or later isotretinoin exposure during embryogenesis (Rosa et al., 1986; Nau et al., 1994; Adams and Holson, 1998). The mildly affected children may show a spectrum of dysmorphologies, mainly of the face and ears, or they may appear normal and have only functional disturbances. Some of the affected children show cortical blindness and many have intellectual deficits and behavioral abnormalities without overt neurological symptoms. Pathological observations on cerebral cortices include reduced size, focal agyria and heterotopias (Rosa et al., 1986; Clotman et al., 1998).
Because so little is known in experimental animals about later RA effects on the forebrain, Figure 1 illustrates examples of RA exposure during the period of cerebral cortex formation in mice. Although these mice survived (but viability was reduced), their brains and skulls show changes that are similar to those in humans. For these demonstrations, mice at different stages of pregnancy were injected with a single dose of RA which raises the tissue RA levels for a few hours (Ward and Morriss-Kay, 1995; Yamamoto et al. , 1998). The two brains on top were dissected from normal weight newborns, the left one from a mouse exposed to RA at E12.5 and the right one at E16. While the size of the right telencephalon is similar to normal, the left one is clearly smaller. RA exposure early during cortex neurogenesis thus seems to produce a reduction in telencephalic size, whereas RA towards the end of this stage has no obvious effect on the overall size of the brain. When other RA-exposed mice were allowed to grow to adulthood, they reached normal body size, but their skulls appeared deformed and noticeably smaller (Fig. 1, bottom).
MATERIALS AND METHODS
Experiments were done on B6/D2 outbred mice and a RA reporter strain generated by J. Rossant (Rossant et al., 1991). Embryos were staged according to Theiler, where the day following mating is defined as E0 (Theiler, 1972). Pregnant mice were killed by cervical dislocation and the embryos were dissected in cold tissue culture medium. We used standard in situ hybridization protocols (Barthel and Raymond, 1993; Riddle et al., 1993) and followed the northern blot protocols of Sambrook et al. (Sambrook et al., 1989). To illustrate RA-induced malformations, aliquots from a stock solution of 0.1 M all-trans RA (Sigma) in dimethyl sulfoxide were emulsified in 300 μl of complete culture medium and injected i.p. RA doses were adjusted to the weight of the mother mouse, at 1 μl stock/g. The adult skulls were cleaned by Dermestid beetles.
Retinoic Acid Synthesis in the Early Embryonic Head Outside the Brain
RALDH2 is the first RA synthesizing enzyme expressed in the embryo and for ~1 day in the mouse it remains the only local RA source (McCaffery et al., 1993; Dräger and McCaffery, 1995; Niederreither et al., 1997). It appears around E7.5 in the future trunk region and maintains an abrupt expression border at the anterior end of the future spinal cord (Fig. 2a). The rostral part of the early embryo from the future medulla upwards contains no trace of RA synthesis, until RALDH2 begins to be expressed in the head process along the anterior neural ridge early on E8 (Wagner et al., 2000). Over the next few hours RALDH2 mRNA levels increase rapidly to high amounts (Fig. 2b), but then are equally rapidly degraded. Later during E8, all RALDH2 mRNA disappears from the head (Fig. 2c), but some enzyme activity can still be measured (McCaffery et al., 1993). The transient RALDH2 mRNA is expressed in cells between the surface and neural ectoderm; they are probably mesenchymal cells, but some could represent a subpopulation of cranial neural crest. The location of the RALDH2 activity includes part of the eye field and the precursor region for the ventral telencephalon and it is likely to represent a late component of the inductive activity at the anterior neural ridge (Rubenstein et al., 1998; Inoue et al., 2000).
The next RA synthesizing enzyme in the head is RALDH3, which appears at E8.5 in the surface ectoderm behind and over the optic recess (McCaffery et al., 1992, 1993; Grün et al., 2000; Li et al., 2000; Mic et al., 2000; Suzuki et al., 2000). Within the following few hours RALDH3 expression increases to extremely high levels. By the time of rostral neural tube closure (E9) the solid RALDH3-expressing surface area covers a large part of the face (Fig. 2c). Over the next 60 h the RALDH3-positive surface ectoderm constricts down to separate fields over the emerging eye and olfactory organs and RALDH3 expression moves deeper into ectodermal parts of the eye and the olfactory epithelium. Up to ~E11 all RALDH3 expression in the head remains outside the brain (Li et al., 2000). Expression of the third RA synthesizing enzyme in the face, RALDH1, begins early on E9 in the dorsal eye vesicle (McCaffery et al., 1991). Because RALDH1 is much less effective in RA synthesis than the other two enzymes, its early influence is likely restricted to the face.
Between E8 and E11 the embryo thus contains two foci of very high RA synthesis, one in the trunk and the other in the rostro-lateral head, which are separated by the RALDH-free brain. These two RA foci are directly visible in embryos of RA reporter mice, whose tissues indicate the responses to endogenous RA with synthesis of β-galactosidase (Rossant et al., 1991). When early RA reporter embryos are briefly flooded with a high RA dose injected into the mother, the reporter gene is activated everywhere, indicating that all early embryonic tissues can respond to RA if the concentration is high enough. The natural labeling in the reporter embryos does not reflect the RA distribution, which is presumably graded, but the distribution of RA sensitivity mediated by RA receptors, which follows the neuromeric boundaries in the brain (Zimmer and Zimmer, 1992).
The comparisons of RALDH and RA reporter expression in Figure 2a–c show that the two patterns are similar, except for spatial discrepancies at the borders: from the trunk region upwards, RALDH2 expression extends only to the upper end of the spinal cord, but the reporter labeling extends further up into the medulla to about the level of the otic vesicle (arrows); rostrally in the head, RALDH expression is limited to the face, but the reporter is also strongly activated in the emerging telencephalon. These two discrepancies are likely to indicate RA diffusing from adjacent RALDH-rich regions into the RALDH-free early brain. In order to visualize the spatial relationship between the forming telencephalic vesicle and RALDH3, the principal RA source at this time, heads of RA reporter mice were cut along the mid-sagittal plamne to expose the ventricular sides of the brain and compared with RALDH3-reacted embryos of similar ages (Fig. 2d–f). At E9, when the RALDH3-positive surface ectoderm covers a large part of the rostral neural tube, RA reporter labeling is equally strong in surface and ventricular views and it extends from the optic recess throughout the entire telencephalic anlage except for the ventral midline. Half a day later the extent of the RA-responsive surface ectoderm roughly matches the RALDH3-expressing field in the face. Above the face, however, the RA reporter is only expressed in the telencephalic neural ectoderm, located some distance away from any RALDH expression, and not in the surface epithelium. Telencephalic reporter labeling has a sharp border at the ridge forming between the diencephalon and telencephalon and it begins to retract from the ventral telencephalon. This gradual restriction of RA-responsive cells to the dorsal telencephalon becomes more obvious at E10.5. At this age and through the following several days all RA-responsive cells are concentrated at the ventricular side of the dorsal telencephalic wall.
Activation of RALDH3 within the Telencephalon and Tangential Distribution of RA Actions
Around E11 RALDH3 expression is activated at the outer surface of the telencephalic vesicle in a rostro-lateral location that was earlier close to the strong RA source in the face (Li et al., 2000). This RALDH3 patch is illustrated on a telencephalon whole mount shown both from its lateral (~outer) and medial (~inner) aspects (Fig. 3a). The RALDH3 patch appears external to the site where the lateral ganglionic eminence is going to form. Over the following day RALDH3 expression intensifies and the labeled somata move inwards through the telencephalic wall into the lateral ganglionic eminence and ventral septum (Li et al., 2000). This inward movement is illustrated on coronal sections through the rostral telencephalon at three ages (Fig. 3b): the perikarya that contain the RALDH3 mRNA move centripetally across the ventral telencephalic wall from the outer surface to the subventricular zone. The labeled profiles move slowly and as a solid front, rather than showing signs of interkinetic radial movements. Although identification of the cell type is not possible by in situ hybridization since no processes are labeled, the location of the perikarya in the subventricular zone at later stages is consistent with radial glia identity (Rakic, 1972; Toresson et al., 1999). In the embryonic forebrain RALDH3 can only be detected in the rostro-ventral telencephalon, which is apparent in Figure 3, as well as in various other preparations not shown. At the end of embryonic development the RALDH3 signal in the ventral telencephalon decreases to trace amounts.
RA actions are closely linked to two types of transcription factors. The first is Pax6, which is required for RALDH3 expression (Suzuki et al., 2000) and whose own expression can be stimulated by RA (Gajovic et al., 1997). Although most Pax6 in the telencephalic vesicle is expressed dorsally and inside in the ventricular zone, a minor expression site is ventrally and at the outer surface (Stoykova et al., 1997) (see Fig. 3a). This surface labeling covers a broad rostro-caudal region. Comparisons of the preparations in Figure 3a show that the RALDH3 patch is located within the Pax6-positive surface area. To illustrate the spatial layout of Pax6 labeling on the inner, ventricular side the telencephalic vesicle was flattened out by radial cuts through its medio-dorsal wall. The flat-mount in Figure 3a, which is viewed from its ventricular side and in the same orientation as the other preparations, shows restriction of Pax6 to the dorsal telencephalon (Stoykova et al., 2000), where it forms a rostro-caudal gradient at this early age (Bishop et al., 2000).
The other transcription factors that are directly relevant to RA actions are the chick ovalbumin upstream promoter transcription factors (COUP-TFs). The COUP-TFs function mainly as repressors of RA-stimulated transcription (Qiu et al., 1996). Mice have two related forms, COUP-TFI and COUP-TFII, which are widely expressed throughout the embryonic brain in partially overlapping but distinct patterns; in general, the spatial expression of COUP-TFII is more restricted than COUP-TFI (Qiu et al., 1994). In the developing telencephalon both forms are expressed in a high caudal to low rostral distribution (Qiu et al., 1994; Liu et al., 2000), with the COUP-TFII gradient being steeper than that of COUP-TFI (Fig. 3c). The combined expression patterns of RALDH3 and the three transcription factors, Pax6 and the COUP-TFs, indicate that RA actions in the early telencephalon must be higher rostrally and that RA effects are suppressed caudally.
RALDH2 is Expressed in the Meninges Covering the Telencephalon
As shown in Figure 2c, all RALDH2 mRNA disappears from the head following its brief expression at the rostral edge (Wagner et al., 2000). When the telencephalon is enlarging RALDH2 is very slowly activated in its meninges. This is shown (Fig. 4) for three embryonic ages on pairs of cerebral hemispheres that are viewed both from their lateral and medial sides; the meninges are left in place on the left samples (+), but have been dissected off the right ones (–). Only traces of meningeal labeling are visible at E13. At E14 the labeling has intensified, which is particularly obvious at the edges where the hippocampi and tips of the olfactory bulbs have been cut off for better access of the in situ reagents. By E16.5 the telencephalon is covered with darkly labeled meninges. Apart from the overall slow pace of meningeal RALDH2 activation, there are regional variations in expression levels. For instance, RALDH2 appears first in the meninges covering the ventral and caudal telencephalon and expression here remains stronger over the next days. The whole mount preparations, as well as sections not shown, reveal no RALDH2 expression anywhere but in the meninges.
RALDH1 Expression in the Nigro-striatal System and Meninges
The main expression site of the third RA generating enzyme, RALDH1, in the brain is in a subpopulation of dopaminergic cells within the substantia nigra and adjoining ventral tegmentum (McCaffery and Dräger, 1994a). These cells send ascending axons through the medial forebrain bundle into the ganglionic eminence and subsequent striatum of the telencephalon. RALDH1 mRNA is located in the cell bodies, as shown by in situ hybridization on an embryonic brainstem (Fig. 5, left), and RALDH1 protein is transported anterogradely into the telencephalon. By immunohistochemistry a few RALDH1-positive axons can be detected in the forebrain by E13 (McCaffery and Dräger, 1994a). In the later embryonic and postnatal brain, as the dopaminergic neurons sprout massive axonal arborizations in the striatum, the amounts of anterogradely transported RALDH1 enzyme become huge. This is illustrated in a coronal section through the rostral telencephalon of an adult mouse reacted with a RALDH1 antiserum (Fig. 5, right). This section shows labeling in an additional telencephalic site: RALDH1 is also expressed in the meninges, like RALDH2. This meningeal RALDH1 expression begins in the embryo, but it is significantly weaker than RALDH2. No other site of RALDH1 expression has yet been detected in the embryonic telencephalon.
Spatio-temporal Analyses of RALDH Distributions by RNA Blots
The histological screens thus show only two major sites of RALDH expression in the embryonic telencephalon: the lateral ganglionic eminence with adjacent ventral telencephalon and the meninges covering the entire telencephalon. In Figure 6a (left lanes) we compare the expression of the three RALDHs by northern blotting on concentrated samples from two embryonic ages: dissected lateral ganglionic eminences (LGE) and isolated meninges. Consistent with the histological observations, RALDH3 is only detectable in the LGE samples and its level decreases with embryonic age, and both RALDH2 and RALDH1 are expressed in the meninges at rising levels, with the RALDH2 signal being much stronger than that of RALDH1. Because RALDH1 is much less effective in RA synthesis than the other two enzymes (McCaffery et al., 1992; Grün et al., 2000), the main local RA sources for the embryonic telencephalon are thus RALDH2 and RALDH3. RALDH1 reaches high levels here only at perinatal to adult stages.
In order to estimate how total telencephalic levels of RALDH2 and RALDH3 change over time, we dissected entire cerebral hemispheres at different developmental ages and compared the relative changes in mRNA levels by northern blotting (Fig. 6b). Telencephalic RALDH3 levels rise early, reach a maximum around E14 and decline to undetectable amounts in the late embryo. RALDH2 appears later, its amounts rise more slowly, peak in the early postnatal telencephalon and then decline to lower adult levels. Postnatally, the huge amounts of anterogradely transported RALDH1 protein represent a dominant RA source in the rostral telencephalon (McCaffery and Dräger, 1994a). In addition, mRNAs of all three enzymes are expressed in the adult. To estimate their spatial distribution, we compared four samples dissected along the adult rostro-caudal telencephalic axis by northern blotting: the olfactory bulb and the rostral, intermediate and caudal telencephalon (Fig. 6a, right lanes). The mRNA levels of RALDH2 and RALDH3 are highest in the olfactory bulb samples, while RALDH1 is distributed relatively more evenly across the rostro-caudal axis. Histological screens (not illustrated) show that both RALDH1 and RALDH2 mRNAs are mainly located in the meninges at the outer surface, but RALDH3 is strongly expressed in cells within the olfactory bulbs; a detailed analysis of these expression patterns will be described elsewhere.
Modes of RA Dispersion by Tissue Diffusion Gradients and via the Circulation
While RA is well known as an in vitro reagent for induction of neuronal differentiation in embryonic stem cells (Jones-Villeneuve et al., 1983 ;Gottlieb and Huettner, 1999), very little information exists about its natural role in the developing forebrain. The principal reasons for this are technical difficulties in detecting endogenous RA at sufficiently high resolution. When bulk RA levels in whole organs are measured the compound appears much more important in RA-rich tissues like the pancreas than the RA-poor brain. Nevertheless, malformations induced in the embryo by too little or too much vitamin A indicate that RA must serve critical functions in the developing forebrain and that its natural distribution must matter. We have here described for the developing telencephalon the changing expression patterns of three RALDHs, which represent the major local sources of RA in the developing embryo. Whereas the capacity to respond to RA is widely distributed in the brain — every location expresses at least one and usually several RA receptors (Ruberte et al., 1993; Krezel et al., 1999)—most telencephalic sites are completely devoid of RA synthesis (Li et al., 2000) and RALDH expression is remarkably sparse. This indicates that most telencephalic locations depend on RA which is either supplied by the circulation or which diffuses from RALDH-expressing sites nearby.
As guides for estimates of the relative RA contributions from these two sources, only measurements on adult mammals are available (Kurlandsky et al., 1995), but no direct information exists for the embryonic brain. Although RA is much more soluble in lipophilic than aqueous media, it can be dispersed in protein-containing aqueous solutions. Serum levels in adult rats are very low (low nanomolar range) and the total serum RA pool makes up ~1% of total body RA. This serum pool is highly dynamic: it turns over every 2 min. In some organs, e.g. testis and pancreas, almost all RA is synthesized locally, but the adult brain obtains almost 90% of its RA content from the circulation (Kurlandsky et al., 1995). Since the RALDH expression sites are spaced widely apart, RA levels in most locations of the adult brain must represent a direct equilibrium with circulation levels.
RA has a high affinity for solid tissues and diffuses through cell membranes without a carrier. However, while RA dispersion via the circulation is very rapid, diffusion through tissues is likely to advance at chromatography speed and the resulting tissue gradients ought to be relatively stable. In addition to tissue solubility properties, the ranges of the diffusion gradients ought to be influenced by exchange with the circulation and the presence of RA degrading enzymes. Although it is impossible to make reliable numerical estimates, it seems plausible that in the early embryo the hypothetical gradients extend over several hundred micrometers to a millimeter. A RA gradient of such range is likely to exist in the early hindbrain, where distinct RA levels are known to elicit distinct morphogenetic effects (Gavalas and Krumlauf, 2000). The early hindbrain contains no trace of local RA synthesis, but it is located adjacent to the RALDH2-rich trunk (McCaffery and Dräger, 1994b; Zhao et al., 1996) and the RA responses in RA reporter embryos (Rossant et al., 1991) extend further rostrally beyond the RALDH2 border (Fig. 2a–c). RA-responsive genes in the hindbrain are known to be expressed in a stacked manner, with those requiring higher RA levels closer to the trunk and the lower concentration requiring genes extending successively more rostrally. Exposure to exogenous RA causes posteriorization of the hindbrain, which means that rhombomeric identities can be switched to more caudal ones or to spinal cord characteristics (Gavalas and Krumlauf, 2000). Conversely, vitamin A deficiency causes a dose-dependent acquisition of rostral characteristics and shortening of the caudal hindbrain (White et al., 2000) and a similar effect is seen in RALDH2 null mutants (Niederreither et al., 2000).
RA Synthesized in the Face Affects the Early Telencephalon
RA diffusing from rostral sources is bound to also influence the early stage of telencephalon formation, before RALDHs are expressed here. Transient expression of RALDH2 at the rostral edge of the head fold is probably required for RALDH3 activation (Marsh-Armstrong et al., 1994). In addition, it is likely to influence the telencephalic anlage directly, but the nature of its effect needs to be analyzed in an isolated assay (Shimamura and Rubenstein, 1997). In RALDH2 null mutants the neural plate does not fold up and all embryos are dead by E10 from heart defects (Niederreither et al., 2000). At the time when the rostral neural tube closes (E9) all RALDH2 mRNA has disappeared from the face, but some enzyme activity, probably located in mesenchymal cells, can still be detected in biochemical assays (McCaffery et al., 1993; Wagner et al., 2000). The main local RA source at this stage is RALDH3, located in the ectoderm of the face in close proximity to the forebrain. In RA reporter embryos strong labeling appears in the emerging telencephalic vesicle (Fig. 2c–f). When the reporter mice were crossed with mutants that express no RALDH3 in the face due to lack of functional Pax6 protein (Suzuki et al., 2000), little RA response remains in the early telencephalon (Anchan et al., 1997; Enwright and Grainger, 2000). Because the geometrical relationships between the RALDH3 expression sites and the beginning telencephalic vesicle are broad and change rapidly (Fig. 2d–f), it is impossible to indicate a dominant spatial orientation of tissue RA gradients. Moreover, it seems likely that a part of the RA dispersion at this early age occurs via the prosencephalic lumen, which connects the RALDH-rich optic vesicle with the RALDH-free telencephalic anlage. For these reasons, it seems plausible to assume that early RA effects on the emerging telencephalic vesicle do not include a spatial dimension, comparable to the caudo-rostral RA dimension in the hindbrain. The rostro-caudal orientation of RA actions illustrated in Figure 3 is probably set up after the telencephalic axes are already established. This axial asymmetry in RA actions is only formed rostrally by RA diffusing from the rostro-lateral RALDH3 site and most of its extent is created by the graded expression of RA repressors, the COUP-TFs.
The proposal that RA diffusing from the surface ectoderm of the face influences telencephalon formation extends an old concept of inductive effects of the olfactory periphery on the brain. When the olfactory placode was removed in very young Xenopus embryos the telencephalon failed to form in a fraction of the operated cases (Graziadei and Monti-Graziadei, 1992). Because RALDH3 expression at the corresponding stage in the embryonic mouse extends further than the olfactory placode, the success rate of Graziadei's experiment might have been complete had the entire RALDH3-expressing ectoderm been deleted. The notion of a rostral RA source also agrees with LaMantia, who proposed that RA synthesized in the olfactory mesenchyme induces the olfactory bulbs and acts on the telencephalon (LaMantia et al., 1993; LaMantia, 1999).
Reciprocal Relationship between RALDH3, Pax6 and RA
An intriguing reciprocal relationship exists between RA actions and the transcription factor Pax6. LaMantia's group observed that head defects in Pax6 mutants are similar to malformations caused by a chemical RALDH inhibitor and they demonstrated that RA levels are significantly lower in heads of homozygous Pax6 mutant mice (Anchan et al., 1997). This observation was confirmed and expanded upon by Enwright and Grainger (Enwright and Grainger, 2000). Suzuki et al. showed that no RALDH3 is expressed in the face of Pax6-negative rat embryos (Suzuki et al., 2000). In the early head process RALDH3 is first activated in the Pax6-positive surface ectoderm behind the optic recess (Grindley et al., 1995; Li et al., 2000), which is close to the high RA source from transiently expressed RALDH2. When RA synthesis in the surface ectoderm has increased to high levels, RALDH3 expression seems to jump to the Pax6-positive eye vesicle, but spares its dorsal part, where RA is removed by a RA degrading enzyme (Wagner et al., 2000). A little later RALDH3 again seems to jump from the RA-rich face into a neighboring Pax6-positive area on the telencephalic surface (Fig. 3), without touching the mesenchyme in between (Li et al., 2000). Moreover, in both Pax6 and, very likely, RALDH2 null mutants RALDH3 is not expressed in the face and probably also not in the brain (Niederreither et al., 1999; Suzuki et al., 2000). These morphological observations indicate that the initial RALDH3 activation must require a combination of Pax6 and high RA. Conversely, Pax6 expression is up-regulated by RA in neuronal stem cells in vitro (Gajovic et al., 1997); whether this also applies to ventricular cells in the dorsal telencephalon is not yet known.
Early RA Synthesis in the Telencephalon by RALDH3
Following its transient expression on the rostro-lateral telencephalic surface, RALDH3 moves deeper through the wall into the lateral ganglionic eminence with adjoining entorhinal cortex and septum (Li et al., 2000). It is probably expressed by radial glial cells (Toresson et al., 1999) which span the wall of the ventral telencephalon. The largest surface area of the RALDH3-positive telencephalic wall is with the ventricular lumen, since the lateral ganglionic eminence protrudes into the ventricle. Because the liquid in this space must have the same protein concentration as serum (the blood–brain barrier forms only postnatally), this high RALDH3 expression is likely to raise RA levels in the ventricular lumen above serum levels. A similar effect occurs in the anterior chamber of the adult eye, where the RA-rich ciliary body is surrounded by the aqueous humor: RA concentrations in this confined liquid compartment are about seven times higher than serum levels (Wakabayashi et al., 1994). While rostral RALDH3 expression should first create a tangential RA gradient along the rostro-caudal telencephalic axis, RA dispersion within the ventricular lumen ought to then set up a radial inside-out gradient across the telencephalic wall (Fig. 7). An effect of endogenously synthesized RA could be the initiation of neuronal fates in mitotic precursors, similar to RA effects in vitro: RA exposure of P19 stem cells terminates mitosis and initiates neuronal differentiation (Jones-Villeneuve et al., 1983; Gottlieb and Huettner, 1999). Neurogenesis in mouse cerebral cortex begins at E11 at a rostro-lateral location (Caviness et al., 2000), which coincides spatially and temporally with RALDH3 appearance and might be triggered by it. Through the neuronogenetic period over the following 6 days, final mitoses take place in the ventricular or subventricular zone, where they are likely promoted by RA from the lumen. An interference with this proposed normal RA role could explain some of the RA-induced cases of microcephaly in humans (Rosa et al., 1986) and mice (Fig. 1): transient flooding of the embryo causes premature initiation of differentiation in too many precursors throughout the telencephalic wall, resulting in a reduction in the remaining mitotic pool for cerebral cortex formation.
The hypothesis that early in corticogenesis endogenous RA acts on the mitotic population in the ventricular layer is consistent with observations on RA reporter embryos: all early RA responses in their forebrains are located in the ventricular layer. In Pax6-negative mutants the cortical phenotype could be explained by a deficiency of RA action on neuronal precursors, because proliferative rates in the early pallium are increased, neuronal differentiation is defective and undifferentiated neuroblasts pile up in the subventricular zone (Warren et al., 1999). If part of the Pax6 phenotype is indeed caused by relative RA deficiency, some of it ought to be mitigated by moderately high vitamin A doses given to the mother.
RA Synthesis by RALDH2 in the Meninges Creates an Outside-in Gradient
Later in corticogenesis RALDH3 expression decreases and RALDH2 in the meninges increases. This should result in a slow reversal of the radial RA distribution from an inside-out to an outside-in pattern (Fig. 7). RALDH2 expression in the meninges remains high through the period of neuronal migration and terminal differentiation, which extends postnatally in mice and then declines to lower adult values. The spatial arrangement of a very strong RA source at the outer margin of the developing cortex indicates that endogenous RA may serve a role in the termination of migration and elaboration of neuronal processes. It is well known that RA stimulates neurite outgrowth in culture (Quinn and De Boni, 1991). A premature stop signal and ectopic initiation of neurite outgrowth could explain the heterotopias and other signs of neuronal migration defects in the cortices of RA-exposed children (Rosa et al., 1986).
RA Functions as a Tool for the Coordination of Neuronal Differentiation Processes across Extended Distances in Vertebrate Brains
Many observations on how RALDH levels change in the developing nervous system indicate that they are high during the periods and close to the locations of neuronal growth and differentiation and that mature levels are much lower (McCaffery et al., 1993, 1996; McCaffery and Dräger, 1994). A similar obser-vation was recently described in the adult brain: the renewal of song neurons in the brains of male finches is initiated by a rise in local RALDH2 expression (Denisenko-Nehrbass et al., 2000). In adult mammals the olfactory bulb represents a major site of ongoing renewal. A precursor population in the subventricular zone of the rostral telencephalon generates neuroblasts throughout life which migrate into the bulb (Luskin, 1993). The high rostral RALDH levels in the adult telencephalon, from the striatum to the olfactory bulb, are likely relevant for this process of ongoing neuronal renewal.
The use of RA as a transcriptional activator is relatively recent in evolution: it is fully evolved only in vertebrates, but has no role in Drosophila development. The transcriptional function probably originated in the eye, where the RA precursor retinaldehyde serves as the visual chromophore (McCaffery et al., 1996). The diffusible character of RA makes it a unique tool in larger animals for ordered interactions across an intermediate range distance. This spatial function must be used mainly in RA-poor regions like the brain, because significant RA gradients can only form on a low background. The use of a food-derived compound as a concentration-dependent signaling molecule represents, however, a potentially very dangerous situation for the embryo. To ensure a reliable supply of RA vertebrates have developed, on the one hand, enormous retinoid storage capacities in egg yolks and livers (which cause the orange color of yolk and the toxicity of ingested liver). On the other hand, the need for perfectly regulated RA availability probably explains the vast and partially redundant network of protein factors involved in RA actions: many different binding proteins and a diverse set of RA receptors (Chambon, 1996). Null mutations for most of these protein factors give only subtle or no apparent pathologies, but they tend to make the animals more vulnerable to abnormal vitamin A supply and defects are brought out in double or triple mutants (Lufkin et al., 1993; Lampron et al., 1995; Chiang et al., 1998; Krezel et al., 1998; Ghyselinck et al., 1999; Mark et al., 1999). Although mutations in all of these factors must occur in humans, no genetic retinoid diseases of the brain have yet been identified. Considering that RA probably exerts several critical functions in brain development, some human diseases ought eventually to be found. A characteristic of retinoid-related forebrain diseases has to be a broad symptomatology linked to diverse genetic loci combined with a strong environmental component. One of the diseases which fits these characteristics and for which retinoid involvement is being discussed is schizophrenia (LaMantia, 1999; Goodman and Pardee, 2000).
We thank J. Rossant for the RA reporter mice, T. Glaser for the Pax6 probe, H.O. Nornes and M.J. Tsai for the COUP probes, R. Lindahl for the RALDH1 antiserum and J. Chupasako for the Dermestid beetles. This work was supported by NIH grants EY01938, HD05515 and HD01179. The first two authors contributed equally to this work.