Pax2 contributes to inner ear patterning and optic nerve trajectory

In vertebrates, the nose, the eye and the ear house the highly specialized sensory apparatuses for olfaction, vision, balance and hearing. Classical embryological studies have shown that both autonomous and inductive cues contribute to specify and regionalize the developmental fields that constitute the complex architecture of sense organs (for review see Noden and Van de Water, 1986; Saha et al., 1992). However, very little is known about the molecular nature of the processes involved. Several families of transcription factors have been found in a variety of organisms, which provide tissue- or region-specific autonomous determination during development and thus represent valid candidates for autonomous regionalizing signals. The Pax6 gene, encoding a paired domain-containing transcription factor, is the only example of a molecule with such a role in sense organ patterning (Walther and Gruss, 1991; Hill et al., 1991; Ton et al., 1991; Quinn et al., 1996). Its function has been strikingly conserved throughout evolution, as demonstrated by the ability of the mouse Pax6 to promote eye development in flies (Halder et al., 1995). In general, Pax genes are specifically expressed during the development of a wide range of structures and organs (Gruss and Walther, 1992). As indicated by the phenotypes resulting from specific mutations in both man and mouse, members of the Pax family have critical morphogenetic functions during the development of complex tissues and organs in vertebrates (for review see Chalepakis et al., 1992; Strachan and Read, 1994). Two of the Pax genes; Pax2 and Pax6, are restrictedly expressed in developing sense organs (Nornes et al., 1990; Walther and Gruss, 1991). While Pax2 is expressed in the developing inner ear and Pax6 is expressed in the developing nose, both are expressed in the developing eye. During development, the embryonic primordium of the eye and optic tracts, consisting of the optic evagination and the contacting surface ectoderm, undergoes morphogenetic movements that delineate the different fields composing the eye (Saha et al., 1992). The different fields subsequently differentiate into the definitive components of the eye and optic tracts. Expression of Pax6 in the visual system is restricted to regions fated to constitute the eye deriving from the distal optic vesicle and the ectodermal lens placode. In contrast, Pax2 expression is initially confined to the ventral optic vesicle, without proximodistal restriction, later being confined mostly to the proximal regions destined to contribute to the optic nerve and optic chiasm. Interestingly, Pax2 is also regionally expressed within the otic vesicle primordium confined mostly to regions of the vesicle fated to have an auditory function. Characterization of spontaneous mouse and human Pax6 mutations has demonstrated a critical role for Pax6 in the differentiation of structures derived from the distal optic vesicle and lens as well as nasal structures (Hill et al., 1991; Ton et al., 1991; Quinn et al., 1996). Spontaneous mutations affecting Pax2 function have been described in both humans and mice (Krd mutants) (Keller et al., 1994; Sanyanusin et al., 1995). Heterozygosity for these two mutations produced eye and kidney defects, suggesting the relevance of Pax2 in eye and kidney development, showing the functional conservation between species and demonstrating haplo-insufficiency for the locus. However, the study of the mutation only in the heterozygous condition in the mentioned reports, precluded the complete definition of Pax2 functions. Mutation of Pax2 by homologous recombination (Torres et al., 1995) has allowed its detailed analysis in the complete loss-of-function condition, showing the important role of this gene during urogenital development, and reproducing Krd and human Pax2 haplo-insufficient phenotype for kidney development.

We present here for the first time the analysis of the phenotype of Pax2 null mutants in the central nervous system (CNS) and sense organ development. Our results show that Pax2 plays important roles in morphogenesis and regional specification of the eye, in which the heterozygous phenotype partially reproduces the human syndrome. In addition, we uncover novel functions for Pax2 in the development of optic tracts, auditory regions of the inner ear and closure of the neural tube at the midbrain level.

A previously described Pax2 null mutation obtained by homologous recombination was used for this analysis (Torres et al., 1995).

Embryo dissections: fetus were collected in PBS and yolk sacs or tails were removed for DNA analysis. Genotyping of embryos and adults was performed by Southern blot as previously described (Torres et al., 1995).

E12.5 embryos were dissected out in PBS and embryos were fixed in 4% paraformaldehyde (PFA) and, after dehydration, embedded in paraffin and sectioned at 8 to 10 μm. Sections were stained with Haematoxylin and Eosin. E17.5 embryos were treated similarly except they were fixed in Bouin’s fixative and stained by Mallory’s tetrachromic procedure.

E13.5 embryos were fixed overnight in PFA, immersed in sucrose 30% for 24 hours, embedded in Tissue tek and cryostat sectioned at 10 μm. Primary antibodies: anti-Pax2 (Dressler and Douglas, 1992), anti-Neurofilament 160 (Sigma) and anti-E-Cadherin (Sigma). Secondary antibodies: Cy3-conjugated (Jackson ImmunoResearch Laboratories) and peroxidase-conjugated (Vectastain) antibodies.

Embryos were processed as described (Wilkinson, 1992) using probes previously described for Wnt-1 (Wilkinson et al., 1987), Pax5 (Asano and Gruss, 1992), En-1 (Davis and Joyner, 1988), Pax2 (Dressler et al., 1990), Shh (Echelard et al., 1993). After in situ hybridization embryos were dissected and photographed as whole mounts.

Pax2 is dynamically expressed during neurulation in the prospective brain neuroepithelium. Initially Pax2 is expressed at E7.5 (embryonic day 7.5) in a broad region encompassing most of the prospective forebrain down to the prospective anterior hindbrain (Püschel et al., 1992; Rotwich and McMahon, 1994). As neural tube closes, the expression rapidly narrows in the anteroposterior (A-P) direction and, by E9.5, it is reduced to a thin stripe at the mid-hindbrain boundary. Pax2 homozygous mutant embryos show exencephaly, resulting from the failure of the neural folds to close at the midbrain region (Fig. 1). At E9.0, when the neural tube has just closed in control embryos, the normal disposition of early A-P brain subdivisions are found in mutant embryos, but the neural folds are not fused (Fig. 1). At later stages, the non-fused neural folds proliferate abnormally and overgrow outwards. The exencephaly is also found in heterozygous embryos, albeit with a low penetrance and in a background-dependent manner. Among the defective embryos, the degree of exencephaly is not dependent on the genotype, appearing as an all-or-nothing phenomenon. The incidence of exencephaly in heterozygous embryos is 11 out of 59 cases analyzed in a 129sv × NMRI mixed background, 1 out of 29 in a 129sv × C57/B6 mixed background, 2 out of 11 in a 129sv × C3H mixed background and none out of 14 in a 129sv inbred background. The extension of the exencephaly shows a low degree of variation among individuals that does not correlate with the genotype. The severity of the phenotype, when present, is similar between homozygous and heterozygous animals.

Wnt-1, a member of the wingless family of signalling molecules, Pax5, a member of the paired box family closely related to Pax2, and the homeobox genes En-1 and En-2 are expressed around the mid-hindbrain boundary area (Wilkinson et al., 1987; Davis et al., 1988; Asano et al., 1992). Loss-of-function mutations in these genes result in deletions affecting different portions of the meten-mesencephalic areas (McMahon and Bradley; 1990, Joyner et al., 1991; Urbanek et al., 1994; Wurst et al., 1994). Pax2 is expressed in the same region and its expression precedes the onset of the above mentioned genes. We have analysed the expression of Wnt-1, En-1 and Pax5 in mutant embryos in order to establish whether Pax2 activates the expression of these genes in this region. Despite the abnormal opening of the neural tube in mutant animals, anteroposterior and dorsoventral patterns of expression of Wnt-1, Pax5 and En-1 appear normal at E9.5 (Fig. 2). Examination of the expression of these markers at later stages is not accessible due to the gross alteration of the brain morphology in midbrain and anterior hindbrain.

Eye formation in vertebrates is mediated by multiple early inductive events. After these initial inductive events, eye formation involves two distinct processes: firstly, morpho-genetic processes, through which the basic elements of the eye are produced and shaped, and, secondly, differentiation processes, through which cells allocated to the different regions during the first phase, differentiate into the appropriate fate according to their position. One of the main events during the morphogenetic phase is the invagination of the ventral portion of the optic vesicle. At the distal-most part of the stalk, the invaginated tissue encircles the ectoderm-derived forming lens. The result of this process is the formation of the optic globe with its three main components, the outer pigmented retina, the neural retina and the lens. Ventrally, the eye remains transiently open at the optic fissure through which vascularization occurs. The ventral invagination and the presence of the optic fissure also extends into the optic stalk region. Closure of the optic fissure marks the end of the morphogenetic phase. Early in morphogenesis, Pax2 is expressed in the ventral half of the optic vesicle (Dressler et al., 1991). The expression is very dynamic, being repressed in the prospective neural retina shortly after its invagination but persisting during the whole morphogenetic phase at the lips of the optic fissure (Fig. 3E) and extending ventrally into the optic stalk up to the brain. We have examined the evolution of the morphogenetic phase of the eye in the mutants. Formation of the optic stalk, invagination of the prospective neural retina and lens segregation occurs normally. However, homozygous Pax2 mutants fail to close the optic fissure, showing complete bilateral coloboma at birth (Fig. 3A-D). During normal eye development, the optic fissure closes by contact-dependent dissolution of the basal lamina at the contacting neuroepithelial lips. We have followed the basal lamina in mutant embryos by the presence of laminin at the lips of the optic fissure (Fig. 3F,G). We found that, despite the apparent contact between the lips, the basal lamina persists precluding the closure of the optic fissure. Despite the open optic fissure, both retinal layers appear normal at birth with regard to cell type composition (data not shown). Therefore the requirement for Pax2 in the optic cup appears to be restricted to morphogenetic events occurring in the most ventral cells around the optic fissure, where its expression is persistent throughout all the morphogenetic period.

Although Pax2 expression is initially restricted to the ventral half of the optic evagination, after the optic cup is formed it extends dorsally in the optic stalk. As a result, it labels the prospective optic nerve, stopping sharply at the prospective pigmented retina-optic nerve boundary (Fig. 4A-D). After the morphogenetic phase, the prospective outer retinal layer is continuous with the optic stalk epithelium, nevertheless they both give rise to very different cell types during the differentiation phase. While the prospective outer retinal cells differentiate into pigmented cells, the optic stalk cells proliferate and differentiate into glial cells that populate the optic nerve. Pigmentation of the outer retina cells progresses in a distal-to-proximal direction and stops at the Pax2-expressing cells in the optic stalk. The limit is sharply defined such that no cell expressing Pax2 shows pigmentation (Fig. 4B,D). In agreement with this expression pattern, we have observed defects in the differentiation of stalk cells. In Pax2 mutant animals, the pigmentation does not stop at the eye-stalk limit but rather extends abnormally into the stalk region. Thus in the mutants, cells located in the stalk region differentiate into cells that can not be morphologically distinguished from pigmented retinal cells (Fig. 4E-H). The extension of the pigmentation into the optic stalk affects, depending on the genetic background, from 50% to practically all of the stalk extension up to the diencephalon.

Optic nerve genesis starts when the axons coming from the retina grow into the stalk region on their way to the base of the diencephalon. Growing axons interact tightly with stalk cells, which in turn, differentiate into glial cells whose processes hold the axon bundles and provide the outer cell layer of the nerve (Horsburgh and Sefton, 1986). Optic nerve glial cells originate from the in situ proliferation and differentiation of optic stalk cells, except for the oligodendrocytes, which migrate postnatally into the optic nerve from diencephalic regions. At the time of the analysis, the only glial cells expected to be in the optic nerve are those derived from the optic stalk. In the mutant, axons migrate as expected through the optic disc, however, we do not detect any cell body in the optic nerve between the axons (Fig. 4F,H). The mutant optic nerve consists of a single bundle of axons and has a reduced diameter, probably because of the absence of cell bodies between the axons, but a lower number of axons is also possible. Therefore, all glial cells derived from the optic stalk are lost in the mutant optic nerves; the outer sheath of glial cells is replaced by pigmented cells and the stalk-derived inner glia is missing.

We have examined the trajectory of the optic fibers in Pax2 mutants as they travel towards the midline to insert into the brain. Normally, leading migrating optic fibers reach the midline at the optic recess region at the base of the diencephalon (Horsburgh and Sefton, 1986; Marcus and Mason, 1995). At this point, axons make a ‘decision’; some fibers will cross the midline and constitute the contralateral projection and some will turn and form the ipsilateral projection; altogether they constitute the optic chiasm (Godement et al., 1990). In the mouse, roughly 95% of the fibers project contralaterally in normal animals. In contrast, Pax2 mutants have no optic chiasma and all axons project ipsilaterally without reaching the midline (Fig. 5). The optic fibers appear to turn without reaching the region of the optic chiasm, such that the insertion of the nerve occurs more laterally than in the wild type. We have also performed retrograde labelling experiments at E16.5, to explore the possibility of the presence of a minority of axons projecting contralaterally, but were unable to identify any (data not shown). The observed phenotype is not likely to result from defects in the choice between the ipsilateral or the contralateral side that takes place at the midline, since axons never reach the optic chiasma region. Rather, it appears to result from the abolishment of the ventral pathway to the midline so that the fibers never have the chance to choose sides and all project ipsilaterally.

Cellular cues in the optic stalks and at the ventral midline of the diencephalon have been implicated in the guidance of growing retinal axons towards the optic recess area to form the optic chiasma (Marcus and Mason, 1995). We have analyzed genes expressed at the ventral midline of the diencephalon in order to determine the possible defects in this area. Hedgehog proteins are a family of signalling molecules with patterning activities in a wide range of organisms (Fietz et al., 1994). Some of the members of this family are expressed in the ventral midline of the vertebrate neural tube. We have analyzed the expression of one of the members of the family, Sonic-hedgehog (Shh) (Echelard et al., 1993), which is expressed in the ventral forebrain area and compared it to Pax2 expression. At E9.5, Shh is expressed in the ventral forebrain from the ventral hypothalamus continuously up to the preoptic area. Pax2 is expressed in the optic stalks, stopping sharply at the limit with the diencephalon, thus avoiding the hedgehog-expressing area (Fig. 6C). At this early stage, we did not find any difference in the expression pattern of Shh between mutant and control embryos (Fig. 6A,B). At E11.5, just before the first axons start to migrate out from the retina towards the optic recess, the expression patterns of both Shh and Pax2 change drastically. Shh expression has split in two, leaving a gap devoid of expression, whereas Pax2 expression has extended medially into the gap left by Shh, so that the expression pattern is now continuous ventrally, yet still not overlapping the Shh expression domain (Fig. 6D,F). This gap is the only interruption of the Shh expression domain along the A-P axis and corresponds to the optic recess, the area through which the first retinal axons cross the midline. Strikingly, in Pax2 mutants, the gap, which is normally devoid of Shh expression, is lost, so that the Shh-expressing regions at the preoptic area and the ventral hypothalamus join together and there is no longer an interruption in Shh expression along the A-P axis (Fig. 6E,F). Furthermore, some examples of ectopic Shh-expressing cells in the ventroproximal region of the optic stalks can be identified (Fig. 6I). This phenomenon occurs in the absence of any obvious A-P extension of the normal Shh expression domains, either in the preoptic area or in the ventral hypothalamic area, suggesting that both domains come together as a consequence of the loss or displacement of the Shh non-expressing cells from the optic recess area.

Pax2 begins to be expressed very early during inner ear development. From the onset of its expression, it is restricted to the ventral half of the developing vesicle from which the cochlear and saccular regions derive (Nornes et al., 1990; Torres and Gruss, unpublished data). Invagination of the otic vesicle progresses normally in the mutants and neuroblastic precursors of the vestibulo-acoustic neurons segregate normally from the vesicle (not shown). However, as shaping of the vesicle proceeds to give rise to the different elements of the ear, the vestibular regions of the ear differentiate properly with only minor shape differences compared to controls, but the cochlear duct fails to extend from the ventral portion of the vesicle, thus remaining unshaped (Fig. 7G,H). As cell differentiation progresses, sensory cells of the vestibular portion of the inner ear differentiate normally and establish their innervation with the vestibular ganglion (Fig. 7A,B). In contrast, the cochlear region remains as a ball-shaped simple epithelium, where none of the organ of Corti cell types can be recognized (Fig. 7C,D). In addition, the cochlear ganglion is absent and therefore no innervation of the cochlear region occurs (Fig. 7E,F). The defects correlate with homozygosity for the Pax2 mutation and not with the exencephaly, since heterozygous exencephalic embryos do not show obvious alterations of inner ear development.

We have analyzed the phenotype of a complete loss-of-function mutation for the paired box-containing gene Pax2 in the central nervous system and sense organs. The defects observed in mutant mice severely affect the regionalization and morphogenesis of the eye, inner ear and midbrain. Pax2 is required within these organs for at least two apparently unrelated functions: morphogenetic functions, such as the closure of the neural tube and the optic fissure, and regional specification, such as the determination of fates along the proximodistal axis of the optic evagination or the development of auditory regions of the inner ear. Outside the nervous system, Pax2 function is essential for the differentiation of all intermediate mesoderm-derived components of the urogenital system (Torres et al., 1995). Multiplicity of functions is not infrequent among transcription factors with roles in embryonic development and suggests that many of them may act as multipotent switches such that a single molecule is able to instruct cells into different pathways depending on the cell history and position in the embryo. Pax genes represent a particularly good example of this multiplicity since many of its members are involved in multiple apparently unrelated developmental functions (see Chalepakis et al., 1992 and Strachan and Read 1994 for review). Pax6 has been shown to be essential for eye and nose development (Hill et al., 1991; Ton et al., 1991). Pax3 is required for neural crest and premyoblast migration (Epstein et al., 1991; Bober et al., 1994). Pax5 is required for midbrain patterning and B cell development (Urbanek et al., 1994) and Pax1 is required for paraxial mesoderm (Balling et al., 1988) and thymic development (Wallin et al., 1996). Phenotypic analysis of Pax2 mutations show that this gene also performs multiple functions in the development of the CNS, sense organs and the urogenital system and, in addition, that Pax2 locus is haplo-insufficient (Keller et al., 1994; Sanyanusin et al., 1995; Torres et al., 1995 and this report). Haplo-insufficiency is a characteristic shared by other three members of the family, suggesting that proper dosage of Pax proteins is essential to provide wild-type function.

Comparison between the three Pax2 mutant models is hampered by the different nature of the mutations, the constrain to the heterozygous condition in two of the models and the different level of the analysis. In the three cases, the neural phenotype found in the heterozygous condition is extremely variable and background-dependent. The main defect observed in the human mutant condition is the optic nerve coloboma, a defect present in some Pax2^+/− mice (data not shown) but not found in the Krd mouse. Both the Krd mice and the human mutants show abnormal electroretinograms indicating retinal defects, which appear to be more extensive in the mouse, and related to hypocellularity in the nuclear layers of the retina (Keller et al., 1994; Sanyanusin et al., 1995). In addition, we have found background-dependent exencephaly in Pax2^+/− mice, a defect not found in either the Krd mice or in humans. A possible explanation for these differences is that, besides Pax2, other genes deleted in the Krd mutation are contributing to its mutant phenotype. In addition, species-specific characteristics may explain the differences observed between mice and humans. Finally, it is likely that the contribution of genetic modifiers, differently represented in mouse lines, enhance different aspects of Pax2 function. This aspect is clearly illustrated by the different phenotypes found in the urogenital and nervous systems for Krd and Pax2-targeted mice under different backgrounds (Keller et al., 1994; Torres et al., 1995).

Our analysis uncovers new functions related to the regional specification within the eye and the inner ear. Inner ear development is extremely complex as it depends on both morpho-genetic events and a variety of cell differentiation programs. Added to this complexity, very little is known about the early patterning molecules involved. Pax2 mutation completely blocks the differentiation of the auditory portion of the inner ear, where under normal circumstances, it is expressed from the otic placode stage onwards. Several homeobox-containing genes expressed in the vestibular portion of the inner ear have been described (Ekker et al., 1992; Deitcher et al., 1994; Rinkwitz-Brandt et al., 1995); however, Pax2 is the only transcription factor expressed in the entire auditory part of the ear. Interestingly, a vestibular-specific gene, Nkx-5.1 is expressed very early in the otic placode in a complementary manner to Pax2 (Rinkwitz-Brandt et al., 1995). In vitro culture of explanted portions of the early otic vesicle, and assesment of their developmental potential, has suggested the early segregation of the domains and developmental programs for vestibular and auditory regions of the inner ear (Streeter, 1914; Fell, 1928; Li et al., 1978). Even though Pax2 is expressed in some vestibular regions, like the sacculus, we have not observed defects in the differentiation of saccular macula and its innervation; therefore, the role of Pax2 appears restricted to cochlear regions and it is likely that other transcription factors are specifically involved in the development of vestibular regions. The strong agenic phenotype of the Pax2 mutation and its early expression in the auditory primordium suggests a high position for this gene in the hierarchy of genes controlling the developmental program of this region. In addition, our results strongly suggest that the hearing defects observed in the human syndrome (Sanyanusin et al., 1995) are due to Pax2 malfunction.

As in the inner ear, Pax2 is expressed in a restricted manner beginning very early in the development of the eye primordium. Its expression is initially confined to the ventral half of the optic evagination, a region that does not correlate with any of the classical developmental compartments of the eye. The expression is very dynamic: in the optic cup, it is rapidly restricted to the cells arround the optic fissure as it forms. The phenotype observed suggests that the only role of Pax2 in the optic cup is related to the closure of the optic fissure and the ventral restriction of its early expression only indicates the future position of the fissure. Besides this role in the closure of the optic fissure, the functions of Pax2 in eye development appear mainly related to the development of the optic stalk into the optic nerve. After the formation of the optic cup, Pax2 is expressed in the entire optic stalk with a sharply defined limit at the pigmented retina. The mutant phenotype shows that Pax2 defines the optic stalk domain, restricting the pigmented retinal cell type to the optic cup. This may result from a transformation in the fate of cells located in the optic stalk region. Alternatively, it may be due to overproliferation of the retinal compartment, with subsequent invasion of the optic stalk region. In either case, Pax2 is required in stalk cells for them to originate glial cells, either by determining the precursors or by building up the region from which those precursors derive.

Most interestingly, our results show a role for Pax2 in the establishment of axonal pathways along the optic stalks and ventral diencephalon. The expression of Pax2 in the optic fissure extends well into the optic stalk where it strongly labels the invaginated ventral cells. These cells play an important role in the formation of the optic nerve since the first ingrowing axons from the retina interact tightly with their membranes on their way to the optic chiasm (Horsburgh and Sefton, 1986). Retinal neurons transplanted to different regions of the head project their axons directly to the tectum without crossing the optic chiasma, unless the transplantation is made close to the optic stalk, in which case some are incorporated into the optic stalk and cross the midline at the optic chiasma. Therefore, cellular cues rather than long-range attractant substances appear to guide the retinal axons to the optic chiasma (Harris, 1989). Pax2 expression correlates well with the distribution of stalk cells involved in the interaction with the first retinal axons. Moreover, its expression extends across the ventral midline at the optic recess area just before the axonal growth into the midline. In Pax2 mutants, optic tracts do not grow towards the midline but grow directly towards the midbrain. A possible explanation for this phenotype is the extension of pigmented cells into the optic stalk. Pigmented cells are repulsive for retinal axons (Silver and Sapiro, 1981) and, therefore, it is not surprising that axons grow in the mutant as a single bundle without interaction with the surrounding pigmented cells. The situation in Pax2 mutant mice would therefore be similar to retinal transplantation experiments, where axons lack the proper cellular environment and choose the shortest way to the tectum. This explanation would not imply any direct function for Pax2 in the establishment of the optic pathway, instead the abnormal trajectory would be a consequence of the extension of the pigmented cells into the stalk region. However, the pigmentation of optic stalk cells is only partial in some genetic backgrounds, but still the projection of the optic fibers is completely abnormal. In addition, Pax2 expression appears specifically in those optic stalk cells that interact tightly with pioneering axons and extends into the optic recess region just before axons enter it. Besides, pigmentation of the optic stalks never extends into the optic recess, but still there is a deletion or displacement of tissue at the optic recess region that brings together the preoptic area with the ventral hypothalamic region. Therefore, the results do not support a major involvement of the abnormal pigmentation in the aberrant projection of the nerves. The agenesis of the optic chiasma in Pax2 mutants is most likely due to the early deletion of the ventral region of the optic stalks and diencephalon, illustrated by the collapse of the Shh-negative area at the optic recess region.

We summarize our conclusions in a model for the role of Pax2 in the development of the optic chiasma represented in Fig. 6J. According to this model, Pax2-expressing cells represent a favourable environment to sustain axonal growth. Before axons reach the optic recess area, a Pax2-expressing domain has to build up across the midline in order to allow axons to reach the chiasma region. Based on the dynamics of Pax2 and Shh expression and on the mutant phenotype, we propose that the generation of the chiasma domain occurs by intercalation of Pax2-expressing cells from the optic stalks across the midline resulting in a separation of the Shh expression domain. In the absence of Pax2, cells from the stalk are not be able to intercalate due to either migration or proliferation defects. Besides creating a favourable environment for axonal growth, the establishment of a bridge of Pax2-expressing cells across the midline may build up a region devoid of signals inhibitory for retinal axon growth. In fact, negative signals have been shown to play a critical role in determining the pathway of the growing fibers into the chiasma region. The cell adhesion molecule CD44 is expressed at E12 in the ventral diencephalon just posterior to the optic recess region in an inverted V shape, strikingly coincident with the expression domain of Shh (Sretavan et al., 1994; Marcus and Mason, 1995). CD44 represents a repulsive signal for retinal axons that grow along the edges of the CD44-positive region thus forming the typical X-shaped optic chiasma. The deletion of the optic recess area in the mutants places the Shh-expressing cells, and presumably those expressing CD44, at the region where axons should cross. Considering the repulsion that CD44 exerts on retinal axons, the presence of CD44-positive cells in that region would be sufficient to explain the mutant phenotype; however, with Shh being a signalling molecule, we can not exclude that it may also represent a negative signal for retinal axon growth.

Ectopic expression of Shh in zebrafish has suggested a model for long-range patterning of the optic vesicle (Ekker et al., 1995; Macdonald et al., 1995). In this model, Shh would promote fates proximal to the midline by activating Pax2. The situation appears to be more complicated than suggested in this model since it does not account for the observation that, at least in mice, Pax2 expression extends all along the proximodistal axis. In addition, we show here that the mRNA expression domains of Pax2 and Shh do not overlap at any time of development suggesting that Shh expression may result in switching off Pax2.

The boundary between hindbrain and midbrain is a major landmark within the developing neural tube. It is characterized by both morphological traits and the remarkable segmental expression of several regulatory molecules in the area like Wnt-1, En-1, En-2 and Pax5 (Wilkinson et al., 1987; Davis et al., 1988; Asano et al., 1992). Targeted mutations for these genes produced deletions of different portions of tissue around the meten-mesencephalic area, demonstrating their role in the regional specification of this area of the brain (McMahon and Bradley, 1990; Joyner et al., 1991; Urbanek et al., 1994; Wurst et al., 1994). The early expression of Pax2 in the mid-hindbrain region led to the proposal that it may act on the top of the hierarchy of genes involved in the regional specification of the area (Rotwich and McMahon, 1995). Our results challenge this view, since the Pax2 mutation does not produce any alteration in the early expression pattern of regulatory genes expressed in the area. Besides, the Pax2 mutant phenotype results in the opening of the neural tube, showing that the main function of Pax2 in the brain is related to the morphogenetic movements of the neural plate that allow its closure. In fact, the expression of Pax2 is extraordinarily dynamic in the area and disappears as the neural tube closes. Tissue recombination experiments have shown that the onset of the expression of En genes is dependent on the interaction with the underlying mesendoderm (Ang and Rossant, 1993). Similarly, it may be possible that the initial activation of other genes expressed in this area is achieved independently by extrinsic factors, and subsequent regulatory cross-talk between them, refines and stabilizes the expression patterns. Our results do not exclude a possible role for Pax2 in the maintenance of gene expression at this boundary. The identification of such a role would only be possible in the analysis of a later stage, not accessible in Pax2 mutants due to the exencephaly.

We are grateful to S. Mashur, R. Scholz and S. Geisendorf for invaluable technical assistance and R. Altschäffel for excellent photographic work. We thank E. Stuart and P. Bovolenta for critical reading of the manuscript and A. Stoykova for stimulating discussions. M. T. received support from EMBO and European Community Human Genome Analysis program.