Midline signalling is required for Pax gene regulation and patterning of the eyes

The development of the vertebrate eye involves a hierarchy of inductive interactions by which the retina, optic stalk and lens progressively become specified within the developing forebrain (review: Saha et al., 1992). Inductive events during gastrulation establish the rostrocaudal and dorsoventral axes of the neural plate (Doniach, 1993; Ruiz i Altaba, 1994) and by the end of gastrulation, further specification results in the local-isation of the presumptive retinal fields to lateral regions of the rostral neural plate (Saha et al., 1992). During neurulation, the optic vesicles evaginate from the forebrain and make contact with the overlying surface ectoderm, remaining attached to the rostral diencephalon via the optic stalks. The optic vesicles subsequently invaginate to form optic cups with their charac-teristic double layer of neural retina and pigment epithelium (Grant et al., 1980) and the surface ectoderm which contacts the optic cups gives rise to the lens placodes (Grainger, 1992). The optic stalks progressively decrease in size (Schmidtt and Dowling, 1994) such that eventually only the optic nerve connects the eyes to the forebrain.

The identity of molecules involved in the early inductive events which determine the eye fields are largely unknown, although a number of potential regulatory genes are expressed in optic tissue during the formation of the eye (reviewed by Beebe, 1994). These include two members of the paired box (Pax) gene family, pax6 and pax2. In zebrafish (Krauss et al., 1991b,c; Püschel et al., 1992a; Macdonald and Wilson, unpub-lished results) and mice (Püschel et al., 1992; Walther and Gruss, 1991), pax6 is expressed throughout the optic vesicles in all cells of the prospective retina, pigment epithelium and lens epithelium. Mutations in Pax6 result in severe visual defects in Small eye mice (Hill et al., 1991; Hogan et al., 1988; Schmahl et al., 1993), rats (Fujiwara et al., 1994; Matsuo et al., 1993) and in humans with aniridia syndrome (Glaser et al., 1994; Jordan et al., 1992; Ton et al., 1991) demonstrating that this gene is crucial for normal eye development. Recently, it has been shown that the Drosophila homologue of pax6, eyeless, is also required for eye development (Quiring et al., 1994) and ectopic eye structures can be induced by mis-expression of eyeless in various imaginal disc primordia (Halder et al., 1995). Although Pax6 is required for eye devel-opment, its precise role and mechanism of action are not under-stood. In contrast to Pax6, Pax2 is restricted to the optic stalks and retinal cells around the choroid fissure (Krauss et al., 1991a; Püschel et al., 1992b). The function of Pax2 during development of the visual system is not known although it has recently been shown that mutations in the PAX2 gene can cause optic nerve colobomas in humans (Sanyanusin et al., 1995).

Signals derived from axial midline mesoderm and ventral midline CNS induce the differentiation of ventral cell types in the spinal cord (Placzek et al., 1993; Yamada et al., 1993). The possibility that similar signals may be involved in patterning the CNS in more rostral regions is suggested by the phenotype of zebrafish embryos homozygous for the cyclops mutation in which midline signalling is disturbed and ventral forebrain development is abnormal (Hatta et al., 1991, 1994; Macdonald et al., 1994). Mutant embryos exhibit fusion of the eyes, reduction in size of the diencephalon and loss of the floorplate in more caudal regions. Cell transplantation experiments have suggested that the cyclops gene product is involved in a sig-nalling pathway between mesoderm and ventral neuroecto-derm (Hatta et al., 1991, 1994). Supporting this hypothesis, there is no expression of the midline signalling molecule shh within the neuroectoderm of cyclops mutant embryos during early stages of development (Krauss et al., 1993; Barth and Wilson, 1995). shh is a vertebrate homologue of the Drosophila hedgehog gene which encodes a secreted protein involved in a variety of signalling events including floorplate induction and anteroposterior patterning of the limb bud (Fietz et al., 1994). Shh and other Hedgehog (Hh) family proteins undergo autoproteolysis to generate two smaller peptides and it appears that all signalling activity resides in the amino-terminal cleavage product (Lee et al., 1994; Bumcrot et al., 1995; Fan et al., 1995; Fietz et al., 1995; Marti et al., 1995; Porter et al., 1995; Roelink et al., 1995).

The absense of shh from the neurectoderm of cyclops mutant embryos together with the defects in eye development raises the possibility that Shh may regulate some aspects of eye development. We have tested this hypothesis by examining the role of Shh in the regulation of Pax6 and Pax2 protein distri-butions and in the morphogenesis of the eyes. We find that in cyclops mutant embryos, cells in the normal position of the optic stalks differentiate as retina and not as optic stalks resulting in fusion of the eyes. At the molecular level, this is reflected by the near absence of Pax2 and the presence of Pax6 in the region that forms retina in place of optic stalk. Con-versely, overexpression of shh results in increased numbers of Pax2-containing cells and hypertrophied optic stalk-like tissue together with decreased numbers of Pax6 containing cells and reduced pigment epithelia and retinae. We suggest that Shh is involved in regulating the spatial distribution of Pax2 and Pax6 which in turn may regulate the subdivision of the optic primordia into optic stalks and retinae.

Breeding fish were maintained at 28.5°C on a 14 hours light/10 hours dark cycle. Embryos were collected from the colony by natural spawning and raised in 10% Hank’s saline. Embryos up to 24 hours (30 somites) were staged according to ‘The Zebrafish Book’ (Westerfield, 1993). Cyclops^b16 mutant embryos were obtained from a stock origi-nally provided by C. Kimmel and C. Nüsslein-Volhard. The cyclops mutation has a slightly variable phenotype in the head – the majority of the mutant embryos from our stock exhibit partial fusion of the two eye fields (synopthalmia) and many such embryos retain two lenses.

Standard procedures were used for antibody labelling (Wilson et al., 1990), frozen sectioning (Westerfield, 1993) and in situ hybridisation (Xu et al. 1994). For primary incubations, anti-Pax6 antibody (Macdonald et al., 1994) was diluted 1:400, anti-Pax2 antibody (Mikkola et al., 1992) was diluted 1:3000 and HNK1 was diluted 1:20. Control and experimental embryos were usually processed in the same tubes to ensure identical labelling conditions.

Living embryos were viewed in their chorions under differential inter-ference contrast optics as described by Xu et al. (1994).

shh RNA for injections was transcribed from the pSP64T-shh plasmid (kindly provided by J.-P. Concordet and P. Ingham; see Krauss et al., 1993). Methods for RNA preparation and injection are described by Barth and Wilson (1995). The cytoplasm of individual blastomeres of embryos at the 1-4 cell stage were injected with several picoliters at a concentration of 0.1 mg/ml which results in widespread although mosaic distribution of the injected RNA (see Barth and Wilson, 1995). For control injections, RNA encoding β-galactosidase was injected at the same or higher concentration as shh RNA.

Cyclops mutant embryos exhibit abnormal specification of the ventral CNS and fusion of the eyes. These defects are likely to arise because of deficiencies in axial mesoderm and in the sig-nalling pathway between dorsal midline mesoderm and ventral CNS (Hatta et al., 1991, 1994; Thisse et al., 1994). Supporting this hypothesis, shh expression initially fails to be induced within the ventral CNS of cyclops mutant embryos (Krauss et al., 1993). The large deficencies in ventral forebrain tissue of cyclops embryos have raised the possibility that the cyclopic eye might arise by fusion of the two eye primordia, ventral to the diencephalon (Hatta et al. 1994). However, we find that the initial fusion of the two eye primordia is around the rostral pole of the neural keel in the position normally occupied by the optic stalks and rostral hypothalamus and that the fused eye only later lies ventral to the brain due to subsequent rotation of the eye. Below, we describe the morphogenesis of the fused eye of cyclops mutant embryos as an understanding of this process is critical to the interpretation of changes in Pax protein distributions seen in these embryos.

Comparison of eye development in living wild-type (Fig. 1A,B) and cyclops mutant embryos (Fig. 1C,D) shows that the prospective retinae are fused around the rostral pole of the neural keel in mutant embryos (Fig. 1C,D). However, at early stages, more caudal/distal regions of the eyes remain in their normal positions lateral to the forebrain and are not fused ventral to the diencephalon (compare Fig. 1C,D with 1A,B) despite the large deficiency of CNS tissue that is present ventral to the mid- and caudal diencephalic neuroepithelium (Fig. 1E,F). The bridge of presumptive retinal tissue lies ventral to the telencephalon (Fig. 1E,F) and occupies the position of cells that normally form the optic stalks and ventral/anterior hypothalamus. Although invagi-nation of the retina is most prominent in the lateral regions of the fused eye vesicle, some invagination usually occurs all the way around the rostral pole of the forebrain, thus forming a horseshoe shaped retina continuous from the lateral parts of the retinal primordia around the rostral pole of the brain (Fig. 1D,F). Histological analysis (see below) indicates that the rostral bridge of tissue linking the lateral retinal primordia forms retina rather than optic stalk and rostral hypothalamus which normally occupy this region in wild-type embryos. Indeed, in cyclops mutant embryos, the rostral hypothalamus is absent (Fig. 1C) and we see no identifiable optic stalks such that the fused retina is in direct contact with the forebrain neuroepithelium (Fig. 2F and data not shown).

Although the initial fusion of the cyclopic eye is around the rostral pole of the brain, the fused eye eventually becomes positioned ventral to the forebrain (Hatta et al., 1994 and see Fig. 2F). The primary reason for this repositioning is the rotation of the eye in relation to the main body axis during the second day of development, such that the anterior region of the optic primordia becomes positioned ventrally (Ross et al., 1992; Schmitt and Dowling, 1994). The reduction of mid-dien-cephalic neuroepithelium in the brains of cyclops mutant embryos exacerbates the repositioning of the fused eye such that the temporal retina eventually comes to lie adjacent to the midbrain tegmentum (Fig. 2F).

In wild-type embryos, Pax6 is present within prospective eye tissue from the end of gastrulation, and continues to be expressed in all cells of the developing pigment epithelium and neural retina throughout the first day of development (Macdonald and Wilson, unpublished results and see Fig. 2A). In contrast to wild-type embryos (Fig. 2A), the Pax6 expression domain of cyclops mutant embryos extends from the lateral retinal primordia around the rostral pole of the neural keel including the medial regions of the optic primordia which normally consist of presumptive optic stalk cells (Fig. 2B-E; see also Hatta et al., 1994). The bridge of prospective retinal tissue linking the retinal primordia of cyclops mutant embryos colocalises with this region of ectopic Pax6 expression (compare Fig. 2B,C with 1C,D). In accordance with morphological observations, the bridge of Pax6-containing tissue linking the two eyes is restricted to the anterior/ventral neural keel and there is no retinal fusion ventral to the mid- or caudal diencephalon (compare Fig. 2D with Fig. 1F).

Despite the abnormal appearance of the fused eye, immuno-histochemical analysis with antibodies that reveal lamination and neuronal differentiation indicated that retinal differentiation did occur in the fused neural retina of cyclops mutant embryos (data not shown). However, within the bridge of retinal tissue across the midline, the pigment epithelium was incomplete, there was some disorganisation of photoreceptors and the fused neural retina was in direct contact with forebrain neuroepithelium.

The results presented above suggest that cells within medial regions of the optic primordia of cyclops mutant embryos contain ectopic Pax6 and may be respecified to forming retina in place of optic stalk tissue. If optic stalk tissue is replaced by retina, it might be expected that genes normally expressed in the optic stalks would fail to do so in cyclops embryos. In wild-type embryos, Pax2 protein is detected at low levels in a small number of cells at the anterior/medial region of the optic vesicle from the 6-8s (somite) stage and continues to be expressed at higher levels within the presumptive optic stalks and some cells of the ventral/anterior retina throughout the first day of devel-opment (see for example Figs 2G and 7A). Thus the restriction of Pax2 protein to the medial part of the optic primordium is almost complementary to the distribution of Pax6 within more distal cells (Fig. 7A and compare Fig. 2G with 2A; Krauss et al. 1991a; Mikkola et al., 1992; Püschel et al., 1992b). As predicted, cyclops mutant embryos exhibit an almost complete absence of Pax2 protein in the optic primordia (Fig. 2H) while the midbrain expression of Pax2 appears unaffected (Fig. 2H). This reduction in Pax2 is more dramatic than has previously been observed (Hatta et al., 1994). While we are not certain of the reason for this difference, it could be that the study by Hatta et al. used cyclops mutant embryos which exhibited a less severe phenotype than the embryos from our own stock of cyclops carrier fish.

Our results indicate that the initial site of retinal fusion in cyclops mutant embryos is around the rostral pole of the neural keel, linking the anterior/medial retina on both sides of the brain. If this is correct, then the cyclops mutation might not affect the development of posterior parts of the retina. To determine if this is indeed the case, we examined the expression of two genes, rtk2 and msxC, normally transcribed in restricted domains of the developing retina. Rtk2 is a member of the Eph family of receptor tyrosine kinases expressed throughout the temporal half of the retina from the posterior groove dorsally to the choroid fissure ventrally (Fig. 3A; Macdonald et al., 1994; Xu et al., 1994). msxC (Ekker et al., 1992) is a homeobox-containing gene that is expressed in the caudal part of the retina, around the posterior groove.

The boundary of rtk2 expression between nasal and temporal retina at the posterior groove was unaffected by the cyclops mutation (compare Fig. 3A and B). However, in support of our interpretation of the cyclops retina, rtk2 expression in the anterior/ventral region of the temporal retina was continuous around the rostral pole of the neural keel (Fig. 3C). Similarly, msxC expression around the posterior groove of the retina was unaffected in cyclops mutant embryos (Fig. 3F).

The results presented above suggest that a signal emanating from the midline and absent in cyclops mutant embryos may be required to regulate the distribution of Pax2 and Pax6 within the developing eye primordia. Shh is a signalling molecule (Johnson and Tabin, 1995) that is expressed along the ventral midline of the CNS (Fietz et al., 1994) including cells at the base of the optic stalks (Fig. 7A; Krauss et al., 1993; Barth and Wilson, 1995). In cyclops mutant embryos, shh transcripts are absent from the neu-roepithelium at early developmental stages (Krauss et al., 1993). To test whether Shh is able to regulate the spatial distribution of Pax6 and Pax2 in the optic primordia, it was overexpressed in the developing CNS. 380 embryos injected with shh RNA were analysed for changes in gene expression and/or changes in eye morphogenesis. In support of previous observations which showed that overexpression of shh leads to eye defects (Krauss et al., 1993; Barth and Wilson, 1995), we found that approxi-mately 90% of injected embryos had readily detectable changes in gene expression and morphogenesis of the eyes (see below). In contrast, embryos injected with RNA encoding β-galactosi-dase failed to show any comparable alterations in eye develop-ment or in gene expression patterns (data not shown and see Barth and Wilson, 1995).

Although optic vesicle evagination invariably did occur in shh injected embryos, the size of the optic vesicle was generally reduced as compared to wild type embryos (compare Fig. 4B with 4A). Confirming previous observations (Barth and Wilson, 1995), injection of shh RNA led to a failure of separation of the eye primordium from the diencephalon such that the eye and the diencephalon were fused over a large region. As we describe in detail below, we interpret this phenotype as being caused by hypertrophy of the optic stalks at the expense of pigment epi-thelium and neural retina. Defects in retinal development were readily apparent in living shh injected embryos such that retinal tissue, most obviously pigment epithelium, was reduced and occasionally absent (Fig. 4C-E).

Pax6 and Pax2 distributions were analysed in shh-injected embryos fixed between 6-8s and 30s (24 hours). In 89 of 101 embryos examined in detail, overexpression of shh led to major reduction in the number of Pax6-containing cells within the optic primordia at all stages of development examined (Fig. 5A and compare Fig. 5C with 5B). The remaining Pax6 protein was distributed in more distal/posterior regions of the optic primordia (Fig. 5C). Pax6 was also reduced in other regions of the embryo including the lens epithelium, dorsal diencephalon (Fig. 5C) and rhombomeres of the hindbrain (not shown).

In marked contrast to the observed reduction in Pax6, Pax2 was widely ectopically expressed in the optic primordia in 80 of 91 shh-injected embryos (Fig. 5D and compare Fig. 5F with 5E). Ectopic Pax2 was present thoughout much of the optic primordia although it was usually absent from the posterior/dis-talmost regions that retained pax6 expression (compare Fig. 5F with 5C). Widespread ectopic Pax2 expression was not observed elsewhere in the embryo although we did not analyse other regions of Pax2 distribution in detail (Fig. 6A).

To confirm that the changes in Pax protein distribution reflected changes in RNA distribution, we examined a further 100 embryos using in situ hybridisation to determine the dis-tribution of pax6 and pax2 transcripts. Complementing the changes in protein distribution in the optic primordia, the number of pax6-expressing cells was reduced whereas the number of pax2-expressing cells was increased (Fig. 5G-J).

The region of attachment of the optic primordia to the diencephalon was greatly increased in shh-injected embryos as compared to wild-type embryos (compare Fig. 5F with 5E). During later stages of development this region of the optic primordia of injected embryos continued to express Pax2 and remained as a grossly oversized optic stalk-like structure (Fig. 6E). The optic stalks of wild-type embryos are only a few cell diameters thick by 24 hours (Fig. 6B,D) whereas the stalks linking the reduced retinae of injected embryos to the forebrain are much broader (compare Fig. 6C with 6B, and 6E with 6D). Optic cup and lens formation were restricted to areas of the optic primordia distal and posterior to the hypertrophied optic stalk-like structures (Fig. 6C).

To determine if the reduced domain of Pax6-expressing cells differentiated as retina in shh-injected embryos, we examined eyes at stages when pigment epithelium and neural retina have normally differentiated. We found that differentiation of both layers of the retina did occur (Fig. 6F-H). However pigment epithelium was more severely reduced than neural retina (Fig. 6F-H). This is perhaps not surprising as prospective pigment epithelial cells are likely to arise from more medial parts of the optic primordium (which showed the most severe changes in Pax protein distribution) than prospective neural retinal cells (Fig. 7A). The proximal extent of pigment epithelial differentiation coincided with the distal limit of Pax2 expression (Fig. 6F). The reduced neural retina showed normal lamination, however, the retinal ganglion cell layer appeared reduced as compared to the photoreceptor and inner nuclear layers (Fig. 6G,H).

Based upon their location in the distal/posterior region of the optic primordia, we assumed that the retinal cells that remained in shh-injected embryos may cor-respond to cells in the dorsal and temporal region of the normal eye. To test this possibility we assessed the expression of rtk2 and msxC in the eye primordia of shh-injected embryos. rtk2 was widely expressed in the eye primordia of all embryos examined suggesting that retinal cells with temporal identity were present (n=48, Fig. 3E). In contrast, msxC, which is normally expressed in a narrow region of the dorsal retina, was absent from the optic primordia in 23 out of 24 shh-injected embryos (Fig. 3G,H). However, this result may not imply that the cells that normally express msxC do not form retina in injected embryos as there was considerable downregulation of msxC thoughout the embryo, particulary in the head (Fig. 3G,H).

The distributions of Pax proteins in wild-type, cyclops and shh-injected embryos are summarised in Fig. 7.

In this study we show that the partitioning of the optic primordium into optic stalk and retina is dependent upon midline signalling. Two genes potentially involved in subdividing the optic primordia are pax6, which is expressed in the presumptive pigment epithelia and neural retinae, and pax2, which is expressed primarily in the presumptive optic stalks. We show that expression of these two closely related genes is affected in a complimentary manner by the midline signalling molecule Shh. Overexpres-sion of shh leads to ectopic expression of pax2 and complimentary loss of pax6 expression. In contrast, cyclops mutant embryos which lack early CNS expression of shh (Krauss et al. 1993) exhibit reciprocal changes in gene expression such that the domain of pax2 expression is reduced whereas pax6 expression is expanded.

If Shh is directly responsible for regulating Pax2 distribution then it may act over considerable distance as the distalmost Pax2-containing cells within the optic primordia are many cell diameters from the site of shh expression at the base of the optic stalks (Fig. 7A and see Barth and Wilson 1995). However, it is possible that presumptive optic stalk cells may be closer to the site of shh expression at the time at which they initiate pax2 expression and that subsequent migration takes them to more distal regions of the optic primordia. There is evidence that members of the Hedgehog family in vertebrates and inverte-brates can have both short range and long range signalling abilities. For instance, the amino-terminal peptide of Drosphila Hh acts as a short range signal that regulates the transcription of wingless (wg) and decapentaplegic (dpp) (Basler and Struhl, 1994; Ingham and Fietz, 1995; Fietz et al., 1995; Porter et al., 1995) and in vertebrates, Shh has been implicated in the contact-dependent induction of floorplate cells by the notochord (Roelink et al., 1994, 1995; Marti et al., 1995). Several lines of evidence have recently suggested that the amino-terminal cleavage product of Shh may be diffusible and may therefore also have longer range signalling ability. For instance, Shh can induce sclerotomal markers in adjacent presomitic mesoderm over distances up to about 150 μm and can still induce these markers when separated from the presomitic tissue by a nucle-opore filter (Fan and Tessier-Lavigne, 1994; Fan et al., 1995).

Furthermore, Shh can induce motor neuron differentiation in a dose-dependent manner that is independent of cell contact (Marti et al., 1995; Roelink et al., 1995; Tanabe et al., 1995). In addition, it has recently been shown that cells expressing Shh can also induce neuronal differentiation within forebrain tissue in the absence of cell contact, suggesting that Shh may be able to diffuse through the forebrain neuroepithelium (Ericson et al., 1995) and hence may be able to regulate pax gene expression at a distance from the site of shh expression.

An additional mechanism by which Hh proteins may elicit long range responses is through the local induction of other sig-nalling molecules that subsequently either diffuse over greater distances or establish a cascade of cell to cell signalling events (Johnson and Tabin, 1995). However, although homologues of the signalling molecules wg and dpp, targets of Hh regulation in Drosophila (Ingham and Hidalgo, 1993; Basler and Struhle, 1994; Ingham and Fietz, 1995) are expressed in the developing vertebrate embryo, we are currently unaware of any that are expressed in regions of the forebrain appropriate to play a role in the spatial regulation of pax gene expression in the eyes.

It is interesting to note that some of the changes in eye devel-opment that occur following shh injection resemble the changes that occur in embryos that have been exposed to altered levels of retinoic acid during the early period of eye development. For instance, inhibition of retinoic acid synthesis results in embryos which lack ventral/anterior retina (Marsh-Armstrong et al., 1994) while treatment of embryos with retinoic acid results in thickened optic stalks (Marsh-Armstrong et al., 1994; Dowling, Hyatt and Schmitt, personal communication). However, exogenous retinoic acid also causes retinal duplication, a feature we have not observed in our experiments. Despite this difference, the fact that retinoic acid can induce Shh expression within the limb bud (Riddle et al., 1993) raises the possibility that at least some of the effects of retinoic acid upon the developing eye may be elicited through alterations in the levels of Shh.

The domain of Pax6 protein distribution in the forebrain of cyclops mutant embryos is expanded to include cells of the rostral neural keel which normally form structures such as the optic stalks and rostral hypothalamus. One consequence of ectopic Pax6 expression appears to be that these rostral cells are respecified to become retina. An alternative possibility is that presumptive optic stalk cells are not present in the neural keel of cyclops mutant embryos and that Pax6 containing cells migrate from the existing retinal fields to the rostral midline thereby fusing the two eye primordia. We favour the first of these possibilities as the complementary changes in pax gene expression observed in cyclops and shh injected embryos favour a model by which midline signals regulate the expression of pax genes rather than the survival of retinal or optic stalk precursor cells. Supporting this possibility are experiments in the chick which have suggested that signals emanating from the notochord can inhibit pax6 expression in the ventral spinal cord (Goulding et al., 1993) – in the absence of these signals, pax6 expression spreads to more ventral cells which normally do not express the gene.

Our results suggest that the presumptive retinae of cyclops mutant embryos are expanded whereas the retinae of shh-injected embryos are reduced. These changes in the extent of retinal tissue are reflected by alterations in the distribution of Pax6 in the optic primordia, suggesting that Pax6 is a key regulator of retinal development. This conclusion is supported by analysis of the phenotype of Small eye mice which carry a mutation in the Pax6 gene and exhibit severe retinal defects in both heterozygous and homozygous embryos (Hill et al., 1991; Hogan et al., 1988; Schmahl et al., 1993). It is interesting that although retinae were always reduced in shh injected embryos, outgrowth of the optic primordia did occur even though Pax6 was reduced or absent from most presumptive retinal cells. This suggests that Pax6 may not be required for the initial formation of the optic vesicle, and indeed the optic vesicle does initially evaginate in Pax6 mutant mice (Hogan et al., 1988, Schmahl et al., 1993).

The relatively normal spatial distribution of rtk2 and msxC transcripts and appropriate lamination of neurons in the retinae of cyclops mutant embryos indicates that while the signalling pathway(s) affected by the cyclops mutation may be required to partition the optic primordia, they appear to be less critical for the differentiation of the retina. The consequences of shh over-expression upon the regional patterning of the remaining retinal tissue in shh injected embryos is less clear. Based upon the mor-phogenesis of the eye of shh-injected embryos, we believe that dorsal and temporal regions of the neural retina are least affected by Shh (this study and see Barth and Wilson, 1995). Supporting this interpretation, we found that the retinal tissue of shh-injected embryos always expressed the temporal retinal marker, rtk2. However, msxC, a gene normally expressed in the dorsal retina adjacent to the posterior groove, was absent from the retinae of shh-injected embryos. This could mean that cells that normally express msxC do not form retinae in shh-injected embryos. However, because we observed widespread overall reduction in msxC expression in dorsal regions of the embryo, we favour an alternative interpretation: that ectopic Shh may directly or indi-rectly downregulate the expression of msxC. Supporting the notion that ectopic Shh can elicit widespread inhibition of genes normally expressed in dorsal tissue, it has recently been shown that Shh can inhibit the expression of Pax3, Sim1 and Pax7 within dorsal regions of the developing somites (Fan and Tessier-Lavigne, 1994; Johnson et al., 1994).

The expression of Pax2 in the developing optic stalks raises the possibility that this protein may play a role in optic stalk differentiation analogous to the role of Pax6 in retinal devel-opment. Indeed, the expanded domain of Pax2 distribution following shh injection correlates with enlarged optic stalk-like structures. However, it is difficult to assess just how closely the hypertrophied optic stalk-like structures of shh-injected embryos resemble the normal optic stalks of wild-type embryos, as the normal development of optic stalk cells has not been well studied. Although it is known that the optic stalks of wild-type embryos decrease in diameter over time (Schmidt and Dowling, 1994), their ultimate fate has not been conclu-sively determined. We believe that it is likely that the medial regions of the optic primordia of shh-injected embryos do not form simply enlarged but otherwise normal stalk tissue. For instance, we know of at least one other gene which is abnor-mally expressed in this tissue: the homeobox gene nk2.2 is normally restricted to the most proximal cells of the optic stalks but following shh injection it is widely expressed in the hypertrophied optic stalk-like tissue (Barth and Wilson, 1995). Given the complementary distributions of Pax6 and Pax2 in wild-type, cyclops and shh-injected embryos it is tempting to speculate that there may be some regulatory interactions between these two closely related proteins. Indeed, in vitro binding assays of Pax6 and Pax2 have shown that the proteins can bind to similar DNA sequences (Epstein et al., 1994) raising the possibility that they may compete for the same target sequences in vivo. Perhaps the simplest model for an interac-tion between Pax2 and Pax6 would be one in which Pax2 represses the expression of Pax6 thus allowing cells towards the midline of the optic primordia to differentiate as optic stalk tissue and not as pigment epithelium or neural retina.

The effects of ectopic expression of Shh have been assessed at various different sites in the developing vertebrate embryo (for example, Roelink et al., 1994; Johnson et al., 1994). In each group of cells that are affected, the alterations in gene expression resulting from ectopic expression of shh appear to be quite specific. For instance, Pax2 is ectopically expressed within cells of the optic primordia but not in cells of the adjacent diencephalic tissue, and nk2.2 is ectopically expressed in both diencephalic cells and the optic primordia (Barth and Wilson, 1995), whereas the HNF-3β homolog, axial, is never expressed in the optic primordia but is ectopically expressed in other regions (Krauss et al., 1993; Barth and Wilson, 1995). These observations suggest that Shh can evoke a response in cells that lie outside its normal domain of function and that the consequences of Shh signalling are critically dependent upon the position of a cell within the developing embryo; thus the same signal can have very different consequences dependent upon the context of the responding cell.

Thanks to Steve Easter, Peter Hitchcock, Kohei Hatta, Chuck Kimmel, Lisa Leonard, John Scholes and Monte Westerfield for comments on various aspects of this work, and Phil Ingham and Jean-Paul Concordet for Shh and Monte Westerfield for msxC plasmids. This study was supported by grants from the SERC and Wellcome Trust. I. M. was supported by the Research Council of Norway. N. H. is a BBSRC Senior Fellow. S. W. was an SERC Advanced Research Fellow and is a Wellcome Trust Senior Research Fellow.

A recent independent study by Ekker et al., presents similar data and conclusions to the work in this paper (Ekker, S. C., Ungar, A. R., Greenstein, P., von Kessler, D. P., Porter, J. A., Moon, R. T. and Beachy, P. A. (1995). Patterning activities of vertebrate hedgehog proteins in the developing eye and brain. Current Biology (in press).