Requirement for the zebrafish mid-hindbrain boundary in midbrain polarisation, mapping and confinement of the retinotectal projection^*

Establishment of cell type diversity in the vertebrate neural plate requires an interplay between intrinsic and extrinsic molecular mechanisms, mediated by transcription factors and secreted patterning molecules (review: Lumsden and Krumlauf, 1996). In the embryonic midbrain, these mechanisms cause different cell types to arise at precise anterior-posterior and dorsal-ventral positions. In addition, cells of a given type often show a rostrocaudal gradient of cytodifferentiation, particularly in the midbrain tectum (La Vail and Hild, 1971), where they receive spatially ordered afferent inputs from the retinal ganglion cells (RGCs) and thereby form a retinotopic map (review: Udin and Fawcett, 1988; Holt and Harris, 1993; O’Leary et al., 1999). At the molecular level, midbrain polarity is reflected in the graded distribution of the Engrailed (En) homeobox transcription factors (Martinez and Alvarado-Mallart, 1990; Davis et al., 1991; Itasaki et al., 1991), and of ephrin-A2 (ELF-1) and ephrin-A5 (RAGS/AL-1), two glycophosphatidyl-inositol (GPI)-linked ligands for Eph family receptor tyrosine kinases (Cheng and Flanagan, 1994; Drescher et al., 1995). Via complementary gradients of tectal ligands and their receptors on ingrowing RGC axons, ephrins and their Eph receptors are thought to mediate the retinotopically organized projections of RGCs to their postsynaptic target cells in the tectum (reviews: Rétaux and Harris, 1996; Orioli and Klein, 1997; Flanagan and Vanderhaeghen, 1998; O’Leary et al., 1999). Misexpression of En-1 or En-2 in chick tecta suggest that they function upstream of the Ephrins (Logan et al., 1996; Friedman and O’Leary, 1996; Rétaux and Harris, 1996).

The molecular mechanisms which set up the graded distribution of these molecules are not known, but appear to be related to the initial formation of midbrain polarity. Tectum rotation experiments previously suggested that polarity becomes established prior to actual ingrowth of axons into the tectum, due to influences from adjacent cell populations (Nakamura et al., 1994). Two candidate cell populations located adjacent to the developing midbrain are at the forebrain-midbrain boundary (Chung and Cooke, 1978) and the midbrain-hindbrain boundary (MHB). When MHB tissue is transplanted to ectopic locations in the posterior forebrain, it can induce midbrain differentiation in surrounding host cells, which in turn can act as targets for RGCs. In addition to its inductive abilities, the MHB tissue also exerts a polarizing influence on the induced tissue: for instance, the induced host cells express Engrailed antigens at levels that decrease with distance from the graft. We will refer to this polarizing activity as midbrain polarizing activity, or MPA. The effects of transplanted MHB tissue can be mimicked by inserting beads soaked with Fgf8 (or Fgf4) protein (Crossley et al., 1996; Shamim et al., 1999), two secreted members of the fibroblast growth factor family, which are expressed in MHB tissue, but not by the adjacent tectum (Crossley and Martin, 1995; Reifers et al., 1998; Shamim et al., 1999; unpublished observations). Together with the results of misexpression experiments of Fgf8 in mice (Lee et al., 1997), this raises the possibility that Fgf8 and/or Fgf4 may correspond to the midbrain inducing activity and/or the MPA. While these experiments establish that MHB tissue or Fgfs are able to polarize in ectopic sites, the requirement for the MHB organizer or for Fgf8 in midbrain polarisation has not been examined.

Zebrafish embryos homozygous for the recessive acerebellar (ace) mutation lack the morphogenetic constriction (isthmic constriction), which we here refer to as MHB, between the midbrain and rhombomere one, whereas the adjacent midbrain is still present (Brand et al., 1996; Reifers et al., 1998; Brand, 1998 and unpublished observations). fgf8 and wnt1 expression, normally seen in a subset of the MHB cells which may mediate its organizing potential, is absent in acerebellar mutants, indicating that the MHB organizer itself is absent in acerebellar mutants. We have previously shown that acerebellar is a loss-of-function mutant of fgf8, and argued that Fgf8 functions during maintenance, rather than initial induction, of midbrain development (Reifers et al., 1998). In particular, the observation that Fgf8 is required to maintain expression of the three zebrafish engrailed genes (eng1 to eng3), wnt1 and other genes suggests that Fgf8 might be a component of the MPA emanating from the MHB.

Here, we use the retinotectal system as a fine-grained readout to assess midbrain polarity in acerebellar mutant embryos. The tectum is the largest retinofugal target of RGC axons in zebrafish, and as in other vertebrates, the projection is retinotopically organized, such that neighbouring RGCs in the retina connect to neighbouring tectal target cells (Stuermer, 1988; Burrill and Easter, 1994). During embryonic development, RGCs grow through the optic nerve, chiasm and optic tract to reach the tectum, where they synapse at the correct retinotopic target site. Initial growth to the target site is direct (Kaethner and Stuermer, 1992) and does not involve competition for target area or electrical activity (Harris, 1980; Stuermer et al., 1990; Kaethner and Stuermer, 1994). Within the tectum, three zebrafish ephrins, ephrin-A5a and ephrin-A5b, both related to mammalian ephrin-A5 (RAGS/AL-1) and ephrin-A2, related to murine ephrin-A2 (Elf-1) are distributed in increasing anterior to posterior gradients. Consistent with a function for these molecules in map formation, in vitro stripe assays (Walter et al., 1987) have shown that the zebrafish ephrins repel both temporal (Ephrin-A5b and Ephrin-A2) and nasal axons (Ephrin-A5b; Brennan et al., 1997). Moreover, targeted inactivation of murine ephrin-A5 leads to ectopic termination of RGCs in the superior colliculus, and to overshooting projections into the inferior colliculus (Frisén et al., 1998).

By studying the midbrain of acerebellar mutants, we show here that the MHB is required for anterior-posterior polarization of the midbrain retinotectal map, to restrict growth of RGC axons to the tectum, and for graded expression of ephrin ligands in the midbrain neuroepithelium prior to axonal ingrowth. Since Fgf8 is mutated in acerebellar, Fgf8 itself might be involved in establishment of midbrain polarity. Unexpectedly, our results also suggest that Fgf8 is required for normal retinal patterning and dorsal-ventral polarization of the tectum.

Zebrafish were raised and kept under standard laboratory conditions at 27°C (Westerfield, 1994, as described in Brand and Granato, 1999) and heterozygous carriers were identified by random intercrosses. To obtain homozygous mutants, carriers were crossed to each other. Embryos were incubated at 28.5°C in embryo medium with 0.2 mM PTU to prevent melanization and fixed according to hours of development and morphological staging criteria (Kimmel et al., 1995).

Whole-mount RNA in situ hybridisation (ISH; Reifers et al., 1998), receptor alkaline phosphatase staining with the chick EphA3/AP, zebrafish EphB4b/AP and Ephrin-A5b/AP fusion proteins on whole-mount embryos (Cheng and Flanagan, 1994; Brennan et al., 1997) and antibody stainings against acetylated tubulin (Macdonald et al., 1997) were described previously. Zebrafish ephrin-A5a, -A5b and -A2 were previously designated zfEphL2, -L4 and -L3, respectively (Brennan et al., 1997). Probably due to a partial genome duplication (Postlethwait et al., 1998), ephrin-A5a and ephrin-A5b are two separate genes that are both related to mammalian ephrin-A5.

For RGC axon tracing, larvae were fixed in 4% paraformaldehyde in PBS at 4°C over night, embedded in 2% LMP agarose (GIBCO BRL) on slides and labelled by inserting glass needles covered with either molten DiI or DiO (Molecular Probes D-282, D-275) into the retina. After overnight storage at room temperature in a dark moist chamber, larval brains were dissected and further processed for microscopy. Whole-eye-fills were done by pressure injection of saturated DiI or DiO solutions in chloroform into the eyes of fixed larvae. To determine their morphology, larval brains were dissected using watchmaker forceps and stained for DNA with 0.5 μM SYTOX (Molecular Probes S-7020) in PBS for at least 24 hours at room temperature. Brain morphology, retinotectal projection and ephrin RNA expression were analysed on fluorescent preparations with a LEITZ DM IRB confocal microscope, equipped with a TCS 4D Argon/Krypton laser and SCANware 5 software.

In situ hybridisations on high pec stage embryos (42-44 hours) were performed as described previously (Xu et al., 1994) except that embryos were developed after ISH with FastRed (Boehringer) as a fluorescent substrate. Thick sections, which bissected the tecta along the anterior-posterior axis in the direction of entry of the axons to the tectum, were cut using a sharpened tungsten needle and mounted for confocal analysis. The gradient along the anterior-posterior tectal axis at approximately the mid-point of each tectum was determined (mean of 5 scans) for at least ten mutant embryos and ten siblings for each probe. Confocal data were quantitated using NIH image and are presented in graphical form.

Fertilized eggs, to be used as donors, were fluorescently labelled by intracellular pressure injection of 10% tetramethylrhodaminedextran, 10,000 MW (Molecular Probes D-1817) in 0.25 M KCl and after 50% epiboly raised together with unlabelled host at 18°C to the 8-10 somite stage. For grafting, live embryos were embedded in 1.2% ultra pure LMP agarose in Ringer at 42°C (GIBCO) and operated in sterile zebrafish Ringer (Westerfield, 1994) in a Petri dish cover under a dissecting microscope. To access the optic vesicles, the epidermis was locally destroyed with a drop of light mineral oil (Sigma M-8410) and removed. Optic vesicles were grafted with electrolytically sharpened tungsten tools. Within the first hour after transplantation, when healing was complete, the embryos were removed from the agarose and raised under standard conditions in sterile Ringer until day 6. One day after transplantation, embryos were scored for position, orientation and vitality of the graft, using a fluorescence microscope.

We examined the retinotectal projection of ace mutant larvae after its establishment on day 6 of development, although similar defects to those described here are also apparent on day 3, albeit in a much smaller tectum (unpublished results). At day 3 of development, the mutant larvae were divided into two classes of different phenotypic strength: type A larvae (78%, n=235) show little overall malformations, whereas type B larvae (22%) are overall retarded compared to the wild type of the same age, probably as a secondary consequence of variably abnormal circulation (Brand et al., 1996; Fig. 1A,B). We examined the overall morphology of wild-type and mutant tecta in whole-mounted brains stained with the fluorescent DNA stain Sytox (Fig. 1C-G). As described previously, the cerebellum is missing in the mutants (Reifers et al., 1998). In type A larvae, the tectum is slightly altered in its shape, but not affected in its anterior-posterior length along the dorsal midline. Since the brain of type B larvae is overall reduced in size (Fig. 1E), tectal defects in type B larvae are likely to be influenced by secondary effects. Our subsequent analysis therefore focusses on type A larvae, although type B individuals show similar retinotectal defects (data not shown). We find that the termination zone of RGC axons in the tectum (neuropil) of type A individuals is reduced in size and located more posteriorly (Fig. 1F,G), as confirmed by anterograde labelling of the retinofugal projection with DiI, which also reveals that axons enter the tectum via abnormal brachia (Fig. 1H,I). In addition, the arborization field AF-7 at the diencephalic-mesencephalic boundary (Burrill and Easter, 1994) is enlarged in 8 of 11 mutants, and the space between AF-7 and the anterior neuropil margin is wider in 10 of 11 mutants examined. We believe that this phenotype reflects a more anterior character of the tectum in ace mutants (see below).

To assess the size and position of RGC axon termination fields on the tectum, we labelled subpopulations of RGCs anterogradely with DiI or DiO and found a pronounced, but partial disruption of the retinotectal map in ace tecta compared to the wild type. The terminations of wild-type RGC axons form an inverted map on the tectum, such that dorsal and ventral RGCs project to the ventral and dorsal tectum, respectively, and nasal and temporal RGCs project to the posterior and anterior tectum, respectively. Whereas the anterior-posterior position at which ventrotemporal RGC axons terminate in the dorsoanterior neuropil is almost correct (Fig. 2A,B), the tectal termination field of dorsotemporal RGCs is expanded from its normal ventroanterior position throughout the posterior neuropil in 12 of 17 ace larvae (Fig. 2C,D). Importantly, both temporal RGC subpopulations have partially delocalized projection fields along the dorsal-ventral axis of the tectum (Fig. 2B,D): dorsotemporal RGCs misproject to the dorsal neuropil in 13 of 17, and ventrotemporal RGCs to the ventral neuropil in 12 of 24 mutant larvae.

Whereas the termination fields of especially dorsotemporal RGCs are clearly delocalized along the anterior-posterior axis, nasal RGCs are predominately affected along the dorsal-ventral axis and appear to follow aberrant routes through the tectum. Whereas wild-type ventronasal RGCs terminate only in the dorsoposterior neuropil, ectopic terminations are also found in the ventroposterior tectum in 12 of 17 mutants (Fig. 2E,F), independently of the brachium they have chosen (Table 1, see below). Similarly, the dorsonasal RGCs, which in the wild-type terminate in the ventroposterior region of the neuropil, in 14 of 19 ace mutants form ectopic axon terminations in the dorsoposterior tectal neuropil (Fig. 2G,H).

To further investigate the relative localization of tectal termination fields in ace mutants, we performed double labellings of RGCs axons from neighbouring quadrants of the retina. In the wild type, RGCs from the dorsonasal and dorsotemporal retinal quadrants form non-overlapping tectal termination fields in the ventroanterior and ventroposterior neuropil (Fig. 2I). In ace mutants, the termination fields of these two RGC populations overlap in a medial position along the anterior-posterior axis of the neuropil (Fig. 2J). In contrast, the anterior-posterior extent of the termination fields in the dorsal tectum of ace mutants is similar to the wild type, and not overlapping, despite a prominent dorsal-ventral delocalization (Fig. 2K,L). In summary, the exact 1:1-representation of retinal positions on the tectum is lost in ace for all four analysed retinal quadrants due to mapping errors along both the anterior-posterior and dorsal-ventral axis of the optic tectum. The changes in anterior-posterior polarity are more pronounced in the ventral than the dorsal tectum.

In addition to mapping errors we find that all RGC subpopulations show ‘overshooting’ axonal projections beyond the posterior border of the tectal neuropil into the hindbrain (Fig. 2F,H). Overshooting RGC axons exit the neuropil preferentially at the ventroposterior margin of the tectal neuropil in 16 of 18 mutants, and only rarely in dorsoposterior regions of the neuropil (2 of 18 mutants); here, filopodia sometimes abnormally explore beyond the neuropil, although they are not elaborated into complete overshooting projections (not shown). Overshooting is most frequent for the dorsonasal RGC axons (12 of 19 mutants, Fig. 2H), which normally project to the ventroposterior part of the neuropil and thus are spatially closest to the favoured exit point. Ventronasal RGCs overshoot in 8 of 17 (Fig. 2F), dorsotemporal RGCs in 5 of 17 and ventrotemporal RGCs in 5 of 24 analysed mutant larvae. These findings suggest that a repulsive property of the MHB might be absent in acerebellar mutants. Interestingly, the trochlear nerve, which normally grows through the posterior, cerebellar part of the MHB, abnormally invades the tectal area in ace mutants (Fig. 3J,K), suggesting that the repulsive property of the MHB is not specific for RGC axons.

Shortly before reaching the anterior border of the tectal neuropil the optic tract normally separates into two fascicles, the dorsal and ventral brachium. RGC axons from the ventral retina are invariably sorted into the dorsal brachium and dorsal RGC axons into the ventral brachium (Stuermer, 1988). This sorting is variably abnormal in ace larvae. For all four retinal quadrants, many larvae show axonal projections via the wrong brachium and ectopic tracts at intermediate dorsal-ventral positions along the anterior boundary of the neuropil (e.g. Fig. 2F). Since brachial missorting might cause the abnormal dorsal-ventral topography of termination fields on the ace tectum, we investigated whether dorsal-ventral errors are also seen in larvae with normal brachial sorting (Table 1). In mutant individuals where temporal RGC axons enter the tectum through their normal brachium, dorsal-ventral mapping mistakes are only rarely seen (2 of 12, respectively 1 of 11 larvae examined). Conversely, even if nasal RGC axons enter through their normal brachium, dorsal-ventral delocalization nevertheless occurs in most individuals, and is therefore probably due to mapping errors in the posterior tectum (9 of 11, respectively 11 of 12 larvae examined).

To examine the molecular basis of the retinotectal defects in ace, we studied expression of ephrin-A5a, ephrin-A5b and ephrin-A2. mRNAs of all three ephrins are expressed in anterior to posterior increasing gradients in the midbrain and MHB of wild-type embryos at 24 and 48 hours (Fig. 3) at least until 84 hours, the latest stage we have examined (unpublished observations). The gradients of ephrin-A5a and ephrin-A2 reach into the anterior tectum, whereas ephrin-A5b expression is confined to the posterior midbrain and MHB (Fig. 3; Brennan et al., 1997), where ephrin expression overlaps with fgf8-expressing cells (Fig. 3G-I). In ace mutants, expression of ephrin-A5b is reduced at the MHB from the 16 somite stage onwards and is not detectable beyond 24 hours (Fig. 3A,B), whereas ephrin-A5a and ephrin-A2 are still expressed at low levels, but at an even distribution compared to the wild-type tecta (Fig. 3C-F). We quantitated relative ephrin mRNA expression levels along the anterior-posterior axis of the tectum after fluorescent ISH at 42-44 hours, a stage directly prior to RGC axon ingrowth to the tectum (Fig. 3L). Whereas the graded distribution is clearly evident in wild-type tecta, the gradients are absent or flattened in the mutant tecta; in particular the expression of ephrin-A2 and ephrin-A5a appear reduced to a level characteristic for anterior tectal areas.

To ensure that alterations in RNA levels reflected changes in protein distribution, we used an EphA3/AP-fusion protein that recognizes all known ephrin-A proteins (Brennan et al., 1997). Ephrin-A protein level is strongly reduced in mutant tecta at 24 and 44 hours, and the normally graded distribution of the ephrins again appears flattened along the anterior-posterior (Fig. 4A-H) and, notably, also the dorsal-ventral axis in ace mutants (Fig. 4I,J). The residual protein appears somewhat concentrated in the dorsoposterior tectum, especially at 44 hours (Fig. 4E-H). In summary, in the wild type, the shape and steepness of the gradient differs with the ligand considered. In ace mutants, ephrin mRNA and protein expression is either eliminated or flattened in the tectum as a consequence of the missing MHB, but a minor amount is still detectable in the dorsoposterior tectum.

Apart from the MHB, Fgf8 is also expressed in the retina and optic stalk (Reifers et al., 1998), suggesting that the retinotectal projection phenotype of ace could be partially due to autonomous defects in RGCs. To distinguish the contributions of Fgf8 function to retinal and tectal patterning, we performed optic vesicle transplantation experiments to combine eyes with a brain of different genotype in one chimaeric individual, using a novel technique developed by Chi-Bin Chien (personal communication). Optic vesicles were transplanted at the 8-10 somite stage from fluorescently labelled donor embryos into unlabelled hosts, from which the optic vesicle had been removed on one side (Fig. 5A). The unoperated contralateral side served as an internal control, and donors were raised to determine the genotype of the transplant. Transplanted eye vesicles heal in smoothly about 1 hour after transplantation, and by 30 hours of development, prior to RGC axon outgrowth from the eye, transplanted and unoperated eyes are indistinguishable, indicating that healing is complete (Fig. 5B-F). After 6 days of development the chimaeric larvae were analysed for the topography of the retinotectal projection between the donor eye and the host optic tectum by anterograde dye-labelling.

In control transplantations of wild-type eyes into a wild-type host, the manipulation does not affect the topographic order of the retinotectal map (Fig. 6A-C; Table 2). Similarly, control transplantations of ace eyes into ace hosts result in the same altered topography of the retinotectal projection, overshooting, brachial missorting and pathfinding errors as in non-manipulated ace larvae (Fig. 6D-F and not shown).

Transplantation of wild-type eyes into ace hosts reproduces the aberrations in retinotectal topography of ace mutants (Fig. 6G-O): axon projections from the dorsotemporal and dorsonasal retina strongly overlap along the anterior-posterior axis of the ventral tectum and their termination fields are expanded along the dorsal-ventral axis (Fig. 6G-I; see also Fig. 3C), whereas mapping of the ventrotemporal and ventronasal RGCs is only affected along the dorso-ventral axis of the tectum (Fig. 6M-O; see also Fig. 3D). In addition, missorting in the optic tract, enlargment of AF-7 (not shown) and overshooting (Fig. 6J-L) are observed. In the reciprocal chimaeras, wild-type embryos with an ace eye, dorsotemporal axons terminate normally in the ventroanterior tectum and do not overshoot (Fig. 6P-R).

The above findings show that the defects in the chimaeras depend on the genotype of the brain. However, analysis of dorsonasal RGCs also demonstrates an autonomous requirement for fgf8 in retinal development: dorsonasal axons that should only terminate in the ventroposterior tectum additionally project to the ventroanterior tectum (Fig. 6P-R; Table 2). Posterior overshooting of RGC projections towards the hindbrain is not detectable.

In conclusion, the transplantation experiments show that disturbed mapping, overshooting and brachial missorting of the RGC axons in ace are mainly due to alterations in the ace brain (Fig. 7). Nevertheless, an autonomous function for Fgf8 in retinal development is evident which requires further investigation.

Our study on the requirement of the MHB in acerebellar/fgf8 mutants demonstrates a dual function: one, in generating polarity of the retinotectal map, and a second, in confining growth of RGC axons to the tectum at the posterior tectal margin (Fig. 7). Transplantation studies in chick have previously identified two potential sources for a midbrain polarising activity (MPA) which influences the graded distribution of the Engrailed transcription factors: (i) the diencephalic-mesencephalic boundary negatively influences Engrailed expression (Chung and Cooke, 1978; Itasaki and Nakamura, 1992), and (ii) the MHB organizer can activate Engrailed expression when placed ectopically (Gardner and Barald, 1991; Martinez et al., 1991). This is reflected in the topography of RGC projections: RGC axons are able to recognize ectopically formed tecta with some topographic specificity (Itasaki and Nakamura, 1992). The early requirement for murine En-1 in development of the midbrain primordium has so far precluded studying the tectum of these mutants directly (Wurst et al., 1994; Joyner, 1996); En-2 mutants appear to have no defects in midbrain development (Millen et al., 1994). However, misexpression of En-1 and En-2 in anterior chick tectum leads to upregulation of Ephrin ligands and avoidance by RGC axons (Logan et al., 1996; Friedman and O’Leary, 1996). These findings support a central role for the Engrailed gradient in controlling downstream properties in the midbrain tectum, which is generally confirmed for the zebrafish by our study. Expression of all three zebrafish Engrailed genes is lost specifically in the tectum and MHB during midsomitogenesis in acerebellar mutant embryos (Brand et al., 1996; Reifers et al., 1998). Our present results show that the tecta of ace mutants lack Ephrin-A5b ligand, and have more even, instead of graded, distribution of Ephrin-A5a and Ephrin-A2 ligand. In the ventral tectum, absence of ephrin ligands correlates with the severe anterior-posterior delocalization we observe for the projection fields of nasal and temporal RGCs. This is consistent with a functional requirement for the zebrafish ephrin-A5 and -A2 homologues in mediating mapping along the anterior-posterior axis of the tectum, as suggested previously on the basis of their distribution and in vitro activity (Brennan et al., 1997). However, anterior-posterior polarity is not completely lost in the ventral tectum, and is surprisingly normal in the dorsal tectum, in spite of the absence of the MHB and the severe reduction of graded ephrin expression. The weaker effect on the dorsal tectum may be due to the residual dorsoposterior Ephrin expression we have observed. Alternatively, axons could also respond to other graded, non-Ephrin molecules that are unaffected by the absence of the MHB (Rétaux and Harris, 1996). Residual midbrain polarity could for instance be due to the presence of other, unaffected cell populations in the mutants, such as the cells of the di-mesencephalic boundary (Chung and Cooke, 1978; Itasaki and Nakamura, 1992), or cells of the hindbrain which are not normally in touch with the midbrain.

Although all Ephrin ligands we studied are influenced in their expression, they differ in their mode of regulation: ephrin-A5b is strictly dependent on ace function for its expression, whereas both ephrin-A5a and ephrin-A2 depend on ace for their graded distribution, but not for expression per se. In mature chick tecta, Engrailed and ephrin-A5 are coexpressed in glial cells, whereas ephrin-A2 is expressed in neurons (Millet and Alvarado-Mallart, 1995; Monschau et al., 1997). However, we observe the differences in Ephrin regulation already in the undifferentiated tectum, where these genes are expressed in all tectal cells (Brennan et al., 1997). Differences in cell type specificity are therefore unlikely to account for the differences in regulation. More likely, the different Ephrins may respond to different levels of Engrailed, possibly mediated by different levels of Fgf8 (see below). In this model, Ephrin A5b would only be activated at the highest Engrailed concentrations found in posterior tectum, whereas progressively lower and more anterior Engrailed levels would only suffice to activate Ephrin-A5a and Ephrin-A2 in a graded fashion. Gradient formation of Ephrin ligands in the midbrain would depend directly on the anterior-posterior polarization of Engrailed expression set during earlier stages in neuroepithelial precursors by a posterior MPA (Fig. 7). We suggest that the MPA is non-functional in acerebellar mutants, and tectal cells thus express Ephrins at levels characteristic for the anterior tectum.

A second function suggested by our observations is that the MHB prevents RGCs from posterior overshooting into the hindbrain. Overshooting could either be due to an endogenous increased ability of mutant RGC axons to extend, or due to lack of a repulsive activity from the MHB. The fact that overshooting is absent in chimaeras with ace RGCs projecting into a wild-type brain argues for a repulsive function of the MHB. Overshooting is seen for all retinal subpopulations, but most frequently for nasal RGCs that normally project to the posterior ventral quadrant of the tectum, close to the site where most overshooting from the tectum takes place. We suggest that axons exit most easily from the tectum at this position, because they can reach alternative axonal tracts which guide them into the hindbrain and beyond. Nasal RGCs may overshoot more frequently in ace simply because their normal tectal targets lie closest to the exit point.

Although Fgf8 from the MHB could itself influence axonal growth (McFarlane et al., 1996), we think it more likely that the repulsive activity is Ephrin-A5b, since (i) it acts in vitro to repel axons in chick and zebrafish (Drescher et al., 1995; Brennan et al., 1997) (ii) it shows the most posterior and steepest gradient of expression overlapping the MHB, (iii) it is missing in ace mutants, and (iv) mouse embryos lacking Ephrin-A5 show similar overshooting RGC projections (Frisén et al., 1998). We therefore reinforce our previous suggestion (Brennan et al., 1997) that the more posterior localization of ephrin-A5b expression, relative to ephrin-A5a and ephrin-A2, reflects a different function of ephrin-A5b, which is to ensure the proper definition of the posterior boundary of the tectal neuropil. In contrast, Ephrin-A5a and Ephrin-A2 might be involved in the relative spatial definition of termination fields. Together, the findings in mice and zebrafish suggest that the MHB, mediated by Ephrin-A5b, constitutes a repulsive zone for RGC axons that is oriented orthogonally to the midline. The repulsive activity is probably not specific for RGCs, since the trochlear nerve misnavigates in acerebellar mutants into the posterior tectal area, which it normally never enters (Fig. 3J,K). We suggest that the same signal may serve to prevent trochlear axons from entering the tectum. Trochlear axons would then be steered by at least two repulsive signals: netrin-dependent repulsion away from the ventral midline would steer the nerve dorsally (Colamarino and Tessier-Lavigne, 1995), while Ephrin-A5b-dependent repulsion would keep the axons on a track orthogonal to the midline. A similar arrangement of repulsive and attractive stripes of tissues that help to keep axons of the postoptic commissure and optic nerve on track has been proposed to exist at the midline surrounding the optic chiasm of the zebrafish (Macdonald et al., 1997).

The early onset and asymmetric expression relative to the midbrain, and the secreted nature of Fgf8 protein and phenotypic requirement for Fgf8 are all consistent with the possibility that Fgf8 itself might be the posterior MPA, or a component of it. This question cannot at present be decided, however, because in acerebellar mutants the MHB does not develop, probably because this tissue is transformed into more anterior midbrain tissue (Reifers et al., 1998; this paper), which could thus eliminate the true MPA, or other components of it apart from Fgf8. Also, it is unclear whether Fgf8 can diffuse over a sufficient distance to fulfill the function of the MPA. Several additional observations suggest, however, that Fgf8 is the MPA or a component of it. (i) Fgf8 is required to maintain polarized expression of neuroepithelial markers in the midbrain and MHB (Brand et al., 1996; Reifers et al., 1998; Lun and Brand, 1998). (ii) Misexpression of Fgf8 in the dorsal midbrain of mice, chick and zebrafish causes ectopic activation of Engrailed and ephrin-A2 and -A5 (Lee et al., 1997; Shamim et al., 1999; unpublished observations). (iii) FgfR4, a high-affinity receptor for Fgf8 is present in the tectum (Thisse et al., 1995). (iv) In several tissues, Fgf8 and ephrin family ligands are expressed in close temporal and spatial association. (v) Delocalized Fgf8 expression in mouse embryos lacking the gbx2 gene, normally expressed in the anterior hindbrain, causes lack of ephrin-A2 expression in the tectum (Wassarman et al., 1997), although it is not yet known how this affects the retinotectal map.

An unexpected finding of our study is that dorsal-ventral polarity of the map and ephrin-A expression is disturbed in acerebellar mutants. Dorsal-ventral delocalization is particularly strong for termination fields of nasal RGCs at the posterior tectal margin. Given the in vitro repulsive nature of Ephrin-A5b on axonal growth (Brennan et al., 1997), one possibility is that it acts as a generalized stop signal for axonal growth which itself is not directionally sensitive. In this view, nasal RGCs would continue dorsal-ventral growth along the posterior tectal margin in ace because they fail to encounter a signal that generally discourages axon growth. Alternatively, the posterior dorsal-ventral delocalization may reflect a simultaneous alteration of dorsal-ventral positional identities in tectal cells of ace mutants. Consistent with this possibility, the extent of ephrin-A ligand expression is reduced not only along the anterior-posterior, but also the dorsal-ventral axis of the tectum, and other, as yet unknown molecules involved in dorsal-ventral mapping may be correspondingly altered. Indeed, we have previously observed transiently altered dorsal-ventral patterning during early somitogenesis stages in acerebellar mutants, indicating that patterning along the two axes may be linked (Reifers et al., 1998; Lun and Brand, 1998). The absence of intertectal commissures in ace embryos (Fig. 3J,K and unpublished results) that has also been reported after antisense inhibition of engrailed expression (Rétaux et al., 1996) may be a reflection of altered dorsal-ventral organisation in the tectum. Further studies of the dorsal-ventral patterning cues for RGC projections, e.g of EphB receptors and their ligands (Braisted et al., 1997), are required in wild type and in acerebellar mutants to distinguish between these possibilities.

Although the map alterations in acerebellar mutants are due to changes in overall midbrain polarity, the Ephrin gradients are changed in a coherent way that yields insights into their function. First, our observations agree with previous suggestions that relative, rather than absolute, concentrations of tectal guidance cues are recognized by ingrowing RGC axon growth cones during mapping (Baier and Bonhoeffer, 1994). In the dorsoposterior tectum of ace mutants, we find strongly reduced, albeit still graded expression of ephrin-A ligands at the time of RGC axon ingrowth (Fig. 4). In spite of the low Ephrin levels, RGCs segregate normally along the anterior-posterior axis of the dorsal tectum. This supports the idea that the slope of the gradients in which guidance cues such as the Ephrins are distributed across the tectum governs mapping, rather than absolute protein concentrations. Second, nasal and temporal RGC axons respond in vitro differentially to gradients of Ephrin expression, with temporal axons being repelled by Ephrin-A2 and only weakly by Ephrin-A5 and nasal axons strongly by Ephrin-A5 but not by Ephrin-A2 (Monschau et al., 1997; Brennan et al., 1997). The difference between temporal and nasal RGCs may be due to nasal-temporally different expression levels and/or differences in the phosphorylation status of the Eph receptors on RGCs of the embryonic retina (Connor et al., 1998; Holash and Pasquale, 1995). Our chimaera analysis provides evidence for the in vivo relevance of these differences, since overshooting wild-type RGC axons in an ace mutant brain are predominately derived from the nasal retina. This effect is more pronounced in the chimaeras than in acerebellar mutants, probably because in the ace mutant dorsonasal RGCs are themselves abnormal.

We have also observed abnormal development of the optic tract in acerebellar mutants. In teleosts, RGC axons normally project through the optic tract in order and sort into dorsal and ventral brachia before reaching the tectum (Maggs and Scholes, 1986; Stuermer, 1988). ace mutants are variably defective in pathfinding, fasciculation, brachial sorting as well as mapping. Abnormal pathfinding of RGC axons is probably due to forebrain defects that are found particularly in the optic chiasm of the mutants (M. B., S. Shanmugalingam, R. Mcdonald, A. P., F. R., S. W. Wilson, unpublished observations) which may contribute to the reduced size of the tectal neuropil in ace mutants. Alternatively, Fgfs have been suggested to be involved in axonal growth, optic nerve fasciculation and tectal target recognition (Walz et al., 1997; Saffell et al., 1997), and it is possible that some or all of the pathfinding, defasciculation and brachial missorting phenotypes we see reflect this. Previous mistakes along the way to the tectum are, however, unlikely to account for the mapping defects we observe when axons have reached the tectum. In newt and axolotl, RGC axons project topographically correctly even when growing into the tectum via ectopic routes (Fujisawa, 1981; Harris, 1982). Similarly, in the zebrafish mutants boxer, dackel and pinscher brachial sorting is abnormal, but axons are nevertheless correctly targeted in the tectum (Karlstrom et al., 1996; Trowe et al., 1996). Generally, sorting within the optic tract differs widely for different vertebrates, suggesting that its importance for correct mapping is minor (Udin and Fawcett, 1988). Indeed, the brachial missorting of nasal RGCs in ace mutants is unlikely to account for their altered dorsal-ventral mapping, since axons entering the tectum through the normal brachium commit the same mapping mistakes as do those entering the tectum via the ectopic brachium. In summary, we believe that the mapping defects we observe are largely due to altered polarized mapping cues on the tectum itself, rather than indirect consequences of abnormal growth trajectories on the way to the tectum.

The defects we observed in formation of the retinotectal map in ace mutants are largely, but not completely restored when ace RGCs project into a wild-type brain in chimaeric animals, since dorsonasal RGCs form two foci on a wild-type tectum (Fig. 6P-R); the defect in this RGC subpopulation is less apparent in non-chimaeric ace mutants, perhaps due to the simultaneous alteration of tectal guidance cues. Fgf8 is expressed and could therefore function at several stages of retinal development from the early neural plate stage onwards, via the optic vesicle stage, and finally in the retina (Reifers et al., 1998, and unpublished observations). Moreover, retinal patterning is altered by Fgf8 overexpression that occurs in the zebrafish aussicht mutant (Heisenberg et al., 1999). It is an intriguing possibility that Fgf8 expression, found on both ‘ends’ of the visual system, would be available to co-ordinate patterning of retinal ganglion cells and their target field in the midbrain tectum.

We thank our colleagues in the Brand and Holder labs, as well as Uwe Drescher, Suresh Jesuthasan and Torsten Trowe for help with embryo manipulation, discussions and reagents. We are especially indepted to Chi-Bin Chien for teaching us optic vesicle transplantations, and to him and Steve Wilson for comments on the manuscript. We gratefully acknowledge support by grants from the Förderprogramm Neurobiologie Baden-Württemberg, Graduiertenkolleg Neurobiologie and SFB 317 (to M. B.) and from the Wellcome Trust (to N. H.). C. B. was funded by a BBSRC RoPA award to N. H. and S. Wilson.