Mutations disrupting the ordering and topographic mapping of axons in the retinotectal projection of the zebrafish, Danio rerio

Most sensory inputs to the central nervous system are topographically arranged (reviewed by Udin and Fawcett, 1988), that is, the spatial order of neurons is reflected by the order of their axon terminals in the target organ. Such order is well illustrated in the extensively studied retinotectal projection of lower vertebrates, where axons of dorsal retinal ganglion cells (RGCs) project to the ventral (lateral) tectum while ventral axons project to the dorsal (medial) tectum. Likewise, nasal RGC axons project to posterior (caudal) tectum while temporal axons project to the anterior (rostral) tectum (Fig. 1). Thus a doubly inverted map of the retina is created on the tectum.

Several mechanisms have been proposed to account for the establishment of this highly ordered projection. Roger Sperry suggested the existence of graded cues in the tectum and in the retina that could serve as positional markers (Sperry, 1963). Other models propose that interactions between fibers give rise to the observed topography (e.g. Hope et al., 1976; Horder and Martin, 1978). Still other models assume that synaptic activity is used to order an initially random or crudely sorted projection (Willshaw and von der Malsburg, 1976; Whitelaw and Cowan, 1981).

However, these mechanisms on their own are unable to explain certain features of regeneration and regulation shown by the projection (Hope et al., 1976; Fraser, 1980; Fraser and Perkel, 1990; Whitelaw and Cowan, 1981; Gierer, 1983). If, for example, the temporal retina is ablated, the anterior tectum will remain devoid of retinal innervation. The remaining nasal RGC axons will pass the anterior tectum and still project correctly to the posterior tectum, demonstrating that they can distinguish between both (Sperry, 1963). Given sufficient time, however, the nasal arbors will expand to cover the entire tectum (Schmidt, 1978; Schmidt et al., 1978; Yoon, 1972). If the tectum, or a piece of it, is rotated prior to innervation by retinal axons, the resulting projection is also rotated, demonstrating the existence of stable positional cues (Yoon, 1973, 1975). If the posterior tectum is ablated prior to innervation by the retinal axons, the nasal axons will innervate the posterior part of the remaining anterior tectum and, together with the other RGC axons, form a compressed but topographic map on the remaining tectal half (Gaze and Sharma, 1970). Finally, if axons of two eyes are forced to innervate the same tectal lobe, they will terminate in alternating stripes, reminiscent of ocular dominance columns (Constantine-Paton and Law, 1978; Levine and Jacobson, 1975). The formation of these columns is an activity-dependent process (Meyer, 1982).

To explain most or even all of the known properties of the projection, a model has to incorporate several mechanisms. Most models combine global positional cues, as proposed by Sperry, with one or more additional mechanisms, such as repulsion between fibers (Fraser, 1980; Fraser and Perkel, 1990; Gierer, 1983), activity-dependent strengthening of synapses (Whitelaw and Cowan, 1981), or all of the above (Fraser and Perkel, 1990).

Up to now, several laboratories have identified molecules with a graded expression (reviewed by Sanes, 1993), thus strengthening the idea that graded cues are used to set up the projection, but strict evidence for their involvement is still lacking. The best candidate molecules for positional signals are probably two recently cloned ligands for receptor tyrosine kinases of the EPH family, RAGS and ELF-1 (Drescher et al., 1995; Cheng et al., 1995; reviewed by Orike and Pini, 1996), and RGM, an antigen not yet characterised at the molecular level (B. Müller, unpublished observations). All three are expressed in gradients increasing from anterior to posterior tectum. There is evidence from in vitro studies that retinal axons read two of these graded molecules, RAGS and RGM, as repulsive signals.

One suitable organism in which to study the retinotectal projection is the zebrafish, Danio rerio. The development of its projection has been investigated in detail (Burrill and Easter, 1995; Kaethner and Stuermer, 1992; Stuermer, 1988; Stuermer et al., 1990). Unlike other vertebrates, the zebrafish offers the possibility of a genetic approach. In the hope of gaining new insights into the mechanisms that lead to the establishment of an ordered neuronal connection, we participated in a large-scale screen for mutations affecting the retinotectal projection of the zebrafish. Mutations were induced in male zebrafish by treatment with a point mutagen. The resulting mutations were made homozygous by inbreeding (Haffter et al., 1996). In the third generation, the retinotectal projections of the fish were analyzed. In order to make the retinal axons visible, one of the eyes of the fish was injected with two lipophilic fluorescent dyes at two specific positions (see Baier et al., 1996, for a detailed description of the method). The dyes are taken up by the RGCs and diffuse within the membranes of the axons, labeling them along their entire length. The pattern of the labeled axons is very characteristic. Mutations that resulted in a deviation from that pattern could be detected by visual inspection with a fluorescence microscope.

We identified 114 mutations that affect different aspects of the development of the projection. This paper will concentrate on those mutations that affect the sorting of axons into the two branches of the optic tract, the topography of the projection, and the formation of termination fields. Mutations affecting the pathfinding of RGC axons are described by Karlstrom et al. (1996).

Fish were raised and kept under standard conditions, as previously described (Mullins et al., 1994). Embryos were kept at 28.5°C in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 33 mM MgSO₄). Mutant carriers were identified by random intercrosses. To maintain a stock, identified carriers were outcrossed to wild-type fish. For complementation analysis, carriers of different mutations were crossed to each other. If the embryos of such a cross did not show a mutant phenotype, the mutations were considered to complement, and were assigned to different complementation groups.

Dye injections were done as described by Baier et al. (1996). Briefly, fish were fixed 5 days postfertilisation in 4% paraformaldehyde. After a minimum fixation time of 12 hours, they were mounted at defined positions in an agarose gel (1.2% in PBS diluted 1:3 in water). Axons of RGCs were labeled by intraocular injections of DiI (1,1′diocta-decyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate) and DiO (3,3′-dioctadecyloxacarbocyanine perchlorate). The brain was labeled by incubating the fish in DAPI (4,6-diamidino-2-phenylindole) (5 mg/l) overnight.

Photoconversion of DiI was done as described by Westerfield (1993).

For sectioning, fish were dehydrated through an alcohol series. Afterwards, they were first transferred to dry acetone, then to diluted araldite (50% acetone, 50% araldite). The acetone was allowed to evaporate overnight at room temperature, and the fish were transferred to pure araldite. After polymerization of the araldite (60°C, 48 hours), the fish were sectioned using a Reichert-Jung microtome (Reichert-Jung 2050). The thickness of the sections was 7 μm.

In zebrafish, the first RGC axons leave the eye about 34 hours postfertilization (Burrill and Easter, 1995; Stuermer, 1988). 3 days after fertilization, the first nasal axons arrive at the posterior tectum. In zebrafish, the mapping of RGC axons is precise from the onset.

To inspect the retinotectal projection, we injected two dyes, DiI and DiO, into the temporoventral and the nasodorsal quadrant of the retina, respectively. In a 5-day-old wild-type zebrafish, this yields the following pattern:

Nasodorsal axons and temporoventral axons project through the optic nerve and tract to the contralateral tectal lobe (Figs 1, 2). In the optic nerve, the two labeled axon populations occupy separate portions. Axons of neighboring cells grow next to each other (Stuermer, 1988) (Figs 1, 2). The optic tract is split into a dorsal and a ventral branch. Dorsal and ventral axons segregate into the ventral and the dorsal branch of the tract, respectively (Figs 1, 2). Nasodorsal axons, upon arriving at the tectum, continue their ventral trajectory and terminate in the posteroventral tectum (Figs 1, 2). Temporoventral axons terminate in the anterodorsal tectum (Figs 1, 2). The size of the labeled termination fields varies, depending on the number of axons labeled by the injection.

During the screen, we could not see the outline of the tectum.

To judge whether the labeled axons terminated retinotopically, we investigated the position of the labeled axons relative to each other. Nasodorsal axons of wild-type fish always terminate posterior and ventral to the temporoventral axons (Fig. 2C).

We keep a stock of 114 mutants. They can be grouped according to their phenotypes into four different classes: pathfinding mutants in which RGC axons have abnormal trajectories, mutants that are affected in the sorting of axons in the optic tract, mutants in which the RGC axons do not map retinotopically on the tectum and mutants showing aberrant arborization patterns in the target. This paper will focus on the last three classes. The pathfinding mutants are described by Karlstrom et al. (1996).

The mutants described in this paper are listed in Table 1, and schematic drawings of their retinotectal phenotypes are shown in Fig. 3.

We found 14 mutations in four loci, boxer (box), dackel (dak), pinscher (pic) and nevermind (nev), which affect the sorting of axons in the optic tract. All show a similar phenotype: dorsal axons grow through both the ventral and the dorsal branches of the optic tract, instead of selectively using the ventral branch (Fig. 4). Mapping is not affected in box, dak and pic. The misrouted axons are able to locate their retinotopic target area in the tectum. Mapping of retinal axons is affected in nev. This mutation is therefore discussed in detail with the mapping mutants. All of these mutants have defects in other organs as well. box and dak mutants have reduced pectoral fins, and box, dak and pic mutants have jaw and brain defects.

In fish with mutations in either box or dak, nasodorsal axons grow into both branches of the optic tract (Fig. 4B).

The nasodorsal axons that enter the tectum through the ventral branch grow in a wild-type manner along the ventral tectal side and terminate retino-topically in the posteroventral tectum (Fig. 4C,E). Nasodorsal axons entering the tectum through the dorsal branch at first continue their dorsal trajectory. Near the posterior edge of the tectum they turn and grow to their retinotopic posteroventral target area (Fig. 4C,E). In box, the ventral location of the termination field has been confirmed by analysis of transverse araldite sections (Fig. 5B).

The nasodorsal axons that grow along the inappropriate dorsal side of the tectum are more fasciculated than the ones on the ventral side (Fig. 4E). Some misrouted axons leave the tectal neuropil (Fig. 4C, arrowhead). These axons invariably grow toward the ipsilateral tectal lobe. If they reach it, they enter it from the dorsal side, grow to its posterior edge, turn, and terminate retinotopically posteroventrally (data not shown).

The sorting of ventral axons is apparently not affected in box and dak mutants. Temporoventral axons grow mostly along the dorsal branch. We analyzed the behaviour of nasoventral axons in box fish to determine if they behave differently from temporoventral axons. In 33% of box mutant fish, nasoventral axons grow exclusively along the dorsal branch of the optic tract (n=14). This number is very similar in wild-type animals (25%, n=119). In the remaining box and wild-type fish, ventral axons were also seen in the ventral branch of the optic tract. In contrast, when dorsal axons are labeled in wild-type fish, they grow in 97% of cases in the ventral branch only (118 injections). That the sorting of ventral axons appears to be less precise than that of dorsal axons might simply reflect a weakness of our injection method. Using our apparatus, the nasoventral retina is more difficult to inject than the nasodorsal retina.

Dorsal axons that enter the tectum through the dorsal branch of the optic tract have two possible ways to reach their ventral target area. One is to grow around the periphery of the tectal lobe, and the other is to grow directly across it to the appropriate site. To distinguish between these two possibilities, we labeled middorsal axons (see Fig. 1). Like nasodorsal axons, middorsal axons grow into both branches of the optic tract. But instead of growing to the posterior margin of the tectal lobe, most ectopic middorsal axons make an abrupt turn in the middorsal lobe (Fig. 4D, arrow). They grow across the lobe directly to their retinotopic midventral target area.

In the optic nerve, the order of axons appears to be normal.

Transverse sections through the brain of box mutants at 6 days postfertilization show that the dorsal diencephalon and the dorsal mesencephalon are compressed along the superficial-deep axis (Fig. 5 and data not shown). The cartilage that forms the jaw is more massive than usual, due to irregular shaped cartilage cells (data not shown) (for a further description of jaw defects, see Schilling et al., 1996). The pectoral fins are reduced in size. This fin abnormality is the first visible deviation from normal development in box, and is apparent at 2 days postfertilization.

In dak mutants, the forebrain, especially the telencephalon, is smaller than in wild-type fish and the olfactory placodes are reduced in size or missing. The jaw cartilage is severely affected by this mutation. Gill arches two to five are reduced, as is the hyoid. The pectoral fins are missing. For a further description of these defects, see Schilling et al. (1996) and also van Eeden et al. (1996). Obviously, box and dak not only share a similar retinotectal phenotype, but also a similar combination of other defects. This suggests that the genes affected in box and dak are involved in the same pathway(s).

As in box and dak, nasodorsal axons in pic mutant embryos grow through both branches of the optic tract. However, in pic, the ectopic nasodorsal axons grow onto the tectum in only about 75% of cases (n=16). In more than 50% of cases, a fraction or all of the ectopic nasodorsal axons are deflected medially before reaching it (Fig. 4F, arrowhead). The deflected axons sometimes enter the ipsilateral tectal lobe. On the ipsilateral lobe, they can navigate to their retinotopic position (data not shown). Ventral axons appear not to be affected in their choice of the correct branch of the optic tract by the mutation. But frequently, ventral axons leave the optic tract and grow medially as well (Fig. 4F). The gill arches in this mutant are massive and abut each other (Schilling et al., 1996). The pectoral fins are normal. The telencephalon is smaller than normal and does not protrude anterior to the eyes.

Five mutations in four genes, nevermind (nev), who-cares (woe), gnarled (gna) and macho (mao), affect the mapping of retinal axons. The first two affect mapping primarily along the dorsoventral dimension of the tectum. Dorsal axons in nev and woe grow and terminate on the ventral and on the dorsal side of the tectum. The remaining two mutants affect mapping along the anteroposterior dimension. In gna and to a lesser extend in mao, nasodorsal axons defasciculate and terminate too soon in the anterior tectum. Woe and mao mutants have expanded melanophores. nev, woe and mao show abnormal motility.

As shown in Fig. 6C,D, nasodorsal axons in nev terminate in a large zone that extends from the posteroventral to the posterodorsal side of the tectum. The termination field of the nasodorsal axons usually does not extend as far dorsally as it does ventrally, and judged by the intensity of labeling, fewer axons terminate dorsally than ventrally. On transverse sections, the intensity of labeling can be seen to decrease smoothly from dorsal to ventral (Fig. 5C). As in box, dak and pic, dorsal axons grow through both branches of the optic tract (Fig. 6D), although a majority of the labeled axons usually use the proper ventral branch. Dorsally growing nasodorsal axons in nev do not leave the tectum, as they do in box. They defasciculate normally upon approaching their target area.

Ventral axons terminate properly in the dorsal tectum, but unlike axons in wild-type animals, they do not grow in a smooth curved line toward their target area, but appear to meander instead (Figs 6E, 7B). The order of axons in the optic nerve is changed in a characteristic fashion (Fig. 6C).

Sections through 5-day-old embryos homozygous for nev show that as in box, the dorsal di- and mesencephalon are smaller than in wild-type animals (Fig. 5 and data not shown). nev fish fail to develop a swim bladder and die, usually around day 8. The mutant embryos show abnormal motility. They rotate around their long body axis when swimming.

In woe mutants, nasodorsal axons terminate posteroventrally and posterodorsally in the tectum (Fig. 6F,G). Judged by the intensity of labeling, the axons distribute themselves about evenly between the two termination fields. Analysis of transverse araldite sections shows two distinct termination fields, one in the ventral and one in the dorsal tectum (Fig. 5D). The actual position of the dorsal termination area in the dorsal tectal half still has to be accurately determined. Due to the shape of the tectal lobe, sections might give a wrong impression of the position of an termination field. Unlike in nev mutants, nasodorsal axons grow appropriately in the ventral branch of the optic tract, and it is only on entering the tectum that some of them grow across the tectal lobe to its dorsal side (Fig. 6F,G). These axons grow in several widely spaced fascicles toward their posterodorsal termination zone (Fig. 6G). The two branches of the optic tract itself are not separated as widely in the mutant as in a wild-type fish.

The sorting of axons in the optic nerve appears to be normal, although the nasodorsal and the temporoventral fascicle cross further away from the eye than usual. Ventral axons terminate retinotopically in the dorsal tectum.

Mutant individuals have a pigment phenotype at 6 days post-fertilization. Their melanophores show no background adaptation when exposed to bright light. About 50% of embryos homozygous for woe develop a swim bladder. These embryos show very little spontaneous movement, but react normally to touch. They die around 12 days postfertilization.

In gna mutant fish, nasodorsal axons split abruptly into several diverging fascicles in the anterior tectum (Fig. 6H). Some of the fascicles turn dorsally before continuing their posterior course. A fraction of the nasodorsal axons terminates prematurely anteroventrally, shortly after arriving on the tectum (Fig. 6H,I). The area covered by the termination field(s) of the nasodorsal axons varies between individuals. In extreme cases, it encompasses almost the entire tectum (Fig. 6I). Ventral axons project retinotopically to the dorsal tectum. In particular, the nasoventral axons also terminate normally in the posterodorsal tectum (data not shown). No further abnormalities have so far been identified in this mutant.

In mao mutant fish, the nasodorsal axons defasciculate and terminate prematurely in a more anterior position on the tectum than usual. The termination field of the nasodorsal axons is enlarged compared to a wild-type, but the enlargement is less pronounced than in gna (Fig. 6J). The termination zone of temporoventral axons appears to be enlarged as well.

mao mutants do not have a touch response at day 2 (Granato et al., 1996) and at day 4 their melanophores are expanded.

We have kept 17 mutants in which mapping of retinal axons onto the tectum is little or not at all disturbed but in which the size of the tectum itself is greatly reduced (Table 1). Often, eye size is diminished as well (Table 1). Two examples of such mutants are shown in Fig. 8.

In the two alleles of noi mutants we found, tb21 and ty31z, temporoventral and nasodorsal axons project near the posterior part of a small tectum (Fig. 8B). The tectal remnant in noi probably has anterior identity, since it does not express posterior markers (Brand et al., 1996). Nasodorsal axons often continue to grow beyond the posterior margin of the tectum (Fig. 8C). Part of the labeled axons project frequently to the ipsilateral tectal lobe (data not shown).

The earliest visible defect in noi mutant embryos is the absence of the midbrain/hindbrain boundary (Brand et al., 1996).

In tin mutant embryos, the tectal neuropil is tiny. Nasodorsal and temporoventral axons are consequently barely separated on the tectum (Fig. 8D). But even in this tiny tectum, nasodorsal axons terminate posterior of the temporoventral axons. Clearly, tectal development rather than mapping of retinal axons is affected. We have kept another ten mutants with small tectal neuropils (Table 1). For several reasons it is not yet clear to what extent mapping is affected in these mutants. Small eyes make the localization of the dye injections less precise, thus introducing a great variance into the location of labeled termination fields on the tectum. The small tecta further increase the difficulty of judging the retinotopy of the projections. However, during the screen we found many mutants with small tecta displaying a clearly retinotopic map, showing that a reduction of tectal size does not necessarily disturb retinotopy (data not shown).

In this group we placed mutants whose RGC axons project to their normal target area, but fail to develop a normal termination field. 27 mutants in six complementation groups fall into this class. Mutants in two complementation groups, braindead (brd) (Fig. 9E) and delayed fade (dfd), show a degenerating tectum around day 3 (Furutani-Seiki et al., 1996), and the retinotectal phenotype in these cases is likely to be a consequence of this degeneration.

In esr, til and tof mutants, which show an identical projection phenotype, RGC axons form a thickening right behind the optic papilla, and fewer axons than normal project to the tectum (Fig. 9B). Usually, most of the nasodorsal axons stop in the anterior tectum, and only a few axons project posteriorly. However, this seems not to reflect a topographic but an outgrowth defect, since, judged by the intensity of labeling, the axon bundles thin out already before reaching the tectum. In some mutant individuals, an apparently normal number of axons arrives at the tectum. In these cases, the nasodorsal axons spread out to cover a big part of the tectal neuropil evenly (Fig. 9C).

All three mutations also result in a reduced xanthophore pigmentation (Odenthal et al., 1996).

Nasodorsal and nasoventral axons in blu mutants defasciculate strongly upon reaching the posterior tectum. The axons appear to turn away from each other. They arborize in a cauliflower-like fashion and end up covering a much larger area than usual (Fig. 9D). The melanophores of blu fish show no background adaptation when exposed to bright light.

In a large-scale screen in zebrafish, 45 mutants were identified that affect either the order of retinal axons en route to the tectum, the establishment of a retinotopic map on the tectum or the formation of proper termination fields. The analysis of the mutant phenotypes allows a dissection of the process of map formation.

We found 11 mutations in three loci that affect the sorting of retinal axons into the branches of the optic tract in an identical manner: dorsal axons, instead of using only the ventral branch, split and grow into both the ventral and the dorsal branch. Nasodorsal axons that grow in the inappropriate dorsal branch continue their dorsal trajectory on the tectum. These misrouted axons can locate their appropriate target area, demonstrating that the sorting of axons onto the two branches of the optic tract is not required for proper mapping.

This finding is consistent with earlier observations, which indicated that the point of entry of retinal axons into the tectum is not crucial for the establishment of a normal map. If, for example, retinal quadrants are directly explanted onto the tectum, retinal axons will grow toward their target area from a variety of positions (DeLong and Coulombre, 1967; Fujisawa, 1981; Thanos and Dütting, 1987). Axons from eye transplants that grow to the tectum on abnormal paths also map normally (Constantine-Paton and Law, 1978; Harris, 1980, 1982; Sharma, 1972).

In box mutants ectopic middorsal axons continue their dorsal trajectory after entering the tectum. After reaching the middorsal tectum, a retinotopic position with respect to the anteroposterior tectal axis, they make a sharp turn and grow across the tectum directly towards their midventral target area, thus correcting their dorsoventral displacement. This observation is in agreement with experiments in chick showing that retinal axons that have been displaced by antibody perturbations first establish their position along the anteroposterior axis, and only then proceed to seek out their proper dorsoventral position (Thanos et al., 1984). The course correction of ectopic axons, as shown for example by dorsal axons in box mutants, is easily explained by the operation of graded positional cues (Gierer, 1983, 1987).

Two of the genes identified in this screen, nev and woe, are important for mapping in the dorsoventral dimension. If either of these genes is mutated, dorsal axons terminate on the ventral and on the dorsal side of the tectum. In nev, the RGC axons are also missorted in the optic nerve and tract, indicating that nev affects the positional identities of ganglion cells in the retina. In contrast, retinal axons in woe mutant fish are sorted normally in the optic nerve and tract.

Gradient models can account for an extension or duplication of termination fields in several ways (Fig. 10). For example, if one assumes that linear gradients specify dorsoventral positions in the retina or the tectum (Fig. 10A), a decrease in slope (Fig. 10B) or a duplication of gradients (Fig. 10C) can in principle explain the observed dorsoventral extension or duplication of termination fields, respectively. On the other hand, radial gradients (Fig. 10D) may be used to specify the positions in the retina or the tectum. In this case, one additional cue is required to define dorsal and ventral (and/or nasal and temporal) sides (Fig. 10E). The functional disruption of such a symmetry-breaking cue could also explain the phenotypes of nev and woe. This cue might be a dorsoventral gradient, a circumferential gradient, or simply a step, that is, a cue expressed in either the dorsal or the ventral retinal/tectal half (Fig. 10E). Steps between the nasal and the temporal and between the dorsal and ventral retina have been reported (e.g. Macdonald et al., 1994; McLoon, 1991; Rissi et al., 1995; Walter et al., 1987), and it has been suggested already that in the retina, a step function along the nasotemporal axis together with a radial (i.e. a centroperipheral) and a dorsoventral gradient could be used to label position (Holt and Harris, 1993).

In nev and woe mutants, ventral axons map retinotopically on the dorsal side of the tectum. Theoretical gradient models designed to account for topographic projections with a minimal number of cues use the same gradients for guiding both the dorsal and the ventral axons. Hence we take the result, that mapping of dorsal axons is affected while that of ventral axons is normal, as an indication that more gradients are involved than are minimally required in theory. Furthermore, ventral axons in nev mutants, although projecting retinotopically to the dorsal tectum, reach their target area by tortuous paths, unlike axons in wild-type animals. Although the ventral axons still possess sufficient information to reach their target area, the precision of their tracking has been affected. It is possible that the ventral axons use multiple cues to orient with respect to the dorsoventral axis.

In gna and mao, nasodorsal retinal axons terminate in a more anterior position on the tectum than in wild-type animals. This defect is much more pronounced in gna than in mao. In gna, the nasodorsal but not the nasoventral axons split abruptly into several diverging fascicles upon arriving on the tectum. Some fascicles frequently enter the dorsal tectum. The ventral axons map normally.

In gna and other mutants, for example nev, changes in the mapping of RGC axons always go together with changes in their degree of fasciculation. This argues that both processes are connected. Defasciculation could be a direct consequence of the increasing permissiveness of the tectal environment once axons approach their target. This permissiveness is perhaps mediated by the same graded cues that direct the formation of a topographic map. The degree of fasciculation of ectopic nasodorsal axons in box and nev mutants correlates with their site of termination, supporting this hypothesis. In box mutants, ectopic nasodorsal axons terminate in the ventral tectum. They are highly fasciculated while growing through the dorsal tectum. In contrast, ectopic nasodorsal axons in nev mutants terminate in the dorsal tectum. In this mutant, the ectopic axons defasciculate gradually while approaching their dorsal target.

Three of the four genes that affect the mapping of retinal axons onto the tectum also affect the motility of the embryos. nev mutants rotate around their long axis when swimming, which might be a direct consequence of their retinotectal phenotype. However, the relative position of the dorsoventral axis of fish to the gravitational axis is affected by vision as well as gravitation (Parzefall, 1993). The relationship between the behavioural and the retinotectal phenotype is probably less direct. The behavioural defects of the other mutants are not explicable by their retinotectal phenotypes in an obvious way. In this context it is important to note that the tectum is a multimodal structure that also receives sensory inputs other than vision (Udin and Fawcett, 1988). Mutations affecting tectal polarity can probably influence the behaviour of fish, not only by changing the formation of visual, but also of other sensory maps.

Transplantation experiments in the chick have shown that the polarity of the anteroposterior axis of the tectum anlage, the mesencephalic alar plate, correlates with the expression of en, which is expressed in an increasing anterior to posterior gradient on the tectal primordium (Itasaki et al., 1991; Itasaki and Nakamura, 1992; Martinez and Alvarado-Mallart, 1990; Patel et al., 1989). The tectum of noi mutants does not express en (Brand et al., 1996), consequently, one might expect it to have lost anteroposterior polarity. The tectum of noi mutants is very small and its innervation pattern (see Results) is indeed consistent with the idea that this remnant has anterior identity.

By screening a large number of mutagenized zebrafish lines, we obtained a set of mutants with defects in retinotectal mapping. We expect that the molecular and functional analyses of these mutants will help us to understand the principles underlying the formation of neuronal projections.

It is a pleasure to acknowledge the help of many individuals who made this study possible. Christian Bayertz, Eckard Dominik, Dr Uwe Drescher, Carmen Gitter, Claudia Handwerker, Ingrid Horschke, Julita Huf, Dr Hirojuki Ichijo, Kurt Knauf, Dr Sigrun Korsching, Ulrike Nell, Thomas Stroh, Dr Thomas Voigt and Franko Weth helped with the processing of the fish. Under the supervision of Franz Endress, our machine shop patiently developed the gadgets needed to mount and inject the fish. Jürgen Jung and Jürgen Löschinger helped with the computers. We would also like to thank Drs D. Dütting, S. S. Easter, A. Gierer, S. Jesuthasan, S. Thanos and two anonymous referees for helpful comments on the manuscript.