astray, a Zebrafish roundabout Homolog Required for Retinal Axon Guidance

See allHide authors and affiliations

Science  20 Apr 2001:
Vol. 292, Issue 5516, pp. 507-510
DOI: 10.1126/science.1059496


As growing retinotectal axons navigate from the eye to the tectum, they sense guidance molecules distributed along the optic pathway. Mutations in the zebrafish astray gene severely disrupt retinal axon guidance, causing anterior-posterior pathfinding defects, excessive midline crossing, and defasciculation of the retinal projection. Eye transplantation experiments show that astrayfunction is required in the eye. We identify astray as zebrafish robo2, a member of the Roundabout family of axon guidance receptors. Retinal ganglion cells express robo2 as they extend axons. Thus, robo2 is required for multiple axon guidance decisions during establishment of the vertebrate visual projection.

During development of the retinotectal projection, retinal ganglion cell (RGC) axons navigate through a series of environments. They first grow toward the optic disc, exit the eye at the ventral fissure, and grow along the base of the ventral diencephalon, where they meet the contralateral retinal axons, forming the optic chiasm. They then progress into the optic tract and project topographically to their central targets, primarily the optic tectum in nonmammalian vertebrates. There is a growing understanding of the molecular mechanisms that specify retinotectal topography (1–3) and that guide retinal axons within the eye (4, 5), but to date few axon guidance molecules have been shown in vivo to function between eye and tectum (6, 7).

We have used the zebrafish to identify such guidance molecules and study their functions in vivo. The transparency of the larvae in combination with the amenability of this model organism to genetics has made it possible to directly visualize the retinal projection and isolate genes required for RGC axon guidance (8).astray (ast) (9) is one of the key genes isolated in this large-scale screen. Four alleles (ti272z, te378, tl231, andte284) have been found, all with similar phenotypes. Compared to wild type (WT), in which the retinotectal projection at 5 days postfertilization (dpf) is exclusively contralateral, RGC axons in ast/ast embryos exhibited misprojections to ipsilateral tectum and several extratectal targets (9) (Fig. 1).

Figure 1

ast mutants show multiple retinal axon guidance defects. (A to F) Three retinal projections are shown in dorsal (A, C, and E) and lateral (B, D, and F) views. Eyes have been removed, and brain outlines are shown in white; boxed insets in (C) are at high gain to reveal faint projections. (G) ant, anterior; dor, dorsal; ch, chiasm; di, diencephalon; ep, epiphysis; ey, eye; hb, hindbrain; ln, lens; mb, midbrain; ob, olfactory bulb; oe, olfactory epithelium; tec, tectum; tel, telencephalon. (A and B) WT axons cross the ventral midline (arrowheads), form the dorsal (db) and ventral (vb) brachia of the optic tract, and arborize in contralateral tectum. (C and D) Weakast phenotype (te284). Most axons project normally, but a small fascicle projects anteriorly (arrowheads), and several fibers recross in the PC (long arrow). Of three fascicles that leave the ventral brachium (short arrows), two correct and enter the tectum, whereas the third continues into the VHB. (Inset C1) Fibers recrossing in the PC arborize in the ipsilateral tectum. (Inset C2) Posteriorly projecting fibers recross in the VHB. (E and F) Strong ast phenotype (ti272z). Few axons reach the contralateral tectum. Some anterior axons recross near the AC (arrowheads), whereas others continue contralaterally. Other axons recross in the PC (long arrows). The optic tract is severely defasciculated. (H andI) WT and ast optic chiasms. Arrows indicate where axons exit right (red) and left (green) eyes. (I) Axons from both eyes split near the midline (asterisks) into several bundles.

Confocal analysis after lipophilic dye labeling at 5 dpf (10) showed that after ast RGC axons exited the eye, they made a wide variety of pathfinding errors at multiple locations (Fig. 1), predominantly at or after the midline. Anterior-posterior guidance defects were common, including anterior projections into diencephalon and telencephalon, and posterior projections to both ipsilateral and contralateral ventral hindbrain (VHB) (Fig. 1, C to F). Anteriorly projecting axons often reached as far as presumptive olfactory bulb; they also often recrossed the midline, then continued anteriorly or turned posteriorly (Fig. 1E). Axons recrossed the midline in at least three distinct locations: VHB, ventral telencephalon near the anterior commissure (AC), and dorsally in the posterior commissure (PC); the latter was a common route by which retinal axons reached the ipsilateral tectum (Fig. 1C). In strong phenotypes, the optic tract showed severe defasciculation (Fig. 1, E and F). Occasionally, the optic chiasm formed abnormally: The optic nerve was unusually distant from the contralateral optic nerve, or split into two or more parts at the midline (Fig. 1I). Retinal axons also sometimes projected into the opposite eye. Axons from all four retinal quadrants showed pathfinding errors; however, axons that reached the tectum appeared to project topographically (9,11). Thus, ast function is required for midline crossing as well as several other axon guidance decisions.

All four ast alleles are recessive and cause a similar array of phenotypes, suggesting that they are loss-of-function mutations. Phenotypic strength varied widely between embryos, even within clutches, and the two eyes sometimes showed different phenotypes. Even in the embryos with the strongest phenotypes, a subset of axons reached the contralateral tectum. Scoring ast/+ × ast/+ incrosses (10) showed three completely penetrant alleles [ti272z (24.5% mutant, n = 335 embryos),te378 (25.8%, n = 159), andtl231 (25.3%, n = 182)] and one weaker allele [te284 (18.0%, n = 133)]. Supporting the conclusion that te284 is weaker was the observation that often only a single eye of te284homozygotes showed a detectable phenotype, whereas in allti272z homozygotes examined, both eyes showed mutant phenotypes. Projections to VHB (86%) and across the PC (78%) were more common than anterior projections (64%) (ti272zhomozygotes, n = 175 eyes).

The requirement for ast function is highly specific. Overall body shape appears normal, and astti272z homozygotes are partially adult viable and fertile. The anterior and postoptic commissures form normally (9). Stainingastti272z (n = 9) with the zn-5 antibody (10) at 52 hours postfertilization (hpf) showed that ast RGC axons oriented locally toward the optic disc and left the eye in a single bundle as in WT (10).

To determine whether ast function is required in the eye or in the brain, we used eye transplantation (10) to create chimeric animals with an ast mutant eye in an otherwise WT embryo, or vice versa. Eye primordia were transplanted at 12 hpf, immediately after optic vesicle formation and well before retinal axon outgrowth. Both host and transplanted eyes were labeled (Fig. 2A) at 4 dpf, when the retinotectal projection is mature. In controls (n = 10), WT eyes in WT hosts showed strong projections to contralateral tectum (10/10 transplants) (Fig. 2B). Occasionally there was a weak projection to ipsilateral tectum (4/10), presumably representing axons that encountered surgically disturbed tissue and were diverted upon exiting the eye. Controls did not show anterior or VHB projections (0/10). WT eyes transplanted into astti272z homozygotes showed strong contralateral projections (6/6), and no anterior or VHB projections (0/6), despite the mutant environment through which their axons grew (Fig. 2C). In contrast astti272z eyes transplanted into WT hosts showed a clear ast phenotype (Fig. 2, D and E), projecting anteriorly to forebrain (4/8), posteriorly to VHB (8/8), or across the PC (5/8). Thus, astfunction is required in the eye, most likely in RGC axons.

Figure 2

ast is required eye-autonomously. (A) Transplant scheme. (B to E) Retinal projections in chimeric animals; dorsal views. Axons from transplanted eyes are shown in red; axons from host eyes are in green. Boxed inset in (E) is shown at high gain. (B) WT eye in WT host projects normally. (C) WT axons in ti272z host project normally despite lack of ast function in the brain.ast host axons project anteriorly into the diencephalon where they recross the midline (arrowhead), and posteriorly to the VHB (short arrow). (D and E) Two examples of ti272z eyes in WT hosts, showing that ast axons project aberrantly despite the WT environment. Transplanted ast eyes project anteriorly into the diencephalon and across the midline (arrowhead); posteriorly toward the VHB (short arrows); and across the PC (long arrows). WT host axons project normally.

The ast phenotype exhibits similarities to loss-of-function mutations in Caenorhabditis elegans sax-3 and itsDrosophila homolog roundabout (robo).sax-3 mutants show abnormal axon crossing of the ventral nerve cord, abnormal guidance to the ventral nerve cord, anterior and posterior projection defects, and fasciculation defects (12, 13). In robo mutants, axons inappropriately cross and recross the CNS midline (14). sax-3 and robo are members of the evolutionarily conserved family of Robo-like receptors with five immunoglobulin (Ig) domains, three fibronectin type III (FNIII) repeats, a single transmembrane domain, and four conserved cytoplasmic motifs. Given the similarity in phenotypes and our eye transplantation results, we examined whether ast is a zebrafish robo. Our group has cloned three zebrafishrobo homologs (15), of which one,robo2 (an ortholog of mammalian robo2), is expressed at the appropriate time in RGCs and was therefore a likely candidate for ast.

To determine whether the ast phenotype could be caused by mutations in robo2, we placed both on the zebrafish genetic map (10, 16, 17).astti272z mapped to linkage group (LG) 15 between simple sequence repeats (SSRs) z13822 (0.3 cM) and z26441 (0.1 cM) on one side and z7381 (3.8 cM) on the other (Fig. 3A). robo2 mapped to LG15 on the radiation hybrid (RH) map, 6 cR3000 (centiRay 3000) from z13822 (Fig. 3A). The 0.3-cM distance ofastti272z to z13822 corresponds to ∼222 kb and the 6 cR3000 distance of robo2 to z13822 corresponds to ∼366 kb [1 cM = ∼740 kb (16); 1 cR3000 = ∼61 kb (17)]. This linkage supported the possibility thatast is robo2.

Figure 3

Cloning of ast. (A)astti272z is tightly linked to robo2.astti272z maps 0.3 cM from z13822 (2 recombinations in 596 meioses), 0.1 cM from z26441 (1 recombination in 762 meioses), and 3.8 cM from z7381 (28 recombinations in 762 meioses).robo2 maps 6 cR3000 from z13822 on the same side as astti272z . (B) Identification of mutations in astti272z andastte284 . Schematic representation of zebrafish Robo2: semicircles, Ig domains; filled rectangles, FNIII domains; open rectangles, cytoplasmic motifs. astti272z encodes an Arg635 to Stop mutation before the second FNIII domain. astte284 changes Gly882 to Asp in the transmembrane domain. Chromatograms show WT (left) and mutant (right) sequences. Numbers indicate amino acid positions.

To further examine whether ast is robo2, we sequenced the robo2 coding sequence from a strong allele,astti272z , and a weak allele,astte284 , seeking mutations that could account for the ast phenotype. Because the mutations were induced by ethylnitrosourea (18), we expected to find point mutations in the affected gene. Although robo2 mRNA was detectable at 5 dpf by in situ hybridization, we were unable to producerobo2 cDNA from phenotypically identified 5-dpf mutants after fixation, embedding, and dye injection. Therefore, we identified homozygous mutant embryos at 36 hpf by genotyping with the closely linked SSR marker z13822. To identify the mutation inastti272z , we sequenced the robo2coding region in two presumed homozygous mutant embryos and one presumed homozygous WT embryo from a ti272z/+ ×ti272z/+ incross (10). Comparison ofrobo2 cDNA sequences revealed four differences between theastti272z embryos and their WT sibling that cause predicted amino acid changes and were therefore candidates to be the astti272z mutation.

To distinguish between an induced mutation and naturally occurring polymorphisms, we sequenced the regions of the four candidate changes in genomic DNA from the mutagenized founder fish that gave rise to theastti272z allele (10). Only one candidate differed between the founder andastti272z , an A1903T transversion, which is a nonsense mutation changing Arg635 to Stop before the second FNIII repeat (Fig. 3B). The other three candidates were identical between the founder and astti272z and are therefore naturally occurring polymorphisms between the strains used to generate the astti272z /+ parents. An identical analysis for astte284 revealed a G2645A transition, a missense mutation changing Gly882 to Asp in the transmembrane domain (Fig. 3B). Three other candidate changes found in the te284/+ × te284/+ incross were identified as naturally occurring polymorphisms. The mutations 1903T and 2645A were only found in astti272z andastte284 , respectively, and in no other clone, and were the only sites where the sequence differed between these two alleles. Taken together, these data show that the astphenotype is caused by mutations in zebrafish robo2. Theastti272z allele encodes a truncated Robo2 protein, which could potentially be secreted but cannot function as a receptor, whereas astte284 changes an uncharged Gly to a charged Asp in the transmembrane domain.

In situ hybridization for robo2 with embryos from anastti272z heterozygous incross revealed a slightly reduced RNA level in 8 of 37 embryos (21.6%) at 36 hpf. Genotyping these embryos with z26441 revealed that those with reduced RNA level were ast/ast whereas the rest wereast/+ (n = 19; 51.4%) or +/+ (n = 10; 27.0%). This is consistent with reports that a premature stop codon can cause nonsense-mediated RNA decay (19). No difference in RNA level was detected between 36 hpf embryos of an astte284 heterozygous incross.

The te378 and tl231 alleles were originally assigned to the same gene as te284 and ti272z(8) by virtue of their similar phenotype and noncomplementation. To rule out the possibility of intergenic noncomplementation, we tested for linkage to robo2 and found 0 recombinations out of 86 meioses between astte378 and z26441, and 0 recombinations out of 44 meioses betweenasttl231 and z13822 (10). Therefore, we conclude that they are indeed mutant alleles of robo2.

To determine when robo2 could act to guide RGC axons, we conducted a detailed spatiotemporal expression analysis using whole-mount in situ hybridization (10). We were unable to detect robo2 mRNA in the retina at 28 hpf. At 31 hpf, weak expression was detectable in a ventronasal patch of cells adjacent to the ventral fissure (Fig. 4A). These are presumably the first-born RGCs, which appear in this location between 27 and 28 hpf (20) and project axons across the midline at 33 to 35 hpf (21). At 36 hpf (Fig. 4B), the expression had spread dorsally and temporally, reflecting the pattern of early RGC differentiation (20, 21). By 41 hpf,robo2 was expressed in all quadrants of the RGC layer (Fig. 4, D and E). Although the RGCs expressing robo2 were initially located centrally (Fig. 4C), expression later became peripherally restricted, and we were unable to detect expression in the older central RGCs at 72 hpf (Fig. 4F). Thus, robo2 is first expressed in RGCs shortly after their differentiation and turns off later, consistent with what might be expected for an axon guidance receptor. We have not yet successfully generated antibodies to study Robo2 protein regulation. Intriguingly, robo2 was also expressed at certain points adjacent to the retinotectal projection (10) and in the inner nuclear layer (INL) (Fig. 4, B, E, and F).

Figure 4

Spatiotemporal expression of robo2 mRNA in the eye. (A, B, D, and E) Lateral views between 31 and 72 hpf. (C and F) Transverse sections through the retina at 36 and 72 hpf. (A) A ventronasal patch of RGCs are the first to express robo2mRNA at 31 hpf. (B and C) At 36 hpf, expression in the RGC layer has spread dorsally and temporally. RGCs expressing robo2 are in the central retina (C). A ventronasal patch of cells are the first to express robo2 in the INL. (D) At 41 hpf, RGC expression has completed a full circle and INL expression has spread nasodorsally. (E and F) At 72 hpf, expression of robo2 in the RGC layer and INL is detectable in all four retinal quadrants but not in the central retina (F). ln, lens.

Our genetic mapping and allele sequencing data show that theast phenotype is caused by mutations in zebrafishrobo2. Together with the phenotypic analysis, this shows that ast/robo2 is essential for establishing the retinotectal projection. Because transplantedastti272z RGC axons navigate incorrectly in a WT environment, and from its structural similarity toDrosophila Robo, we conclude that Ast/Robo2 acts as a guidance receptor in RGC axons. Conversely, because WT axons project normally in an astti272z host, it is likely thatast/robo2 function is not required in the environment. The only caveat is that astti272z homozygotes still express some robo2 mRNA, and thus could produce a secreted Robo2 fragment encoded byastti272z . We cannot exclude the possibility that this truncated protein could mimic a normal non–cell autonomous function of Ast/Robo2 and thus guide the transplanted WT axons.

In Drosophila, Robo acts as a guidance receptor that recognizes the repulsive signal Slit, produced by midline glia, and prevents inappropriate crossing of the midline (14,22), whereas the combination of different Robos determines the medial-lateral position of the longitudinal fascicles (23,24). In zebrafish RGC axon guidance,ast/robo2 functions not only to prevent inappropriate midline recrossing, but also to form the optic chiasm and prevent abnormal anterior and posterior projections and optic tract defasciculation. The ast phenotype is thus more reminiscent of the C. elegans sax-3 axon guidance phenotype (12, 13).

Coculture experiments have shown that mammalian Slit2 can repel RGC axons (25), inhibit RGC axon outgrowth (26,27), and cause tighter fasciculation of retinal axons (27). The complex pattern ofslit1, slit2, and slit3 expression along the optic pathway suggested that mammalian Slits might guide retinal axons at positions other than the midline (25–27). From its expression in RGCs, mammalian Robo2 is likely to be the receptor that mediates their response to Slits (26, 27). Together with preliminary observations that two zebrafish Slits are expressed along the optic pathway (28), our Astray/Robo2 functional data suggest a conserved role for this ligand-receptor system in the vertebrate visual system.

  • * To whom correspondence should be addressed at Department of Neurobiology and Anatomy, 401 MREB, University of Utah Medical Center, 50 North Medical Drive, Salt Lake City, UT 84132, USA. E-mail: chi-bin.chien{at}


View Abstract

Navigate This Article