Volume 221, Issue 2 p. 216-230
Free Access

Cloning and expression of three zebrafish roundabout homologs suggest roles in axon guidance and cell migration

Jeong-Soo Lee

Jeong-Soo Lee

Department of Neurobiology and Anatomy, University of Utah, Salt Lake City, Utah

Search for more papers by this author
Russell Ray

Russell Ray

Molecular Biology Program, University of Utah, Salt Lake City, Utah

Search for more papers by this author
Chi-Bin Chien

Corresponding Author

Chi-Bin Chien

Department of Neurobiology and Anatomy, University of Utah, Salt Lake City, Utah

Dept. Neurobiology and Anatomy, 401 MREB, University of Utah Medical Center, 50 North Medical Drive, Salt Lake City, UT 84132Search for more papers by this author
First published: 17 April 2001
Citations: 76


We report the cloning and expression patterns of three novel zebrafish Roundabout homologs. The Roundabout (robo) gene encodes a transmembrane receptor that is essential for axon guidance in Drosophila and Robo family members have been implicated in cell migration. Analysis of extracellular domains and conserved cytoplasmic motifs shows that zebrafish Robo1 and Robo2 are orthologs of mammalian Robo1 and Robo2, respectively, while zebrafish Robo3 is likely to be an ortholog of mouse Rig-1. The three zebrafish robos are expressed in distinct but overlapping patterns during embryogenesis. They are highly expressed in the developing nervous system, including the olfactory system, visual system, hindbrain, cranial ganglia, spinal cord, and posterior lateral line primordium. They are also expressed in several nonneuronal tissues, including somites and fin buds. The timing and patterns of expression suggest roles for zebrafish robos in axon guidance and cell migration. Wiley-Liss, Inc. © 2001 Wiley-Liss, Inc.


The guidance of growing axons to their appropriate targets is a crucial process for the proper function of the nervous system. Growth cones encounter a series of choice points, at which they interpret a combination of local and long-range cues, which may be either attractive or repulsive (reviewed in Mueller, 1999; Tessier-Lavigne and Goodman, 1996). As the molecular mechanisms of axon guidance have begun to be elucidated, it has become clear that axon guidance molecules are highly conserved across species (reviewed in Chisholm and Tessier-Lavigne, 1999). The axon guidance cues described to date include several major classes: Semaphorins, Netrins, Ephrins, and, more recently, Slits. Each class of cues has a cognate class of receptors; for the Slits, the receptors are the Roundabouts (Robos).

robo was originally identified in a large-scale screen in Drosophila as a mutant affecting the development of CNS axon pathways (Seeger et al., 1993). Robo function is critical for a particular choice point, the midline. In robo mutants, ipsilateral axons, which normally do not cross the midline, now cross the midline, while commissural axons, which normally cross the midline only once and subsequently remain contralateral, now cross the midline several times (Kidd et al., 1998b; Seeger et al., 1993). robo has been shown to encode a transmembrane receptor protein that belongs to the immunoglobulin superfamily and is highly expressed in growth cones (Kidd et al., 1998a). The regulation of Robo protein levels depends on the comm gene, through a mechanism that is not totally clear (Kidd et al., 1998b; Tear et al., 1996). Robo receives a repulsive signal from the midline and prevents inappropriate midline crossing (Kidd et al., 1998a). Robo homologs have been cloned in mammals (human and rat Robo1 and Robo2: Kidd et al., 1998a; Sundaresan et al., 1998; mouse Rig-1: S.S. Yuan et al., 1999) and C. elegans (Sax-3: Zallen et al., 1998). Their structural similarity suggests that their molecular function is conserved across species, but the in vivo function of vertebrate Robos has not yet been characterized.

Slit was identified as a secreted extracellular protein expressed in midline glia and suggested to be involved in commissural axon development in Drosophila (Rothberg et al., 1988, 1990). Recently, Slit has been shown to be the chemorepulsive midline signal for the Robo receptor in Drosophila (Kidd et al., 1999). In vertebrates, three Slit homologs have been cloned in mammals (Brose et al., 1999; Holmes et al., 1998; Itoh et al., 1998; Li et al., 1999), while only one Xenopus Slit (Li et al., 1999), one partial chick Slit (Li et al., 1999), and two zebrafish Slits (Yeo et al., 2001) have so far been described. Vertebrate Slits have been shown in vitro to act as repellents for several classes of axons (e.g., Brose et al., 1999; Chen et al., 2000; Li et al., 1999; Nguyen Ba-Charvet et al., 1999; Niclou et al., 2000). Intriguingly, Slits also can repel several neuronal types as well as muscle precursors during migration (Hu, 1999; Kidd et al., 1999; Wu et al., 1999; Zhu et al., 1999), and Slit2 can promote branching of sensory axons (Wang et al., 1999), suggesting multiple roles during embryogenesis. It is not clear at present whether Robo mediates all of the actions of Slits. More in vivo studies are required to fully understand the functions of Robos and Slits, especially in the vertebrate embryo.

The zebrafish is an excellent vertebrate model system to study developmental processes because its embryo is transparent and can be easily manipulated (see, for example, Halloran et al., 2000). The embryonic nervous system is relatively simple, early axons grow in a highly stereotyped fashion, and neuronal subtypes and their projections have been well characterized for forebrain (Chitnis and Kuwada, 1990; Ross et al., 1992; Wilson et al., 1990), hindbrain (Mendelson, 1986a,b; Hanneman et al., 1988), and spinal cord (Bernhardt et al., 1990; Kuwada et al., 1990; Myers et al., 1986). The relatively short generation time (∼3 months) and large clutch size facilitate genetic screening in a vertebrate. Many axon pathfinding mutants have been found in a large-scale screen (Granato et al., 1996; Karlstrom et al., 1996), but few have yet been cloned (ace: Shanmugalingam et al., 2000; noi: Brand et al., 1996; yot: Karlstrom et al., 1999). It is, therefore, of interest to identify zebrafish axon pathfinding molecules, not only to study their function directly, but also as candidate genes that may be affected in known mutants. While many axon pathfinding molecules have been identified in zebrafish (reviewed in Bernhardt, 1999), no zebrafish Robo homologs have yet been described in the literature.

To begin to study the roles of zebrafish Roundabouts, we cloned and analyzed sequences of three zebrafish Robos, which we have named Robo1, 2, and 3. Phylogenetic analysis shows that Robo1 and Robo2 are the orthologs of mammalian Robo1 and Robo2, respectively, while Robo3 is likely to be an ortholog of mouse Rig-1, a more distantly related Robo family member (S.S. Yuan et al., 1999). Cytoplasmic motifs shared among the other Robo family members are also found in the zebrafish Robos, suggesting their conserved roles in intracellular signaling.

In order to see where zebrafish Robos may play roles in development, we have characterized their expression patterns in detail. Whole-mount in situ hybridization shows high expression both in the nervous system and in several non-neuronal tissues during embryogenesis. The expression patterns of the three Robos are distinct in some structures, but overlapping in others. These data suggest that zebrafish Robos play roles in axon guidance and cell migration. A role in axon guidance has been confirmed by our recent demonstration (Fricke et al., in press) that the robo2 gene is mutated in the retinal axon pathfinding mutant astray.


Cloning of Zebrafish Roundabout Homologs

To clone zebrafish Roundabout homologs, we designed degenerate primers against highly conserved residues in the immunoglobulin domains of known Robos (see Experimental Procedures for details), and carried out RT-PCR on total RNA isolated from zebrafish embryos at 36 hr postfertilization (hpf). This yielded partial clones for three independent roundabout (robo) genes. Based on inferred orthologies with mammalian Robos (see below) and following the established nomenclature for zebrafish, we have named these genes zebrafish robo1, robo2, and robo3. For clarity, we will refer to the previously characterized Drosophila, human, and rat Robos as DRobo, HRobo, and RRobo throughout this paper. By combining cDNA library screening and 5′ RACE, we obtained full-length coding sequences for robo2 and robo3 (see Experimental Procedures for details). Full-length robo2 and robo3 encode deduced proteins of 1,513 and 1,389 amino acids, respectively. Start codons were predicted based on preceding stop codons and putative hydrophobic signal sequences near the N-terminal ends of the deduced proteins. Both Robo2 and Robo3 have the structure characteristic of the Robo family (Kidd et al., 1998a), comprising an extracellular domain containing five immunoglobulin (IG) domains and three fibronectin type III (FN) domains; a single transmembrane domain; and a long cytoplasmic domain containing several conserved cytoplasmic motifs (Figs. 1, 2A). The partial clone for robo1 encodes 376 amino acids spanning the first four IG domains.

Details are in the caption following the image

Deduced amino acid sequences of zebrafish Roundabouts aligned against human Robo1 (HRobo1) and Drosophila Robo1 (DRobo1) (Kidd et al., 1998a). The characteristic Robo structure includes five immunoglobulin (IG) domains, three fibronectin type III (FN) domains, a single transmembrane domain (TM), and four conserved cytoplasmic motifs (CM). Dark and pale gray boxes represent identical and similar amino acids, respectively.

Details are in the caption following the image

A: Diagrammatic comparison of Robo family members. The full-length sequences of rat Robo2 and zebrafish Robo1 are not yet available; missing regions are shown as dashed lines. For Robo3 and MRig-1 (Mouse Rig-1), the half “2” indicates that only the first half of CM2 is conserved. zf = zebrafish; IG = immunoglobulin domain; FN = fibronectin type III domain. B: The phylogeny of the Robo family based on the alignment of conserved IG domains. Percentages indicate amino-acid similarities. Robo1 and Robo2 are closest to HRobo1 and RRobo2, respectively, while Robo3 is equally distant from HRobo1, RRobo2, and MRig-1. HCDO is a human protein that also has a 5 IG + 3 FN extracellular structure but is not a Robo family member (Kang et al., 1997). C: Comparison of cytoplasmic motifs among Robo family members. Consensus residues are shown in bold. – denotes an absent motif and ? denotes lack of information due to partial sequence. h = hydrophobic residue. Motifs 1 and 3 are highly conserved and motif 0 is fairly well conserved between all Robos, whereas only the first half of motif 2 is conserved in Robo3 and MRig-1. The CM0 squence in C. elegans Sax-3 is tentative due to its weak conservation.

Extracellular and Cytoplasmic Domains Are Characteristic of the Robo Family

To assign orthologies with known vertebrate Robos, and search for evolutionarily conserved motifs, we analyzed the extracellular and intracellular domains of the zebrafish Robos. As for other Robo family members, the extracellular domains are highly conserved (Figs. 1, 2A). Phylogenetic analysis of the IG domains (Fig. 2B) indicates that the zebrafish Robos are all clearly members of the Robo family, for two reasons. First, they cluster with the three known mammalian Robo family members, Robo1, Robo2, and Rig-1 (Kidd et al., 1998a; S.S. Yuan et al., 1999). Second, they are much more similar to DRobo1 (Kidd et al., 1998a) and its C. elegans homolog, Sax-3 (Zallen et al., 1998) than they are to the human CDO protein, which also has a 5 IG + 3 FN extracellular structure but is not a Robo family member (Kang et al., 1997). This IG domain analysis (Fig. 2B) shows that Robo1 is clearly most similar to HRobo1 (79.8% amino-acid similarity) and Robo2 is clearly most similar to RRobo2 (79.9% similarity). Robo3 is equally distant from HRobo1, RRobo2, and mouse Rig-1 (MRig-1) (71.0, 67.6, and 68.4%, respectively).

The Robo cytoplasmic domains are relatively divergent, but contain four conserved cytoplasmic motifs (CMs): CM0, 1, 2, 3 (Bashaw et al., 2000; Kidd et al., 1998a). CM0 and CM1 are potential tyrosine phosphorylation sites, and CM2 and CM3 are proline-rich motifs involved in protein-protein interactions (Bashaw et al., 2000; Kidd et al., 1998a). Zebrafish Robo2 and Robo3 also contain these four cytoplasmic motifs, which are almost identical to the known motifs in mammalian Robo2 and MRig-1, respectively (Figs. 1 and 2C; note that the partial Robo1 clone does not cover the cytoplasmic domain). This confirms that zebrafish Robos are bona fide Robo family members, and suggests conserved downstream signaling pathways among Robos. There are, however, subtle variations in CM2 of Robo3/MRig-1, and CM3 of vertebrate Robos, which may have implications for intracellular signaling (see Discussion).

Taken together, the similarities in the extracellular domains and cytoplasmic motifs suggest that Robo1 and Robo2 are the orthologs of mammalian Robo1 and Robo2, respectively. Similarities in the cytoplasmic domains and genetic mapping data (data not shown) suggest that Robo3 is most likely the ortholog of mammalian Rig-1 (see Discussion for more details). Our nomenclature for the zebrafish Robos reflects these inferred orthologies.

Overview of robo Expression Patterns

All three robos are expressed in the nervous system (consistent with their inferred role as axon guidance receptors) as well as some non-neuronal tissues. They are initially expressed throughout the rostrocaudal axis, becoming rostrally restricted as development progresses. The three robos each have distinct expression patterns, which are complementary in some tissues and overlap in others.

The earliest expression of robo1 and robo3 is detected uniformly throughout the gastrula around 8hpf, becoming restricted to the neural keel at 12hpf, with strongest expression in the head and tail (data not shown). At 24hpf, robo1 shows widespread expression in many tissues, with distinct expression in ventral hypothalamus, tegmentum, hindbrain, spinal cord, somites, and the migrating lateral line primordium (Fig. 3A). At 48hpf, it is expressed in the brain, including tectum, and the pectoral fin buds, with no expression in the somites or spinal cord (Fig. 3D). Its expression becomes weaker and is barely detectable by 72hpf (data not shown). At 24hpf, robo3 shows distinct expression in telencephalon, ventral hypothalamus, parts of the midbrain and hindbrain, spinal cord, and somites (Fig. 3C). At 48hpf, it is expressed in the brain and in pectoral and caudal fin buds (Fig. 3F). robo3 expression is also barely detectable by 72hpf (data not shown).

Details are in the caption following the image

Expression patterns of zebrafish robos during embryonic development shown by whole-mount RNA in situ hybridization. Lateral views, rostral to the left. A–C: 24hpf; D–F: 48hpf. All three robos are initially expressed throughout the rostrocaudal axis, but become largely restricted to the head at later stages. All three are expressed in both the nervous system and in nonneuronal structures. Expression in the indicated structures is discussed in detail in the text. cfb, caudal fin bud; e, eye; hb, hindbrain; op, olfactory placode; ov, otic vesicle; pfb, pectoral fin bud; prim, posterior lateral line primordium; sc, spinal cord; som, somites; tec, tectum; tel, telencephalon; tg, trigeminal ganglion; vht, ventral hypothalamus. Scale bars = 240 μm in A–C, 300 μm in D–F.

robo2 expression begins 3–4 hr later than the other two robos (around 12hpf), with weak expression in the hindbrain (data not shown). At 24hpf, robo2 is distinctly expressed in olfactory placode, telencephalon, ventral hypothalamus, hindbrain, trigeminal ganglia, lateral line ganglia, and spinal cord (Fig. 3B). At 48hpf, it is expressed in tectum, hindbrain, and retina, but no longer in spinal cord (Fig. 3E). Its expression is still high at 72hpf with a pattern similar to 48hpf (data not shown).

Zebrafish robos Have Distinct Expression Patterns in Hindbrain and Spinal Cord


Three pairs of regularly-spaced ventrolateral cell clusters express robo1 in the hindbrain at 24hpf (Fig. 4A,D). robo2 is expressed in four pairs of cell clusters in rostral hindbrain at 24hpf (arrowheads in Fig. 4B) and is also highly expressed medial to the otic vesicle in the ventrolateral hindbrain (Fig. 4B,E). robo3 is also expressed in ventrolateral hindbrain at 24hpf (Fig. 4F), but differs from robo1 and robo2 in that its expression appears to be continuous throughout the rostrocaudal extent of the hindbrain (Fig. 4C).

Details are in the caption following the image

Expression of robos in the developing hindbrain at 24hpf. A–C: Dorsal view of whole-mounts, rostral up; D–F: transverse sections, dorsal up. Double-headed arrows in A–C indicate the approximate level of transverse sections shown in D–F, respectively. Arrowheads indicate robo-expressing cells. A,D: robo1 is expressed in three pairs of cell clusters. B,E: robo2 is expressed in four segmentally repeated pairs of clusters in rostral hindbrain as well as in trigeminal ganglia (arrows in B), anterior lateral line and acoustico-vestibular ganglia (asterisks in B). C,F: robo3 is broadly expressed throughout the hindbrain. All three robos are expressed primarily in mediolateral or ventrolateral hindbrain (D–F). cb, cerebellum; ov, otic vesicle; tg, trigeminal ganglia. Scale bar = 50 μm in A–C; 25 μm in D–F.

At 24hpf, the first reticulospinal (RS) neurons and commissural neurons in the hindbrain have been born (Hanneman et al., 1988; Kimmel, 1993; Mendelson, 1986a), and subsets of RS neurons have begun to project axons (Mendelson, 1986b). RS neurons are known to develop from ventrolateral portions of the hindbrain and to later be displaced dorsally (Mendelson, 1986b). Based on these previous observations, it is likely that some of the robo-expressing cells represent RS neurons due to their locations and timing of axogenesis. However, it will be necessary to double-stain with antibodies or in situ probes that recognize specific hindbrain subpopulations in order to identify the robo-positive cells definitively.

Cranial ganglia.

At 24hpf, three pairs of ganglia adjoining the hindbrain express robo2 intensely. These are the trigeminal ganglia, just posterior to the cerebellum (marked “tg” in Fig. 4B), and the anterior lateral line and acoustico-vestibular ganglia, near the otic vesicle (asterisks in Fig. 4B). The trigeminal ganglion gives rise to some of the earliest axons in the embryo at about 16.5hpf (Metcalfe et al., 1990). Trigeminal axons pioneer the lateral longitudinal fascicle in hindbrain (Metcalfe et al., 1990) and synapse with the Mauthner neuron immediately after axogenesis (Kimmel et al., 1990). The acoustico-vestibular and lateral line ganglia project their first axons and then synapse with the Mauthner neuron at 23hpf and 25hpf, respectively (Kimmel et al., 1990). Thus, robo2 is expressed in the cranial ganglia at an appropriate time to be involved in axon guidance.

Spinal cord.

All three zebrafish robos are distinctly expressed in subsets of spinal cord cells at 24hpf. robo1 is expressed in a pair of cells in mediolateral spinal cord (Fig. 5D) which is segmentally repeated (Fig. 5A; also see asterisks in Fig.7A). robo2 and robo3 are expressed in spinal cord cells that are dispersed irregularly along the rostrocaudal axis (Fig. 5B,C). robo2-positive cells are located in lateral spinal cord including dorsolateral, mediolateral, and ventrolateral positions (Fig. 5E), while robo3-positive cells are mostly found in dorsolateral spinal cord (Fig. 5F). The development of spinal neurons has been well documented in embryonic zebrafish, and they are distinguishable by cell body location and especially axonal trajectory (Bernhardt et al., 1990; Kuwada et al., 1990; Myers et al., 1986). Based on the relatively lateral location of the somata, it is likely that the robo-positive cells in lateral spinal cord represent a subset of the spinal interneurons, while some of the dorsolateral cells may be Rohon-Beard neurons, consistent with the fact that their earliest axons are growing at this stage (Kuwada et al., 1990). However, double-label analysis will be required to identify the robo-expressing cells definitively.

Details are in the caption following the image

Expression of robos in developing spinal cord at approximately the level of the 10th somite at 24hpf. A–C: Dorsal view of whole-mounts, rostral up. D–F: Transverse sections, dorsal up. Arrowheads indicate robo-expressing cells. Lines in A denote the somite boundaries. A,D: robo1 expression appears to be in a single pair of cells per segment in the mediolateral spinal cord. B,E: robo2 is expressed in cells irregularly dispersed along the rostrocaudal axis in the lateral spinal cord. Some of the robo2-positive cells in B are out of focus due to their position along the dorsoventral axis. C,F: robo3 is also expressed in cells irregularly dispersed along the rostrocaudal axis, but these cells are mostly restricted to the dorsolateral spinal cord. n, notochord; som, somite; sc, spinal cord. Scale bar = 25 μm.

Expression of robos in the Developing Visual System

The robos are differentially expressed in the developing visual system. robo2 is strongly expressed in the developing RGC layer in 48hpf retina (Fig. 6B). While robo2 is expressed throughout the RGC layer at 36hpf (data not shown), its expression is restricted to the peripheral retina by 48hpf. robo2 is also expressed in the inner nuclear layer at this stage (inl in Fig. 6B). In contrast to robo2, robo1 and robo3 do not appear to be expressed in the retina, although robo3 may be present at a very low level at 48hpf (Fig. 6C). The first retinal ganglion cells (RGCs) exit the retina at 32hpf and project across the optic chiasm at 36hpf, reaching their target, the optic tectum, at 48hpf (Stuermer, 1988; Burrill and Easter, 1995). Later born RGCs are added to peripheral retina and send their axons along the path of the earliest retinal axons. The peripheral expression of robo2 at 48hpf thus suggests that it is transiently expressed only in young RGCs and later turns off. The expression of robo2 in RGCs suggests that this gene could play an important role in retinal axon guidance. Indeed, we have recently shown (Fricke et al., in press) that robo2 is the gene disrupted in astray, an axon guidance mutant isolated in a large-scale screen for retinotectal mutants (Karlstrom et al., 1996), confirming that robo2 is necessary for retinal axon guidance.

Details are in the caption following the image

Expression of robos in the developing visual system at 48hpf. Transverse sections through the retina at the level of the optic nerve, also showing hypothalamus and optic tectum. Lateral to the right; dorsal up. Arrowheads indicate ventral hypothalamus. A: robo1 is weakly expressed in the ventral hypothalamus and broadly in the tectum, but not in the retina. B: robo2 is expressed in the retinal ganglion layer and inner nuclear layer of the peripheral retina. robo2 is also highly expressed in the ventral hypothalamus and broadly in the tectum. C: robo3 mRNA is barely detectable in the retina. robo3 expression in the tectum is restricted to a subset of superficial cells in the tectum. ht, hypothalamus; inl, inner nuclear layer; ln, lens; on, optic nerve; rgc, retinal ganglion cell layer; tec, tectum; teg, tegmentum. Scale bar = 50 μm.

In addition, robo2 is highly expressed in ventral diencephalon just dorsal to the optic chiasm; robo1 and robo3 are also expressed in this structure, although very weakly (arrowheads in 6A–C). Transverse sections also show distinct expression patterns of robos in the optic tectum, the primary target of the RGC axons. robo1 and robo2 expression is widespread in the tectum, with robo2 expressed more strongly than robo1(Fig. 6A,B). In contrast, robo3 is expressed in only a subset of superficial cells in the tectum (Fig. 6C).

robo1 Is Expressed in the Migrating Lateral Line Primordium

robo1 shows a striking expression pattern in the posterior lateral line (PLL) primordium. The PLL is a sensory system that detects water flow via hair cells in the neuromasts. The neuromasts are positioned in a line along the body and arise from a migrating cell mass called the PLL primordium (Metcalfe, 1985; Metcalfe et al., 1985). At 24hpf, robo1 expression is found in the primordium at the level of the 5th somite (Figs. 3A, 7A). The primordium continues to express robo1 as it migrates caudally (Fig. 7B,C). PLL staining is no longer detected by 48hpf (see Fig. 3D). Neither robo2 nor robo3 expression is detected in the primordium during these stages. This expression suggests a possible role for robo1 in guiding the cells of the PLL primordium during migration.

Details are in the caption following the image

robo1 is expressed in the migrating posterior lateral line primordium. Lateral views of whole-mounts, rostral to the left. A: 24hpf; B: 30hpf; C: 36hpf. Inset in A: a higher magnification view of the primordium. Arrowheads indicate the migrating primordium. Asterisks in A denote the segmentally repeated spinal cord expression of robo1 (see Fig 5A). cfb, caudal fin bud; ye, yolk extension. Scale bar = 200 μm.

robos Are Differentially Expressed in Non-Neuronal Tissues Including Somites and Fin Buds


robo1 is broadly expressed in nearly all somites at 19hpf (data not shown). At 24hpf, robo1 somite expression has become restricted to the caudal two-thirds of the embryo (Figs. 3A, 8A). By 30hpf, robo1 expression in somites has nearly disappeared (Fig. 7B). Developing somites in the caudal half of the embryo express robo3 at 19hpf (data not shown), and by 24hpf its expression has become restricted to the more caudal somites (Figs. 3C, 8C; compare Fig. 8F to 5F). At 30hpf, only the caudalmost somites express robo3 (data not shown). robo2 is not detectably expressed in the somites during these stages (Figs. 3B, 8B,E). The somites begin to develop at 10hpf, with one pair generated every 20–30 min by the progressing somitic furrow, eventually yielding approximately 30 somite pairs at ∼24hpf (van Eeden et al., 1996). Zebrafish robos may be involved in somite development since their expression appears to be correlated with the wave of somite formation.

Details are in the caption following the image

Differential expression of robos in somites at 24hpf at approximately the level of the 20th somite. A–C: Lateral views of whole-mounts, rostral to the left. D–F: Transverse sections, dorsal up. Arrowheads (A,C) and brackets (D,F) indicate robo-expressing somites. A,D: robo1 is broadly expressed in developing somites. B,E: robo2 is not expressed in somites, in contrast to its distinct expression in the spinal cord. C,F: robo3 is highly expressed in developing somites, but is more distinct in the dorsalmost and ventralmost portions. n, notochord; sc, spinal cord; som, somite. Scale bar = 50 μm in A–C; 25 μm in D–F.

Fin buds.

Another example of the expression of robos in non-neuronal tissues is in the pectoral and caudal fin buds. In the pectoral fin buds, all three robos are expressed at 36hpf, but robo2 expression later turns off (data not shown). At 48hpf, robo1 is widely expressed throughout the pectoral fin bud (Fig. 9A,C), while robo3 expression is restricted to the distal pectoral fin bud (Fig. 9B,D). At this stage of development, distal mesenchymal cells in the pectoral fin bud are beginning to migrate into the apical fold along the developing actinotrichia, transforming the apical fold into the fin fold (Grandel and Schulte-Merker, 1998). By its location, robo3 expression appears to be in the migrating mesenchymal cells. In the caudal fin bud, robo3, but not robo1 or robo2, is expressed in the mesenchymal cells at 36hpf (Fig. 9E) and at 48hpf (Fig. 3F). The apical ridge for the caudal fin bud appears at ∼24hpf and the subepidermal space and adjacent actinotrichia are formed between 27–36hpf (Dane and Tucker, 1985), suggesting that this is when the caudal fin mesenchymal cells begin to migrate. In both pectoral and caudal fin buds, the location and timing of robo3 expression are correlated with mesenchymal cell migration events, suggesting a role for robo3 during fin morphogenesis.

Details are in the caption following the image

Expression of robo1 and robo3 in developing pectoral (A–D) and caudal (E) fin bud. A,B: Dorsal views of whole-mounts, rostral up; C,D: transverse sections, dorsal up; E: lateral view, rostral to the left. A–D at 48hpf; E at 36hpf. Brackets in A–C and arrowheads in D,E indicate robo-expressing regions. Double-headed arrows in A and B indicate the approximate level of transverse sections in C and D, respectively. A,C: robo1 is broadly expressed throughout the pectoral fin bud. B,D: robo3 is expressed in the distal portion of the developing pectoral fin, most likely representing migrating mesenchymal cells. E: At 36hpf, the caudal fin bud mesenchyme shows high expression of robo3. The orientation of the pectoral fin buds follows the definition of Grandel and Schulte-Merker (1998). som, somite; pfb, pectoral fin bud. Scale bars = 100 μm in A,B; 25 μm in C,D; 50 μm in E.


Robo and Slit form one of the principal receptor-ligand pairs for axon guidance. Their function is best understood in the midline crossing decision in Drosophila. Slit, a repellent signal, is secreted from the midline, while the Robo receptor, expressed in the growth cones, recognizes Slit in order to keep axons from crossing the midline inappropriately (Kidd et al., 1998a, 1999). In vertebrates, Slits have been shown to repel several types of extending axons as well as migrating neurons (Brose et al., 1999; Hu, 1999; Li et al., 1999; Nguyen Ba-Charvet et al., 1999; Niclou et al., 2000; Wu et al., 1999; Zhu et al., 1999). The complementary expression of Slits and Robos during vertebrate development (Brose et al., 1999; Holmes et al., 1998; W. Yuan et al., 1999) and the binding of Slit and Robo in cellular assays (Brose et al., 1999; Li et al., 1999; W. Yuan et al., 1999; Chen et al., 2000) suggest that the Robo-Slit receptor-ligand relationship is conserved in vertebrates, but the functions of vertebrate Robos are largely unproven, especially in vivo.

To begin to analyze the function of zebrafish Robos, we have cloned three zebrafish Roundabout homologs and characterized their expression patterns by whole-mount in situ hybridization. Zebrafish Robo1, Robo2, and Robo3 appear to represent the orthologs of mammalian Robo1, Robo2, and Rig-1, respectively. Robo2 and Robo3 share conserved cytoplasmic motifs with Robos in other species, suggesting phylogenetic conservation of intracellular signaling pathways. All three robo mRNAs show very dynamic and distinct, but overlapping, expression patterns in a variety of neuronal cell types, both in the CNS and in cranial ganglia. They are also differentially expressed in non-neuronal tissues including somites and fin buds, suggesting additional functions besides axon guidance.

Orthologies of Zebrafish Robos With Other Vertebrate Robos

We have deduced orthologies for the zebrafish Robos based primarily on analysis of their sequences. Phylogenetic analysis of the immunoglobulin (IG) domains shows that Robo1 and 2 cluster closest to mammalian Robo1 and 2, respectively (Fig. 2B). The four cytoplasmic motifs (CMs) are nearly identical between Robo2 and mammalian Robo2 (Fig. 2C). Based on these observations and on genetic mapping data (Fricke et al., in press), we conclude that zebrafish Robo1 and Robo2 are orthologs of mammalian Robo1 and Robo2, respectively.

Robo3 is more problematic: it is clearly a Robo family member, but does not seem to be an ortholog of mammalian Robo1 or Robo2. MRig-1 is a third Robo family member, which was found in a screen for genes upregulated in the Rb-deficient mouse (S.S. Yuan et al., 1999). Robo3 is equally distant from HRobo1, RRobo2, and MRig-1 (71.0, 67.6, and 68.4% amino acid similarity in the IG domains, respectively) (Fig. 2C). Might Robo3 be the zebrafish ortholog of mammalian Rig-1? We conclude that this is likely, for two reasons. First, the cytoplasmic motifs of Robo3 and MRig-1 are almost identical, including similar variations in CM2 (Fig. 2C). Unlike Robo3, MRig-1 lacks CM1, but this might be explained by the existence of many alternatively spliced forms of MRig-1 (S.S. Yuan et al., 1999). Second, the EST clone fb78e10, which corresponds to the 3′ region of zebrafish robo3, maps approximately 337cR from the top of LG10 on the zebrafish hybrid map (see Experimental Procedures). This region shows conserved synteny with the region containing human rig-1 on chromosome 11 (data not shown). Therefore, we conclude that Robo3 is most likely the ortholog of mammalian Rig-1.

Intracellular Signaling by Robos

The guidance cues to which the growth cone responds during navigation must be transduced to affect actin polymerization and cell adhesion, and ultimately to steer the growth cone. Signal transduction by Robo is carried out through its cytoplasmic domain, as shown by domain swapping experiments (Bashaw and Goodman, 1999). The four conserved cytoplasmic motifs have been described based on conservation between C. elegans, Drosophila, and mammalian Robos (Kidd et al., 1998a; Bashaw et al., 2000). CM0 and CM1 are potential tyrosine phosphorylation sites, while CM2 and CM3 are proline-rich sequences that are involved in protein–protein interactions. While no function has yet been assigned to CM0, CM0 of HRobo1 has been shown in vitro to be a site for tyrosine phosphorylation (Bashaw et al., 2000), and the CM0 sequence is fairly well conserved between Robos, including zebrafish Robos (Fig. 2C). The function of the other three motifs has begun to be elucidated in Drosophila (Bashaw et al., 2000), and comparison of these motifs in the zebrafish Robos suggests possible conserved roles in signaling.

In Drosophila, it appears that CM1 in DRobo1 is phosphorylated by Abl, a cytoplasmic tyrosine kinase, and Robo function is antagonized by this phosphorylation (Bashaw et al., 2000). In addition, Drosophila receptor protein tyrosine phosphatases (RPTPs) interact genetically with robo and appear to be positive regulators of Robo function (Sun et al., 2000). As discussed by Bashaw et al. (2000), the phosphorylation status of CM1, controlled by tyrosine kinases and RPTPs, may be essential for the function of Robos. CM1 is nearly identical in zebrafish Robo2, Robo3, and all other known Robos (except MRig-1, which lacks CM1; Fig. 2C), suggesting that CM1 in zebrafish Robos may be regulated by the same intracellular proteins, such as c-Abl homologs. If Robo3 is, indeed, the ortholog of MRig-1, it would imply either that its intracellular signaling has diverged during evolution, or that there may be an undescribed spliced form of MRig-1 that contains CM1.

CM2 in DRobo1 has been shown to bind Ena, a cytoplasmic adapter protein, and to partially mediate Robo function (Bashaw et al., 2000). The LPPPP sequence in CM2, which was shown in vitro to bind the mouse Ena homolog, Mena (Niebuhr et al., 1997), is completely conserved in all the known Robo family members including zebrafish Robo2 and 3 (Fig. 2C). This suggests conservation of signaling pathways across species, probably through Ena homologs. However, it should be noted that in contrast to its nearly absolute conservation in other Robos, the second half of the CM2 motif is not conserved in Robo3 and MRig-1 (Fig. 2C), suggesting a divergent signaling function for CM2 in Robo3 and MRig-1. One possibility is that the Robo3/MRig-1 CM2 may bind a distinct Ena-containing complex, conferring a different downstream signaling capability.

CM3 can bind the Abl SH3 domain and is at least partly necessary for DRobo1 function (Bashaw et al., 2000). Interestingly, there is a subtle difference between the invertebrate and vertebrate CM3 motifs: all known vertebrate CM3 sequences contain the conserved Mena binding site LPPPP, but this sequence is not present in CM3 of Sax-3 or DRobo1 (Fig. 3C). Perhaps vertebrate CM3 motifs can bind Ena homologs, which would represent a difference in signaling pathways between invertebrates and vertebrate Robos.

In comparing CM sequences between species, zebrafish is especially useful as a vertebrate that is phylogenetically distant from mammals. For instance, all four CMs are identical between HRobo1 and RRobo1, but this is not surprising given the 96% overall amino-acid similarity between these two orthologs. Our analysis of the Robo cytoplasmic motifs in zebrafish suggests that some signaling pathways have been conserved between invertebrates and vertebrates, but that there may be subtle vertebrate-specific differences, especially for CM2 and CM3.

What Are the Roles of Zebrafish Robos During Embryogenesis?

In vitro and in vivo studies in other species have suggested that Robos and Slits may be involved in several aspects of development including axon guidance, axon branching, and cell migration. Our group has recently shown that the robo2 gene is disrupted in the astray retinal pathfinding mutant (Fricke et al., in press). However, the astray phenotype is remarkably specific: astray homozygotes are adult viable (our unpublished data) and the forebrain axon scaffold is normal (Karlstrom et al., 1996). Indeed, the retinal projection defect is the only known phenotype of astray at present. The expression analysis of robo2 presented here will be critical in guiding future experiments to understand when and where robo2 acts.

While the expression patterns of slit1, 2, and 3 have been examined in detail (e.g., Itoh et al., 1998; Holmes et al., 1998; Li et al., 1999; W. Yuan et al., 1999), the expression studies of vertebrate robos are much less complete. The expression patterns of mammalian robo1 and 2 have been studied in the developing spinal cord (Brose et al., 1999; Wang et al., 1999; W. Yuan et al., 1999), eye (Erskine et al., 2000; Niclou et al., 2000; Ringstedt et al., 2000), kidney (Piper et al., 2000), olfactory bulb, and hippocampus (Nguyen Ba-Charvet et al., 1999). Mammalian robo1 expression also has been described in somites, limb buds, and trigeminal ganglia (W. Yuan et al., 1999). rig-1 expression is much less characterized: expression patterns have been briefly described in mouse hindbrain and spinal cord (S.S. Yuan et al., 1999).

The expression patterns that we have observed for the three zebrafish robos are largely consistent with those previously seen in spinal cord, eye, somites, limb buds, and trigeminal ganglia. However, we have described for the first time expression of robos in olfactory placode, lateral line primordium, and other cranial ganglia, and analyzed the expression of robo2 and 3 in detail including their expression patterns in hindbrain, fin buds, and somites.

Axon guidance.

In Drosophila, genetic evidence clearly shows that Robo acts as an axon guidance receptor for the repulsive signal Slit, expressed by the ventral midline (Kidd et al., 1998a, 1999). In vertebrates, Slit proteins have been shown in vitro to repel rat spinal motor axons, chick olfactory bulb axons, mouse hippocampal axons, and mouse and rat retinal ganglion cell (RGC) axons, while Robos are expressed in these neuronal tissues and presumed to be the relevant receptors (Brose et al., 1999; Erskine et al., 2000; Li et al., 1999; Nguyen Ba-Charvet et al., 1999; Niclou et al., 2000; Ringstedt et al., 2000).

The expression of zebrafish Robos in the nervous system suggests possible roles as axon guidance receptors for several neuronal types. For instance, robo2 is expressed in a subset of interneurons in lateral spinal cord (Fig. 5). Among the described interneurons are CoPA and CiA, which project long contralateral and ipsilateral axons, respectively (Kuwada et al., 1990). As described in rat (Brose et al., 1999) and mouse (W. Yuan et al., 1999), zebrafish slits are expressed in the floor plate of spinal cord (Yeo et al., 2001), a structure analogous to the ventral midline of Drosophila. Thus robo2 may act to determine whether the axons of the interneurons cross the midline, as in the Drosophila midline. The expression of zebrafish robos in interneurons (Fig. 5) is consistent with the expression of robo1 in commissural neurons in the developing mammalian spinal cord (Brose et al., 1999; W. Yuan et al., 1999). While robo1 and robo2 are expressed in rat motoneurons (Brose et al., 1999; W. Yuan et al., 1999), we do not find significant motoneuron expression at 24hpf (Fig. 5).

Another example is in the developing retina: robo2 is expressed in the RGC layer (Fig. 6B), as is mammalian robo2 (Erskine et al., 2000; Niclou et al., 2000; Ringstedt et al., 2000), suggesting a role in guiding RGC axons. Indeed, retinal axons in astray display multiple midline crossing, as well as several other dramatic guidance defects. This confirms the role of robo2 in axon guidance in zebrafish. Interestingly, while mammalian robo1 is expressed in scattered cells in the developing RGC layer (Erskine et al., 2000; Niclou et al., 2000; Ringstedt et al., 2000), neither zebrafish robo1 nor robo3 appear to be expressed in the RGCs (Fig. 6).

Cell migration and somitogenesis.

Besides their proven role in axon pathfinding, several pieces of evidence implicate Robos in cell migration. First, Slits are involved in several mesodermal and neuronal cell migrations. In Drosophila, muscle precursor cell migration away from the midline is defective in slit mutants (Kidd et al., 1999). In vertebrates, Slit can act as a repellent for several classes of migrating neurons in vitro, and is expressed appropriately to control those migrations in vivo (Hu, 1999; Wu et al., 1999; Zhu et al., 1999). Second, although it is not yet proven that Robo mediates all of Slit's effects on cell migration, interfering experiments in which the extracellular fragment of Robo was used to nullify Slit function suggest that Robo is responsible for some cell migration events (Wu et al., 1999; Zhu et al., 1999). Third, the sax-3 mutation in C. elegans disrupts the cell migrations of some neuronal types (Zallen et al., 1999) and is involved in sex myoblast migration (Branda and Stern, 2000). Finally, Ena and Abl, which as discussed above are downstream effector proteins for Robo signaling, are involved in cell migration (reviewed in Machesky, 2000).

Zebrafish robo1 is expressed in the cells of the lateral line primordium as they migrate (Metcalfe et al., 1985), and thus could play a role in this migration. Zebrafish robo3 is expressed (Fig. 9) in what are likely to be migrating mesenchymal cells in both pectoral and caudal fin buds (Dane and Tucker, 1985; Grandel and Schulte-Merker, 1998). Zebrafish slit3 is also expressed in the fin buds at the same stage (Yeo et al., 2001), suggesting that zebrafish robo3 and slit3 are involved in cell migration events during fin formation. Consistent with this idea, mouse robo1 and slits are expressed in the forelimb (W. Yuan et al., 1999), a structure homologous to the pectoral fin.

In addition, zebrafish robo1 and robo3, but not robo2, are expressed in somites at 24hpf (Fig. 8) and a zebrafish slit is expressed in the somites at the same stage of development (our unpublished data). Mouse robo1 and slits are also expressed in somites (W. Yuan et al., 1999). One possible role of robos and slits in the somites is that they may be involved in the formation of somite boundaries, as has been suggested for Eph signaling (Durbin et al., 1998).

Implications of Overlapping robo Expression

While the three robos have distinct expression patterns, at certain places during development their expression domains appear to overlap. All three robos appear to overlap in the ventrolateral hindbrain (Fig. 4D–F), as do robos in the lateral spinal cord (Fig. 5E,F), and robo1 and robo3 in caudal somites (Fig. 8D,F). Some individual cells in these tissues are likely to express more than one robo. What is the function of this overlap? The simplest possibility is redundancy: two or three robos may be expressed concurrently in order to assure that Robo functions properly without errors. Redundancy with other robos may explain the apparent specificity of the astray/robo2 phenotype. Another intriguing possibility is a “combinatorial effect”': different Robos may heterodimerize with each other to create a receptor with new specificity for ligands or intracellular components. The possibility of dimerization is supported by the Robo protein structure: the immunoglobulin domains and fibronectin type III repeats have been shown in some cases to be involved in protein-protein interactions (for instance, Silletti et al., 2000). Dimerization is also suggested by the dominant-negative action of a RoboΔC construct lacking the cytoplasmic domain (Bashaw and Goodman, 1999). Using the molecular and genetic tools available in zebrafish to elucidate how the three Robos signal in vivo, whether alone or together, and how they interact with Slits, will be important for understanding their roles during embryogenesis.

Experimental Procedures


Fish were maintained using standard methods. For RT-PCR, 5′-RACE, and in situ hybridization, wild-type embryos from the WIK L11, Tübingen, and AB*/gol strains were raised at 28.5°C, and staged according to time after fertilization and morphology (Kimmel et al., 1995).

cDNA Cloning and Sequence Analysis

To isolate zebrafish Roundabout homologs, we performed degenerate RT-PCR with oligonucleotide primers designed based on the conserved IG domains of human, rat, and Drosophila Robo1 (Kidd et al., 1998a) using the CODEHOP program (Rose et al., 1998) (two forward primers, 5′-GGG CGA GCC TGC CAC NYT NAA YTG-3′, 5′-CGC CAT GCT GCG AGA NGA YTT YMG-3′ and two reverse primers, 5′-CGT AGT TGG ACT CTC GCT CTC CNA CNA DRT T-3′, 5′-GGT GCC GTC CAC CAT CAV NGT YTG RTT-3′). Total RNA from 36 hr postfertilization (hpf) wild-type embryos was prepared and reverse transcribed, using either random hexamers or a gene-specific primer, followed by PCR: 60 s at 94°C, then 35 cycles of 30 s at 94°C, 60 s at 55°C, 90 s at 72°C. Amplified PCR products were cloned into the pCR2.1Topo vector (InVitrogen), restriction-analyzed, and sequenced, yielding PCR fragments for Robo1, 2, and 3. A longer Robo2 clone covering the 3′ coding region was obtained by screening a lambda cDNA library from 22–26hpf embryos (kindly provided by Motoko Aoki and Hitoshi Okamoto) at high stringency using the Robo2 PCR fragment as a probe. Membranes were hybridized at 42°C in 50% formamide, 5×SSPE, 5×Denhardt's solution, and 0.5% SDS with 100 μg/ml yeast RNA for 18 hr with α-32P-dCTP labeled probe, and then washed with 2×SSPE/0.1%SDS at 50°C, 0.5×SSPE/0.1%SDS at 55°C, and 0.1×SSPE/0.1%SDS at 60°C two times (30 min for each wash). A positive clone containing a ∼3.8-kb insert was cloned into pBluescript SKII (Stratagene). To obtain the 5′ end of Robo2, 5′ RACE was carried out using the 5′ RACE Marathon kit (Clontech): the poly(A) mRNAs were selected and reverse transcribed using a mixture of random hexamers and a gene specific primer (5′-GTT TCG ACT GAC TGC TTC CCC CAA G-3′) located in the most 5′ end of the Robo2 PCR fragment. cDNAs were amplified by touchdown PCR according to the manufacturer's instructions. A full-length Robo3 clone (ICRFp524B1824Q8) was obtained by screening a high-density gridded cDNA library (German Human Genome Project, library no. 524, late somitogenesis stages, constructed by Matthew Clark) using the same conditions described for Robo2. Robo2 and Robo3 clones were completely sequenced in both directions. Searching the zebrafish ESTs accessible through http://zfish.wustl.edu revealed EST clone fb78e10, which covers the 3′ end of Robo3 (Research Genetics: independently confirmed by our sequencing). The mapping position for fb78e10 was obtained from the EST database (done by Dr. Zon's lab: http://zfrhmaps.tch.harvard.edu/ZonRHmapper)

Alignments of the deduced amino acid sequences and a phylogenetic tree were made with the Clustal method using MacVector and MegAlign, respectively. Immunoglobulin and fibronectin type III domains were predicted by searching the Pfam database (Bateman et al., 2000) and transmembrane domains were predicted using the TMHMM and TopPred 2 algorithms (Sonnhammer et al., 1998; von Heijne, 1992). The GenBank accession numbers for the nucleotide sequences of robo1, robo2, and robo3 are AF337034, AF337035, and AF337036, respectively.

Whole-Mount In Situ Hybridization

Wild-type embryos were grown in 0.1M phenylthiourea to block pigmentation, then dechorionated, fixed in 4% PFA in PBS overnight, and stored in 100% methanol before use. Whole-mount in situ hybridization was performed as previously described (Schulte-Merker et al., 1992) with slight modifications. Briefly, Robo RT-PCR fragments or the Robo2 lambda clone were in vitro transcribed using T7 polymerase and labeled with digoxigenin (DIG) to prepare the antisense RNA probes. Embryos were washed with PBS containing 0.1% Tween-20 and treated with Proteinase K (10 μg/ml) for 5 to 20 min depending on stage. After postfixation with 4% PFA, embryos were treated with acetic anhydride in 0.1M triethanolamine, hybridized at 65°C overnight with antisense RNA probe, washed with SSC containing 0.1% Tween-20 at 65°C, and blocked with 2% blocking reagent (Boehringer Mannheim) for 1 hr to overnight. Embryos were incubated in alkaline phosphatase-conjugated anti-DIG Fab fragment, washed, and then developed by incubating in AP substrate (Boehringer Mannheim).

Paraffin Sections

Overstained whole-mount in situ hybridized embryos were dehydrated through a graded ethanol series, pre-infiltrated with paraffin/isopropanol (1:1) at 60°C, then infiltrated with paraffin. They were then embedded, allowed to harden overnight, and sectioned at 10–20 μm. Sections were deparaffinized in Histosol and mounted in Pro-Texx mounting medium (Baxter). Scale bars for sections have not been corrected for shrinkage.


We thank Drs. Maureen Condic, Jeff Essner, David Grunwald, and members of the Chien lab for comments on the manuscript and helpful advice, and Amy Kugath for technical assistance. We also thank Dr. Hitoshi Okamoto for sending a cDNA library and for sharing data before publication. Supported by a seed grant from the University of Utah Research Foundation and by NIH grant EY-12873 to C.-B. C. Note added in press: another group has independently isolated zebrafish robo1 and robo3 and observed similar expression patterns during embryogenesis (Challa et al., 2001).