The zebrafish unplugged gene controls motor axon pathway selection

During embryogenesis motor neurons extend axonal growth cones to establish precise and stereotyped neural connections with the body musculature. A major challenge for an individual motor growth cone is to select its appropriate axonal path among many possible paths present for similarly searching growth cones. This process of axonal pathfinding is facilitated by intermediate targets and choice points, often located at positions where axonal pathways diverge. At intermediate targets and choice points, specialized cells or extracellular matrix provide a mixture of guidance cues controlling two features of axonal pathfinding: growth cone guidance between intermediate targets (or choice points) and pathway selection at the choice point. Specialized cells at the midline of flies, worms and vertebrates control such growth cone behavior through a sophisticated system of attractive and repellent signals (Terman and Kolodkin, 1999; Van Vactor and Flanagan, 1999). In contrast, the identity of local signals and the guidance capacity of cells at motor pathway choice points are less clear. In the Drosophila embryo, motor neuron growth cones of the segmental and intersegmental nerves encounter multiple well-characterized choice points before reaching their muscle targets (Jacobs and Goodman, 1989; Thomas et al., 1984; Van Vactor et al., 1993). Some of these choice points are associated with identified mesodermal cells, reminiscent of guidepost cells in other organisms (Bate et al., 1991; Van Vactor et al., 1993). Classical and molecular genetic studies have identified numerous genes controlling many steps of this process, including cell surface molecules (e.g. Sema I, Yu et al., 1998; Plexin A, Winberg et al., 1998; DLAR, Desai et al., 1996; Krueger et al., 1996), transcription factors (e.g. lim3, Thor et al., 1999) and intracellular signaling molecules (e.g. Drac1, Kaufmann et al., 1998), all of which function on or within motor neurons. Surprisingly, none of the known Drosophila choice point genes encodes a cue provided by mesodermal cells at the choice point and capable of controlling pathway selection.

In the vertebrate embryo, different populations of spinal motor neurons initially share common paths, but diverge at specific choice points. For example, in the avian hindlimb, motor growth cones extend within a common path to a plexus region at the base of the limb, where individual axonal populations diverge into the appropriate trunk nerve path, and then diverge further into the correct muscle nerves (Tosney, 1991). When lumbosacral motor neurons were displaced by two segments along the neuraxis, they exited the spinal cord via the adjacent segmental nerve and selected their appropriate path within the plexus region (Lance-Jones and Landmesser, 1980). These and other elegant manipulations have provided evidence that motor growth cones respond to environmental cues that might be located within the plexus region (Lance-Jones, 1988; Tosney, 1991). Recent progress has identified factors that act within vertebrate motor neurons to control cell specification and target selection (Appel, 1999; Eisen, 1998; Pfaff and Kintner, 1998), as well as intrinsic factors converting external cues into growth cone motility (Lanier et al., 1999). In contrast, little is known about guidance cues that direct pathway selection of motor growth cones locally at choice points.

The zebrafish embryo is an excellent model system in which to study pathway selection of spinal motor neurons (Eisen, 1994, 1998). As in other vertebrates, zebrafish spinal motor neurons have distinct identities and targets. Each somitic hemisegment is typically innervated by three pioneering motor neurons: the caudal primary (CaP), middle primary (MiP) and rostral (RoP) primary neurons (Fig. 1A). They first project their axons in common, segmentally reiterated nerves, and then pioneer individual paths towards their somitic muscle targets (Eisen et al., 1986; Myers et al., 1986; Westerfield et al., 1986). All three motor growth cones initially extend on a common path along the medial surface of the somites (Bernhardt et al., 1998; Eisen et al., 1986). Migration of all three motor growth cones along this common path is dependent on diwanka gene activity. In diwanka mutant embryos, primary motor axons are able to navigate correctly within the spinal cord, but fail to project on the common path from the spinal cord to the somites (Zeller and Granato, 1999). At the distal end of the common path, CaP, MiP and RoP growth cones contact a group of specialized cells called the muscle pioneers (Felsenfeld et al., 1991; Melançon et al., 1997). At this choice point, they pause before selecting cell-type-specific paths to ventral, dorsal and medial myotomal regions, respectively (Fig. 1A). Thus, the region where motor growth cones commonly pause before they pioneer independent pathways might provide signals for such divergent pathway decisions. Laser ablation experiments have shown that muscle pioneer cells are not critical for pathway selection (Melançon et al., 1997), suggesting that other cells may guide the decisions of motor axons at the choice point.

We examined a collection of mutants with presumed defects in the neural circuitries underlying coordinated body movements in order to identify genes controlling motor axon pathfinding (Granato et al., 1996; M. G., unpublished results). Here, we show that unplugged embryos display axonal guidance defects specific for motor neurons. We focused on the development of primary motor axons because they can be individually identified and because they pioneer the peripheral motor nerves (Eisen et al., 1986; Myers et al., 1986; Westerfield et al., 1986). In unplugged mutants, pathway selection of two motor neurons, CaP and RoP, is affected while pathway selection of MiP neurons appears unaffected. In mutant embryos, CaP and RoP growth cones navigate correctly from the spinal cord to the somitic choice point, but fail to select their subsequent cell-type-specific path. We show that unplugged-dependent pathway selection is provided by a small and selective population of somitic cells, and we identify these cells as dorsal adaxial cells, flanking the choice point. Moreover, the discovery that motor growth cones require guidance cues from adaxial cells, both for their migration from the spinal cord to the choice point (Zeller and Granato, 1999) and for pathway selection, suggests a central role of adaxial cells in motor axon guidance. Our results identify the unplugged gene as generating or encoding one of the first external cues that controls motor axon pathway selection. We hypothesize that unplugged is part of a local signaling mechanism that evokes different pathway decisions in CaP and RoP growth cones.

Zebrafish were raised and maintained as described by Mullins et al. (1994). A single unplugged allele, unp^te314b, was identified in a large scale genetic screen (Granato et al., 1996). This allele was used in a subsequent non-complementation screen (van Eeden et al., 1999), which yielded six additional unplugged alleles: unp^tbr307, unp^tbo81, unp^tbt187, unp^tbb72, unp^tbr160 and unp^tdp253. Experiments described here were performed using the unp^te314b allele in a mixed TL/WIK genetic background or the stronger allele unp^tbr307 in a mixed TL/Tü genetic background. The p⁴ allele is a γ-ray-induced deficiency, which fails to complement the unplugged phenotype. Homozygous p⁴ embryos lack chromosomal markers to both sides of the unplugged locus, suggesting that the p⁴ allele lacks the entire unplugged gene. The phenotypic and molecular characterization of the p⁴ allele will be published elsewhere.

Individual CaP, MiP and RoP motor neurons were labeled by single blastomere injection, as described in Zeller and Granato (1999) and by cell transplantation (see below). We took advantage of the fact that unplugged is required cell non-autonomously for motor axon pathfinding (see Table 2A). Thus, when genotypically wild-type cells developed in mutant unplugged host embryos, they formed mutant CaP and RoP trajectories, indistinguishable from those obtained by single blastomere injection of unplugged mutant embryos. Embryos with motor neurons and adaxial cells derived from transplanted, genotypically wild-type cells were not included in this analysis. CaP, MiP and RoP motor neurons were identified by (a) their dorsoventral position within the spinal cord, (b) their rostrocaudal position within a spinal hemisegment, (c) the size of their soma, and (d) by their axonal trajectories before reaching the choice point. Other ventral neurons present at 27 hours postfertilization (hpf) include secondary motor neurons, KA neurons and VeLD neurons (Bernhardt et al., 1992; Kuwada et al., 1990a,b; Myers et al., 1986; Pike et al., 1992). Using the criteria given above, we can distinguish these cell types from mutant primary motor neurons. The secondary motor neurons can be distinguished by their smaller soma size (∼1/2 the size of primary motor neurons, Myers et al., 1986) and the more ventral position of their somata in the spinal cord (Myers et al., 1986). KA neurons are located close to the floorplate and extend their axons rostrally (Bernhardt et al., 1992). Although VeLD neurons are located in the same region as the primary motor neurons, at 26 hpf VeLD growth cones have extended up to 10 segments caudally (Kuwada et al., 1990). In about 50% of the wild-type hemisegments an fourth primary motor neuron with a variable fate, VaP, is present (Eisen et al., 1990). The soma of VaP is adjacent to the CaP soma and like CaP neurons, they extend a growth cone ventrally towards the choice point (Eisen et al., 1990). Wild-type CaP and VaP can be distinguished because at 20 hpf CaP growth cones have extended into the ventral somite, while VaP growth cones never extend ventrally from the choice point (Eisen et al., 1990). This may lead to the conclusion that the presumed CaP growth cones that aberrantly stop at the choice point and branch, are in deed VaP growth cones. This seems unlikely because, in unplugged mutant embryos, the vast majority of potential CaP or VaP growth cones stall or branch aberrantly at the choice point (85%, Table 1), while we observe the normal number of VaP neurons in unplugged mutant hemisegments (50%, Fig. 3F).

Axonal trajectories of identified spinal neurons were labeled as follows: Rohon-Beard mechanosensory neurons, DoLA, VeLD and CoSA by cell transplantion and antibody stainings using an anti-acetylated tubulin antibody (Piperno and Fuller, 1985), KA neurons by cell transplantation and antibody staining using an anti-L-GABA antibody (Bernhardt et al., 1992), CoPA neurons by cell transplantation and antibody staining using the 3A10 antibody (Hatta, 1992), and the hindbrain commissural Mauthner neurons by antibody staining using the 3A10 antibody (Hatta, 1992). Labeled neurons were examined at 26-27 hpf and their identities were determined by (a) their dorsoventral position within the spinal cord, (b) the size of their soma, and (c) by the path of their axonal trajectories as previously described by Bernhardt et al. (1990) and Kuwada et al. (1990a,b). For each neuronal class, we analyzed at least 7 axonal trajectories in at least 7 embryos.

Antibody staining was performed essentially as described by Zeller and Granato (1999). The following primary antibodies were used: anti-acetylated tubulin antibody (1:1000; Piperno and Fuller, 1985; Sigma), znp-1 (1:200; Trevarrow et al., 1990; Antibody Facility, University of Oregon), 3A10 (1:50; Hatta, 1992; kindly provided by T. Jessell), prox-1 (1:2000; Glasgow and Tomarev, 1998; a gift from S. Tomarev), 4D9, which recognizes zebrafish Engrailed proteins (1:200; Hatta et al., 1991; Developmental Studies Hybridoma Bank, University of Iowa), zn-5, which recognizes DM-GRASP (1:500; Fashena and Westerfield, 1999; Trevarrow et al., 1990; Antibody Facility, University of Oregon), and F59 (1:10; Crow and Stockdale, 1986; Devoto et al., 1996; kindly provided by F. Stockdale).

In situ hybridization was performed as previously described in Odenthal and Nusslein-Volhard (1998) using the following probes: islet-1, islet-2 (Appel et al., 1995; kindly provided by I. David), twist (Morin-Kensicki and Eisen, 1997; kindly provided by B. Riggleman), and wnt 11 (Makita et al., 1998; kindly provided by H. Takeda). Stained embryos were dehydrated, viewed and documented as described below.

Chimeric embryos were generated as previously described in Zeller and Granato (1999). Analysis of chimeric embryos was performed as previously described by Zeller and Granato (1999), with the following modification for the simultaneous analysis of chimeric cells, motor axons and adaxial cells. After F59 antibody staining, chimeric embryos were fixed at room temperature for 1 hour in 4% PFA and rinsed three times with PBST before further processing.

Stained embryos were dehydrated through a MeOH series (10% / 30% / 50% / 70% / 90% / 100%), cleared in a mixture of benzylbenzoate and benzylalcohol (2:1), mounted in Canada Balsam/methyl salicylate (10:1) and viewed using Nomarski optics on a Zeiss Axiophot microscope. DIC images were acquired via a digital camera (Progres 3012, Kontron), saved on a Macintosh computer and processed with Adobe Photoshop 4.0.1 software. For fluorescent microscopy, stained embryos were sectioned, mounted in Vectashield (Vector Laboratories) and analyzed using a Zeiss Axiophot microscope. For DIC combined with fluorescent microscopy, stained embryos were sectioned, mounted in Vectashield (Vector Laboratories) and analyzed with a Hamamatsu Color Chilled 3CCD camera. Simultaneous visualization of transplanted cells and adaxial cells (Fig. 6) was performed on a Leica confocal microscope.

We used a motor axon-specific antibody (znp-1; Trevarrow et al., 1990) to examine the trajectories of spinal motor neurons in unplugged mutant embryos. At 27 hpf, the znp-1 antibody labels CaP, MiP and RoP axons along their common path, from the spinal cord to the choice point, as well as CaP and MiP axons along their cell-type-specific path into ventral and dorsal somites, respectively (Melançon, 1994; Melançon et al., 1997). In over 97% of the somitic hemisegments that we examined in wild-type embryos, primary motor axons projected along the common path from the spinal cord to the choice point and continued along their cell-type-specific path (segments 6-15, n=600, Fig. 1B). In unplugged mutant embryos, znp-1-positive axonal trajectories from the spinal cord towards the choice point appeared unaffected whereas, in the region of the choice point, over 84% of the motor axon projections were severely affected (n=539, Fig. 1C,D). In the majority of the cases (76%), a CaP-like axonal projection invaded the ventral somite and ectopic caudal and/or rostral projections emerged at the choice point (Fig. 1C). In about 7% of the mutant hemisegments, znp-1-positive motor axons failed to extend into the ventral somite, but formed ectopic caudal and rostral branches at the choice point (Fig. 1D). Conversely, in all mutant hemisegments analyzed, pathway selection of presumptive MiP neurons into the dorsal somite appeared unaffected (Fig. 1C,D). Thus, znp-1 antibody stainings in unplugged mutants revealed pathfinding defects predominantly at the choice point, indicating a potential role for unplugged in axonal pathway selection.

To define the role of the unplugged gene in more detail, we asked which of the znp-1-positive primary motor neurons in unplugged mutants were affected and how their trajectories were altered. Given that the znp-1 epitope is expressed on CaP, MiP and RoP axons, znp-1 stainings did not allow us to determine which of these three motor axons contributed to the mutant axonal phenotype. To visualize individual CaP, MiP and RoP trajectories, we labeled these neurons in wild-type and unplugged mutant embryos by transplanting vital-dye-labeled cells and by single blastomere injections (Fig. 2, Table 1; for details see Materials and Methods). Labeling of CaP, MiP and RoP neurons in wild-type embryos revealed their well-described stereotypic axonal trajectories (Fig. 2; Eisen et al., 1986; Myers et al., 1986; Westerfield et al., 1986). At 26 hpf, CaP and MiP growth cones in unplugged siblings extended into ventral and dorsal somites, respectively, while RoP growth cones always stopped and paused at the choice point (Fig. 2A,D,E).

In unplugged mutant embryos, MiP growth cones always selected their correct path, while CaP and RoP growth cones failed to make appropriate pathway decisions at the choice point. Labeled CaP growth cones navigated correctly to the choice point but instead of migrating into the ventral somite, they stalled at the choice point, where they often formed short, aberrant branches (71%; Fig. 2B; Table 1). In a small fraction of unplugged mutants, we observed CaP axons forming two long and distinct branches at the choice point (14%; Fig. 2C; Table 1).

One branch extended past the choice point into the ventral somite, while a second branch extended caudally or rostrally at the level of the choice point (Fig. 2C). Conversely, labeled RoP growth cones did not pause at the choice point, but formed aberrant branches at the choice point (38%; Fig. 2G; Table 1) or even invaded the ventral somite (23%; Fig. 2F; Table 1). This observation is significant since, in wild-type embryos, RoP growth cones never extend into the ventral somite (n=52, compare Fig. 2E to 2F). Overall, in unplugged mutant embryos, more than 70% of CaP and RoP growth cones failed to make their appropriate pathway choice at the choice point. Moreover, at 26 hpf 71% CaP growth cones made a RoP-like pathway choice and 23% of RoP growth cones made a CaP-like pathway choice at the choice point (Fig. 2H; Table 1). Together, these results demonstrate that unplugged is essential for CaP and RoP growth cones to select their cell-type-specific path at the choice point.

To investigate if mutations in the unplugged gene affect axonal guidance and pathway selection throughout the nervous system or specifically affect pathway selection of CaP and RoP motor neurons, we examined axonal trajectories of various neuronal cell types. We first asked whether axonal trajectories of four identified spinal neurons, Rohon-Beard mechanosensory neurons (RB; Kuwada et al., 1990), dorsal longitudinal ascending (DoLA; Kuwada et al., 1990), ventral longitudinal descending (VeLD; Kuwada et al., 1990) and Kolmer Agduhr (KA; Bernhardt et al., 1992) neurons were affected in unplugged mutant embryos. Axonal trajectories of these neurons were visualized by antibody staining or by transplantation of vital-dye-labeled cells (for details see Materials and Methods; n refers to the number of axonal trajectories analyzed). In unplugged embryos, axonal projections of RB (n=81), DoLA (n=22), VeLD (n=7) and KA (n=10) neurons appeared indistinguishable from those in wild-type embryos (data not shown).

We then asked whether mutations in the unplugged gene affect axon guidance of commissural neurons at the central nervous system (CNS) midline. Floor plate cells at the vertebrate CNS midline serve as intermediate targets to attract growth cones of commissural neurons and to repel those of non-crossing interneurons (Bovolenta and Dodd, 1991; Kennedy et al., 1994; Nguyen Ba-Charvet et al., 1999). We reasoned that, if unplugged participates in general axonal pathway selection, then axon guidance of commissural interneurons at the CNS midline might also be affected. Analysis of commissural Mauthner neurons in the hindbrain (n=10; Metcalfe et al., 1986), spinal commissural primary ascending neurons (CoPA; n=10; Kuwada et al., 1990) and spinal commissural secondary ascending neurons (CoSA; n=65; Kuwada et al., 1990) in unplugged mutant embryos revealed axonal trajectories indistinguishable from those in wild-type embryos (Fig. 3A,B and data not shown). Together, these results strongly suggest that unplugged is not required for general axonogenesis or general pathway selection, but is essential for guiding CaP and RoP growth cones at the choice point.

In unplugged mutant embryos, a significant number of neurons with soma positions consistent with CaP neurons made a RoP-like pathway decision and, conversely, neurons with soma positions consistent with RoP neurons made a CaP-like pathway choice at the choice point (Fig. 2). To examine if this defect is caused by a switch in cell type identity between CaP and RoP neurons, we analyzed the expression of the LIM domain genes islet-1 and islet-2. Similar to other vertebrates, individual subpopulations of zebrafish spinal motor neurons express specific combinations of LIM homeobox genes (Appel et al., 1995; Eisen, 1998). Moreover, transplantation of individual motor neurons to new spinal cord positions resulted in the initiation of a new program of LIM gene expression, as well as axonal trajectories characteristic for their new positions, suggesting that LIM gene expression reflects the specification of individual motor neurons (Appel et al., 1995). In wild-type embryos two adjacent cell bodies (RoP and MiP) at the rostral somite boundary express islet −1 (see Fig. 3C), whereas a single cell body (CaP), located in the center of the somite, expresses islet-2 (Fig. 3E). In unplugged mutant embryos, we observed wild-type expression patterns of islet-1 and islet-2 mRNA, demonstrating that unplugged CaP and RoP neurons express their specific complement of LIM genes (Fig. 3D,F).

We also examined the number, position and specification of muscle pioneer cells located at the choice point. Muscle pioneer cells are a specialized group of somitic cells that may influence motor axon pathway selection (Melancon, 1994). To determine if muscle pioneer cells were present and properly specified in unplugged mutants, we examined the expression of engrailed and DM-GRASP, two molecules characteristic for muscle pioneer cells (Fashena and Westerfield, 1999; Hatta et al., 1991; Kanki et al., 1994; Trevarrow et al., 1990). The expression levels and patterns of Engrailed and DM-GRASP proteins in unplugged muscle pioneer cells were indistinguishable from those in wild-type muscle pioneer cells (Fig. 3G,H and data not shown). Thus, our analysis shows that mutations in the unplugged gene do not overtly affect motor neuron or muscle pioneer specification, supporting the idea that unplugged functions primarily in motor axon pathway selection.

In order to understand the biological mechanism of how unplugged influences axonal pathway selection, we asked whether unplugged activity is provided by motor neurons themselves, or by surrounding cells. Pathway selection requires the specific interaction of axonal growth cones with a multitude of external cues (Goodman and Tessier-Lavigne, 1996). Accordingly, genes controlling this event can be divided into two main groups depending on where they function: in neurons or in cells of the embryonic environment. In the zebrafish, this question can be addressed through the analysis of chimeric embryos (Ho and Kane, 1990). To generate chimeric embryos, we transplanted genotypically wild-type, dye-labeled cells at the dome stage into age-matched embryos derived from heterozygous unplugged fish. At 27 hpf, we identified unplugged mutant embryos by their motility phenotype (Granato et al., 1996) and examined these embryos for the presence of labeled, genotypically wild-type CaP and RoP neurons. In unplugged host embryos, genotypically wild-type CaP and RoP motor neurons developed mostly mutant motor axonal trajectories (Table 2A). Reciprocally, we transplanted dye-labeled cells from homozygous mutant unplugged embryos into wild-type embryos. In these host embryos, genotypically mutant CaP and RoP motor neurons developed wild-type-like axonal trajectories (Fig. 4B,D; Table 2A). These results demonstrate that the unplugged gene is not required in CaP and RoP motor neurons, but instead functions cell non-autonomously in the embryonic environment to mediate correct pathway selection.

To identify the cell type(s) that provides unplugged activity, we generated a second set of chimeric embryos and examined cell types in the vicinity of motor neurons for their ability to restore pathway selection of unplugged motor axons. In this set of experiments, wild-type cells were transplanted at the dome stage into unplugged embryos and, at 26 hpf, chimeric embryos were stained with znp-1 antibody to visualize motor axonal projections. We then correlated the identity of wild-type-derived cells with their ability to restore motor pathway selection. It was important to analyze a large number of chimeric unplugged embryos because in non-chimeric homozygous mutant unplugged embryos, we observed that 16% of the somitic hemisegments contained motor axonal projections indistinguishable from those in wild-type embryos. Consistent with this observation, in chimeric host embryos 8% of all somitic hemisegments lacked wild-type-derived cells but displayed wild-type-like axonal trajectories. Given the incomplete penetrance of the unplugged phenotype, we therefore only considered values well above 16% to be significant for the ability of a particular cell type to provide unplugged activity.

Analysis of 143 chimeric unplugged embryos revealed that wild-type-derived notochord, floorplate, neural crest and epithelial cells failed to rescue the axonal phenotype (Fig. 5A-C; Table 2B). Even when genotypically wild-type cells populated combinations of tissues, such as the notochord and floorplate (n=14; Table 2B), they failed to restore motor axon trajectories (Fig. 5B). In contrast, genotypically wild-type cells located in the somites were able to rescue the axonal phenotype, although to a variable degree, depending on their positions. Medially located somitic cells were not able to rescue the axonal phenotype (Fig. 5E,F; Table 2B), while approximately 34% of the somitic hemisegments with wild-type cells populating medial and lateral regions were capable of restoring axonal trajectories (Table 2B). Moreover, when wild-type cells were predominately located in lateral regions of the somite, rescue of the axonal phenotype was evident in 46% of the hemisegments (Fig. 5G,H; Table 2B). Interestingly, wild-type-derived muscle pioneer cells, located at the motor axon choice point, did not rescue the axonal phenotype (Fig. 5A,D; Table 2B). Our results demonstrate that cells in the lateral somite can provide unplugged activity to motor neuron growth cones.

The position of the somitic cells capable of rescuing the unplugged motor axon phenotype is consistent with the position of adaxial cells (Devoto et al., 1996). Adaxial cells are initially located in the medial region of the somite, adjacent to the notochord. After somite formation, adaxial cells organize into a single cell layer along the medial surface of the somite and later migrate radially through the somites to form a single layer of slow muscle fibers (Devoto et al., 1996). During radial migration, adaxial cells maintain their dorsoventral positions, and by 26 hpf, they have migrated to the lateral portion of the somite (Devoto et al., 1996). A subset of adaxial cells, the muscle pioneers, do not migrate to the lateral surface of the somite, but instead span the depth of the somite at the region of the future choice point (Devoto et al., 1996). To determine if the lateral somitic cells capable of restoring pathway selection in unplugged mutants were adaxial cells, we generated a third set of chimeric embryos by transplanting wild-type cells into unplugged mutant host embryos. In these chimeric embryos, we monitored simultaneously the position of the transplanted, genotypically wild-type cells (rhodamine dextran labeled, red in Fig. 6), the axonal path of primary motor axons (znp-1 antibody labeled, blue in Fig. 6), and the position of adaxial cells using an adaxial cell-specific antibody, F59 (Devoto et al., 1996; green in Fig. 6).

In 100 chimeric unplugged embryos, we examined a total of 148 somitic hemisegments in which wild-type-derived cells contributed to somitic tissue. In 108 hemisegments, wild-type-derived cells were distributed throughout the somite, but did not contribute to adaxial cells (Fig. 6A-C′). In the vast majority of these hemisegments (89/108, 82%), we observed the characteristic unplugged phenotype (Fig. 6D), showing that non-adaxial somitic cells were unable to restore pathway selection. In contrast, wild-type cells that developed into three or more dorsal adaxial cells were able to rescue the unplugged axonal phenotype very efficiently (16/17, 94%; Fig. 6E-H). These rescuing adaxial cells were located in a dorsal domain, adjacent to the muscle pioneers and the choice point, encompassing a region of about 6 to 7 cells (arrows in Fig. 6G). Wild-type-derived adaxial cells located further dorsally or in the ventral half of the somite failed to rescue the mutant phenotype (1/23, 4%; Fig. 6I-L). Moreover, we find that, in addition to their positions, the number of wild-type-derived domain were sufficient to rescue the mutant phenotype (16/17, 94%; Fig. 6E-H). In summary, our chimeric analyses demonstrate that a small group of dorsal adaxial cells, confined to a region adjacent to the choice point, is crucial for unplugged-mediated pathway selection of motor axons.

We used the well-characterized neuromuscular system of the zebrafish embryo to study pathway selection of motor growth cones. Our phenotypic analyses implicate the unplugged gene in a local, rather than a global mechanism specialized in motor axon guidance. We demonstrate that an exclusive set of adaxial cells in the zebrafish embryo provides unplugged-dependent signaling, essential for two different subtypes of motor growth cones to select their appropriate paths. To our knowledge, this represents a unique example in vertebrates where a group of identified somitic cells influences divergent growth cone decisions at a motor axon choice point. Our studies identify the unplugged gene as generating or encoding one of the first external cues required for pathway selection of motor axons. We propose a model in which unplugged activity governs pathway selection by eliciting different responses in CaP and RoP growth cones.

A total of seven recessive mutant unplugged alleles were isolated in a genetic screen for zygotic mutations affecting zebrafish embryogenesis and in a subsequent screen for non-complementation of the unplugged phenotype (Granato et al., 1996; van Eeden et al., 1999). Mutants of the different unplugged alleles are readily identifiable by their reduced motility between 26 and 48 hpf (Granato et al., 1996). Analysis of embryos from all seven ENU-induced unplugged alleles reveals variation in the penetrance of the axonal phenotype (Table 1; data not shown), possibly reflecting the presence of residual unplugged activity in mutant embryos. However, embryos carrying one ENU-induced allele, unp^tbr307, and the p4 allele in which the unplugged locus dorsal adaxial cells was equally important. The presence of only one or two wild-type-derived adaxial cells within the dorsal domain was insufficient to rescue the mutant phenotype (3/17; 17%; Fig. 6M-P), while three or more cells in the same is entirely deleted, display a similar range of motor axon defects as observed in unp^tbr307/unp^tbr307 mutant embryos, indicating that the unp^tbr307 allele is a strong hypomorphic or amorphic allele (data not shown).

Analysis of axonal trajectories of identified spinal cord interneurons, sensory neurons and primary motor axons indicates that the unplugged gene specifically acts in the guidance decision of CaP and RoP growth cones at the choice point. In unplugged mutant embryos, axonal growth and pathfinding of all sensory and interneuronal cell types examined appear normal (e.g. Fig. 3A,B). Furthermore, in unplugged mutant embryos MiP motor neurons always select their appropriate path, while CaP and RoP motor neurons fail to make their appropriate pathway selection at the choice point (Fig. 2; Table 1). As in wild-type embryos, unplugged CaP growth cones navigate correctly to the choice point. However, unplugged mutant CaP growth cones stall at the choice point, while wild-type CaP growth cones extend into the ventral somite (Fig. 2; Table 1). Often mutant CaP growth cones extend multiple small branches at the region of the choice point, reminiscent of transient CaP branches observed in 20 hpf wild-type embryos (see Fig. 5 in Liu and Westerfield, 1990).

In wild-type embryos CaP motor neurons initially form one or two short branches at the region of the choice point, but these branches are later retracted (Liu and Westerfield, 1990; Myers et al., 1986; Westerfield et al., 1986). By 27 hpf, CaP branches at the choice point are not detectable in wild-type and unplugged sibling embryos (see Figs 2A, 4A), while they are present in unplugged mutant embryos (Fig. 2B). Thus, it is possible that, in mutant embryos, transient CaP branches at the choice point persist or even extend, reflecting a random search for alternative cues in the absence of unplugged-dependent guidance.

While wild-type RoP growth cones always stop and pause at the choice point, in unplugged mutant embryos, RoP growth cones explore the region around the choice point or extend far ventrally into the somite. Thus, unplugged activity enables CaP growth cones to pioneer the ventral somite path and prevents RoP growth cones from inappropriately crossing the choice point. Our results demonstrate that the unplugged gene specifically controls growth cone behavior of CaP and RoP neurons at the choice point. Moreover, the unplugged mutant phenotype suggests that pathway selection of primary motor neurons requires different cues at the choice point. While CaP and RoP neurons depend on unplugged activity to make their appropriate guidance decisions, MiP neurons appear to utilize a separate, yet unknown signaling mechanism, to direct the MiP growth cone towards the dorsal somite.

En route to their synaptic partners, pioneering growth cones encounter specialized cells that serve as intermediate targets or flank choice points. Specialized cells at the CNS midline provide a well-studied repertoire of cues controlling both axonal guidance towards the midline and pathway selection at the choice point (for reviews see: Chien, 1998; Flanagan and Van Vactor, 1998; Terman and Kolodkin, 1999; Van Vactor and Flanagan, 1999). In contrast, the guidance capacity of cells at choice points along motor pathways is less clear (Tosney, 1991). In the chick embryo, the plexus region at the base of the limb bud serves as an intermediate target, where individual motor neuron populations make their first pathway selection (Lance-Jones and Landmesser, 1981; Tosney and Landmesser, 1985a,b). In the absence of all limb bud tissue distal to the plexus, motor growth cones correctly navigate through the plexus region, suggesting that the cues for specific pathway decisions are target independent and, thus, may reside within the plexus region (Tosney and Landmesser, 1984). In the zebrafish, all primary motor growth cones contact a group of muscle pioneer cells at the choice point, suggesting that these cells may provide important signals for proper pathway selection. However, in somitic hemisegments in which all muscle pioneer cells were ablated, CaP and MiP growth cones select their appropriate pathways (Melançon et al., 1997). These results raise the intriguing possibility that other cells near the choice point are important for zebrafish motor neuron pathway selection. In summary, many studies suggest that cells at motor choice points provide cues for pathway selection; however, the identity of such cues and the mechanisms by which they control pathway selection have remained elusive.

Our analyses of chimeric embryos revealed that only a selective group of adaxial cells is capable of restoring unplugged-dependent pathway selection. About 20 adaxial cells are present in each somitic hemisegment. They first elongate to span the entire length of the somite and then migrate radially through the somites to form a single layer of superficial myotomal cells (Devoto et al., 1996). A subset of adaxial cells, the muscle pioneer cells, do not migrate radially, but extend from the notochord to the surface of the myotome, at the level of the future choice point (Devoto et al., 1996). Consistent with previous studies in which laser ablation of muscle pioneers did not severely affect pathway selection (Melançon et al., 1997), muscle pioneer cells were unable to rescue unplugged-dependent pathway selection (Table 2B). Adaxial cells capable of restoring pathway selection in unplugged mutants were confined to a dorsal domain encompassing 6 to 7 cells, adjacent to muscle pioneer cells and the choice point (Fig. 6).

Our chimeric analysis of the unplugged gene identifies exactly the same set of adaxial cells that are crucial for motor axon guidance towards the choice point (Zeller and Granato, 1999). We had previously shown that in diwanka mutant embryos, CaP, MiP and RoP growth cones failed to migrate along their common path from the spinal cord to the choice point and that diwanka activity is provided by a defined set of adaxial cells (Zeller and Granato, 1999). Our chimeric analysis of the unplugged gene identifies the same group of adaxial cells to be important for pathway selection. While the presence of only one or two wild-type-derived adaxial cells is sufficient to rescue the axonal diwanka phenotype, at least three of the six or seven adaxial cells had to be of wild-type origin to correct the unplugged phenotype. Our previous results showed that these adaxial cells prefigure the common path to the choice point (Zeller and Granato, 1999). As the first motor growth cone approaches, these adaxial cells migrate away, suggesting that they do not simply provide a substratum for migrating growth cones, but provide a source of guidance information (Zeller and Granato, 1999). Together, our results define two important roles for dorsal adaxial cells in controlling growth cone behavior. They provide local guidance information required for all primary motor growth cones to exit the spinal cord and to migrate towards the choice point (Zeller and Granato, 1999). Secondly, they control pathway selection of both CaP and RoP growth cones at the choice point. Thus, dorsal adaxial cells adjacent to the choice point might play dual roles in axonal guidance, similar to specialized cells governing growth cone behavior at the CNS midline.

We demonstrate that pathway selection of CaP and RoP growth cones at the choice point is guided by unplugged gene activity emanating from an identified group of mesoderm-derived adaxial cells. These adaxial cells delineate the common motor path before growth cone extension and start their radial migration as the first motor growth cone (CaP) pioneers the common path (Zeller and Granato, 1999). By the time RoP growth cones arrive at the choice point, these adaxial cells have reached the lateral surface of the somite. Based on these observations, we conclude that adaxial cells do not influence pathway selection through extensive cell-cell contacts with CaP and RoP motor growth cones, but might provide a source of guidance information, enabling motor axons to select their appropriate paths. What information could this be?

We can exclude several simple models. First, unplugged activity might exclusively control CaP pathway selection while RoP pathfinding defects are only a secondary consequence of CaP growth cones failing to select their appropriate path. For example, in the absence of CaP growth cones invading the ventral somite, RoP growth cones might extend into this territory simply because it is vacant. This is unlikely, because in the absence of CaP neurons, RoP growth cones select their appropriate path (Eisen et al., 1989; Pike and Eisen, 1990). Alternatively, unplugged might play a role in dorsoventral somite patterning. Such patterning might confer polarity to the ventral somite, rendering it permissive for CaP growth cones and repulsive to RoP growth cones. We find that wnt11r (Makita et al., 1998) and twist (Morin-Kensicki and Eisen, 1997) mRNA, which are both predominately localized to the ventral somite of wild-type embryos, are properly expressed in unplugged mutant embryos, suggesting that unplugged does not affect overall somite polarity (data not shown). Finally, our experiments show that CaP and RoP pathway selection requires unplugged gene activity, provided by a selective group of adaxial cells. This raises the possibility that mutations in the unplugged gene affect specification or migration of these adaxial cells, thereby affecting pathfinding of CaP and RoP motor axons. Analysis of two different adaxial cell-specific markers (prox-1, Glasgow and Tomarev, 1998; and F59), did not reveal any difference in the total number of adaxial cells or in their radial migration pattern, suggesting that in unplugged mutant embryos proliferation, migration and specification of adaxial cells are unaffected (data not shown). How, then, does unplugged influence CaP and RoP pathway selection at the choice point? Our results show that adaxial cells located dorsal to the choice point are responsible for unplugged-dependent pathway selection, suggesting that unplugged activity influences CaP and RoP growth cones before they reach the choice point and make their pathway decision. One attractive model that takes these findings into account is that the dorsal adaxial cells adjacent to the choice point deposit unplugged activity on the extracellular matrix of the overlying somite surface (Fig. 7A). This unplugged activity generates or functions as a bifunctional cue, eliciting different responses in CaP and RoP growth cones (Fig. 7B). Consistent with our results, unplugged activity may instruct CaP growth cones to extend into the ventral somites while preventing RoP growth cones from inappropriately crossing the choice point (Fig. 7C). Such bifunctionality could be achieved if unplugged activates a single receptor mechanism that is coupled to different downstream effectors present in CaP and RoP growth cones. Alternatively, unplugged could activate different types of receptors in CaP and RoP growth cones. From biochemical and genetic studies, both types of mechanisms have been shown to operate (Hong et al., 1999 and references therein). Regardless of the exact mechanism, mutations in the unplugged gene abolish its bifunctional effects and consequently the activation of downstream effectors within CaP and RoP growth cones. As a result, unplugged CaP growth cones are not directed into the ventral somite and stall at their intermediate target (Fig. 7D). Similarly, RoP growth cones are not restricted to the choice point and explore adjacent regions (Fig. 7D). Thus, it is conceivable that unplugged is part of a somite-derived signaling mechanism evoking differential pathway decisions in CaP and RoP growth cones.

Our studies demonstrate that a selective group of mesodermal cells provides unplugged-dependent guidance at a motor pathway choice point. To our knowledge, this analysis of the unplugged mutant phenotype provides one of the few examples demonstrating the potential of external cues in controlling pathway selection at motor neuron choice points. Our analysis does not allow us to distinguish if unplugged acts indirectly (e.g. as transcriptional regulator in adaxial cells), or directly with CaP and RoP growth cones (e.g. as a secreted cue). To address this question, we have undertaken a positional cloning approach to identify the gene mutated at the unplugged locus, which ultimately will reveal the mechanism by which unplugged activity guides the decision of motor growth cones at the choice point.

We would like to thank the following colleagues: Tom Jessell for 3A10 antiserum and Slava Tomarev for prox-1 antiserum; Qiang Chang, Steve DiNardo, Gerald Downes, Mary Mullins, Jonathan Raper, Julie Waterbury and Jörg Zeller, for critical comments; and Chapell Miller and Helena Robinson-Bey for care of the fish. We like to thank Qiang Chang for his help with the confocal microscope and Marnie Halpern (Carnegie Institution of Washington) for providing γ-ray-treated fish amongst which we identified the p⁴ allele. This work was supported by grants from the University of Pennsylvania Research Foundation, The Whitehall Foundation (A#98-19) and the National Institute of Health (HD37975-01) to M. G.