Drosophila Rac1 controls motor axon guidance

In order to reach appropriate target tissues in the developing embryo, neuronal growth cones must navigate through a complex cellular terrain in response to various guidance cues. This process requires an interplay between information outside the growth cone and intracellular motility machinery that drives the growth cone forward. Although rapid progress has been made in our knowledge of extracellular axon guidance signals and their receptors (reviewed by Tessier-Lavigne and Goodman, 1996), the biochemical pathways that convey and integrate this information within the growth cone are poorly understood.

Analysis of motor axon guidance in the Drosophila embryo provides a convenient system to identify molecules that control growth cone behavior in vivo (reviewed by Bate and Broadie, 1995; Keshishian et al., 1996). Motor growth cones must pathfind across a series of different surfaces and navigate a number of important choice points along the way to correct muscle targets (Van Vactor et al., 1993). Choice points are special landmarks in the developing embryo where growth cones reorganize their cell contacts, shifting from one cellular substratum to another, presumably in response to a combination of cues presented on the cells or extracellular matrix that they encounter in these regions. Using both classical and molecular genetics to examine one particular choice point, a number of genes encoding cell surface and secreted proteins have been shown to be necessary for specific motor growth cones (ISNb) to enter their correct target domain (Van Vactor et al., 1993; Lin and Goodman, 1994; Desai et al., 1996; Fambrough and Goodman, 1996; Krueger et al., 1996). Although different receptor-like proteins control each step in this process, none of the known choice point genes encodes an intracellular signaling molecule capable of linking cell surface to the motility machinery within the growth cone.

Growth cone exploration and response to the extracellular environment is believed to occur primarily at the leading edge membrane, a dynamic structure supported by a complex cytoskeletal architecture composed largely of actin microfilaments and associated proteins (reviewed by Smith, 1988; Stossel, 1993; Mitchison and Cramer, 1996). Studies of growth cone behavior in culture and in situ suggest that focal actin reorganization at the point of contact between the leading edge and other attractive cell surfaces plays an important role in guidance (O’Connor and Bentley, 1993; Lin and Forscher, 1993). Complete disruption of actin structures with high doses of cytochalasin toxins stops growth cone advance (Yamada et al., 1971). However, when actin function is compromised with low doses of toxin, growth cones continue to extend but lose directional specificity (Letourneau and Marsh, 1984; Bentley and Toroian-Raymond, 1986; Chien et al., 1993). These observations suggest that guidance cues exert their effects by controlling actin assembly.

The small GTPases Cdc42, Rac and Rho control actin cytoarchitecture in leading edge structures (reviewed by Hall, 1990; Chant and Stowers, 1995; Luo et al., 1997) and coordinate cell polarity and morphogenesis (reviewed by Drubin and Nelson, 1996). Modeled on studies of Ras, dominant interfering mutations (Diekmann et al., 1991, Ridley et al., 1992) reveal that Cdc42, Rac and Rho control distinct actin-based motility behaviors (Ridley et al., 1995; Nobes and Hall, 1995a). In Drosophila, homologues of Rac and Cdc42 are involved in dorsal closure (Harden et al., 1995), wing bristle formation (Eaton et al., 1995), myoblast fusion and sensory neuron morphogenesis (Luo et al., 1994). Recently, constitutively active GTPase mutants were shown to block Drosophila motor axon outgrowth (Sone et al., 1997), presumably via massive and inappropriate actin polymerization (Luo et al., 1994). This indicates that GTPase effectors can be activated in motor growth cones, however, the normal roles of Rac and Cdc42 in motor growth cones were not addressed with loss-of-function mutations. To better understand how these GTPases function in motor growth cones, we examined dominant loss-of-function phenotypes during neuromuscular development. These phenotypes suggest that motor growth cones require Rac activation for directional specificity and target recognition, whereas Cdc42 is required for motor axon extension.

Gal4 expression was controlled by the neuronal specific P[GAL4-C155] (Lin and Goodman, 1994), P[GAL4-1407] (Luo et al., 1994), or P[GAL4-elav] driver chromosomes (1407 was obtained from Dr Y. N. Jan; the others were obtained from Dr C. S. Goodman). Wild-type, (N17), (L89) and (V12) mutant P[UAS-Drac1], as well as P[UAS-Dcdc42] reporter lines are described in Luo et al. (1994). Dlar-P[Drac (N17.1)] double mutant combinations were created by crossing flies heterozygous for the Dlar alleles and homozygous for either P[UAS-Drac(N17.1)] or P[GAL4-C155]. All flies were maintained at 25°C.

Motor axon pathways were visualized using mAb 1D4 as previously described (Van Vactor et al., 1993). mAb 3C10 (anti-even-skipped) and mAb In 4D9B1F1 (anti-engrailed) were used to assess cell fate and patterning in the CNS (obtained from Dr Goodman). mAb 22C10 (Zipursky et al., 1984) was used to asses PNS development. When homozygous mutants could not be made, the correct genotypes were chosen by staining with anti-β-galactosidase antibodies (Sigma) to recognize embryos carrying lacZ-expressing balancer chromosomes (CyO-P[actin5C-LacZ]). Prior to evaluation, stage 17 embryos of the appropriate genotypes were selected and filleted under a dissection microscope. Embryos were staged according to Campos-Ortega and Hartenstein (1985). Each genotype was coded and scored blind. When some ISNb axons bypassed the choice point and others entered correctly, the segment was scored as a ‘partial bypass’; when all axons bypassed, with no evidence of appropriate target domain entry, then the segment was scored as ‘full bypass’ (as scored in Krueger et al., 1996; note: total bypass = full + partial). Since the ‘full bypass’ phenotype is more accurate for quantitative comparisons, we scored only full bypass when evaluating genetic interactions.

Carefully staged, wild-type (Canton-S) embryos were dechorionated and aligned on tape under a thin layer of Halocarbon oil (series 700); these were injected on an inverted microscope with diamond-beveled micropipettes. Cytochalasin D (CD; Sigma) was dissolved in DMSO prior to dilution in standard embryo injection buffer (0.1 mM phosphate buffer, pH 6.8, 5 mM KCl). The final concentration of DMSO was 0.01% in all samples, including buffer controls. Non-toxic green food dye was added to help visually evaluate the success of injections; this had no effect on ISNb development in control injections. In order to access the internal core of the developing embryo without disturbing morphogenesis, fixed volumes of sample were injected into the perivitelline space at embryonic stage 13. By stage 15, the injected sample was incorporated into the gut cavity, providing a central source of membrane permeable toxin to diffuse throughout the embryo during the key stages of ISNb pathfinding. At estimated final CD concentrations approaching 1.0 mg/ml, muscle development was often abnormal; thus phenotypic analysis was restricted to concentrations at or below 0.5 mg, where the intersegmental nerve (ISN) and muscle development appeared completely normal. Injected embryos were allowed to mature to stage 17, rinsed from the slide surface and stained with mAb 1D4. Muscle morphology was evaluated with Nomarski optics.

To disrupt GTPase function in neurons during axon pathfinding, without affecting the development of other tissues or previous embryonic stages, dominantly activated (V12) and dominant negative (N17 and L89) GTPase mutations were placed under the control of the yeast transcriptional activator GAL4 (Luo et al., 1994; see Materials and Methods). Combination of a neuronal GAL4 ‘driver’ with a GTPase cDNA ‘reporter’ under control of the GAL4 upstream activator sequence (UAS) results in specific cDNA expression (Brand and Perrimon, 1993); three different neuron-specific GAL4 drivers were used (see Materials and Methods). Although gain-of-function and loss-of-function GTPase mutations have similar effects on sensory axon outgrowth (Luo et al., 1994), they have distinct effects on motor axon pathways, which reveal a role for Rac in executing specific guidance decisions (see below).

In each abdominal hemisegment, Drosophila motor growth cones exit the central nervous system (CNS) via two pathways: the segmental (SN) and intersegmental (ISN) nerve roots (Fig. 1). In the periphery, motor axons divide into five major nerve branches, each destined for a different region of body wall muscle (Johansen et al., 1989). Intersegmental nerve b (ISNb, previously known as SNb) development has been most extensively characterized, in both cellular and genetic terms. To reach its correct muscle targets after emerging from the CNS as part of the ISN, ISNb defasciculates at the exit junction to form a distinct bundle of nine axons, and then leaves the ISN axon pathway to explore ventral muscle surfaces at the ventral choice point (Fig. 1A). Target entry at the ventral choice point is very precise: after leaving the common motor pathway, wild-type ISNb axons always follow the internal surface of muscle 14 in order to enter the ventral domain (Van Vactor et al., 1993) (Fig. 1B).

Neuronal-specific expression of either Drac1(V12) or Dcdc42(V12) causes ISNb motor growth cones to arrest outgrowth just as axons begin to explore the periphery [Fig. 2B shows the Drac1(V12) phenotype; the Dcdc42(V12) phenotype is identical] (also see Sone et al., 1997). This highly penetrant motor phenotype (96%, n=134 abdominal segments under the control of the C155-GAL4 driver) is analogous to the growth cone arrest observed in sensory neurons when they express the same GTPase mutations (Luo et al., 1994). The growth cone arrest seen in the V12 backgrounds suggests that hyperactivation of different GTPase pathways disrupts leading edge motility. However, in order to determine when and where GTPases normally act in motor axon guidance, it is necessary to examine loss-of-function phenotypes.

Neuronal-specific expression of dominant negative mutations reveal a difference in the function of Dcdc42 and Drac1. Expression of Dcdc42(N17) has a more subtle effect on axonal development than Drac1(N17); we only observed a Dcdc42(N17) phenotype under the strongest GAL4 driver (elav-GAL4). When UAS-Dcdc42(N17) is combined with elav-GAL4, ISNb axons stop short of their most distal target muscle (muscle 12) in 8% of abdominal hemisegments (Fig. 2C). Interestingly, the Dcdc42(N17) ISNb phenotype closely resembles the growth cone arrest observed in profilin mutants (D. Van Vactor and C. S. Goodman, unpublished observations). In contrast, inhibition of neuronal Drac1 function with Drac1(N17) altered the trajectory of motor axons but did not block their extension (see below). The fact that Cdc42 and Rac phenotypes do not overlap suggests that each N17 dominant negative effect is specific.

Expression of Drac1(N17) caused ISNb axons to bypass ventral muscle targets (Fig. 2D,E; see arrowheads). Instead of turning to grow into the ventral muscle domain along the internal surface of muscle 14, Drac1(N17) ISNb axons extended along the exterior surface of this muscle and followed ISN axons towards dorsal muscles. Although the qualitative nature of these phenotypes remained constant in different GAL4 driver-Drac(N17) combinations, each genotype yielded a different frequency of defects, suggesting a dose-dependent relationship (Fig. 3A). The presence of neuronally expressed GAL4 or the Drac1(N17) transgenes by themselves had no effect on ISNb development (Fig. 3A). As a control, wild-type Drac1 was expressed in the same manner; no defects in motor axon pathways resulted. Thus, the Drac1(N17) phenotypes resulted from the dominant negative mutation and not simply overexpression of Rac protein. The cell fate and patterning of neural soma within the CNS was not altered in the GAL4-Drac1(N17) embryos, as assessed with anti-even skipped and anti-engrailed antibodies (data not shown).

Neuronal Drac1(N17) expression did not cause ISNb growth cone arrest. ISNb axons that enter the ventral domain correctly in Drac1(N17) embryos reach all of their targets. Even under the strongest GAL4 driver, growth cone arrest did not increase above background levels observed in the P[UAS-Drac1(N17.1)] line alone (maximal at 3%). Furthermore, when ISNb bypassed ventral muscles (as often as 73% in the elav-GAL4;Drac1(N17) genotype), these growth cones tended to extend further than their normal counterparts, suggesting that outgrowth and motility were intact even though directional guidance was compromised.

The genetic analysis of Drosophila neuromuscular development has identified several proteins required for ISNb axon guidance (Van Vactor et al., 1993; Lin and Goodman, 1994; Desai et al., 1996; Fambrough and Goodman, 1996; Krueger et al., 1996). Close examination of the ISNb phenotypes in different mutants reveals two steps in choice point guidance: the defasciculation of ISNb from ISN at the exit junction, and target domain entry at the ventral choice point (see Krueger et al., 1996 for a discussion of this distinction). In Drac1(N17) embryos, misguided ISNb axons were most frequently observed as a fascicle distinct from the ISN (Fig. 2D,E). The frequent presence of two separate axon fascicles in the ISN focal plane indicates that failure in target entry does not represent a failure to defasciculate from the common pathway. In this regard, the Drac1(N17) phenotype most closely resembles the phenotype of only one other known protein: the receptor-tyrosine phosphatase Dlar (Krueger et al., 1996).

As expected, the effects of Drac1(N17) are not specific to ISNb; a related phenotype was observed in the SNa projection to lateral muscle targets. The dorsal branch of SNa normally forms terminal contacts with muscles 21-24 in a characteristic region of muscle surface underneath or proximal to the non-target dorsal muscle 4 (Figs 1A,C, 4A). In Drac1(N17) embryos, the dorsal projection of SNa often bypassed this stereotypic target region on lateral muscles (Fig. 3B). In many such cases, SNa axons failed to make normal target contacts and extended beyond lateral muscles to make ectopic contacts onto a variety of cell surfaces. Ectopic contacts were observed on the sheath cells of lateral sensory neurons (chordotonals), on ISN axons or often on non-target muscles in the dorsal region (including muscles 11, 19 and 20; Fig. 4B,C). This particular SNa phenotype has not been described in any previous mutant background; however, it appears analogous to the ISNb bypass phenotype where motor growth cones extend beyond normal target muscles. In addition, Drac1(N17) causes subtle disruption of interneuronal pathways within the ventral nerve cord (data not shown); misguided CNS axons could not be followed individually with mAb 1D4 due to the complexity of the central neuropil.

The N17 dominant loss-of-function mutation is well characterized, reducing GTPase affinity for GTP tenfold relative to GDP binding (Feig and Cooper, 1988; Ridley et al., 1992). N17 is thought to block endogenous GTPase by competing for the binding of cellular activators (e.g. exchange factors). A second class of Rac dominant mutation (L89), inspired by a Ras(F89) allele (Han and Sternberg, 1991), was previously shown to block sensory axon outgrowth (Luo et al., 1994). Although the biochemical effects of the L89 mutation have not been determined, Luo et al. (1994) showed that Drac1(N17) and Drac1(L89) have differential effects, depending on the cells in which they are expressed (e.g. neuron or muscle). For ISNb motor axons, we found that Drac1(L89) has effects similar to both Drac1(V12) and Drac1(N17). Expression of Drac1(L89) under the neuronal GAL4 drivers described above resulted in a consistent ISNb growth cone arrest; this arrest ranged from 6% to 68% (n=172 hemisegments or greater), depending on the combination of GAL4 driver and UAS reporter. Drac1(L89) ISNb axons stop at different points along their normal trajectory to ventral muscles (Fig. 2F). Though the growth arrest phenotype is prevalent in Drac1(L89), we also observed an ISNb bypass phenotype in some segments (ranging from 7% to 16%). Drac1(L89) also caused disruption of CNS axon pathways, consistent with growth cone arrest (Fig. 5B).

Since the Drac1(L89) phenotype shared elements of both loss-of-function and gain-of-function, we asked whether these phenotypes could be rescued by co-expression of wild-type Drac1. Luo et al. (1994) showed that co-expression of wild-type Drac1 could rescue a very weak Drac1(L89) sensory axon phenotype as evidence of a loss-of-function. However, when we examined embryos that co-expressed wild-type Drac1 and Drac1(L89.6) with the mAb 1D4 marker, we found that the majority of embryos displayed a dramatic enhancement of the growth cone arrest phenotype (62%, n=222 segments; Fig. 5D). The three characteristic mAb 1D4-positive interneuronal pathways that form along the longitudinal connectives could not be seen in these segments; there was a massive failure in axonal extension. The few axons that did extend into the periphery in these embryos were highly disorganized; we were unable to identify specific nerve branches in these cases. The dramatic enhancement of axon extension phenotypes in embryos that co-expressed wild-type and (L89) forms of Drac1 indicates that Drac1(L89) is not a simple loss-of-function allele. In contrast, co-expression of wild-type Drac1 and Drac1(N17) showed no enhancement of the N17 phenotype (Fig. 5C). Similar results have been obtained from comparisons of Drac1(N17) and Drac1(L89) during dorsal closure (N. Harden, personal communication), suggesting that Drac1(L89) acts through a different mechanism than N17 and V12. In fact, several dominant alleles of Ras have been described that display combined loss- and gain-of-function phenotypes (Han and Sternberg, 1991). For this reason, Drac1(L89) was not used in the double mutant analysis with Dlar and Beaten Path (see below and discussion).

Since GTPase signaling is known to control processes other than motility behaviors (e.g. Lamarche et al., 1996), we wondered whether the Drac1 loss-of-function phenotype was correlated with a defect in cytoskeletal assembly. Unfortunately, due to the small size of Drosophila motor growth cones and the fact that they extend between actin-rich muscle fibers, we have been unable to obtain high-resolution views of motor growth cone cytoskeleton in vivo by conventional or confocal fluorescence microscopy. As an alternative approach, we compared the GTPase phenotypes to the effects of a pharmacological agent that blocks actin assembly. In order to examine a range of toxin effects without disturbing the development of muscle targets or the common motor pathway, we took advantage of the fact that ISNb growth cones are last to reach their targets. Different concentrations of membrane permeable cytochalasin D (CD) were microinjected into developing embryos after the ISN pathway had extended into the periphery and muscle targets were patterned correctly, but before ISNb axons had begun to explore the ventral choice point (see Materials and Methods).

At low doses of CD, many ISNb projections were normal (48% normal at 0.05 mg/ml or 0.1 μM estimated final concentration, n=234 abdominal hemisegments). At this low concentration of toxin, where CD is known to inhibit membrane ruffling (Cooper, 1987), many ISNb branches displayed bypass phenotypes (32%). Interestingly, both the fusion bypass (failure to defasciculate, as seen in beaten path) and the parallel bypass (failure in target entry, as seen in Dlar) phenotypes were observed (Fig. 6B,C). This suggests that actin assembly is essential for both phases of ISNb guidance. Bypassing axons often reached back from an abnormal trajectory to contact ventral targets, illustrating that the bypassing axons were indeed ISNb (Fig. 6B).

ISNb branches that did enter the ventral domain often stopped early as observed in Dcdc42(N17) embryos (20%; Fig. 6B). At this low dose, filopodia remained visible on ISNb growth cones (lamellipodia are difficult to distinguish in such small growth cones). No ISNb defects were observed in control injections (data not shown).

At high CD concentrations, few ISNb branches were normal (12% normal at 0.5 mg/ml or 1 μM estimated final concentration, n=272 abdominal hemisegments); far fewer ISNb branches bypassed the choice point (19%) than displayed growth cone arrest (69%). Growth cone arrest at high concentrations was often observed at, or prior to, the ventral choice point (Fig. 6D). The tips of axons in these embryos typically lacked filopodia. Previous experiments with CD in vitro have suggested that axon extension can proceed in the absence of actin assembly (Letourneau and Marsh, 1984); this is not the case for ISNb motor axons in vivo. Although toxin effects on peripheral cell types (ie. muscle) were a concern, the ventral muscles appeared morphologically normal after late stage injections of 0.05-0.5 μg/ml CD (see Materials and Methods). The prevalence of ISNb bypass phenotypes at the low toxin dose supports a model where target entry requires active cytoskeletal assembly downstream of Rac activation.

Of all the proteins currently known to control ISNb guidance, Drac1 and Dlar are most likely to mediate target entry as opposed to defasciculation, because both display the parallel bypass phenotype. To better understand the relationship between Dlar and Drac1, we examined double mutant embryos. The penetrance of full ISNb bypass in Dlar hemizygous (Dlar^13.2/+), partial loss-of-function (Dlar^bypass/Dlar^bypass) and null (Dlar^5.5/Dlar^13.2) genotypes is 0%, 7% and 21%, respectively (see Materials and Methods for scoring criteria). However, when combined with C155-GAL4;Drac1(N17.1) which displays 3% full bypass alone, all three Drac1(N17)-Dlar compound mutants display a penetrance twofold to threefold higher than expected if the defects were simply additive (Table 1). Thus, Rac function in ISNb axons is quite sensitive to the dosage of Dlar protein. This synergistic genetic interaction suggests that Rac and Dlar function together, though not in a simple linear pathway. The increase in penetrance observed in the Drac1(N17)-Dlar^null combination, far above the maximum frequency of defects seen in the complete absence of Dlar, suggests that a Dlar-independent pathway mediated by Drac1 contributes to ISNb target entry.

To address the specificity of Rac’s role in target entry, we examined embryos deficient in both Rac and the defasciculation factor Beaten Path. In beat²/beat³ embryos, we find a 69% penetrance of full ISNb bypass (consistent with Fambrough and Goodman, 1996) (Table 1). However in C155; beat²/beat³ ; Drac1(N17.1) double mutant embryos, the penetrance of ISNb bypass increases only to 80% (Table 1). This small increase in penetrance contrasts the synergy seen between Drac1(N17) and Dlar.

The signaling pathways that growth cones use to interpret their environment are not well understood. Rho-family GTPases such as Cdc42 and Rac control distinct actin-dependent motility behaviors in response to extracellular cues, making them attractive candidates to mediate growth cone guidance (Reviewed by Nobes and Hall, 1995b; Machesky and Hall, 1996). We have shown that whereas neuron-specific disruption of Dcdc42 blocks motor axon extension, the equivalent perturbation of Drac1 alters directional specificity without inhibiting extension. Although previous studies have shown that constitutively active Drac1 and Dcdc42 can disrupt motor pathways (Sone et al., 1997; see Results), our results provide the first evidence that Drac1 is required for directional guidance in vivo.

The effects of Dcdc42(N17) are subtle, which may indicate that Cdc42 plays only a minor role in axonogenesis, as suggested by recent experiments in chick neurons (Jin and Strittmatter, 1997). The effects of Drac1(N17) are more dramatic, causing different groups of motor axons to bypass their normal target regions. Since the Drac1(L89) mutation displays features consistent with some gain-of-function activity, we base our interpretation of Rac function on the well-characterized N17 loss-of-function mutation. Our observation that partial disruption of actin assembly consistently mimics the N17 effect further supports the model that Rac conveys guidance information from cell surface to cytoskeleton.

The correct entry of ISNb axons into the ventral muscle domain requires a very precise turn at contacts with specific muscle surfaces; errors are rarely made in wild-type embryos. Genetic analysis of ISNb development has identified several proteins that control different aspects of this guidance process. For example, the adhesion molecule Fasciclin II and the secreted, anti-adhesion protein Beaten path control the defasciculation of ISNb and other axons from the common motor pathway that emerges from the central nervous system (Lin and Goodman, 1994; Fambrough and Goodman, 1996). The receptor-like tyrosine phosphatase (RPTPase) Dlar is required for target entry, but not for defasciculation (Krueger et al., 1996). So far, Rac is the only intracellular signaling molecule that has been shown to function in the target entry pathway. Presumably Rac mutations were not recovered in a previous neuromuscular screen of the third chromosome (Sink et al., 1993) due to requirements early in embryogenesis (e.g. Harden et al., 1995).

Dissecting the relationship between guidance signals that instruct the growth cone and the motility apparatus that implements this information is difficult, because cytoskeletal reorganization is essential for two vital roles in axonal development: axonal extension and guidance. Since different forms of actin cytoskeleton are presumably controlled by different intracellular signals, we would expect that cytochalasin D (CD) should generate a composite phenotype. In fact, at low toxin doses, we observed failures in both ISNb defasciculation and target entry. However, different doses of toxin favor distinct axonal phenotypes: bypass of ventral targets at low doses and growth cone arrest at high doses. This indicates that forward motility and directional specificity demand slightly different levels of actin assembly or polymerization. For example, the predominance of bypass at low CD suggests that the normal response of ISNb growth cones to their target domain (mediated by Dlar, Drac1 and other genes) requires a high level or rate of actin assembly. This is consistent with the observation that growth cone interactions that define directionality often induce large and immediate accumulations of F-actin at the site of cell contact (O’Connor and Bentley, 1993; Lin and Forscher, 1993).

From budding yeast to the Pelvitia zygote, cell polarity relies on focal assembly of actin-based cytoskeletal components, often in response to extracellular cues (Reviewed by Drubin and Nelson, 1996; see also Kropf, 1994). Cell polarity in the Drosophila wing epithelium is oriented by cell-cell communication and is manifested by a burst of actin polymerization at the apex of each cell that leads to the formation of a tiny actin-filled wing hair (Wong and Adler, 1993). Whereas inhibition of Cdc42 stops wing hair elongation, disruption of Rac prevents the coherent focus of actin and randomizes hair initiation, mimicking the phenotypes of mutations in genes (e.g. for the transmembrane receptor Frizzled) that transmit the patterning information (Eaton et al., 1995, 1996). The roles of Cdc42 and Rac in motor growth cones may be analogous. Dcdc42(N17) blocks motor axon extension, while Drac1(N17) interferes with directional guidance. At a choice point, cytoskeletal asymmetry induced by extracellular cues may drive the growth cone in a new direction. Activation of Rac is known to stimulate actin assembly and the formation of lamellipodial membrane ruffles (Nobes and Hall, 1995a). Recent evidence suggests that lamellipodia play a dominant role in growth cone turning behavior in vitro (Wang and Jay, 1997). However, this model assumes an asymmetric activation of Rac at the leading edge. Since removal of all somatic muscle results in the failure of ISNb, and other motor axons, to leave the common motor pathway (Sink, H. and Goodman, C.S., personal communication), we speculate that Rac responds to localized mesodermal signals in choice point regions in order to orient growth cone movement.

The similarity of the Drac1(N17) and Dlar phenotypes suggests that Drac1 functions specifically in the target entry process, as opposed to the defasciculation of ISNb. This is supported by the strong dose-sensitive relationship between Dlar and Drac1. Accordingly, there is no dramatic interaction between Drac1 and beaten path. However, the interaction between Drac1(N17) and Dlar null mutations further suggests that ISNb target entry requires multiple, convergent signals. The existence of multiple inputs during target entry may explain why null mutations in Dlar (and other choice point genes) are not completely penetrant on their own (Desai et al., 1996; Krueger et al., 1996; Korey et al., 1997).

Dlar is a member of an RPTPase sub-family expressed in the developing nervous system, including the human proteins LAR, HPTPδ and HPTPσ (reviewed by Van Vactor, 1998). Recently, the highly conserved D2 PTPase domain of human LAR was shown to interact with a novel protein, Trio, that contains a Rac-specific guanine nucleotide exchange factor domain (Debant et al., 1996). This suggests a model for RPTPase function where activation of the receptor leads to cytoskeletal reorganization via recruitment and activation of the small GTPase pathway. Although a Drosophila Trio counterpart exists (J. Bateman and D. Van Vactor, unpublished observations) and human Trio avidly interacts with the Dlar cytoplasmic domain (N. Kaufmann and D. Van Vactor, unpublished observations), the functional relationships between DTrio and the Drosophila RPTPases remain to be elucidated.

We wish to thank Ryn Miake-Lye, John Flanagan, Rong Li, Liqun Luo and Tim Mitchison for helpful comments and discussions regarding this manuscript. We thank Corey Goodman for providing the C155 and elav-GAL4 drivers, in addition to various antibody stocks. We thank Liqun Luo and Yuh Nung Jan for providing Drosophila stocks carrying dominant mutations in Rac1 and Cdc42 under the control of GAL4 UAS, in addition to the 1407 GAL4 driver. Part of this work was supported by The Esther A. and Joseph Klingenstein Fund for Neuroscience and NIH grant NS35909. Nancy Kaufmann was supported by an NIH/NRSA predoctoral traineeship and Zachary Wills by a National Eye Institute Graduate Fellowship.