Receptor tyrosine phosphatases regulate axon guidance across the midline of the Drosophila embryo

During embryonic development, growth cones, which form the leading edges of neuronal processes, sample the environment and make pathfinding decisions. Contacts between growth cone surface receptors and attractive or repulsive guidance cues on surrounding cells and in the extracellular matrix affect signal transduction cascades within the growth cones (Goodman, 1996). These signaling events change the growth cone’s cytoskeleton, altering its morphology and direction of movement.

One signaling mechanism by which growth cones respond to guidance factors is the control of tyrosine phosphorylation. Phosphotyrosine in growth cones is generated by receptor tyrosine kinases that are activated by direct binding to extracellular ligands and by cytoplasmic tyrosine kinases which interact with a variety of surface receptors. Receptor-linked protein tyrosine phosphatases (RPTPs), which reverse the reactions catalyzed by tyrosine kinases, also regulate tyrosine phosphorylation within growth cones. The five known Drosophila RPTPs have receptor-like extracellular regions consisting of immunoglobulin-like and/or fibronectin type III domains, linked via a single transmembrane domain to a cytoplasmic region containing one or two phosphatase domains. Strikingly, four of these five RPTPs (DPTP10D, DLAR, DPTP69D and DPTP99A) are expressed only on central nervous system (CNS) axons in the embryo (reviewed by Desai et al., 1997b). RPTPs are also localized to axons and growth cones in vertebrate systems (reviewed by Stoker and Dutta, 1998).

Genetic studies in Drosophila have demonstrated that DLAR, DPTP69D and DPTP99A are important regulators of motor axon guidance decisions (Desai et al., 1996, 1997a; Krueger et al., 1996). DPTP69D is required for normal innervation of the lamina by photoreceptor growth cones (Garrity et al., 1999). However, we do not currently have a framework for understanding how RPTPs control growth cone pathfinding in the neuromuscular system or the larval optic lobes. This is because the ligands and substrates that interact with RPTPs in vivo are unknown, and the attractive and repulsive guidance cues that guide motoneuron and photoreceptor growth cones to their targets are not well defined.

In this paper, we link RPTPs to a specific pathfinding event by showing that two of these phosphatases regulate repulsion from the CNS midline. Axon guidance at the midline has been intensively studied, and several of the key molecules controlling attractive and repulsive interactions between neuronal growth cones and the specialized midline glial cells have already been identified. Attractive signals emanating from the midline are required to recruit growth cones into commissural pathways that cross over to the contralateral side of the embryo. In Drosophila, these attractive factors include two Netrins, which are secreted proteins used in both vertebrates and invertebrates for control of guidance at the midline (Harris et al., 1996; Mitchell et al., 1996).

The midline also produces signals that repel growth cones. These are necessary to prevent longitudinal axons that express receptors for attractive factors from crossing the midline. Repulsion also allows the growth cones of commissural neurons to leave the midline and reach the contralateral side of the CNS, and prevents them from returning to the midline after crossing. Recent studies have shown that repulsive signals are transduced via interactions between the Roundabout (Robo) receptor expressed on neuronal growth cones and the extracellular matrix protein Slit, which is produced by midline glia (Kidd et al., 1999). Slit binds directly to Robo and can mediate repulsion of Robo-expressing growth cones (Brose et al., 1999; Li et al., 1999; Nguyen Ba-Charvet et al., 1999; for review see Zinn and Sun, 1999). In slit mutants, all interneuronal axons converge on the midline and remain there, suggesting that repulsion has been eliminated. robo mutants, however, have a weaker phenotype in which longitudinal axons cross to the contralateral side of the embryo and then follow looping pathways that traverse the midline multiple times (Seeger et al., 1993; Kidd et al., 1998a,b). The difference between the robo and slit phenotypes is likely to be due to the existence of a second Robo protein, Robo2, which could mediate repulsion of some growth cones by Slit even when Robo is absent (Kidd et al., 1998a, 1999).

Another component of the repulsive pathway is Commissureless (Comm), a transmembrane protein that is expressed by midline glia and transferred to commissural axons by an unknown mechanism (Tear et al., 1996). Transferred Comm causes downregulation of Robo, neutralizing the repulsive signal and allowing commissural axons to cross. In comm loss-of-function mutants, Robo fails to be downregulated, and as a result all axons are repelled from the midline (Seeger et al., 1993; Kidd et al., 1998b).

Here we report the isolation of mutations in the gene encoding the fourth neural RPTPs, DPTP10D. Ptp10D single mutants are viable and fertile, and display no abnormal embryonic phenotypes. The phenotypes of mutants that lack both DPTP10D and DPTP69D, however, reveal that RPTPs have important roles in the control of axon guidance across the CNS midline. This had not been apparent before now because a triple mutant that lacks all three of the previously characterized RPTPs has a relatively normal CNS axon array (Desai et al., 1997a; Q. S. and K. Z., unpublished results).

When DPTP10D and DPTP69D are removed, we find that a subset of longitudinal axons are rerouted across the midline. Elimination of all four neural RPTPs converts most longitudinal pathways into commissures, indicating that tyrosine phosphorylation controls the manner in which growth cones respond to midline signals. DPTP10D and DPTP69D genetically interact with Robo, Slit and Comm, and these interactions suggest that the RPTPs are positive regulators of Robo signaling pathways.

The Ptp10D gene is contained within three cosmid clones: cos 1, cos 8 and cos F (Fig. 1). EP1332 (BDGP line EP(X)1172) was isolated by Rorth et al. (1998). A short segment of EP1332 flanking sequence is available in GenBank (no. AQ025398). The EP1332 insertion was mapped by Southern DNA hybridization to a 5 kb EcoRI fragment of cos 8 that is 5′ to the breakpoint of Df(1)59 (Bahri et al., 1997). EP1332 is therefore located about 5 kb upstream of the translation start of DPTP10D. Imprecise excision lines were generated as described by Hamilton and Zinn (1994). Since deletions in the Ptp10D/bif region are viable, we were able to screen for deletions by PCR analysis of genomic DNA from single adult flies (Gloor et al., 1993) with primer sets specific to sequence tags at various locations (Fig. 1). The deletion in Ptp10D1 starts from the EP1332 insertion site and ends between primer sets 10D-1 and 10D-2. The P element line P842 and deletion Df(1)59 were previously described by Bahri et al. (1997). Df(1)59 is an approx. 60 kb deletion which removes the entire coding regions of bif and Ptp10D. Df(1)101 was generated by mobilizing P842 and selecting for lines missing the 10D-4 PCR fragment. One of the breakpoints of Df(1)101 falls within a 14 kb region 3′ to 10D-3. Because this region contains repetitive sequences, we were unable to map the precise location of this breakpoint by Southern hybridization. The other breakpoint of Df(1)101 is 3′ to the P842 insertion site. Sequences of primers used for mapping are available on request.

Whole-mount antibody staining of staged fly embryo collections was performed essentially as described by Patel (1994). Monoclonal antibodies (mAbs) 1D4 (Van Vactor et al., 1993), 3F11 (anti-PTP69D; Desai et al., 1994), 45E10 (anti-PTP10D; Tian et al., 1991), 8C4 (anti-DLAR; B. Burkemper and K. Z., unpublished), and 10D3 (anti-Wrapper (Noordermeer et al., 1998) were used at a dilution of 1:5. BP102 (Seeger et al., 1993) was used at dilution of 1:10. For staining with anti-connectin mAb C1.427, embryos were fixed for 20 minutes in 4% formaldehyde/PEM buffer, and incubated with a 1:3 dilution of the C1.427 supernatant (Meadows et al., 1994). mAb staining was visualized using HRP-conjugated secondary antibodies and DAB immunohistochemistry. Mutant embryos were identified by the absence of staining with particular anti-RPTP mAbs, sorted, and restained with 1D4, BP102 or C1.427. Dissected embryos were photographed on a Zeiss Axioplan microscope using DIC optics.

We delivered DiI (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbo-cyanine perchlorate; Molecular Probes, Inc.) to neuroectodermal cells by the method of Bossing and Technau (1994), with modifications described by Schmid et al. (1999). Embryos were labeled at stage 8. After 10 hours at 16°C, embryos developed to stage 11; at this stage, the parasegmental groove, the segmental groove and the midline are used as morphological landmarks for NB identification and GFP-marked balancers are used to distinguish homozygous embryos. By 37 hours after egg laying at 16°C, the embryo is well advanced into stage 17. Dissected stage 17 embryos were imaged on a Biorad 1024 microscope, using a Leitz 50′ water immersion lens, as 1.5 μm step z-series. Data were collected at 568 nm excitation (for DiI), and each z-series was immediately rescanned using Nomarski optics to determine cellular positions within the CNS and identify motoneuronal target muscle(s). Cell and axon measurements were done with Biorad software calibrated to a stage micrometer. Biorad software was used to project each z-series to form 2-dimensional images, which were assembled into figures using Photoshop (v 5.0) and Freehand (v 7.02) software.

To isolate Ptp10D mutations, we employed standard P element mutagenesis techniques. Several P element insertions have been mapped to the cytological location 10D. One of these, EP1332, is inserted less than 5 kb upstream of the translation start of Ptp10D (Fig. 1; see Materials and Methods). Embryos homozygous or hemizygous for EP1332 express very low levels of DPTP10D protein as assayed by staining with mAbs against DPTP10D (Tian et al., 1991). The phenotypic analysis described below indicates that the EP1332 insertion is a hypomorphic Ptp10D mutation. In order to generate mutations that delete Ptp10D coding sequences, we mobilized EP1332 and isolated five imprecise excision derivatives that delete the first coding exon of Ptp10D, which includes the ATG and the DNA encoding the signal sequence. The breakpoints of one of these deletions, Ptp10D1, are indicated in Fig. 1.

Another P element in the region, P842, was previously described by Bahri et al. (1997). P842 is inserted 3′ to bifocal (bif), the gene adjacent to Ptp10D (Fig. 1). We carried out similar imprecise excision procedures for P842 and isolated two deletion lines, Df(1)59 and Df(1)101. Df(1)59 completely removes the Ptp10D and bif genes, while Df(1)101 deletes most of the Ptp10D coding sequence (but does not remove the first exon), and all of the bif coding sequence (Fig. 1).

Ptp10D¹ deletes the ATG and signal sequence, and we were unable to detect any DPTP10D protein in Ptp10D¹ embryos. Thus, it is likely to be a null allele. Ptp10D¹, Df(1)59, Df(1)101, and a transheterozygous combination of Ptp10D¹ and Df(1)101 all produce the same phenotypes when combined with Ptp69D (see below). Ptp10D¹ and Df(1)101 do not overlap, so the only gene likely to be affected in these transheterozygotes is Ptp10D. Finally, the hypomorphic insertion EP1332 is a P element containing multiple binding sites for the yeast transcriptional activator GAL4 (Rorth, 1996), oriented in a manner which would direct GAL4-dependent transcription of Ptp10D. When EP1332 is crossed to C155-GAL4 (Lin and Goodman, 1994), a fly line in which GAL4 is expressed in all postmitotic neurons, DPTP10D is now also expressed at high levels in neurons. This expression rescues the phenotype conferred by EP1332 (see below). All of these data indicate that the phenotypes described below are due to the absence of the DPTP10D protein.

Ptp10D¹, Df(1)59 and Df(1)101 are all homozygous viable and fertile, and we were unable to detect any embryonic phenotypes conferred by these mutations. We focused our analysis on the nervous system, since DPTP10D is expressed only on CNS axons in the embryo (Tian et al., 1991; Yang et al., 1991). Fig. 2B shows the CNS of a Ptp10D¹ embryo stained with the 1D4 mAb, which recognizes the transmembrane form of Fasciclin II (Van Vactor et al., 1993). In late stage 16 and early stage 17 embryos, 1D4 stains three distinct axon bundles in each longitudinal connective of the CNS, but does not stain any commissural bundles. The pattern of 1D4-positive bundles seen in Ptp10D¹ is identical to that observed in wild-type embryos (Fig. 2A; Table 1).

DLAR, DPTP69D, and DPTP99A have partially redundant functions in the control of motor axon guidance (Desai et al., 1996, 1997a; Krueger et al., 1996). A triple mutant lacking all three displays severe motor axon defects, but has a relatively normal pattern of 1D4-positive CNS axons (Desai et al., 1997a; Q. S. and K. Z., unpublished results). This suggests that if these four neural RPTPs are involved in guidance within the CNS, expression of DPTP10D alone is sufficient to compensate for the absence of the other three. DPTP10D is not required when the other three are present, however, since Ptp10D mutants have no detectable CNS phenotypes.

To examine whether DPTP10D has a function in CNS axon guidance that is compensated for by another RPTP, we made double mutant combinations in which Ptp10D mutations were combined with mutations in each of the other three neural Rptp genes. Double mutants lacking DPTP10D and DPTP99A (Hamilton et al., 1995) are viable and exhibit no detectable embryonic phenotypes. Dlar mutations are lethal and confer motor axon guidance phenotypes (Krueger et al., 1996; Desai et al., 1997a), but these are unchanged when DPTP10D is also absent. No CNS abnormalities are detectable by 1D4 staining in Ptp10D Dlar double mutants (Table 1).

Ptp69D mutations are also lethal and cause motor axon phenotypes (Desai et al., 1996, 1997a). The pattern of 1D4-positive axons in the CNS of Ptp69D null mutants is identical to wild type (data not shown). When null Ptp10D mutations (Ptp10D¹, Df(1)59, and Df(1)101) are combined with null mutations in the Ptp69D gene (Ptp69D¹/Df(3L)8ex25 and others; Desai et al., 1997a), a unique and highly penetrant CNS phenotype is observed in double mutant embryos stained with 1D4 (Figs. 2D and 3; Table 1). In these embryos, the longitudinal axon bundles are irregular and often fuse to each other, so that some segments have only one or two bundles instead of the usual three. The outer 1D4-positive bundle is usually missing or reduced to short, discontinuous stained regions, and breaks in the inner two bundles are also observed. Three or more 1D4-positive bundles that cross the midline are observed within the commissural tracts of each segment. These are never seen in wild-type embryos.

We also examined embryos bearing null Ptp69D mutations combined with the hypomorphic Ptp10D insertion EP1332. DPTP10D protein is normally localized in these embryos, but is present at greatly reduced levels. This combination produces a much weaker phenotype, in which only the outer 1D4 bundle is strongly affected. Occasional fusions of the inner two bundles are also observed, and 1D4-positive bundles cross the commissures in a few segments of each embryo (Fig. 2C). As described above, this double mutant phenotype is rescued by driving high-level DPTP10D expression using a neuronal GAL4 source.

To examine the complete axonal array in double mutant embryos, we used the BP102 mAb, which recognizes an epitope present on all CNS axons (Seeger et al., 1993). Late stage 16 wild-type embryos stained with BP102 display a ladder-like pattern of axons, with two commissural tracts, anterior (A) and posterior (P), in each segment, and bilaterally symmetric longitudinal connectives extending the length of the embryo. At this stage, the longitudinal tracts are thicker than the commissural tracts, especially in the neuropilar region between the anterior and posterior commissures (Fig. 2F). Ptp10D Ptp69D double mutant embryos stained with BP102 have broader commissures than wild-type embryos, and the intercommissural space at the midline of the embryo is compressed along the A-P axis (arrow, Fig. 2G). The longitudinal tracts are also somewhat reduced, especially in the intersegmental regions between the neuromeres (arrowhead, Fig. 2G). The 1D4 and BP102 staining patterns are consistent with the hypothesis that some axons that would normally travel within the longitudinal tracts are rerouted across the midline in double mutant embryos.

We also examined double mutant phenotypes using mAb C1.427, which recognizes the Connectin protein (Meadows et al., 1994). In late stage 16 embryos, C1.427 stains two commissural bundles in each segment, two longitudinal bundles, and a small group of neuronal cell bodies (Fig. 2I). In Ptp10D Ptp69D double mutant embryos, the C1.427-positive commissural bundles are much thicker, and the longitudinal bundles are irregular and sometimes broken (Fig. 2J).

Although some longitudinal axons cross the commissures in Ptp10D Ptp69D double mutant embryos, many others remain within the longitudinal tracts. We wondered whether RPTPs might also be involved in defining these other longitudinal pathways, since all four neural RPTPs are expressed on most or all CNS axons. Accordingly, we constructed and analyzed quadruple mutant embryos bearing null mutations in all four Rptp genes. When these embryos were stained with 1D4, a striking phenotype was observed in which most axons cross the midline and all longitudinal pathways are severely disrupted. Only short fragments of longitudinal tracts can be visualized in quadruple mutants (arrowhead, Fig. 2E). Many bundles cross the midline in each segment, and these do not respect the normal boundaries of the commissures (arrows, Fig. 2E). A comparison between Figs 2B and 2E reveals that removing all four RPTPs essentially converts all of the 1D4-positive longitudinal axons into commissural axons.

Quadruple mutants stained with BP102 have fused anterior and posterior commissures, so that a single broad tract crossing the midline is observed in each segment. This phenotype is consistent with the hypothesis that more axons are rerouted across the midline than in double mutants (Fig. 2H). We have also done a phenotypic analysis of all triple mutant combinations, in order to understand how each RPTP contributes to CNS and motor axon guidance, and this will be described elsewhere (Q. S. and K. Z., unpublished results).

1D4-positive axons make a number of different axon guidance errors in Ptp10D Ptp69D double mutants. First, the three longitudinal bundles, which are normally separate, fuse with each other for short regions and then separate, and the outer bundle is usually missing (Fig. 2D). Second, the entire 1D4-positive longitudinal tract is sometimes rerouted across the midline within the domain that is demarcated by the anterior and posterior edges of the commissures (Fig. 3B, arrow). The ectopic commissural tract does not appear to avoid the zone between the commissures, which is normally free of axons; this is reflected in the BP102 staining pattern, which shows expansion of the commissures into this zone (Fig. 2G). Third, a longitudinal bundle occasionally wanders diagonally across the midline, traversing the intersegmental region between neuromeres where no axons normally grow (Fig. 3C, arrow). Finally, longitudinal bundles often stop abruptly, causing breaks in the longitudinal tracts (Fig. 3D, arrows).

Although these phenotypes indicate that 1D4-positive axons have been rerouted, they do not show what changes have occurred in the pathways taken by individual growth cones.

This is because many axons are stained by 1D4 in stage 16 embryos, and staining appears to be primarily restricted to established bundles rather than to growth cones. To analyze the behavior of individual growth cones, we examined 1D4-stained double mutant embryos at much earlier stages. During stage 12, the only neurons stained by 1D4 are the longitudinal pioneer neurons MP1, dMP2, vMP2, and pCC, in addition to the motoneuron aCC (Fig. 3A; Seeger et al., 1993). In normal embryos, the vMP2 and pCC axons fasciculate with each other and extend anteriorly, while MP1 and dMP2 axons fasciculate and extend posteriorly. These axons eventually pioneer two continuous longitudinal pathways. One of these is the inner 1D4-positive bundle, and the other is part of the middle bundle (Hidalgo and Brand, 1997).

The pCC, vMP2, MP1, and dMP2 longitudinal pioneer growth cones follow abnormal pathways in robo and slit mutants, turning toward the midline rather than navigating within the longitudinal tracts (Seeger et al., 1993; Kidd et al., 1998a,b, 1999; Battye et al., 1999). When we examined Ptp10D Ptp69D double mutants at stages 12 and 13, however, we found that these growth cones follow normal pathways (Fig. 3A). The longitudinal pathways established by the pioneer axons develop normally up through stage 14. Early C1.427-positive growth cones also follow normal pathways in the double mutants (data not shown). These data suggest that the axons that cross the midline in Ptp10D Ptp69D mutants are those of later neurons whose growth cones would normally follow established longitudinal pathways. We have been unable to find antibody or enhancer trap markers that allow selective visualization of these growth cones.

To visualize individual axons and growth cones that are affected in Ptp10D Ptp69D mutants, we performed lineage tracing experiments in which the fluorescent dye DiI was used to label all of the progeny of single neuroblasts (NBs) in vivo. Individual neuroectodermal cells were randomly labeled at stage 8, and the embryos allowed to develop until stage 17, after which DiI-labeled NBs arising from the injected cells were identified based on their positions, and the axons and cell bodies of the NB progeny were visualized by confocal microscopy (Schmid et al., 1999).

Analysis of a large number of NB lineages in the double mutants revealed that many CNS axonal pathways are altered in complex ways by the absence of DPTP10D and DPTP69D. A complete description of these changes will be published elsewhere. Here we describe the projection patterns of three sets of neurons that illustrate essential aspects of the phenotype: the progeny of NBs 3-1, 4-2, and 2-5. No alterations in numbers or positions of cell bodies are observed for these lineages in Ptp10D Ptp69D embryos.

The NB 2-5 lineage generates 15-22 cells by stage 17, of which 8-16 are intersegmental interneurons. Some of these (4 to 8 neurons) extend axons across the midline in the anterior commissure; these axons then turn anteriorly in the contralateral longitudinal tract and grow all the way to the brain (up to 10 segments). The remaining intersegmental interneurons (4 to 8 neurons) extend axons anteriorly in the ipsilateral longitudinal tract that stop after projecting about half as far. These contralateral and ipsilateral axons form the most substantial fibers in the longitudinal connectives. There is also a single motoneuron which extends an axon in the ipsilateral ISNd pathway and innervates muscles 15-17 (Schmid et al., 1999; Fig. 4E). In Ptp10D Ptp69D mutants (n=3), the contralaterally projecting interneuronal axons cross the midline and turn anteriorly in a normal manner, but then double back across the midline after about two segments and grow posteriorly in the ipsilateral longitudinal tract (Fig. 4F). The axons of the ipsilateral intersegmental neurons grow anteriorly for a short distance and stop. The ISNd motoneuron extends an axon toward the midline that stalls and never enters the ISN root. This lineage illustrates that interneuronal axons abnormally cross the midline in the Rptp double mutant, and that a motor axon is deflected toward the midline.

The NB 4-2 lineage produces about 22 cells, including the well-characterized RP2 motoneuron, which extends its axon along the ISN pathway and innervates the dorsal muscle 2. It also generates the CoR motoneurons, whose axons constitute all of the SNc motor nerve. All of the interneurons are local; two or three of them extend axons across the anterior commissure that bifurcate in the contralateral connective (Schmid et al., 1999; Fig. 4C). In Ptp10D Ptp69D double mutants (n=6), the RP2 axon stalls before reaching its target, and the CoR axons do not branch onto all of their target muscles. An ipsilateral longitudinal projection is formed that extends anteriorly from the clone and crosses the segment border; this is never observed in wild type. Finally, the local interneuronal projection splits after crossing the midline, so that two pathways form instead of one; this was observed in all lineages examined (Fig. 4D). In summary, this lineage illustrates that abnormal longitudinal pathways are formed in mutant embryos and that pathway selection in the commissures is altered.

NB 3-1 produces the RP1, RP3, RP4 and RP5 motoneurons, which extend axons across the anterior commissure and into the ISNb nerve, eventually innervating the ventrolateral muscles. It also generates a variable number of interneurons, which cross the midline and project both posteriorly (intersegmental interneurons) and anteriorly (local interneurons) in the contralateral connective (Schmid et al., 1999; Fig. 4A). In Ptp10D Ptp69D mutants (n=3), the RP neurons extend axons normally across the commissure and into the ISNb nerve, although they do not form normal synapses. The interneuronal projections, however, are radically altered. They still cross the midline, but do not form defined anterior and posterior projections in the contralateral connective. Instead, they grow anteriorly in a circular path around the neuropil, contacting the midline at the end of their trajectory (Fig. 4B). Like the other lineages, 3-1 illustrates that longitudinal pathways cannot form normally. Both the anterior and posterior interneuronal projections are missing, and are replaced by a swirl of axons that grow to the midline. These kinds of pathway alterations could give rise to the connective breaks that are observed in mutant embryos.

The midline glia that enwrap the commissural tracts express guidance cues for commissural axons such as Netrin and Slit (Harris et al., 1996; Mitchell et al., 1996; Kidd et al., 1999).

To evaluate the origin of the Ptp10D Ptp69D phenotype, we wished to determine whether these cells develop normally in mutants lacking these phosphatases. The RPTPs are not expressed in midline glia, but it is possible that their absence from CNS growth cones could indirectly affect the glia by perturbing neuronal-glial interactions. We visualized glia with a mAb recognizing the midline glial surface protein Wrapper (Noordermeer et al., 1998), which is an excellent marker for glial morphology. Fig. 5A shows a Ptp10D Ptp69D embryo stained with anti-Wrapper. The midline glial pattern is indistinguishable from that seen in wild type (Noordermeer et al., 1998). Midline glia also appeared normal when stained with an anti-Slit mAb (data not shown).

Midline neurons might also be important for normal navigation of commissural growth cones. We evaluated whether these cells develop normally by dye-filling midline lineages. Fig. 5B,C show MP1 and MP2 lineages in Ptp10D Ptp69D mutants. These are also indistinguishable from wild type (Bossing and Technau, 1994; Schmid et al., 1999). Examination of stage 12 embryos with 1D4 (Fig. 3A) also showed that MP1, dMP2 and vMP2 extend axons in a normal manner. Finally, we found that the median neuroblast (MNB) lineage develops normally in these mutants (data not shown).

The interaction of Robo with Slit produces a repulsive signal that keeps longitudinal axons away from the midline and prevents commissural axons from recrossing. robo mutants have a phenotype in which the inner 1D4-positive longitudinal bundle crosses the midline and circles around it. The outer two bundles are less affected (Fig. 6E). Mutations in slit confer a more extreme phenotype in which most axons converge on the midline and do not leave (Fig. 7E).

robo is likely to have a weaker phenotype than slit because Robo2, a second Slit ligand, can mediate repulsion of some axons when Robo is absent. Consistent with this, removal of one copy of slit, which would be expected to reduce repulsion through both Robo family proteins, produces a weak midline crossing phenotype when a copy of robo is also removed (slit/+, robo/+), and enhances the phenotype of robo homozygotes (slit/+, robo/robo) (Kidd et al., 1999). Removing one copy of slit in a wild-type background does not produce any abnormal phenotypes.

Another component of this system, the Comm protein, is transferred from midline glia to commissural axons and downregulates Robo, allowing commissural axons to cross the midline (Tear et al., 1996; Kidd et al., 1998b). In comm mutants, Robo is not downregulated, and all axons are repelled from the midline. This produces a striking phenotype in which no commissures form (Fig. 6A; Table 1). Because Comm acts through Robo, elimination of Robo suppresses the comm phenotype and allows axons to cross the midline; thus, comm robo double mutants have phenotypes like those of robo single mutants (Seeger et al., 1993).

To investigate the relationship between the Robo/Slit/Comm system and the midline crossing phenotype of the Rptp double mutant, we combined the Ptp10D and Ptp69D mutations with comm, robo and slit. In a comm Ptp10D Ptp69D triple mutant, axons are able to cross the commissures. A thick commissural tract, visualized by BP102 (Fig. 6B) or 1D4 (Fig. 6D) staining, is observed in 37% of segments in these triple mutants (Table 1). Note that the 1D4-positive axons would normally not cross the midline even in wild-type embryos. These results suggests that Robo-mediated repulsion from the midline is reduced in the absence of the RPTPs.

When Ptp10D Ptp69D is combined with robo, a severe phenotype is produced in which most of the 1D4-positive axons converge on the midline (Fig. 6F). Some circles around the midline are still seen in these triple mutants, but many axons fasciculate into a single thick bundle that extends up and down the midline. Vestiges of the lateral longitudinal pathways are also present in some segments. The robo Ptp10D Ptp69D phenotype approaches the severity of the slit phenotype (Fig. 7E) when visualized with 1D4, suggesting that most repulsion from the midline has been eliminated in these mutants. We also combined robo with each of the Rptp single mutants, and removed one copy of robo from the Ptp10D Ptp69D double mutant, but found that none of these combinations produced synergistic phenotypes.

To investigate the relationship between the RPTPs and Slit, we removed one copy of slit from the Ptp10D Ptp69D double mutant. As shown in Fig. 7, this produces a significant enhancement of the phenotype. More 1D4-positive axons cross the midline, and the longitudinal tracts move closer together, so that the average width of the CNS axonal array decreases from 15.4 μm (±1 μm) in Ptp10D Ptp69D to 9.4 μm (±1 μm; 32 segments measured for each genotype) in Ptp10D Ptp69D, slit/+ (Fig. 7A,B). The axonal array observed with BP102 is also compressed toward the midline when one copy of slit is removed, and more extensive commissural fusion is observed (Fig. 7C,D). These results indicate that, like reducing or eliminating Robo expression, removal of the RPTPs sensitizes the embryo to a 50% reduction in the amount of Slit-mediated repulsion from the midline.

The Rptp mutations do not appear to affect Robo or Slit expression, because Ptp10D Ptp69D embryos stain in a normal manner with anti-Robo or anti-Slit mAbs (data not shown). Interestingly, however, the expression pattern of two RPTPs in stage 16 embryos may reflect their roles in midline guidance. The Robo protein, although apparently expressed on most or all CNS neurons, is restricted to longitudinal tracts in late embryos due to its downregulation by Comm on commissural axons (Kidd et al., 1998a,b). We found that the DLAR (Fig. 6G) and DPTP10D (Fig. 6H) proteins are also selectively localized to the longitudinal tracts in late stage 16 embryos. This localization, however, is not Commdependent, because ectopic expression of Comm on all neurons, which causes Robo to disappear from CNS axons (Kidd et al., 1998b), does not visibly reduce expression of these RPTPs. Ectopic Comm expression also does not affect the other two RPTPs, DPTP69D and DPTP99A, which do not exhibit selective localization to longitudinal tracts (data not shown). It will be of interest to determine the mechanism by which DLAR and DPTP10D become localized to longitudinal axons.

To examine the specificity of the interaction between the Rptp mutations and the Robo/Slit/Comm system, we also combined the Ptp10D Ptp69D double mutation with a deletion of both Netrin genes. The Netrin deletion alone produces a phenotype in which commissures are reduced but not eliminated (Harris et al., 1996; Mitchell et al., 1996), and this phenotype is unaffected when the RPTPs are also removed (data not shown). Thus, the suppression of the comm phenotype by the Rptp double mutation is due to a selective effect on the repulsion system rather than to a generalized suppression of any phenotype that reduces midline crossing.

We also examined whether the Rptp phenotypes are affected by alterations in interaxonal adhesion. In one model, the longitudinal bundle fusion phenotypes could be caused by a decrease in interaxonal adhesion that would allow longitudinal axons to separate from their normal pathways and fasciculate with other bundles. Conversely, bundle fusion might be produced by an increase in adhesion that would cause two normally separate bundles to adhere to each other. The deviation of longitudinal axons from their normal pathways might also generate midline crossing phenotypes, since it has been shown that ablation of longitudinal pioneer axons causes some follower axons to cross the midline (Hidalgo and Brand, 1997). If this were the case, crossing the midline could be considered to be a ‘default’ pathway that longitudinal axons select when they fail to adhere normally to their appropriate bundles.

To evaluate these models, we used genetic methods to reduce adhesion within the longitudinal tracts, and determined whether this could enhance or suppress Ptp10D Ptp69D phenotypes. In PlexinA mutants, which lack a Semaphorin receptor, the outer 1D4 (Fasciclin II)-positive bundle is discontinuous, as it is in Ptp10D (hypomorph) Ptp69D mutants. This phenotype is suppressed when one copy of the FasciclinII (FasII) gene is removed. Since FasII encodes a homophilic adhesion molecule expressed on these axons, this manipulation would be expected to decrease interaxonal adhesion within the 1D4-positive longitudinal bundles (Winberg et al., 1998). To examine whether reducing adhesion in this manner would affect Rptp phenotypes, we removed one copy of FasII from Ptp10D Ptp69D mutants. We observed no effect on their phenotypes, however (data not shown).

Four Drosophila RPTPs are selectively expressed on CNS axons and growth cones. Genetic analysis of DLAR, DPTP69D and DPTP99A showed that pathfinding by motoneuron and photoreceptor growth cones is dependent on RPTP function (Desai et al., 1996, 1997a; Krueger et al., 1996; Garrity et al., 1999). The signaling mechanisms used by RPTPs to control growth cone guidance are not well understood, although recent studies suggest that DLAR regulates a motor axon pathfinding decision via competitive genetic interactions with the Abl tyrosine kinase (Wills et al., 1999). DLAR also functions in opposition to another RPTP, DPTP99A, at the same growth cone choice point (Desai et al., 1997a). The guidance cues that motoneuron growth cones recognize at such choice points remain unidentified, however, so these results do not define mechanisms by which RPTP activity regulates growth cone navigation in the neuromuscular system.

Here we show that RPTPs regulate growth cone repulsion from the CNS midline. This is a well-characterized decision controlled by identified ligands and receptors. Thus, the finding that RPTPs participate in repulsion should facilitate identification of the components of the signaling pathways in which these phosphatases function.

Although the RPTPs are expressed on most or all interneuronal axons within the CNS, previous genetic studies have not defined CNS phenotypes associated with removal of DLAR, DPTP69D or DPTP99A. In this paper we describe the isolation of mutations in the gene encoding the fourth neural RPTP, DPTP10D, and show that this phosphatase is critical for control of axon guidance at the midline. No embryonic phenotypes are observed in single mutant embryos lacking only DPTP10D. However, when DPTP10D and DPTP69D are both absent, a CNS phenotype is observed in which many longitudinal axons abnormally cross the midline (Figs 2, 3; Table 1).

We visualized ectopic midline crossing in Ptp10D Ptp69D double mutants by staining with mAbs 1D4, C1.427 and BP102 (Figs 2, 3), and by dye-filling lineages of neurons derived from specific NBs (Fig. 4). These results indicate that longitudinal pioneer axons are not affected by the Rptp mutations (Fig. 3A), but many later axons cross the midline or follow abnormal pathways within the longitudinal tracts. The opposite situation is observed in mutant embryos lacking the Robo receptor, which is expressed on neuronal growth cones and mediates their responses to a midline repulsive signal emanating from the midline glia. Here longitudinal pioneer axons in the medial 1D4-positive longitudinal bundle cross the midline and circle around it, while the later axons in the middle and lateral bundles appear to follow relatively normal pathways (Kidd et al., 1998a,b; Fig. 6E). In triple mutants lacking Robo, DPTP10D and DPTP69D, a very strong phenotype is observed in which all three longitudinal 1D4 bundles are severely affected. In some segments, only a single tract running along the midline is observed (Fig. 6F).

This triple mutant phenotype does not necessarily indicate that Robo and the RPTPs are involved in the same pathways, since simply adding together the guidance errors seen in robo and in Ptp10D Ptp69D might be expected to generate a phenotype affecting all three 1D4-positive longitudinal bundles. Furthermore, the phosphatases are not required for Robo function in some axons because the growth cones of longitudinal pioneer neurons are rerouted across the midline in robo mutants but are not affected in Ptp10D Ptp69D embryos (Fig. 3A). We showed that the RPTP mutations do perturb Robo-mediated repulsive signaling at the midline by examining their interactions with two other genes in the repulsive pathway, comm and slit.

Comm is a protein made by midline glia that is transferred to commissural axons and causes downregulation of Robo. The loss of Robo from these axons allows them to ignore the repulsive signal and cross the midline. When Comm is not expressed, Robo cannot be downregulated and no axons are able to cross (Fig. 6A; Table 1; Tear et al., 1996; Kidd et al., 1998b). Because Comm acts through Robo, in a comm robo double mutant the comm phenotype is suppressed and extra axons cross the midline (Seeger et al., 1993). We find that Ptp10D Ptp69D partially suppresses comm, indicating that removal of the RPTPs interferes with reception of the repulsive signal (Fig. 6B,D; Table 1).

The repulsive factor produced by midline glia is likely to be the extracellular matrix protein Slit. In the absence of Slit, all interneuronal axons converge on the midline and remain there (Fig. 7E; Kidd et al., 1999; Battye et al., 1999; for review see Zinn and Sun, 1999). Robo and Slit proteins bind to each other (Brose et al., 1999; Li et al., 1999), and robo and slit also interact genetically. Removing one copy of slit, which produces no phenotype on its own, enhances the robo homozgyote phenotype and confers a weak phenotype on robo/+ heterozygotes (Kidd et al., 1999). Similarly, the Ptp10D Ptp69D phenotype is strengthened by removal of one copy of slit (Fig. 7A-D).

Slit ligand is thought to interact with two Robo receptors: Robo and Robo2. Many axons follow normal longitudinal pathways in robo mutants, probably because repulsive signaling through Slit and Robo2 keeps them from converging on the midline. If Robo and Robo2 are the only receptors for Slit, removing both would be expected to eliminate repulsion, producing a slit-like phenotype (Kidd et al., 1999). The robo Ptp10D Ptp69D triple mutant, when visualized with 1D4, has a phenotype which approaches that of slit (Figs 6F, 7E). This suggests that DPTP10D and DPTP69D are required for signaling through Robo2 (or possibly through other, as yet unidentified, Slit receptors). Our data, however, do not support a model in which the RPTPs are only involved in Robo2 signaling. If this were the case, one would not expect to observe a genetic interaction between the Ptp10D Ptp69D and robo2 mutations. The robo2 Ptp10D Ptp69D triple mutant, however, has a very strong synergistic phenotype, favoring a model in which DPTP10D and DPTP69D function in repulsive signaling through both Robo receptors (Q. S., J. Simpson, T. Kidd, C. S. Goodman, and K. Z., unpublished results).

Although some longitudinal axons follow normal pathways in the absence of DPTP10D and DPTP69D, they apparently still require RPTP function to avoid the midline. In a quadruple mutant embryo lacking all four neural RPTPs, all 1D4-positive longitudinal pathways are converted into commissural pathways (Fig. 2E). Interestingly, however, the quadruple mutant phenotype is quite different from that of slit, because these ectopic commissural axons are still able to leave the midline and cross to the contralateral side of the CNS. Thus, growth to the midline and midline crossing are under separate genetic control in ways we do not yet understand.

The fact that longitudinal axons can be changed into commissural axons by elimination of RPTP activity suggests that tyrosine phosphorylation controls the manner in which growth cones respond to midline repulsive signals. This is consistent with the observation that pharmacological inhibition of tyrosine kinase activity in grasshopper embryos causes a robo-like phenotype in which the longitudinal axon of the pCC neuron crosses the midline and circles back to the ipsilateral side (Menon and Zinn, 1998). Further evidence that the effects of the inhibitor may actually be due to blockage of Robo signaling is provided by the recent observation that the Drosophila pCC axon in robo embryos has a unique branched morphology that is identical in appearance to that of the grasshopper pCC in inhibitor-treated embryos (Murray and Whitington; 1999; Menon and Zinn, 1998).

The repulsive response to midline signals is encoded within the Robo cytoplasmic domain, because chimeric receptors in which the extracellular domain of the attractive Netrin receptor is linked to the Robo cytoplasmic domain mediate repulsion rather than attraction (Bashaw and Goodman, 1999). The cytoplasmic domains of fly, nematode and mammalian Robo family proteins (Robos) contain conserved tyrosine-containing PYATT sequence motifs, suggesting that these domains could be direct targets for tyrosine kinases (Kidd et al., 1998a; Zallen et al., 1998). Phosphorylated tyrosine motifs usually function by binding to SH2 and PTB-domain adapter proteins that mediate downstream signaling events. Robo also contains two proline-rich sequences that could interact with SH3-domain adapters. Robo2 has the tyrosine-containing motif, but lacks the proline-rich sequences (Kidd et al., 1998a).

How are Robo signaling pathways regulated by RPTPs? We have no evidence at present that the RPTPs directly alter signaling by the Robo protein. It is possible that the RPTPs and Robo feed into separate pathways that only intersect after several signaling steps. There is, however, a known mechanism for RPTP-mediated positive regulation of tyrosine kinase pathways that suggests how DPTP10D and DPTP69D could facilitate Robo signaling. During T cell receptor (TCR) signal transduction, the RPTP CD45 removes an inhibitory C-terminal phosphate group from the Src-family tyrosine kinase Lck, thereby activating it and allowing it to phosphorylate the ζ chain of the TCR. The phosphorylated ζ chain in turn binds to an SH2-domain containing tyrosine kinase, ZAP-70, which mediates downstream signaling events. CD45 is required for TCR signaling because in its absence Lck is not activated and thus cannot efficiently phosphorylate the ζ chain (reviewed by Weiss and Littman, 1994). [Interestingly, CD45 may also be involved in the termination of the TCR signaling response, since it can dephosphorylate the ζ chain and prevent it from binding to ZAP-70 (Furukawa et al., 1994).]

Another mammalian receptor phosphatase, RPTPα? also dephosphorylates and activates Src-family kinases (Zheng et al., 1992; den Hertog et al., 1993). Fibroblasts derived from RPTPα knockout mice have reduced Src and Fyn activities, suggesting that RPTPα is an in vivo regulator of Src family kinase function (Ponniah et al., 1999; Su et al., 1999).

By analogy to these pathways, DPTP10D and DPTP69D might regulate growth cone repulsion by activating Src-family tyrosine kinase(s) that phosphorylate Robos. This could explain the genetic data, since the loss of RPTP function would be expected to cause a decrease in the extent of Robo phosphorylation.

One might also propose that positive regulation of repulsion by the RPTPs occurs through direct dephosphorylation of Robos, and that dephosphorylated Robos are more active in signaling. This would be unusual, however, since normally it is the phosphorylated form of a signaling motif that binds to downstream adapters.

A variant of the direct interaction model proposes that Robos become phosphorylated on tyrosines after engagement of Slit, and that DPTP10D or DPTP69D are recruited into a Robo/Slit signaling complex by their interactions with the phosphotyrosine motifs. RPTPs might remain bound to these sites for a significant time period, because they often hydrolyze phosphate-tyrosine bonds quite slowly (for example, see Lim et al., 1998). The RPTPs could then function as adapters themselves, binding to downstream signaling proteins and recruiting them into Robo/Slit receptor complexes. Determining which, if any, of these models is correct will require biochemical or genetic identification of in vivo substrates for RPTPs.

We thank the members of the Zinn group for discussions and comments on the manuscript, and Tom Kidd, Julie Simpson, and Corey Goodman for communication of results before publication. This work was supported by a Human Frontiers Science Project grant (RG0122/1997-B) to K. Zinn and W. Chia, and by an NIH RO1 grant to K. Z.