At the next stop sign turn right: the metalloprotease Tolloid-related 1 controls defasciculation of motor axons in Drosophila

Accurate innervation of somatic muscles is a prerequisite for coordinated movements in any higher organism. During embryogenesis, correct neuromuscular connectivity is established by the developmental processes of axon outgrowth,axon guidance, target selection and synapse formation(Tessier-Lavigne and Goodman,1996). Despite the fact that motor axons initially form coherent nerve bundles, certain growth cones leave these tightly fasciculated nerves at specific peripheral locations in order to migrate into their target regions. Defasciculation, the exit of axons from a nerve track, is therefore a crucial process that needs to be strictly regulated. Failure to defasciculate at these choice points leads to axon guidance defects and muscle innervation errors(Araujo and Tear, 2003). Which molecules regulate the detachment of motor axons from adherent nerve bundles?How are choice points molecularly defined?

The Drosophila neuromuscular innervation pattern provides a powerful experimental system to genetically dissect the molecular mechanisms of defasciculation and motor axon guidance in vivo. Trajectories of motor axons and their branching patterns are stereotypic and segmentally repeated(Sink and Whitington, 1991),as are the positions of neuromuscular terminals on muscle fibers(Hoang and Chiba, 2001). This relatively simple anatomy, together with the available genetic tools in Drosophila, have been of great advantage to identify genes and gene families with roles in motor axon guidance, including the Semaphorins(Kolodkin et al., 1993) and Netrins (Mitchell et al.,1996). Furthermore, three molecules have been identified that seem to specifically control defasciculation of motor axons: the transmembrane tyrosine phosphatase Lar (leukocyte common antigen-related) and the members of the immunoglobulin superfamily Beaten path (Beat) and Sidestep (Side), all three of which give rise to highly penetrant motor axon guidance phenotypes(Desai et al., 1996; Fambrough and Goodman, 1996; Krueger et al., 1996; Sink et al., 2001). Sidestep is expressed in somatic muscles during the period of motor axon pathfinding and is thought to function as a muscle-derived attractant for motor nerves(Sink et al., 2001). In sidestep mutant embryos, motor axons frequently bypass their targets,as they fail to defasciculate from their nerve tracts. As Lar, Beat and Side give rise to similar bypass phenotypes when mutated, they all seem to be involved in the defasciculation process, but how they interact mechanistically, and whether additional regulators are required, is currently unknown.

Here, we describe piranha, a novel axon guidance mutant with defasciculation defects. In piranha mutants, we identified point mutations in the evolutionarily conserved gene tolloid-related 1(tlr1), also called tolkin (tok), which encodes an extracellular metalloprotease (Nguyen et al., 1994; Finelli et al.,1995). Mutations in tlr1, but not in related Drosophila metalloproteases, lead to stable innervation errors that persist into larval stages. Motor nerves stay attached to each other at places where they should diverge and consequently fail to reach their muscle targets or use irregular routes. Genetic rescue experiments demonstrate that Tlr1 functions non-cell-autonomously, possibly in the hemolymph. In addition, the proteolytic activity of Tlr1 is required in cooperation with other axon guidance molecules, such as Sidestep, to terminate coherent axonaxon interactions at choice points.

The mutant alleles of tlr1 (tlr1^D427,tlr1^I678, tlr1^K598 and tlr1^K788) and side (side^C137,side^I306, side^I1563, side^K717) were isolated in an F2 methanesulfonic acid ethylester (EMS)-mutagenesis screen on the third chromosome for recessive mutations affecting the structure of neuromuscular junctions (NMJs) (Aberle et al.,2002). The side, tlr1 double mutants were constructed by recombination using recessive markers of the rucuca-chromosome. UAStlr1 and UAS-tld^23A were a kind gift of Michael O'Connor. The tolloid alleles tld^10E95,tld^9B66, tld^6B69 and tld⁷⁰⁷⁴and Heat shock-Gal4 were provided by the Tübingen stock collection. tlr1¹, tlr1³, kuz^E29,kuz^K01403, Mmp1^k04809, Mmp2^k00604 and Mmp2^KG01263,UAS-kuz^F, UAS-dsRed2 (#6282) and CG-Gal4 were obtained from the Bloomington Stock Center(http://flystocks.bio.indiana.edu/). To assess neuromuscular phenotypes, mutants were crossed into the CD8-GFP-Sh background (Zito et al.,1999). The following Gal4-lines were used: Elav-Gal4, 24B-Gal4(gifts of C. S. Goodman), G14-Gal4 (Aberle et al., 2002), Cha-Gal4(Salvaterra and Kitamoto,2001), Serpent-Gal4 (gift of R. Reuter), PPL-Gal4 (gift of M.J. Pankratz) and OK371-Gal4 (Mahr and Aberle,2006). As wild-type control strains, y w¹ or w;;CD8-GFP-Sh were used. Genotypes of larvae with transgenically labeled NMJs and motor nerves were OK371-Gal4/+; CD8-GFP-Sh,UAS-dsRed2/+ and OK371-Gal4/+; tlr1^K788/tlr1^D427,UAS-dsRed2. Rescue crosses were performed according to the following scheme: UASGeneX; tlr1^D427/TM6B crossed to Tissue-specific-Gal4; tlr1^K788/TM6B. For heat shock experiments, flies were allowed to lay eggs into the food of their vials for 3 hours. Embryos were then stored for 3, 6 or 9 hours or overnight at room temperature before a heat shock (1 hour, 37°C water bath) was applied.

Mutant lines with similar phenotypes were crossed to each other and judged for non-complementation by the presence of mislocalized NMJs. Mutations were mapped by meiotic recombination against the multiply marked rucucachromosome(Bloomington). Mapping was refined using available deficiencies. Df(3R)crb87-4, Df(3R)crb-F89-4 and Df(3R)96B did complement but Df(3R)crb87-5,Df(3R)slo3, Df(3R)XS and Df(3R)XTA1 failed to complement tlr1mutations. Database searches were performed at FlyBase(http://flybase.bio.indiana.edu/)and NCBI(http://www.ncbi.nlm.nih.gov/).

The innervation phenotypes were quantified in intact third instar CD8-GFP-Sh larvae. The locations of NMJs were evaluated through the translucent cuticle in abdominal segments A2-A7 using confocal microscopy. Embryonic guidance phenotypes were quantified in dissected embryos stained with anti-Fas II.

Genomic DNA of the tlr1 alleles was isolated using the QIAamp DNA Mini Kit (Qiagen), amplified by PCR and sequenced on both strands with the BigDye Terminator kit (PE Applied Biosystems). Sequences were analyzed using the Lasergene software package (DNAStar). The partial tlr1 cDNA clone RH04849 was obtained from BDGP(http://www.fruitfly.org/). Digoxigenin-labeled sense and antisense probes for in situ hybridizations were synthesized from RH04849 (Tautz and Pfeifle, 1989). In situ hybridizations were imaged with an Axiophot light microscope (Zeiss) equipped with a CCD-camera(ProgRes3012).

Embryos were stained as described (Mahr and Aberle, 2006), except that fluorescent labeling was performed with the TSA Cyanine-3 System (Perkin Elmer). Stained embryos were dissected with sharpened tungsten needles on microscope slides and imaged with an LSM510 confocal microscope (Zeiss) in the Cy3- and DIC channels. Signals in the Cy3-channel were depicted in black to enhance the contrast in overlays with the DIC images. Wandering third instar larvae were dissected in PBS on Sylgaard plates (Dow Corning Corporation) using spring scissors (Fine Science Tools) and insect pins (Emil Arlt). Larval fillets were fixed in 3.7%formaldehyde/PBS for 15 minutes, washed with PTx (PBS containing 0.1% Triton X-100) and blocked in PTx/5% normal goat serum. Primary antibodies were added overnight at 4°C. Fillets were washed with PTx and incubated with fluorescence-labeled secondary antibodies for 1 hour. Stained fillets were cleared in 70% glycerol/PBS and mounted onto microscope slides. The dilutions of the primary antibodies (supernatants) were as follows: mouse anti-Discs large (4F3) 1:100 and mouse anti-Fasciclin II (1D4) 1:5 (gifts of C. S. Goodman). The Cy3-conjugated secondary antibodies were diluted 1:400(Molecular Probes or Jackson ImmunoResearch). Fluorescently-labeled larvae were examined with a TCS SPL (Leica) or LSM510 (Zeiss) confocal laser scanning microscope. Projections and single images were adjusted for brightness and contrast using Adobe Photoshop.

In Drosophila embryos and larvae, the somatic musculature of abdominal segments A2-A7 consists of 30 muscles that are individually identifiable and innervated at invariant positions(Fig. 1A). Wiring of the neuromuscular system starts at stage 13 of embryogenesis, when motor axons begin to migrate toward the differentiating muscle fields(van Vactor et al., 1993). In each abdominal hemisegment, two major motor nerves defasciculate into five nerve branches (Fig. 1A). The segmental nerve (SN) bifurcates into the SNa and SNc, and the intersegmental nerve gives rise to a dorsal branch (ISN) and two ventral branches (ISNb and ISNd).

In a large-scale mutagenesis screen for genes that affect the structure and maintenance of NMJs (Parnas et al.,2001; Aberle et al.,2002), we discovered four alleles of a novel mutant, piranha, which showed missing and/or mislocalized NMJs. To visualize NMJs in piranha mutants, we used the transgenically encoded fusion protein CD8-GFP-Sh (Zito et al.,1999). CD8-GFP-Sh is controlled by a muscle-specific promotor and consists of the extracellular and transmembrane domain of human CD8, fused cytoplasmatically to GFP and a C-terminal domain of the Shaker potassium channel. It binds to the postsynaptic protein Discs large (Dlg) and highlights all NMJs consisting of type Ib and Is boutons(Zito et al., 1999). Using this synaptic marker, we could evaluate NMJs through the translucent cuticle of intact third instar larvae.

In wild-type larvae, the dorsalmost NMJs on muscle pairs 1/9 and 2/10 are normally arranged symmetrically relative to the axis of the ISN(Fig. 1A-C). In piranha^D427/piranha^K788mutants, however, we found that muscles 1 and 9 either lacked NMJs completely or were innervated at ectopic sites in 38.1% (muscle 1) or 25.0% (muscle 9) of hemisegments (Table 1, Fig. 1D). We observed even stronger defects in ventral muscle regions. In wild type, the ISNb defasciculates from the ISN and forms, among others, NMJs in the cleft between muscles 6/7 and on a ventral and medial position on muscles 12 and 13,respectively (Fig. 1E,F). In piranha mutants, we found that NMJs on muscle 12 and 13 were mislocalized or absent in 64.3 and 56.0% of hemisegments, respectively(Table 1, Fig. 1G). Muscles 6 and 7 showed an erroneous innervation pattern in 23.9 and 34.6% of hemisegments. It is important to note that the pattern and degree of misinnervation varied between hemisegments along the anterior-posterior axis and also between two hemisegments located on contralateral sides. We determined the total number of misinnervated muscles per hemisegment and found that between 7.1 and 46.4% of muscles were affected in piranha mutant hemisegments (n=84)compared with 0-14.3% in wild-type larvae (n=44). Thus, innervation errors occurred in all body wall regions and in every hemisegment, but the innervation pattern was unique for each hemisegment, indicating that the innervation process is largely autonomous within a given hemisegment.

As CD8-GFP-Sh is a transgenically encoded exogenous protein, it is possible that it does not accurately reflect the location of endogenous NMJs. To exclude this possibility, we stained piranha mutant larvae with antibodies that recognize synaptic marker proteins. Dlg is a postsynaptic protein that mediates the anchorage of transmembrane receptors to the underlying cytoskeleton (Budnik et al.,1996). We did not observe neuromuscular Dlg staining in places where CD8-GFP-Sh clusters were absent, neither in wild-type(Fig. 1H-J) nor in piranha mutant larvae (Fig. 1K-M). CD8-Sh-GFP therefore faithfully reflects the presence and positions of NMJs.

As CD8-GFP-Sh visualizes only postsynaptic sides of NMJs but not motor nerves, we asked whether the unusual locations of NMJs in piranhamutants were caused by misguided axons. We stained dissected third instar larvae with anti-Fasciclin II antibodies (anti-Fas II). Fasciclin II is expressed by all motoneurons and accumulates in their axons and presynaptic terminals, making it possible to selectively visualize motor nerves but not the afferent sensory nerves that project along similar pathways(van Vactor et al., 1993). In wildtype larvae, muscles 12 and 13 are innervated by a ventrodorsal projection of the ISNb pathway (Fig. 2A-C). In piranha mutants, we found that muscles 12 and 13 were often innervated via a dorsoventral projection of the ISN or SNa,which induced the formation of NMJs at ectopic sites(Fig. 2D-F). This suggests that motoneuronal growth cones of the ISNb pathway failed to navigate into the ventral muscle field via their normal route. Despite this initial failure,however, some growth cones eventually managed to detach from the ISN or SNa and innervated muscles 12 and 13 from the dorsal side (reach back phenotypes via a different nerve route). These experiments show that motor axons are misrouted in piranha mutants, which leads to the formation of synaptic terminals at ectopic locations.

During manual dissection of larvae, great care must be taken to preserve the trajectories of peripheral nerves. To circumvent this technical hurdle, we devised a genetic method that permits the visualization of both motor nerves and NMJs in living animals. Using the Gal4/UAS-system(Brand and Perrimon, 1993), we expressed the Discosoma Red Fluorescent Protein (UAS-dsRed2)specifically in motoneurons (OK371-Gal4) in the background of CD8-GFP-Sh wild-type or piranha mutant animals. Motor axons and presynaptic terminals were therefore highlighted in red and muscles and postsynaptic terminals in green (Fig. 2G-L). In wildtype larvae, the SN and ISN nerves form a single nerve bundle that defasciculates into five nerve branches at the ventralmost choice point (shown schematically in Fig. 2G). The ISNb and ISNd detach from the ISN, whereas the SNc detaches from the SNa(Fig. 2J). In piranhamutants, we observed that the ISNb frequently stayed attached to the ISN,sometimes as a separately identifiable axon bundle, and reached its target muscles via irregular routes (Fig. 2H,K). We found similar defasciculation errors in the SN pathway. As shown in Fig. 2I,L, the SN nerve often grew too far to the posterior, because the SNa defasciculates much too late. piranha mutant larvae display, therefore, severe bypass and misrouting phenotypes, primarily due to motor axons remaining attached to each other at choice points.

We asked next at which time point these projection errors develop, and we examined motoneuronal trajectories during stages 16-17 of embryonic development, when neuromuscular connectivity is established. The ISN of wild-type embryos has three clearly visible branch points(Fig. 3A). In piranhamutants, we observed in 84.4% of hemisegments(Table 1) that the branch points appeared underdeveloped and/or the ISN was stalled at the second branch point and barely reached the position of the third branch point (arrow in Fig. 3B). Within the developing SNa pathway, axons normally bifurcate into a dorsal and posterior branch at the dorsal edge of muscle 12 (Fig. 3C). In piranha mutant embryos, we found SNa defects in 51.2% of hemisegments, including, for example, two dorsal branches of the SNa(arrowheads in Fig. 3D). In ventral muscle regions, the ISNb failed to defasciculate from the ISN in 96.2%of hemisegments. The ISNb was either tightly attached to the ISN (fusion bypass, see left segment in Fig. 3D and Table 1) or formed a separated parallel nerve bundle (split bypass, middle and right segment in Fig. 3D and Table 1). In addition, we and others (M. Serpe and M. O'Connor, personal communication) observed defasciculation defects in Fasciclin II-positive nerve tracts in the embryonic central nervous system (data not shown). The innervation defects do not strongly compromise embryogenesis and larval development, because we recover piranha^D427 mutant second and early third instar larvae at the expected ratios (data not shown). Based on these observations, we conclude that the piranha mutant guidance phenotype arises during embryogenesis, and the defects are quantitatively stronger but qualitatively similar to the larval defects.

The piranha locus was mapped by meiotic recombination and deficiency mapping to the cytological region 96A2-A21 on the third chromosome. During meiotic mapping, we noticed homozygous mutant adult escapers that lacked the posterior crossvein of the wings. Literature and database searches revealed that mutants in tlr1 [tolkin (tok)],located at 96A18-A19 (Fig. 4A),display a similar wing phenotype (Nguyen et al., 1994; Finelli et al.,1995). To examine if piranha is allelic to tlr1,we performed complementation crosses using existing tlr1 alleles. Both tlr1¹ and tlr1³ did not complement our piranha alleles, and the transheterozygous animals had both axon guidance and posterior crossvein defects. The tlr1 cDNA encompasses 5.4 kb and encodes a predicted extracellular metalloprotease(Nguyen et al., 1994; Finelli et al., 1995). Consisting of 1464 amino acids, Tlr1 is composed of an N-terminal prodomain, a central metalloprotease domain and a C-terminal protein-protein interaction domain containing five CUB (complement subcomponents C1r/C1s, Uegf, BMP-1) and two EGF (epidermal growth factor) domains(Fig. 4B).

To identify potential molecular lesions, we sequenced the Tlr1-encoding exons and found single point mutations in all four piranha alleles. We therefore renamed our piranha alleles to tlr1^K788, tlr1^D427,tlr1^K598 and tlr1^I678. Three of them contain nonsense mutations in the N-terminal prodomain: R409Stop in tlr1^K788, Q464Stop in tlr^D427 and Q472Stop in tlr1^K598(Fig. 4B). Provided that the prodomain functions only in inhibiting the activity of the protease domain,these alleles are probably null alleles. Consistent with this assumption is the previous classification of tlr1¹, which contains a stop codon in the protease domain, as a null allele(Finelli et al., 1995). In the fourth allele, we identified a serine to threonine exchange within the protease domain (S669T in tlr1^I678). This highly conserved serine residue is located within the sequence motif `SIMHY', which forms a characteristic loop, the Met-turn, that positions the terminal tyrosine residue to coordinate the zinc ion in the active site(Gomis-Ruth et al., 1993). Insufficient zinc coordination probably renders the protease domain non-functional in tlr1^I678. Indeed, a point mutation in the very same serine residue of the related metalloprotease Tolloid (S276F in tolloid^10E) has been isolated in a screen for new tolloid alleles (Finelli et al.,1994).

tolloid (tld) is a close sequence homolog and genomic neighbor of tlr1, with less than 1000 base pairs separating the two genes (Fig. 4A). Tolloid is involved in the determination of the dorsoventral body axis and functions as an activator of Decapentaplegic (Dpp) signaling in dorsal regions of blastoderm embryos (Shimell et al.,1991; Marques et al.,1997). In mammals, four relatives of Tlr1 and Tld have been identified (Fig. 4B): mammalian Tolloid (Tld), bone morphogenetic protein 1 (Bmp1), Tolloid-like 1 (Tll1) and Tolloid-like 2 (Tll2), the first two of which are isoforms derived from the same gene (Takahara et al.,1994; Takahara et al.,1996; Scott et al.,1999). The metalloprotease domain of Drosophila Tlr1 is 64.5, 67.5 and 67.0% identical to its mouse relatives Tld, Tll1 and Tll2,respectively. The CUB/EGF protein-protein interaction domain is also well conserved and shows 42.0, 42.5 and 43.9% identity to these proteins.

To determine where tlr1 is expressed during embryonic development,we synthesized Digoxigenin-labeled antisense probes using cDNA clone RH04849. tlr1 showed a dynamic expression pattern during embryogenesis(Nguyen et al., 1994; Finelli et al., 1995). We observed tlr1 expression first in the dorsal blastoderm(Fig. 4C). In the extended germ band stage, it was transcribed in progenitor cells of the visceral mesoderm(Fig. 4D). At stages 13-15, tlr1 was expressed in somatic muscles and the midgut(Fig. 4E). Beginning with stage 15, tlr1 transcripts also appeared in a few cells in the ventral nerve cord, the thoracic peripheral nervous system and the ring gland(Fig. 4F). Interestingly, tlr1 was strongly expressed in muscles but not in neurons when motor axons leave the central nervous system, indicating that muscle-derived Tlr1 may be important for motor axon guidance.

To determine in which tissue and during which developmental period Tlr1 is required, we performed transgenic rescue experiments using the Gal4/UAS system. We induced transcription of wild-type Tlr1 in a tlr1 mutant background during different developmental stages using Heat shock-Gal4. Compared with wildtype animals (Fig. 5A), mutant animals transgenic for UAS-tlr1 and Heat shock-Gal4 displayed strong innervation defects when reared at room temperature, because expression of Tlr1 is not induced(Fig. 5B). By contrast, mutant embryos that received a 1-hour heat shock (37°C) when they were 3-6, 6-9 or 9-12 hours old (stages 6-9, 10-12 or 12-15, respectively) developed into larvae with a normal innervation pattern(Fig. 5C). When we applied the heat shock at the end of embryogenesis (stage 17), however, we were unable to rescue the mislocalized NMJs in tlr1 mutants, indicating that Tlr1 is ineffective after NMJs have formed. Hence, Tlr1 functions during mid- to late-embryonic stages, when neuromuscular connectivity is established.

To examine whether Tlr1 is required in nerves or muscles, we performed rescue experiments using tissue-specific Gal4-lines. As tlr1 is strongly expressed in muscles during motor axon pathfinding, we reasoned that muscle-specific expression of Tlr1 might be sufficient for rescue. Indeed,expression of Tlr1 in all somatic muscles using G14-Gal4 completely restored the innervation pattern in homozygous mutant larvae(Fig. 5D). The rescued larvae were indistinguishable from wild-type animals with respect to the size and localization of their NMJs. Unexpectedly, however, expression of Tlr1 in all postmitotic neurons using Elav-Gal4 also completely restored the wild-type innervation pattern (Fig. 5E). When we expressed Tlr1 only in cholinergic neurons using Cha-Gal4, the mutant phenotype was not rescued, demonstrating that UAS-tlr1 is not unspecifically expressed in the absence of a Gal4 driver(Fig. 5F). From these results,it appeared that Tlr1 could be expressed either in muscles or all neurons to fully rescue the guidance defects. As Tlr1 is likely to be a secreted protease, this finding could be explained by a non-cell-autonomous function of Tlr1. If secretion into the extracellular matrix or hemolymph is sufficient to deliver Tlr1 to the locations where its proteolytic activity is required, then expression of Tlr1 in tissues normally irrelevant for motor axon guidance should rescue the innervation defects. Using fat body-specific Pumpless-Gal4(PPLGal4) or two different hemocyte-specific drivers (CG-Gal4 and Serpent-Gal4), we could completely restore the wild-type innervation pattern in tlr1 mutants (data not shown), supporting a cell non-autonomous function for Tlr1.

To test whether Tlr1 is also sufficient to misdirect motor axons in a wild-type background, we overexpressed it either in the nervous system(Elav-Gal4) or in all muscles (G14-Gal4) but we did not observe any motor axon guidance phenotypes, as visualized with CD8-GFP-Sh (data not shown). We noticed, however, that overexpression of Tlr1 in wing discs (24B-Gal4) leads to a large central wing blister, suggesting that the function of Tlr1 during wing development is dosage sensitive: too little Tlr1 leads to missing posterior crossveins, whereas too much Tlr1 causes loss of adhesion between wing epithelia (data not shown).

As mutations in tlr1 affect a large subset, but not all, NMJs, we wondered whether the proteolytic activities of other metalloproteases are additionally required for pathway selection. As a closely related paralog of Tlr1, Tolloid may be required for the innervation of the remaining muscles. As strong alleles of tld are embryonic lethal, we searched for transheterozygous allelic combinations that would survive to third instar larvae. Compared with tlr1 mutants(Fig. 6A), we were not able to detect mislocalized or structurally abnormal NMJs in tld^6B69/tld⁷⁰⁷⁴,tld^6B69/tld^9B66 or tld^9B66/tld^10E95 mutants, indicating that Tolloid is not required for wiring and maintaining NMJs(Fig. 6B).

Both Tlr1 and Tld are members of the astacin family of metalloproteases, a subgroup of the metzincin superfamily, which comprises in addition the serralysins, the matrixins (matrix metalloproteases) and the adamlysins(ADAMs, metalloproteases with a disintegrin and metalloprotease domain)(Bond and Beynon, 1995; Stocker et al., 1995). Kuzbanian (Kuz), a member of the ADAM family in Drosophila, is expressed in neurons and has been shown to play a role in axon guidance at the ventral midline, raising the possibility that it may also regulate pathfinding of motor axons (Fambrough et al.,1996; Schimmelpfeng et al.,2001). When we examined kuz^E29/kuz^K01403 mutant larvae,however, all NMJs were at their wild-type positions, indicating that motor axons project along correct trajectories(Fig. 6C). Other prime candidates for regulators of axonal pathfinding are matrix metalloproteases(Mmps). Mmps in general process extracellular matrix components, which is necessary for tissue remodeling and cell migration, but has also been shown to be crucial for axon guidance (Webber et al., 2002). Drosophila has only two Mmps, Mmp1 and Mmp2. During the time of motor axon pathfinding, Mmp1 is expressed in midline glial cells and Mmp2 in neurons and mesodermal cells, all tissues that are relevant for navigating growth cones (Llano et al.,2000; Llano et al.,2002; Page-McCaw et al.,2003). When we analyzed Mmp1^k04809, Mmp2^k00604 and Mmp2^KG01263 mutant larvae, however, we failed to detect any neuromuscular abnormalities (data not shown).

We next examined if Tlr1 and Tld could functionally replace each other if expressed in similar tissues. To test for functional redundancy, we crossed UAS-tld or UAS-tlr1 into the tlr1 mutant background and expressed either gene with Elav-Gal4 or 24B-Gal4. In contrast to experiments using UAS-tlr1 (Fig. 6D), Tld was not able to substantially rescue the tlr1mutant phenotype, irrespective of whether we expressed it in neurons or muscles (Fig. 6E). Similarly, UAS-kuz failed to improve the projections errors in tlr1mutants (Fig. 6F), showing that neither Tld nor Kuz could functionally replace Tlr1. Thus, loss- and gain-of-function analysis of metalloproteases belonging to the matrixin,adamlysin and astacin families support a key-regulatory role of Tlr1 in motor axon guidance in Drosophila.

To identify candidate molecules that might act in concert with Tlr1, we searched for mutants with related phenotypes. In our large-scale EMS mutagenesis screen of the larval NMJ, we identified four new alleles of sidestep (side^C137, side^I306,side^I1563, side^K717). Sidestep, originally identified by H. Sink and colleagues (Sink et al., 2001), is a transmembrane receptor of the immunoglobulin superfamily that functions as a muscle-derived attractant for motor axons. Similarly to tlr1 mutants (Fig. 7A,7A'), we found missing and abnormally localized NMJs in side^C137/side^I1563 mutant larvae, regardless of whether we used CD8-GFP-Sh (Fig. 7B) or anti-Fas II antibodies (Fig. 7B') as detection methods. The immunohistochemical stainings also confirmed that the CD8-GFP-Sh marker reliably reflected the axon guidance defects in side mutants. In side mutant embryos, pathfinding of the ISNb was defective in 83.8% of hemisegments(Table 1). tlr1 and side mutants therefore share a variety of phenotypic similarities:(1) embryonic defasciculation defects affect all pathways and persist into larval stages; (2) failures to reach final branch points; (3) innervation patterns vary between segments; (4) ventral muscles are more frequently affected; and (5) NMJs can be mislocalized or absent on an affected muscle fiber. For a majority of muscles, however, the side mutant phenotype was more penetrant (Fig. 7A”,B”).

As mutations in genes that function together in the same biological process often share phenotypic similarities, we created side,tlr1 double mutants using genetic null alleles. If Tlr1 and Side function in a linear pathway, then double mutants would be predicted to exhibit a phenotype that resembles each single mutant phenotype. In tlr1^D427,side^C137/tlr1^K788,side^C137double mutants, however, we observed a strong enhancement of each single mutant phenotype. Double mutant larvae lacked virtually all NMJs on all ventral muscles in all abdominal hemisegments(Fig. 7C,7C', Table 1). Muscles located in lateral or dorsal regions of the body wall also showed an increase in innervation errors, albeit less dramatically(Fig. 7C”, Table 1). The increase in phenotypic strength suggests that tlr1 and side function in parallel pathways, or that their functions converge on shared components further downstream. Interestingly, tlr1,side double mutants were still capable of forming NMJs in lateral and dorsal regions, suggesting that additional gene products are involved in ensuring complete innervation of the musculature.

In this paper, we show that tlr1 is required for motor axons to defasciculate from other motor axons at specific choice points. tlr1encodes a secreted metalloprotease that functions non-cellautonomously,possibly via secretion into the hemolymph. Mutations in tlr1 lead to the permanent formation of NMJs at ectopic sites or to the absence of synaptic terminals. Only a few studies have examined whether embryonic motor axon guidance phenotypes persist into larval stages in Drosophila(Nose et al., 1994; Sink et al., 2001; de Jong et al., 2005). Apart from embryonic lethality, a possible reason is presumably that the detailed analysis of larval motor axon phenotypes is hampered by the availability of suitable methods other than manual dissection. We used transgenic methods for the visualization of the neuromuscular wiring pattern in intact larvae. The fluorescent CD8-GFP-Sh marker specifically highlights the postsynaptic apparatus of each muscle fiber, revealing the structure and location of each NMJ (Zito et al., 1999). As CD8-GFPSh is under the control of the myosin heavy chain promotor, we combined it with the Gal4/UAS-system to simultaneously express the red fluorescent protein dsRed2 in motoneurons. CD8-GFP-Sh by itself indirectly reflects the projection pattern of motor axons by reliably revealing the location of each NMJ. However, the combination of both CD8-GFP-Sh and motoneuron-specific dsRed2 highlights the entire neuromuscular connectivity pattern in living animals.

Supported by these non-invasive methods, we show that tlr1 mutants have defects in motor axon guidance rather than in target recognition or innervation site selection. First, motor nerves show defasciculation defects already during embryogenesis, even before they reach their target muscles. Second, motor axons remain physically attached to each other beyond their destined branch points and until late larval stages. Third, the bypass, stall and misrouting phenotypes are similar to those found in sidestepmutants, a known regulator of motor axon guidance(Sink et al., 2001). As a result, motor axons fail to reach their targets in tlr1 mutants or do so from unusual directions, which leads to anomalous muscle innervation. The formation of NMJs at ectopic locations suggests that the growth direction and entry point of the motor nerve may instruct the final location of an NMJ. Even if the position of an NMJ is predetermined by the muscle fiber, the incoming nerve terminal appears to be able to relocate postsynaptic components.

In tlr1 mutants, all abdominal segments show innervation defects,and all pathways are affected. The intrasegmental phenotype, however, is quite variable, i.e. the innervation pattern of a given hemisegment differs visibly from the adjacent or contralateral hemisegment, indicating that neuromuscular wiring is largely autonomous for a given hemisegment. The variability further implies that Tlr1 may not regulate specific and invariable guidance decisions,but rather plays a general role in defasciculation. Migrating growth cones in tlr1 mutants apparently do not strictly rely on single molecular labels positioned at specific locations. These conclusions support the idea that exact target selection appears to be a rather stochastic process, in which the growth cones integrate attractive and repulsive cues provided by the microenvironment of surrounding cells(Winberg et al., 1998).

Metalloproteases have been implicated in a variety of cellular processes,including cell migration, angiogenesis and metastasis(Sternlicht and Werb, 2001; Yong et al., 2001). Neuronal growth cones migrate through an environment that is rich in different extracellular surfaces and may thus exploit similar molecules and mechanisms as migrating cells. Despite the wealth of data on cell migration, only a few reports have been published implicating metalloproteases in axon outgrowth and guidance (Fambrough et al.,1996; Galko and Tessier-Lavigne, 2000; Hattori et al., 2000; Schimmelpfeng et al., 2001; Webber et al.,2002; McFarlane,2003; Vaillant et al.,2003; VanSaun et al.,2003; Hehr et al.,2005; Jaworski et al.,2006). With regard to motor axons, only ADM-1 (unc-71), a member of the ADAM family in Caenorhabditis elegans, has been shown to regulate pathfinding (Huang et al.,2003). Tlr1 belongs to the astacin family of metalloproteases and is highly related to Drosophila Tolloid. Despite this high degree of conservation, we and others have shown that these two proteins have mutually exclusive functions (Nguyen et al.,1994; Serpe et al.,2005). While Tld cannot functionally replace Tlr1, it is still possible that other metalloproteases with redundant functions assist Tlr1 in defasciculation control, because not all guidance decisions are affected in tlr1 mutants. Our loss- and gain-of-function analysis of related metalloproteases, however, did not support this possibility. In addition, no other metalloprotease has so far been recovered from mutant screens(van Vactor et al., 1993; Kraut et al., 2001; Sink et al., 2001). These observations support the idea that Tlr1 may be a key regulatory member of the metalloprotease family in Drosophila that controls motor axon guidance.

As a secreted metalloprotease, Tlr1 is predicted to function extracellularly, either in the extracellular matrix or in the interstitial fluid. Consistent with this prediction, overexpression of Tlr1 in hemocytes or cells of the fat body could fully rescue the tlr1 mutant phenotype. Neither hemocytes nor fat-storing cells have so far been implicated in axon guidance. Hence, it is unlikely that Tlr1 remained associated with the extracellular matrix of these cells, but probably got released into the hemolymph. The circulating hemolymph then distributed it to where it was required. As endogenous Tlr1 is expressed in developing muscles during the period of axonal pathfinding, it is possibly secreted from there into the hemolymph to either proteolytically activate a repellent on motor nerves to induce defasciculation or to activate an attractant on muscles. The axon guidance receptor Sidestep is expressed on muscles and functions as an attractant for motor axons (Sink et al.,2001). Based on similarities in their loss-of-function phenotypes,Tlr1 could be required to activate the attractive function of Side. The phenotype of tlr1,side double mutants, however, was clearly stronger compared with each single mutant, indicating that they regulate the same biological process but that they function either independently or that the two functions converge further downstream. Tlr1 could therefore regulate the activity of an alternative pathway. Interestingly, M. O`Connor's group suggested that Tlr1 regulates motor axon guidance in part by activating latent TGF-β ligands (M. Serpe and M. O'Connor, unpublished). Although the exact molecular function of Tlr1 is currently not known, the data presented here clearly demonstrate that the evolutionarily conserved metalloprotease Tolloid-related 1 is necessary for motor axon guidance in Drosophila.

We are indebted to Mike O'Connor and Mihaela Serpe for sharing unpublished results. We also would like to thank Corey Goodman, in whose lab the EMS screen was performed, and Christiane Nüsslein-Volhard for support. We are grateful to Corey Goodmann, Mike O'Connor, Toshihiro Kitamoto, Rolf Reuter and Michael Pankratz for providing fly stocks and reagents. We thankfully appreciate the help of Dirk Beuchle for meiotic mapping, and Anne Spang and Matthew Harris for critically reading the manuscript. This project was funded by the Max-Planck Gesellschaft and the Deutsche Forschungsgemeinschaft(SFB446).