Proper axon pathfinding requires that growth cones execute appropriate turns and branching at particular choice points en route to their synaptic targets. Here we demonstrate that the Drosophila metalloprotease tolloid-related (tlr) is required for proper fasciculation/defasciculation of motor axons in the CNS and for normal guidance of many motor axons enroute to their muscle targets. Tlr belongs to a family of developmentally important proteases that process various extracellular matrix components, as well as several TGF-β inhibitory proteins and pro-peptides. We show that Tlr is a circulating enzyme that processes the pro-domains of three Drosophila TGF-β-type ligands, and, in the case of the Activin-like protein Dawdle (Daw), this processing enhances the signaling activity of the ligand in vitro and in vivo. Null mutants of daw, as well as mutations in its receptor babo and its downstream mediator Smad2, all exhibit axon guidance defects that are similar to but less severe than tlr. We suggest that by activating Daw and perhaps other TGF-β ligands, Tlr provides a permissive signal for axon guidance.
Proper wiring of the nervous system during development requires high-fidelity navigation of axons through the extra-cellular environment in both time and space to find their correct synaptic targets. Axon navigation is regulated at several steps by both long- and short-range forces that can either attract or repel the growth cone (Dickson, 2002; Tessier-Lavigne and Goodman, 1996). Guidance cues can include molecules that impart directionality such as chemo attractants and repellents, as well as those that impart permissive or inhibitory effects on general growth cone motility or tracking by affecting adhesive properties. Secreted ligands that appear to act as diffusible guidance cues include members of the Netrin and Slit families that bind to Dcc/Unc-5 and Robo receptors, respectively, whereas Ephrins are linked to the membrane either by glycosylphosphatidylinisotol (GPI) anchors or transmembrane domains and signal through Eph receptors. The fourth and largest class of guidance cues is represented by the semaphorin family which includes both secreted and membrane-bound forms that signal through a diverse array of multimeric receptor complexes that include Plexins, Met, L1, Otk and neuropilins (Huber et al., 2003).
Once pioneer axons have navigated a particular path, other axons follow by adhering to the first axon, forming bundles or fascicules. These secondary axons can depart from the bundle, i.e. de-fasciculate, to initiate their own trajectories at particular choice points. In Drosophila, the cell adhesion molecule (CAM) Fasciclin 2 is a well-studied example of a molecule that is thought to provide cohesion between axons on which it is expressed. In the CNS, Fas2-positive axons form six distinct longitudinal bundles that serve as convenient readouts for the process of fasciculation. In addition, the 40 or so embryonic motoneurons also express Fas2 and exit the CNS at defined intervals to innervate the 30 unique muscle targets in highly reproducible stereotypic patterns. Loss-of-function mutants in Fas2 exhibit detachment of the six longitudinal fascicules, whereas ectopic expression of Fas2 prevents proper de-fasciculation of motoneurons causing the axons to remain abnormally joined and to bypass their choice points in the periphery while enroute to their target muscles (Landmesser et al., 1988; Lin et al., 1994). Likewise, in vertebrates, CAMs play important roles in selecting motor axon fasciculation preferences within the limb plexus (Landmesser et al., 1988).
The fact that overexpression of Fas2 inhibits de-fasciculation suggests that normally there must be mechanisms to decrease adhesiveness at the choice points. In addition, because motor axons still adhere to each other in the periphery of Fas2 mutants, redundant adhesion molecules must exist to maintain peripheral fasciculation. Genetic screens in Drosophila have uncovered several mutations including the beaten path (beat), sidestep (side) and receptor-linked protein tyrosine phosphatases (RPTPs), whose loss-of-function phenotypes are similar to Fas2 gain-of-function phenotypes, suggesting that the products of these loci help drive de-fasciculation (de Jong et al., 2005; Desai et al., 1996; Desai et al., 1997; Krueger et al., 1996; Pipes et al., 2001; Schindelholz et al., 2001; Sink et al., 2001).
Adhesive forces as well as chemo-attractant and repellent cues are likely to be modulated by other secreted factors. There is increasing evidence to suggest that matrix metalloproteases (MMPs) as well members of the ADAMs family of disintegrin-containing metalloproteases are likely candidates for modulating the activity of guidance cues. Mutations in the Drosophila ADAM10 homolog kuzbanian (kuz), for example, exhibit both axon extension and guidance defects where many Fas2-positive axons inappropriately cross the CNS midline (Fambrough et al., 1996). In Xenopus, application of general metalloprotease inhibitors to exposed brain preparations yields severe disruption of the retinal ganglion cell axon projections as they extend through the brain to their targets in the optic tectum. More recent experiments employing selective MMP-inhibitors suggest that axon behavior at specific guidance choice points in this system is likely to involve MMPs (Webber et al., 2002).
How these metalloproteases affect guidance choices is not clear, although one simple model is that they might digest components in the extracellular matrix (ECM) to clear a path for extending axons. However, recent observations suggest that they may play more direct roles by processing components of different guidance pathways. For example, kuz mutations genetically interact with slit and robo mutations suggesting that Kuz might proteolytically modify one of these two components (Schimmelpfeng et al., 2001).
In this report, we demonstrate that mutations in the Drosophila metalloprotease tolloid-related (tlr), also known as tolkin (Finelli et al., 1995; Nguyen et al., 1994), result in fasciculation defects both within the CNS and at choice points in the periphery as motor axons traverse to their target muscles. The choice point defects are similar to those seen in side, beat and lar (leukocyte-antigen-related-like) mutants, suggesting that Tlr functions in conjunction with, or parallel to, these pathways. Tlr is a member of the BMP-1/Tolloid family of Astacin-like metalloproteases. In vertebrates, these proteases process a number of different ECM components (Amano et al., 2000; Kessler et al., 1996; Scott et al., 2000); however, their best-characterized developmental role is to regulate the activity of TGF-β-type ligands. This is accomplished in two conceptually similar ways, either by processing extracellular inhibitors such as Sog or Chordin that normally bind to the ligand and prevent it from binding receptor (Blader et al., 1997; Piccolo et al., 1997; Marques et al., 1997; Serpe et al., 2005), or by processing the N-terminal pro-domains of ligands such as Myostatin and GDF-11 that would otherwise remain bound to the C-terminal ligand domain and prevent it from binding receptor (Ge et al., 2005; Wolfman et al., 2003). We tested if Tlr processes the pro-domains of various Drosophila TGF-β-type ligands and found that in vitro it cleaves Myoglianin (Myo), which is the Drosophila homolog of Myostatin, Activin (Act), and the Activin-like protein Dawdle (Daw). In the case of Daw, we further show that this processing enhances its signaling activity in vitro and we find that daw-null mutants exhibit axon guidance defects similar to, but less severe than, tlr. We also demonstrate that Daw is likely to signal through the canonical Activin pathway to mediate axon guidance because germline mutant clones of babo, the type I receptor for Daw, as well as Smad2 (Smox - Flybase), the main downstream transcriptional mediator of activin-like signaling, also produce axon guidance defects similar to daw mutants. Since Tlr supplied either in motoneurons, muscles or the hemolymph is able to rescue tlr mutants, we suggest that Tlr is likely to provide a permissive signal rather than a spatially instructive cue for guidance, perhaps by activating several TGF-β ligands.
MATERIALS AND METHODS
tlrex[2-41] is a strong hypomorphic allele (Nguyen et al., 1994), whereas tlrE1 is a likely null allele and tldP1∂ is a small deletion that removes both tld and tlr (Finelli et al., 1995). dawex11 and dawex32 were obtained by imprecise excision of a transposomal element (Strain #13221, Japan) inserted in the first intron of the daw gene, 2.5 kb downstream of the transcription start of the Alp23B-RA (Flybase). The molecular lesions were characterized by sequencing: dawex11 lacks 1.8 kb of the daw gene sequence, including the start codon; dawex32 contains remaining fragments of the transposomal element and lacks 1.6 kb from the daw gene, including the start codon. UAS-tld16 and UAS-tld23 lines were obtained from Mike Hoffmann (University of Wisconsin). UAS-tlr-HA has been described previously (Serpe et al., 2005). UAS-cd tlr and pm tlr constructs were generated by insertion of the corresponding cDNAs in the pUAST vector, and germline transformation (Harvard CBRC Fly Core Transgenics Facility).
Other fly stocks collected for this study include: gcm[N7-4]/Cyo (BL-4104), babo52/Cyo (Brummel et al., 1999), babo32/Cyo (BL-5399), punt135-22/TM3 (BL-3100), put62/TM6B (Simin et al., 1998), da-Gal4 (BL-5460), Cg-Gal4 (BL-7011), hml-Gal4 (BL-6396), hs-Gal4 (BL-2077), twist-Gal4 (BL-2517), 24B-Gal4 (BL-1716), Sca-Gal4 (BL-6479). The elav-Gal4, OK6-Gal4, G14-Gal4, 321C-Gal4, 12M-Gal4 and mhc-Gal4 lines were obtained from C. Goodman (UC Berkeley). The phantom-Gal4 is a lab stock.
For generation of V5-His N-terminal ligands, the Tlr signal peptide (M1-A33) (MRRRRGTPLGVSWTNCVLLLATGLLVVLISVHA) followed by the V5 epitope (underlined) and 6xHis (KHRSGKPIPNPLLGLDSTRTGHHHHHHR) were subcloned into the expression vector pAcPA (pBR322 derivative containing an Actin5C promoter and poly-A signal), together with the cDNA ligand sequences, lacking their endogenous signal peptide. Thus, the Tlr signal peptide and the tags were joined with the following amino acids: L16 in Daw, C299 in Act and S101 in Myo. The Tlr processing site mutants in the ligand pro-domains were engineered by site-directed mutagenesis (QuickChange, Stratagene), for which details are available on request.
RNA localization and immunohistochemistry
In situ hybridization was performed with digoxigenin-labeled RNA probes and visualized with alkaline phosphatase precipitates as previously described (Brummel et al., 1999). Immunohistochemistry of Drosophila larvae and embryos was performed as previously described (Patel, 1994), using BP102 at a dilution of 1:10, anti-Fas2 1D4 at 1:5, and anti-csp 6D6 at 1:200 monoclonal sera (Hybridoma Bank), followed by secondary antibodies (1:200) conjugated to Alexa Fluor 488 or biotin, and visualized using the Vectastain ABC Kit (Vector Laboratories). Anti-Tlr antibodies were produced and affinity purified using a C-terminal synthetic peptide (Open Biosystems). Anti-Tlr was used at 1:500 dilution, followed by secondary antibodies (1:200) conjugated to biotin, and visualized using TSA kits (Perkin-Elmer).
Protein production and detection
Drosophila S2R+ and S2* cells were used for producing recombinant proteins as described previously (Shimmi and O'Connor, 2003; Serpe et al., 2005). The cleared supernatant (conditioned medium) was used directly for western blotting or added to cell-based signaling assays. For cleavage assays, expression constructs for both substrate and enzyme were co-transfected. Hemolymph samples were collected by capillary suction from third-instar larvae, mixed on ice with Complete™ protease inhibitor cocktail (Roche), and boiled in SDS loading buffer. Primary antibodies were used at the following dilutions: anti-HA 12CA5 (Roche) 1:2000, anti-V5 (Invitrogen) 1:5000, and anti-β-gal (Promega) 1:1000. Immune complexes were visualized with secondary antibodies IRDye 700 and 800 at 1:5000, followed by scanning with the Odyssey Infrared Imaging System (Li-cor Biosciences).
A cell-based assay for TGF-β signaling was described previously (Shimmi and O'Connor, 2003; Zheng et al., 2003). In brief, S2 cells were transfected with flag-mad or flag-Smad2. Three days after transfection, cells were incubated with ligand for 3 hours, and cell extracts were analyzed by western blotting. The Mad-P and Smad2-P levels were detected with anti-Phospho-Smad1 (gift from C. Heldin, Uppsala, Sweden) and anti-Phospho-Smad2 (Cell Signaling) at 1:2000, and quantified relative to the Flag signal detected with anti-Flag M2 antibody (Sigma) at 1:2000, using IRDye secondary antibodies for simultaneous detection on the Odyssey Infrared Imaging System.
Tlr mutants exhibit fasciculation/defasciculation and innervation defects
Mutants in the metalloprotease tlr cause lethality during larval and pupal stages of development (Finelli et al., 1995; Nguyen et al., 1994); however, the cause of the lethality has not been determined. Since a small percentage of larvae (about 15%) die as soon as they hatch, the need for Tlr may start during embryogenesis. Beginning at stage 13, Tlr protein expression is found in the muscles, a subset of cells in the central nervous system that include many glia (data not shown) and the corpus allatum portion of the ring gland (Fig. 1), matching the previously described mRNA pattern (Finelli et al., 1995; Nguyen et al., 1994). The distinct pattern of expression of tlr in the CNS and muscles, together with the observation that rare tlr mutant escapers exhibit impaired movement, prompted us to examine nervous system development in tlr mutants. To look for global defects, we initially stained all CNS axons in mutant embryos of the strong allelic combination tlrex[2-41]/tlrE1 with the monoclonal antibody mAB BP102. This analysis did not reveal any gross abnormalities (Fig. 2A) in formation of longitudinal or commissural tracts. We next used the Fas2 monoclonal antibody mAb 1D4 which, at stages 16 and 17, highlights motor axon tracts in the periphery and six longitudinal bundles within the CNS (Van Vactor et al., 1993). In tlrex[2-41]/tlrE1 mutants, the Fas2-positive longitudinal bundles were wavy and irregular and the outer bundle was discontinuous or missing (Fig. 2B-D). This phenotype was of variable penetrance because we found embryos that had mild defects (Fig. 2C) as well as embryos with severe, interaxonal adhesion defects (Fig. 2B,D).
In abdominal segments A2-A7, motor axons exit the CNS within the intersegmental nerve (ISN) and segmental nerve (SN) roots; these then split into five pathways that innervate 30 muscle fibers. The ISN develops fairly early and reaches the terminus region near muscle 1 at stage 16 of embryogenesis (Van Vactor et al., 1993). In tlrex[2-41]/tlrE1 mutants, ISN growth appeared delayed: 85% of ISNs (140 hemisegments examined) reached their final destination by late stage 16, whereas 15% of ISNs were still at the secondary branch point, around muscle 2 (Fig. 2F). By stage 17, the ISN reached its terminal position in most hemisegments of the tlrex[2-41]/tlrE1, but the terminal arbors were thin or bifurcated (Fig. 2G).
The SNa has a bifurcated morphology. The posterior branch of SNa innervates muscles 5 and 8, and the anterior branch innervates muscles 21-24. To reach muscle 24, the anterior branch makes a characteristic turn at stage 16. In tlrex[2-41]/tlrE1 mutant animals, SNa did not turn, but instead stalled or produced random branches at this point (Fig. 2H,I, quantified in Table 1).
The SNb branch innervates the ventral muscles 7, 6, 13 and 12, and contains the axons of RP1, 3, 4 and 5. The development of the SNb involves two key sets of contacts, the first at muscle 28 and 14, where SNb axons leave the common pathway and enter the ventral muscle field, and the second near muscle 30, where stage 16 SNb growth cones shift their trajectory to extend along a more interior muscle layer. At early stage 17, the SNb forms a linear synaptic branch at the muscle 6/7 cleft, a `blobby' synapse at the proximal edge of muscle 13 (referred as the 13/30 synapse), and a linear synapse at 12/13. In tlrex[2-41]/tlrE1 mutant animals, the appearance of the SNb proximal synapse (6/7) was normal (black arrow in Fig. 2L); however, the SNb bundles stalled at the 13/30 `blob' with an occasional thin bundle exiting and extending towards the 12/13 cleft. The thinned SNb appeared to reach the target muscles at random locations and produced very short synapses. The overall appearance was that of stalled growth cones with frail axons perhaps trying to achieve some sort of innervation at the 12/13 synapse (Fig. 2L, `stalled' phenotype in Table 1). Such unsuccessful attempts to compensate for the lack of proper 12/13 innervation were observed in 46% of the tlrex[2-41]/tlrex[2- 41] hemisegments, in 48% of the tlrex[2-41]/tlrE1 hemisegments, and in 56% of the tldP1/tlrE1 hemisegments (Table 1). Since tlrP1 is a deletion comprising both tld and tlr genes, and tlrE1 contains a stop codon within the tlr ORF, the slightly lower penetrance of defects in the case of tlrex[2-41]/tlrex[2-41] animals could be due to some residual tlr muscle expression. Such minimal expression might be the result of using an alternative exon, upstream of the breakpoints of the tlrex[2-41] deficiency, that is computationally predicted by Flybase, (http://flybase.bio.indiana.edu/) although it was not recovered in any of the extant tlr cDNAs.
The axon guidance defects of tlr mutants persist until larval stages
Several screens searching for axon guidance regulators have led to the characterization of mutants that exhibit subtle late embryonic nervous system defects. Many of these defects are compensated for later in development, in part due to the plasticity of the guidance response (Dickson, 2002). By contrast, we found that the tlr mutant defects persist until larval stages, when the animals begin to die. As in embryos, staining of the tlrex[2-41]/tlrE1 larvae with either anti-Fas2 or anti-csp revealed relatively normal NMJs on muscles 6 and 7, but abnormal innervation in the cleft between muscles 12 and 13 (Fig. 2M-S). In tlr mutants, the innervation on muscle 12 was entirely absent, reduced (Fig. 2N,O,R) or misrouted (Fig. 2S). In comparison with wild-type larvae (61 A4 hemisegments analyzed), all tlrex[2- 41]/tlrex[2-41] and tldP1/tlrE1 hemisegments we examined exhibited such defects. For example, among 99 tlrex[2-41]/tlrex[2-41] hemisegments analyzed, we found 86 with defective innervation in the cleft between muscles 12 and 13, whereas 13 hemisegments were entirely missing the NMJ on muscle 12. Similarly, from 119 tldP1/tlrE1 hemisegments analyzed, 110 had defective innervation at the 12-13 muscle cleft and 9 hemisegments had none. Similar results have been recently noted by Meyer and Aberle (Meyer and Aberle, 2006).
Rescue of tlr mutants requires embryonic expression of enzymatically active Tlr
Since endogenous Tlr appeared to be expressed in multiple tissues (i.e. muscle, nervous system and ring gland), we carried out rescue experiments to address the tissue requirements for this protein. Previously, we have shown that ubiquitous expression of a C-terminally HA-tagged tlr transgene rescued the viability and fertility of tlr mutants, and completely restored the posterior crossvein formation (Serpe et al., 2005). Expression of transgenes encoding catalytically dead Tlr (cdtlr) or a pro-domain processing site mutant (pmtlr) did not rescue the viability of the tlrex[2-41]/tlrE1 mutants (Table 2). This was not the result of insufficient protein expression because we detected robust protein induction for all variants tested (not shown).
We next examined the tissue and time requirements in more detail using a selection of tissue-specific drivers. Overexpression of the same tlr transgene in the nervous system with a pan-neural driver (elav) or with glial cell drivers (12M and 321C) rescued the viability and fertility of tlr mutants (Table 2) as well as the axon guidance defects of tlrex[2-41]/tlrE1 mutant embryos. tlr mutant control animals carrying only the UAS-tlr(HA) transgene did not express Tlr-HA (see western analysis below, Fig. 4, lane 5) and did not exhibit any improvement in viability or phenotype. Very low levels of tlr expression, such as that provided using the Sca1 driver at 20°C, were insufficient to restore full viability, whereas increasing the level of tlr expression (Sca1-Gal4>UAS-tlr2 at 25°C) lead to the production of more protein and fully restored the viability and fertility of mutant flies.
Intriguingly, overexpression of the tlr transgene in the muscles alone also rescued mutant animals to full viability and fertility (Table 2). However, not all muscle drivers were equally effective. We observed full rescue with the 24B and G14 drivers, but not with mhc- Gal4. Expression with the mhc driver is initially observed during the first instar larval stage (DiAntonio et al., 1999), whereas expression of 24B and G14 begins during embryogenesis at around stage 13. We infer from these results that Tlr is required for proper axon guidance during late embryogenesis, consistent with its expression in developing muscles and glial cells at this stage.
Tlr is a circulatory enzyme secreted into the hemolymph
The fact that Tlr was able to rescue mutant animals when supplied from either the muscle or the nerve suggests that its precise spatial expression pattern may not be important for function. Since Tlr is a secreted protein, it is possible that it could gain access to its substrate from the hemolymph. In fact, in addition to expression in muscle and glia, tlr is also heavily expressed in the corpus allatum of the ring gland (Fig. 1E), a known secretory tissue. To determine if Tlr could rescue mutant animals when expressed exclusively in secretory or circulating cells, we used a series of ring gland or hemocyte drivers (Cg, hml, phantom) (Table 2) and found full rescue of tlrex[2-41]/tlrE1 lethality and axon guidance defects in all cases. Furthermore, as shown in Fig. 3A (compare lanes 3-5 and 2), we found that hemolymph samples from wild-type animals, but not from tlr mutants, contained the processed activated Tlr protein. Moreover, we detected significant levels of HA-tagged Tlr in hemolymph samples collected from animals in which a UAS-tlr-HA transgene was overexpressed in various tissues, including glial cells and muscle. The levels of Tlr species we detected were not due to contamination of hemolymph samples with cells from other tissues, as control overexpressed proteins were detectable only in larval carcasses (i.e. β-galactosidase, Fig. 3B). These results support our hypothesis that Tlr is secreted and circulates in the hemolymph and need not be supplied locally by either the muscle or glial cells in order to promote proper axon guidance.
Identification of new Tlr substrates
Enzymes of the BMP-1/Tolloid family have been shown to process a number of substrates including the BMP inhibitors Sog and Chordin, components of the ECM, and the pro-domains of several TGF-β-type factors including Myostatin, which is a key regulator of muscle mass, and GDF-11, which is a negative feedback inhibitory signal in neurogenesis (Ge et al., 2005; Wolfman et al., 2003). In both cases, the mature ligand remains associated with the pro-domain after secretion in a non-covalent latent complex. Additional processing of the pro-domain by Tld-like enzymes releases the ligand from the complex enabling it to signal.
Since TGF-β-type ligands have been implicated in some aspects of axon guidance in both C. elegans and the mouse (Butler and Dodd, 2003; Colavita et al., 1998), we asked whether the role of Tlr in Drosophila motoneuron axon guidance might involve processing of a latent TGF-β-type ligand. Drosophila has four non-BMP TGF-β-type ligands comprising Act, Activin-like protein (now called Dawdle, see below), Myo and Mav that are, as yet, poorly characterized (Haerry and O'Connor, 2002; Lo and Frasch, 1999; Nguyen et al., 2000). A scan of the pro-domain sequences of these ligands revealed at least one putative BMP-1/Tolloid cleavage site in each pro-peptide that resembled the cleavage sites characterized in Myostatin and GDF11 (alignment in Fig. 4A) and other sites in Sog and Cv-2 (M.S. and M.B.O., unpublished).
To test directly if any of these pro-peptides is a substrate for Drosophila Tolloid proteins, we introduced a V5/His tag at the N-terminus of the pro-domains and produced the tagged proteins in Drosophila S2 cells. As shown in Fig. 4, Act, Myo and Daw are secreted and processed at the predicted multibasic, subtilisin-like maturation site just prior to the ligand domain, yielding a pro-peptide of the appropriate size. This processing is independent of Tlr and occurs for all TGF-β-type ligands. For example, in the case of Myo, we detect a pro-domain of ∼50 kD (subtilisin generated) in the conditioned media, which corresponds to the calculated 48 kD myoglianin pro-domain (Fig. 4C, lane 1). When the ligand was produced in cells co-transfected with a Tlr construct, the pro-domain was further processed at an additional site, producing a ∼30 kD tagged fragment (Fig. 4C, lane 2). The identity of the cleavage site within the pro-domain of each ligand was confirmed by alanine substitution mutagenesis of four residues on each side of the putative cleavage site (Fig. 4A,F). The Tld enzyme also cleaves these sites (Fig. 4F), consistent with our observation that Tld can rescue Tlr axon guidance defects when expressed at a low level (see Table 3).
Processing of the Daw pro-domain by Tlr/Tld activates Daw in a cell-based signaling assay
The stability of the Daw, Myo and Act pro-domains in the absence of Tlr and their rapid disappearance upon induction of protease expression is similar to the behavior of the vertebrate Myostatin and GDF11 pro-domains in the presence of Tld proteases. In these cases, cleavage of the pro-domains results in activation of latent ligand complexes (Ge et al., 2005; Wolfman et al., 2003) leading to enhanced signaling. We asked if cleavage of the Act, Daw or Myo pro-domains by Drosophila Tolloids had any effect on the signaling ability of the ligands. We tested this hypothesis using a tissue culture-based signaling assay (Shimmi and O'Connor, 2003; Zheng et al., 2003). We analyzed both the BMP pathway that signals through Mad, and the Activin pathway that signals through Smad2, by transfecting S2 cells with either tagged Mad or Smad2 constructs and presenting them with various ligands. For example, the control ligand Dpp signals through the BMP pathway: cells treated with 10 ng Dpp had a 5-fold increase in the level of phosphorylated Mad (Mad-P) as compared with mock treated cells (Fig. 5A, lanes 1-3). Under our experimental conditions, none of the other ligands tested appeared to elicit a strong BMP-like signal (Fig. 5A, left panels). Act signaled weakly through Smad2 with or without Tld (Fig. 5A, compare P-Smad2/Smad2-Flag ratios in lanes 13 and 14). Cells treated with Daw, in the absence of Tlr or Tld, increased their level of Smad2-P 2-fold compared with the control. More importantly, cells treated with Daw+Tlr or Tld exhibited a much stronger increase in the relative level of Smad2-P, suggesting that Tld proteins enhanced Daw signaling properties (Fig. 5A,C, lanes 15, 16). Neither Myo (Fig. 5A, lanes 17, 18) nor Mav (not shown) produced any significant signal under our experimental conditions, either with or without Tlr. The signaling advantage of adding Tld or Tlr was completely lost for DawTm, a mutant form of Daw in which the Tld/Tlr cleavage site was destroyed (Fig. 5, compare C to E, quantified in D,F).
To determine if Tlr could enhance Daw signaling in vivo, we examined gain-of-function phenotypes of daw with or without additional tlr in various tissues. Indeed, when daw was overexpressed in hemocytes using the Cg-Gal4 driver we observed less than 10% pupal lethality (in more than 200 progeny); Cg-Gal4>UAS-tlr1 animals had no detectable phenotype. By contrast, when both tlr and daw were overexpressed, the pupal lethality of Cg-Gal4>UAS-tlr1,UAS-daw animals exceeded 50% (180 animals examined). In addition, transheterozygous combinations of tlr and daw showed enhanced axon guidance defects as compared with the single heterozyous animals, consistent with a role for Tlr in activating Daw in vivo (K. Arora, personal communication). Taken together, our data show that Drosophila Tolloids cleave various TGF-β-ligand pro-domains and, at least in the case of Daw, this cleavage activates the ligand, most likely by releasing it from a latent complex with its inhibitory pro-domain.
daw and tlr expression patterns overlap at several developmental stages
To determine if daw is a biologically relevant substrate for the secreted Tlr, we analyzed the daw expression pattern. daw was expressed ubiquitously in the early embryo (Fig. 6A), but in the later stages of embryogenesis the expression was enriched in mesoderm (Fig. 6B,C), muscle (Fig. 6D), and in a subset of cells in the CNS that are, based on position, likely to be glial (Fig. 6E-G). In the third instar larval stage, daw expression was seen in many glia within the brain lobes and ventral ganglion (Fig. 6H). The muscle expression was very strong at later developmental stages, as seen in the ventral muscle field of a third-instar filet (Fig. 6I). We also observed daw expression in all larval imaginal discs (antennal and eye disc shown in Fig. 6J), and in the cells of the tracheal system (not shown).
daw-null mutants exhibit axon guidance defects similar to tlr mutants
The observed pattern of daw expression, especially its expression in the muscles where tlr is abundant, is consistent with the possibility that Daw could be cleaved and activated in vivo by Tlr. If Daw is a physiologically relevant substrate for Tlr, then one might expect that the daw mutant phenotype should exhibit some overlap with the tlr mutant phenotype. To examine this, we generated daw-null mutants by excising a P-element located just upstream of the daw gene (Fig. 6K). We obtained two deletions, dawex11 and dawex32, in which parts of the Daw coding sequence, including the transcriptional and the translational start sites, were deleted. Both mutations produce lethality in the larval-pupal stages and the lethality was rescued by ubiquitous expression of a daw transgene, indicating that no other essential gene was removed by these excisions (data not shown).
Examination of the nervous system of stage-17 daw-null mutant embryos revealed very mild fasciculation defects in the CNS with thinning and breaks of the exterior (third) bundle of pioneer axon fascicles (Fig. 7A,B). Stage-17 daw mutant embryos exhibited various motoneuron misprojections including ISN delays, stalling and misrouting of the SNa, particularly no turning towards muscle 24, and stalling of SNb at the 13/30 synapse accompanied by defective 12/13 innervation (Fig. 7A,C). Overall, the axon guidance defects of daw mutants (quantified in Fig. 7H) were similar in nature, but weaker in intensity and penetrance than the tlr mutant defects. In addition, unlike tlr mutants in which defects persist into the third instar stage, daw mutant third-instar larvae exhibited relatively normal innervation at the 12/13 muscle cleft (135 dawex11/dawex11 hemisegments and 75 dawex32/dawex32 hemisegments scored) (Fig. 7D). The delayed nature of the innervation defects lead us and others (Parker et al., 2006) to rename this locus from alp to dawdle. These results suggest that the inability to properly activate the daw ligand may contribute to the tlr phenotype.
Other TGF-β pathway components exhibit axon guidance defects
To determine if the daw axon guidance defects arise because of a loss of signaling through a canonical TGF-β pathway, we examined several additional pathway components for axon guidance defects. Since our cell culture signaling assay revealed that Daw signals through Smad2, which is phosphorylated by the type I receptor encoded by the babo locus (Brummel et al., 1999; Zheng et al., 2003), we examined babo and Smad2 mutant embryos for axon guidance defects. In both cases, zygotic loss-of-function mutants showed no increase in defects (data not shown). However, when the maternal contribution of each was also removed by making germline clones, we observed a very similar distribution and penetration of defects in stage 17 embryos as was found for daw mutants (Fig. 7E,F,H,I quantified in J).
In Drosophila, two different type II receptors, Punt (Put) and Wishful thinking (Wit), mediate signaling by TGF-β ligands. Since wit is not expressed in S2 insect cells (Marques et al., 2002; McCabe et al., 2003), but these cells are able to transduce a signal to Smad2 (Fig. 5), and as wit mutants have a normal pattern of muscle innervation at the end of embryogenesis (Aberle et al., 2002), we focused our analysis on put mutants. Like babo and Smad2, put is maternally supplied to the embryo (Letsou et al., 1995); its maternal loss results in severely ventralized embryos with a highly disorganized CNS. However, certain combinations of put alleles are temperature-sensitive (Simin et al., 1998), allowing us to eliminate put activity by temperature shift. At permissive temperatures of 18-20°C we did not observe any significant embryonic axon guidance defects in put135-22/put62 animals (not shown). At temperatures above 25°C, put135-22/put62 exhibited gross head involution and dorsal closure defects making it difficult to score for axon guidance defects. However, analysis of put135-22/put62 embryos that were developed at 23-25°C revealed axon guidance defects similar to those seen in tlr, daw, babo and Smad2 mutant embryos (Fig. 7G, quantified in J). We also found no increase in the penetrance or severity of defects in put, wit double-mutant embryos, suggesting that Wit is not redundant with Put for this activity. Together these data suggest that Daw signals through a canonical TGF-β pathway involving Babo, Put and Smad2 to elicit a transcriptional output crucial for fulfilling proper axon guidance in late Drosophila embryos.
In this study we analyzed Drosophila Tlr, an enzyme of the BMP-1/Tolloid family, and its role in nervous system development. We find that mutant embryos exhibit numerous defects in motoneuron axon guidance beginning at embryonic stage 16-17 and persisting into larval stages. Furthermore, we demonstrate that one set of substrates for Tlr are the pro-domains of several TGF-β-type ligands. In the case of Daw, processing of its pro-domain by Tlr leads to enhanced signaling abilities in a cell culture assay. Since daw mutants also exhibit axon guidance errors, albeit less severe and persistent than those of tlr mutants, these observations suggest (but do not prove) that Tlr processing of inhibitory pro-domains enhances TGF-β signaling to regulate nerve branching and innervation, perhaps by altering adhesiveness at particular choice points and target cues on muscles.
Tlr and Daw provide permissive cue(s)
Previous work characterizing the attraction/repression and adhesive/non-adhesion forces acting during axon guidance revealed remarkable plasticity in the growth cone responses to modulation by both intrinsic and extrinsic factors (Dickson, 2002). Specifying an axon's trajectory is therefore not just a simple matter of selecting the appropriate set of guidance receptors and delivering them to the growth cone. The growth cone must also be able to modulate its responsiveness en route. Tlr and Daw appear to be required for a process that allows proper modulation of this en route responsiveness of the growing axons and perhaps also muscle site target selection.
How might Tlr and Daw regulate axon guidance and target site selection? An attractive model is that Tlr processing of Daw and other TGF-βs in muscles might provide a chemoattractant signal that collaborates with other muscle-derived attractants such as Side to guide motor axons to their appropriate innervation sites, similar to the way in which the morphogen sonic hedgehog has been shown to collaborate with netrin 1 in mediating midline axon guidance to the floor plate in mice (Charron et al., 2003). Although Tlr and Daw both show localized expression in muscle, our observation that tlr mutants can be rescued by expression of a tlr transgene in a wide variety of tissues, and that the activated form of the protein is a normal constituent of the hemolymph, suggests that it may not provide a directional cue for axon guidance. Likewise, daw mutants can be rescued by transgene expression from multiple tissues (Parker et al., 2006). These results are more consistent with Tlr and Daw providing a permissive signal rather then providing a directional cue(s).
At least one response to a permissive cue is likely to be mediated through a canonical TGF-β pathway because we show that maternal loss of both babo, the Drosophila type I receptor for Daw, and Smad2, the primary transcriptional transducer of Daw signaling, both produce axon guidance defects similar to daw mutants. Daw signaling is likely to control the transcription of other gene products that directly regulate axon fasciculation/guidance, perhaps by modulating production of an attractant on muscle or a repellent on motor nerves, and thereby to integrate this new pathway with previously identified mediators of attractive and repulsive forces. It is interesting to note in this regard that double mutants of side and tlr show strongly enhanced phenotypes as compared with each single mutant (Meyer and Aberle, 2006), suggesting that these two products act in parallel, as opposed to in a linear pathway.
One last point with respect to Daw function is that it is likely to regulate other aspects of Drosophila development in addition to motor nerve axon guidance. This follows from the fact that mutations in daw lead to lethality in the pupal stage, yet by this time motor axon guidance defects are largely corrected. Therefore, the lethality is likely to result from defects in some other process. This might reflect a more general role of daw in guiding axons other than those from motoneurons to their targets, or maybe Daw and Tlr regulate other functional aspects of motoneurons or glial cells that remain to be identified.
Possible redundancy in the Tlr/Daw pathway
Although the daw and tlr loss-of-function axon guidance phenotypes are similar in both penetrance and the spectrum of defects that they exhibit at embryonic stage 17, there are significant differences between the two. In particular, many tlr defects persist into the larval third instar stage, whereas daw embryonic mutant phenotypes are corrected by this time. In addition, tlr mutant embryos have significant fasciculation defects in the Fas2-positive longitudinal bundles within the ventral ganglia that are not seen in daw mutants.
One possible explanation for these differences is that other TGF-β molecules might act redundantly with Daw. Among the uncharacterized ligands, Myo in particular is intriguing in this regard as its expression overlaps daw in muscle and glia cells (Lo and Frasch, 1999). mav expression also overlaps that of both daw and myo by virtue of being broadly expressed in most embryonic tissues (Nguyen et al., 2000). BMP ligands might also be involved because microarray analysis has shown that the Activin and BMP pathways share several transcriptional targets in brain tissue (Yang et al., 2004). In one simple scenario, daw mutants might be corrected because of redundancy with mav, myo or one of the BMP ligands, whereas tlr mutants exhibit more severe defects because lack of Tlr might affect the activation of multiple ligands. At present, mutants that disrupt mav and myo are not available to assess possible functional overlap of their products with Daw and each other.
Are TGF-βs the only relevant Tlr substrate for proper axon guidance?
Although we believe that Tlr regulates motor axon guidance in part by processing latent complexes of Daw and other TGF-β ligands, it is quite possible that TGF-β-type molecules are not the only Tlr substrates relevant to this process. Metalloproteases may interact directly with the guidance signaling pathway via either ligand modification or processing of their receptors. For example, metalloproteases regulate cell-surface expression of DCC or robo receptors (Galko and Tessier-Lavigne, 2000; Schimmelpfeng et al., 2001), and ADAM family members have been implicated in terminating the interaction between the ephrins and their receptors (Hattori et al., 2000). Alternatively, metalloprotease may cleave components of the ECM thereby clearing a path for axon extension through the extracellular environment (McFarlane, 2003).
TGF-β signaling and axon guidance in other systems
The activation of latent TGF-β-type ligands for regulating axon guidance might be a conserved and ancient mechanism. In C. elegans, the unc-129 locus codes for a TGF-β ligand that is 33% identical to human BMP7 and mediates motor axon attraction to the dorsal midline (Colavita and Culotti, 1998; Colavita et al., 1998). In vertebrates, BMPs are roof-plate secreted chemorepellents for commissural axons (Bovolenta, 2005; Charron and Tessier-Lavigne, 2005). Whether either of these examples also utilizes a Tld-like enzyme in the processing of a latent complex of the BMP ligands attached to their pro-domains remains to be determined.
We thank Hermann Aberle and Kavita Arora for communication of results prior to publication and Mike Hoffmann and Corey Goodman for fly stocks. We thank David Zhitomersky for technical assistance. We also acknowledge Kai Zinn's website http://www.its.caltech.edu/~zinnlab/motoraxons/fma%20home%20page.html for help in identifying branching defects. This manuscript was improved by the thoughtful comments of Guillermo Marques, MaryJane Shimell and Scott Selleck. M.B.O. is an Investigator of the Howard Hughes Medical Institute.
- Accepted October 23, 2006.
- © 2006.