In an effort to understand the mechanism of action of Highwire, we have identified Highwire-binding proteins from the Drosophila nervous system. Tandem affinity purification (TAP) is a powerful two-step purification scheme for isolating a protein of interest and its binding partners 20 . We generated transgenic flies expressing a TAP-tagged Highwire and demonstrated that this fusion protein is functional since it rescues the highwire loss-of-function phenotypes 11 . We also generated a transgene encoding TAP-tagged HighwireΔRING (HiwΔRING), which carries two Cys-to-Ser mutations in the conserved RING domain, and demonstrated that HiwΔRING acts as a potent dominant negative when expressed in neurons 11 . Both highwire transgenes were tested in our TAP system. We found that HiwΔRING shows a much stronger enrichment in the final complex than Highwire, possibly because of a higher stability of HiwΔRING (data not shown). Thus, we expressed TAP-HiwΔRING in the Drosophila central nervous system (CNS), purified full-length HiwΔRING and associated proteins (see Materials and methods for details), and analyzed the purified complex by two independent approaches. First, the eluted proteins were resolved on SDS-PAGE and the sypro-stained protein bands were excised, digested with trypsin, and subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS). The searching program MASCOT was used to identify the protein species. Peptide mass fingerprinting and MS/MS peptide sequencing (data not shown) identified the bait protein Highwire as the 600 kDa band in Figure 1a. In addition, CG4643 was identified as the 27 kDa band indicated in Figure 1a based on the recovery of two unique peptides (SDLLASVSTYK, NVYIKPNGFTLHR) that are present in CG4643. Second, LC-MS/MS was used to directly analyze the purified protein complex without an intermediate SDS-PAGE separation. Again, in addition to Highwire peptides, five unique peptides from CG4643 were recovered (ESLKSDLLASVSTYK, NPVAQSTDAAR, YQVGER, IRVILDCEDNTLSFEK, NYEFLGVAFR).

CG4643 is predicted to encode a 27 kDa protein containing an F-box domain, a WD-40 repeat, and a spla and ryanodine receptor (SPRY) domain. Its closest homolog in C. elegans is FSN-1, which is composed of the same functional domains and which binds to the RPM-1, the C. elegans homolog of Highwire. The two proteins share 41.9% homology, and define a small sub-family of F-box proteins with SPRY domains of which there is only a single member in worms, flies, mice (F-box protein 45 (GenBank:AAH26799)), and humans (hCG1734196 (GenBank:EAW53643)) with high homology across species (Figure 1b). Based on the sequence similarity of CG4643 and FSN-1, we have renamed CG4643 DFsn, for Drosophila Fsn.

The fact that TAP-HiwΔRING can pull down DFsn indicates that an intact RING domain is not required for the Highwire-DFsn interaction. To test whether wild-type Highwire also interacts with DFsn, we performed the following co-immunoprecipitation experiments. Since we have been unable to generate an antibody that recognizes endogenous DFsn, we generated transgenic flies that express His-Myc-tagged DFsn. As described below, the transgene encoding this is functional since it can rescue a DFsn mutant. We expressed His-Myc-tagged DFsn throughout the nervous system and performed immunoprecipitations from third instar larvae brains. Immunoprecipitations (IPs) and western blot analysis was performed with either affinity-purified polyclonal antibodies raised against Highwire 11 or a monoclonal antibody to Myc. As shown in Figure 1c, Myc-DFsn is present when endogenous Highwire is immunoprecipitated. Conversely, endogenous Highwire is detected in the IP fraction when Myc-DFsn is immunoprecipitated by the anti-Myc antibody. In addition, neither endogenous Highwire nor Myc-DFsn is detected when a control protein, Drosophila vesicular Glutamate transporter (DVGLUT), is immunoprecipitated from Myc-DFsn expressing larval brains (data not shown). These data demonstrate that endogenous Highwire physically interacts with DFsn. While eukaryotic genomes can express hundreds of F-box proteins and a similar number of RING domain ubiquitin ligases 18 21 , these data and the previous work of Liao et al. 17 demonstrate that the physical interaction between Highwire/RPM-1 and Fsn is evolutionarily conserved.

Since our efforts to generate an antibody against DFsn were unsuccessful, we generated a functional green fluorescent protein (GFP)-tagged DFsn transgene. When the transgene is expressed by a pan-neuronal Gal4 driver, GFP-DFsn localizes to the NMJ (Figure 1e,h). Double staining with markers for synaptic vesicles (Figure 1e–g) and periactive zones (Figure 1h–j) demonstrates that GFP-DFsn is not restricted to any of these compartments. GFP-DFsn is also detectable in the axon and neuronal cell body (data not shown), although the staining of GFP-DFsn is much stronger in synaptic boutons than in the preterminal axon (Figure 1e, arrow), suggesting a synaptic enrichment. GFP-Highwire encoded by a pan-neuronally expressed transgene has a very similar localization pattern 11 . These data are consistent with the presence of a Highwire-DFsn complex at the NMJ.

C. elegans FSN-1 is required for normal presynaptic differentiation 17 , and its phenotype partially overlaps with the rpm-1 phenotype. Having shown that the Highwire-DFsn interaction is conserved in Drosophila, we wished to test whether DFsn is also required for normal synapse development. We obtained a chromosomal deficiency, Df(2R)7872, that removes DFsn and flanking genes, as well as a transposable-element-insertion mutant in the DFsn locus from the Drosophila stock center. The insert, P(f06595), is a piggyBac P-element inserted in the second exon of the DFsn, 121 base-pairs upstream of the start codon (Figure 2a,b). RT-PCR data demonstrate that transcription of DFsn is disrupted in homozygous f06595 and hemizygous P(f06595)/Df(2R)7872 mutants (Figure 2c). Both the homozygous and hemizygous adult DFsn mutants are viable and male-sterile. Morphological analysis of the NMJ from P(f06595)/P(f06595) and P(f06595)/Df(2R)7872 larvae reveals dramatic synaptic terminal overgrowth (Figure 3). We quantified this phenotype at the NMJ of a single identified motoneuron, MN4-1b 22 , which forms the type Ib innervation onto muscle 4, but all phenotypes are qualitatively similar at every type I NMJ, including the muscle 6/7 NMJ (Additional file 1). We quantified three metrics of synaptic structure: bouton number, number of branch points and synaptic span (the extent of the muscle covered by the synapse). The P(f06595)/Df(2R)7872 mutant shows an almost four-fold increase in both bouton number and number of branch points when compared with the wild-type NMJ (Figure 3). The synaptic span of the P(f06595)/Df(2R)7872 mutant is also increased from 18% in wild type to 30% (Figure 3). This synaptic terminal overgrowth is quantitatively and qualitatively very similar to that observed in a highwire null mutant 11 . The homozygous P(f06595)/P(f06595) mutant also has a large increase in bouton number, branch point number, and synaptic span, but the phenotype is less severe than in the P(f06595)/Df(2R)7872 mutant. Since the mutant phenotype of P(f06595) is enhanced when placed over a deletion, P(f06595) is behaving as a genetic hypomorph.

Highwire regulates synaptic terminal growth presynaptically. If DFsn functions with Highwire to restrain this growth, then it should also be required in the presynaptic motoneuron. We generated transgenic flies capable of expressing DFsn in a tissue-specific manner (UAS-GFP-DFsn). We expressed GFP-DFsn in neurons with elav-Gal4, a pan-neural Gal4 driver, in the DFsn mutant background [P(f06595)/Df7872; UAS-GFP-DFsn/elav Gal4]. To assess rescue, we again quantified bouton number, synaptic branching, and synaptic span at the MN4-Ib NMJ onto muscle 4. All of these aspects of synaptic terminal overgrowth are rescued by the presynaptic expression of DFsn (Figure 3). The presence of the elav-Gal4 transgene without the UAS-GFP-DFsn transgene fails to rescue the DFsn mutant (p > 0.4 for number of boutons and branch points). Furthermore, synaptic terminal overgrowth of the DFsn mutant is also fully rescued when GFP-DFsn is expressed selectively in motoneurons (driven by D42; Additional file 1), indicating DFsn functions cell autonomously in the motoneuron to restrain synaptic terminal growth. Although we only quantified NMJs onto muscle 4, all type I NMJs, including those on muscle 6/7, show qualitatively similar rescue (Additional file 1). The transgene encoding the His-Myc-tagged DFsn also rescued the mutant phenotype (data not shown), so the fusion proteins used for immunoprecipitation and immunolocalization above are both functional. These data demonstrate that DFsn is responsible for the phenotype of the P(f06595) mutant, which we have renamed DFsn^f06595. These data demonstrate that DFsn is required in the presynaptic motoneuron to restrain synaptic terminal growth at the Drosophila NMJ.

Highwire is not only required to restrain synaptic terminal growth, but also to promote synaptic function. Both of these activities require the ligase function of Highwire, but Highwire regulates distinct molecular pathways to affect synapse morphology and function. Since F-box proteins can confer specificity to ubiquitin ligases, we wondered whether DFsn is also required to regulate synaptic release. We analyzed synaptic transmission in DFsn mutants by performing intracellular recordings from muscle 6 (segments A3 and A4) of third instar larvae and measured both spontaneous and evoked neurotransmitter release (Figure 4). Quantal size (mEJP amplitude), the postsynaptic response to the spontaneous fusion of a single synaptic vesicle, is unchanged in DFsn mutants (Figure 4b). However, evoked release is greatly reduced – DFsn mutant larvae have a 70% reduction (p < 0.001) in the excitatory junctional potential (EJP) compared with wild type (Figure 4c). The reduction in evoked release is due to a 66% decrease (p < 0.001) in quantal content, the number of vesicles released by a nerve following an action potential, in the mutant (Figure 4d). Presynaptic expression of the wild-type UAS-DFsn transgene rescues the deficit in both evoked neurotransmitter release and quantal content (Figure 4), indicating DFsn functions presynaptically to regulate synaptic function. The decrease in evoked synaptic release observed in the DFsn mutant is very similar to that seen in highwire null mutants, although the hypomorphic DFsn mutant does not show the same modest decrease in quantal size observed in the hiw null mutant 11 .

The analysis above demonstrates that DFsn and highwire loss-of-function mutants have very similar morphological and physiological phenotypes. Since they physically interact, these similar phenotypes suggest that DFsn and Highwire function together in a complex. To test this hypothesis, we analyzed genetic interactions between DFsn and highwire. If DFsn and Highwire function together in a complex, then we predict that hypomorphic (partial loss-of-function) alleles of DFsn should enhance the phenotype of a hypomorphic highwire mutant. We generated double mutants between the hypomorphic DFsn^f06595 mutant and the weak highwire allele hiw^ND51, which carries an amino acid change at residue 2054 (G2054R) 3 . While hiw^ND51 and DFsn^f06595 mutants each show mild synaptic terminal overgrowth compared to wild type, the double mutant shows dramatically overgrown synapses (Figure 5). In addition to the large increase in bouton number (Figure 5b), there is also enhancement of synaptic branching, with the number of branch points being: wild type, 2.2 ± 0.3 (n = 27); hiw^ND51, 4.3 ± 0.3 (n = 27); DFsn^f06595, 6.5 ± 0.4 (n = 34); hiw^ND51, DFsn^f06595, 10.8 ± 0.6 (n = 28). The enhancement of a weak hiw allele by the DFsn hypomorph is consistent with the two proteins acting in a complex; however, it is also consistent with the two proteins acting in parallel pathways. To distinguish between these possibilities, we generated double mutants between the DFsn^f06595 mutant and hiw^ND9, a genetic null for highwire harboring a premature stop. If DFsn and Highwire function together, then loss of DFsn should not enhance a highwire null, since it is already without function. However, if the proteins work in parallel pathways, then the DFsn phenotype would be additive and it should enhance either a hypomorphic or null highwire allele. We found that DFsn^f06595 does not enhance hiw^ND9. Instead, hiw^ND9 and the hiw^ND9, DFsn^f06595 double mutant have a very similar synaptic terminal overgrowth phenotype (Figure 5). The double mutant also fails to enhance the number of synaptic branch points, with the number of branch points being: hiw^ND9, 9.6 ± 0.8 (n = 22); hiw^ND9, DFsn^f06595, 7.4 ± 0.4 (n = 39). Synaptic terminal growth is not saturated in a strong highwire mutant 13 , so the absence of enhancement by the addition of a DFsn mutant is strong genetic evidence that DFsn and Highwire collaborate to restrain synaptic terminal growth.

How do Highwire and DFsn regulate synaptic terminal growth? In both fly and worm, Highwire/RPM-1 down-regulates the protein levels of the Wallenda/DLK kinase. Increased levels of Wallenda/DLK over-activate downstream MAP kinase cascades and lead to altered synapse development. Mutations in wallenda/DLK suppress the highwire/rpm-1 phenotypes. However, in worm, FSN-1 regulates the levels of the tyrosine kinase ALK and ALK mutants suppress the fsn-1 phenotype, leading to the suggestion that FSN-1 regulates synapse development by down-regulating ALK 17 . If Highwire and DFsn collaborate as part of a ubiquitin ligase complex, then we predict they should both regulate the same target protein. We have found that ALK is postsynaptic at the Drosophila NMJ (AD, unpublished observation), so we have focused our attention on Wallenda.

We first tested if DFsn, like Highwire, regulates the levels of Wallenda in the nervous system. Wallenda protein is undetectable in the nerve cord of wild-type larvae, but is abundant in the synaptic neuropil of ventral nerve cords from DFsn mutants (Figure 6a,b). This staining is due to Wallenda protein since it is absent in DFsn, wnd double mutant animals. Expression of a DFsn transgene rescues the increase in Wallenda levels in the DFsn mutant (Figure 6a,b). Hence, DFsn, like Highwire, is required to down-regulate the levels of the Wallenda/DLK kinase. While Highwire and DFsn are each necessary to down-regulate Wallenda/DLK, their simultaneous over-expression in neurons is not sufficient to reduce the level of Wallenda protein (data not shown), indicating that the Hiw-DFsn ubiquitin ligase complex likely includes other components.

Having demonstrated that DFsn regulates Wallenda levels, we tested whether Wallenda is required for synaptic terminal overgrowth in the DFsn mutant. As shown in Figure 6c–f, loss of wallenda suppresses the synaptic morphology defects of the DFsn mutant, restoring bouton number, branching number, and synaptic span to near wild-type levels. Therefore, Wallenda is necessary for synaptic terminal overgrowth in both the DFsn and highwire mutants. While wnd mutants suppress the morphological defects of the DFsn mutant, they do not suppress the electrophysiological phenotype. Quantal content is not significantly different between DFsn single mutants and DFsn, wnd double mutants, while both show impaired release compared to wild type (p < 0.001; Figure 4d). Therefore, DFsn, like Highwire, must regulate at least two pathways. In both mutants, the down-regulation of Wallenda is necessary to restrain synaptic terminal growth, while a genetically separable pathway is involved in promoting synaptic release.

Transgenic flies carrying the UAS-NTAP-HiwΔRING or UAS-TAP transgene driven by BG380 Gal4 driver were expanded. Fresh adult flies were deep frozen and then vortexed. Fly heads were then collected by passing the flies through sieves that separate the heads from other body parts. The purification procedure was performed at 4°C. Fly heads were homogenized with homogenization buffer (50 mM Tris-HCl pH 7.5, 125 mM NaCl, 5% Glycerol, 0.7% NP40, 1.5 mM MgCl₂, 25 mM NaF, 1 mM EDTA, 0.2 mM DTT, 1 mM Na₃VO₄, 0.05 mM MG1115, 1 mM PMSF, Protease inhibitor cocktail (Roche Applied Science, Indianapolis, IN, USA). Two sequential centrifuge steps (17,000 rpm for 20 minutes and 45,000 rpm for 40 minutes) were used to clean the homogenate before loading the first column. The following purification steps were performed according to the Séraphin lab TAP protocol 25 . Final elutions from the second column were divided into two portions. The first portion was loaded to a SDS-PAGE gel (Criterion XT 4–12%) followed by Sypro Ruby staining. Individual bands were then excised, digested with trypsin, and subjected to LC-MS/MS. The second portion was precipitated, digested with trypsin and then subjected to LC-MS/MS. The searching program MASCOT (Matrix Science Inc., Boston, MA, USA) was used to identify the protein species. As a control, the TAP-only complex was analyzed by SDS-PAGE and no bands higher than 15 kDa were detected by Sypro Ruby staining.

Flies were maintained at 25°C on normal food. The following strains were used in this study: Canton S (wild type), P-f06595 (Exelixis Collection at Harvard Medical School), Df(2R)7872 (Bloomington Stock Centre), elav Gal4 (neuron specific) 26 , BG380 Gal4 (neuron specific) 27 , D42 Gal4 (motor neuron specific) 28 , ELAV-GeneSwitch 29 , hiw^ND51 and hiw^ND9 3 12 , two wallenda alleles: wnd¹ and wnd³ 15 .

Generation of TAP-tagged full-length highwire transgenes UAS-NTAP-Hiw and UAS-NTAP-HiwΔRING was described previously 11 using the TAP vector 20 . A UAS-TAP transgene was also generated for control purposes. A full-length cDNA clone (Clone ID: LD47425) for DFsn (CG4643) was obtained from Research Genetics (Huntsville, AL, USA). The entire coding region of DFsn was subcloned into pUAST-EGFP or pUAST-HM vector 30 to generate UAS-GFP-DFsn and UAS-His-Myc-DFsn transgenes. Transformant lines were generated using standard techniques.

For immunoprecipitations, about 100 larval brains for each sample were collected by manual dissection. Larval brains were then homogenized in 0.5 ml lysis buffer (Tris-HCl 50 mM pH 7.4; NaCl 150 mM; EDTA 1 mM; NaF 2 mM; NP-40 0.2%; NaDOC (Na deoxycholate) 0.2% and protease inhibitors). Clean lysate was then incubated with the rabbit α-Highwire antibody 11 (HIW2b, 1:80), the mouse α-Myc antibody (9E10, 1:100), or the rabbit α-DVGLUT antibody 31 (1:80) for 2 hours. Protein A beads (25 ul; Roche) were added to the lysate. After overnight incubation, the beads were washed four times with lysis buffer before elution. HIW2b (1:250) and the mouse α-Myc antibody (1:1,000) were then used to probe the endogenous Highwire and the neuronally expressed Myc-DFsn (driven by BG380 Gal4), respectively, in the IP complex on western blots.

Endogenous Wallenda was detected using the WndA1 antisera as described 15 . For all other antibodies, larvae were fixed in Bouin's solution (a 1:5:15 ratio of acetic acid/formalin/picric acid) for 8 minutes. The rabbit α-DVGLUT antibody was described previously 31 , and used at 1:5,000 dilution. The α-fasciclin II (FasII; 1D4) antibody developed by Corey Goodman (Renovis, San Francisco, CA, USA) was obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by The University of Iowa, Department of Biological Sciences, Iowa City, IA, USA, and used at 1:5. To examine the localization of GFP-DFsn, GFP-DFsn was expressed in neuronal tissues by driving the UAS-GFP-DFsn transgene with ELAV-GeneSwitch in the presence of RU486. The mouse monoclonal α-GFP antibody (3E6; Invitrogen, Carlsbad, CA, USA) was then used at 1: 800 dilution to detect the GFP-DFsn protein.

Larvae were imaged on a Nikon (Tokyo, Japan) C1 confocal microscope. All images were acquired in third instar wild-type, mutant, and rescued larvae that had been simultaneously stained. The gain was chosen as the maximum gain that did not saturate the signal for each sample. A complete Z-stack was acquired for each NMJ or ventral nerve cord and rendered as a maximum projection. All quantifications were done at muscle 4 in segments A2–A4. The muscle sizes for all genotypes were similar. The number of type I boutons and the number of branch points of type I boutons were counted following α-DVGLUT and α-Fas II staining. A synaptic branch was defined as an arborization containing at least 2 type I boutons. Synaptic span was calculated by dividing the length of type I synapses by the length of muscle. The length of synapses and muscle were measured by an ocular micrometer (Nikon). Statistical analysis was performed and graphs were generated in Origin 7.0 (Origin Lab, Northampton, MA, USA). Each sample is compared to other samples in the group using ANOVA, with n for each condition described in the figure legends. All histograms are shown as mean ± standard error of the mean.

Intracellular electrophysiological recordings were performed as described previously 32 . Both spontaneous mEJPs and evoked EJPs were recorded in HL-3 saline containing 0.47 mM Ca2+ 33 . At least 70 consecutive events were measured per cell using MiniAnal (Synaptosoft, Decatur, GA, USA) and averaged to determine the mean mEJP. Events with a slow rise time were rejected as artifacts from neighboring electrically coupled muscle cells. Evoked EJPs were recorded by stimulating the cut end of the segmental nerve with a 1 ms depolarizing pulse as described by Broadie 34 . Quantal content was estimated by dividing the mean EJP by the mean mEJP without correcting for non-linear summation, which would only have increased the observed differences.