A role for PS integrins in morphological growth and synaptic function at the postembryonic neuromuscular junction of Drosophila

Integrins have long been known to participate in embryonic neurogenesis. Integrin expression in the nervous system correlates with axonal outgrowth, target recognition and sprouting of synaptic arbors (Chiba and Keshishian, 1996; Martin et al., 1996). In recent years, integrins have also been implicated in postembryonic neuronal function. For example, integrins have been shown to participate in the mechanical modulation of neurotransmitter release in mature frog motor nerve terminals (Chen and Grinnell, 1995). Moreover, at least 10 different integrin subunits are expressed in the adult vertebrate brain (Pinkstaff et al., 1999). Functionally, the pharmacological blocking of integrins in rat hippocampal slices interferes with the stabilization of long-term potentiation (LTP), thought to form a mechanistic basis for learning and memory (Staubli et al., 1990). Genetic evidence suggests that a similar mechanism is essential for behavior regulation in Drosophila since mutations in an α integrin, Volado, cause defects in memory formation and retention (Grotewiel et al., 1998). Thus, integrins appear to play multiple roles in synaptic development and perhaps in the maintained developmental potential, or synaptic plasticity, believed to underlie behavioral modulation.

A great deal has been learned about the role of integrins in vertebrate neuronal development and function by studying in vitro preparations. For example, early in synaptogenesis, integrin-cytoskeleton interactions in growth cones are involved in stabilizing intercellular contacts (Schmidt et al., 1995). Following target recognition, integrins interact with laminin and agrin in neurally induced acetylcholine receptor clustering in the postsynaptic membrane (Burkin et al., 1998). However, in vivo confirmation of these mechanisms using genetic analyses has proven difficult, and genetic studies in mice often contradict in vitro results using pharmacological perturbation of integrin function (reviewed in Fassler et al., 1996; Hynes, 1996). These discrepancies might be due to the expression of multiple, potentially redundant, integrin subunits during vertebrate neurogenesis (Martin et al., 1996). However, removal of integrin subunits in less redundant invertebrate systems such as C. elegans and Drosophila has so far also revealed only minor defects in pathfinding, fasciculation and embryonic synaptogenesis (Baum and Garriga, 1997; Huang et al., 1998; Prokop et al., 1998). The question remains, therefore, whether integrins play only subtle, minor roles during synaptic development or whether we have so far failed to appropriately assess the in vivo significance of this class of proteins.

An attractive model to further the genetic analysis of integrin synaptic function is the Drosophila neuromuscular junction (NMJ). This system is a powerful genetic model to analyze synaptic growth and transmission properties regulated through electrical activity, adhesion and signaling molecules (Chiba and Keshishian, 1996). Importantly, only five integrin subunits have been identified in Drosophila, two β (βPS and βν) and three α subunits (αPS1, αPS2, αVolado/scab/αPS3), and numerous mutant alleles of these integrin subunits are available (Brown, 1993; Grotewiel et al., 1998; Yee and Hynes, 1993). This relatively simple integrin complement contrasts favorably with the complex situation in vertebrates. We report here that the βPS, αPS1 and αPS2 integrin subunits become detectable at Drosophila NMJs during first and second larval instars, correlating with the onset of extensive morphological and physiological remodeling at the NMJ (Keshishian et al., 1996). Analyses of larvae carrying regulatory mutations affecting βPS expression or transgenic βPS constructs reveal that integrins impact morphological growth, synaptic specificity and, in some terminals, synaptic function, at the postembryonic terminal. These results provide the first genetic evidence that integrins are involved in postembryonic synaptic development and suggest that integrins may well play an important role in the modulatory mechanisms underlying synaptic plasticity.

Drosophila melanogaster stocks were reared at 25°C unless otherwise noted. Stocks were raised in well-yeasted, uncrowded vials on standard medium (Ashburner, 1989). Oregon-R flies were used as control animals. Three myospheroid (mys) mutant stocks were used: sn v mys^b9 (D. Brower; see Flybase, 1998), a hypomorphic mutant that shows temperature-sensitive viability (D. Brower, personal communication), cm ct⁶ mys^ts1, also a weak hypomorphic mutant (Bunch et al., 1992); mys^XG43, an embryonic lethal genetic null (Bunch et al., 1992). For heat-shock overexpression, we used transgenic lines carrying P[HSPSβ] (Brabant and Brower, 1993). The construct was continually induced from hatching to late 3^rd instar by a 50 minute heat-shock pulse at 37°C every 7 hours in a cycling incubator. Neural overexpression was induced using a P[UASβPS] construct (M. Brabant; constructed by M. Diamond) induced by a variety of previously described and newly isolated GAL4 constructs (Flybase, 1998). See Lindsley and Zimm (1992) for descriptions of the marker mutations used.

Wandering 3^rd instar larvae were dissected and stained as previously reported (Broadie and Bate, 1993). To examine NMJ morphology, preparations were probed with a mouse monoclonal anti-cysteine string protein (CSP) antibody at 1:200 (Zinsmaier et al., 1990; K. Zinsmaier). Staining was visualized using a Vectastain ABC Elite kit with NiCl₂ or CoCl₂ enhancement and specimens mounted in Araldite (Broadie and Bate, 1993). Images were captured with an Optronics video camera and a Scion LG3 capture card. For clarity, different focal planes were occasionally combined in one picture using Adobe Photoshop 4.0.

In confocal preparations, either rabbit anti-DLG (discs-large, 1:1000; Woods and Bryant, 1991; D. Woods) or the mouse anti-CSP was used to mark synaptic arbors. Integrin expression was examined with the following antibodies: mouse monoclonal anti-βPS integrin antibody CF6G11 (1:500) or the CF6G11 ascites (1:300, both; Brower et al., 1984; N. Brown); mouse monoclonal anti-αPS1 antibody DK.1A4 (1:200) or mouse ascites anti-αPS1 antibody (1:500, both; Brower et al., 1984; D. Brower and E. Mart í n-Blanco, respectively); rat polyclonal anti-αPS2, PS2hc2 (1:5; Bogaert et al., 1987) and rabbit rabPS21 (1:1000; Bloor and Brown, 1998; N. Brown); rabbit polyclonal βν 84E (1:600; Yee and Hynes, 1993; R. Hynes). All anti-integrin antibodies were independently tested for their well-characterized embryonic staining patterns prior to probing the larval preparations (Bloor and Brown, 1998; Brower et al., 1984). Fluorescent secondaries used were Alexa-red-594- and Alexa-green-488-conjugated goat anti-mouse and goat anti-rabbit (Molecular Probes). Confocal microscopy was performed on a Bio-Rad MRC-600. NIH Image software was used to quantify staining intensity. Statistical significance of all measurements was determined with the Mann-Whitney-U test using InStat software except for the significance of synaptic specificity, which was determined with Fishers Exact test using InStat software.

Flat dissected larvae (Broadie and Bate, 1993) were fixed for 10-60 minutes with 2.5% glutaraldehyde in 0.2 M phosphate buffer (pH 7.2). Fixation for immunoelectron microscopy: (a) 1 hour 4% paraformaldehyde in phosphate-buffered saline (PBS; 0.02 M phosphate buffer, 0.1 M NaCl, pH 7), wash in PBS-TX (0.1% Triton-X in PBS), normal antibody procedure (see above; sometimes with NiCl₂ enhancement); (b) 10 minutes 1% glutaraldehyde in 0.2 M phosphate buffer (pH 7-7.2), antibody procedure as above but without detergent (instead 0.2 M phosphate buffer). En bloc treatment, embedding, sectioning and microscopy as described previously (Prokop et al., 1998).

Micrographs showing boutons with active zones on muscle fibers 6/7 were taken randomly for statistical measurements. For size and area measurements, outlines of the particular structures were drawn onto transparencies, scanned into a computer, and measured using the histogram function in Adobe Photoshop. Vesicle numbers per bouton area were measured excluding the areas containing mitochondria or axolemma. For SSR analysis, SSR areas (excluding the bouton) were measured, or the SSR area was divided by the bouton area in each sectional plane. Statistical significance of all data was determined by Mann-Whitney-U-test.

Nerve-evoked excitatory junctional currents (EJCs) were recorded at 18°C from muscle 6 or muscle 4 (segment A3) of wandering 3^rd instar larvae using standard two-electrode voltage-clamp techniques as described previously (Rohrbough et al., 1999). Recordings were made in a modified standard Drosophila saline (Jan and Jan, 1976) composed of: 128 mM NaCl, 2 mM KCl, 1.5 mM CaCl₂, 4 mM MgCl₂, 5 mM trehalose, 70 mM sucrose, 5 mM Hepes; pH adjusted to 7.1 with NaOH. EJCs were evoked by stimulating the severed segmental nerve (0.5 milliseconds, 0.2 Hz) with a suction electrode filled with extracellular saline. The stimulation voltage was increased until transmission failures were abruptly replaced by all-or-none responses. In all but a few mutant cases (see Results), increasing the voltage slightly revealed a second, high-threshold input which abruptly increased the peak synaptic current by an additional 2-fold or more. The voltage was raised ∼50% over this level to safely excite both inputs. At the peak of the EJC, the intracellularly recorded voltage deviated by no more than 2-3 mV from the command potential (−60 mV). Current signals were filtered at 0.5 kHz, digitized to disk and subsequently analyzed using PClamp6 acquisition hardware and software (Axon Instruments). EJC mean peak amplitude was determined for each animal by averaging 10-20 consecutive responses. Spontaneous miniature EJCs (mEJCs) were recorded continuously in muscle 4, in 1.5 mM Ca²⁺ saline containing 0.1 µg/ml TTX, and analyzed as reported previously (Rohrbough et al., 1999).

RESULTS

PS integrins are expressed at the postembryonic neuromuscular junction

At the Drosophila larval NMJ, motor axons contact specific target muscles and form characteristic arborizations at defined locations on the muscle surface (Johansen et al., 1989). These arborizations consist of branches forming numerous varicosities (boutons) falling into two primary classes: glutamatergic type I boutons (2-5 µm diameter) innervate all muscles and mediate fast, excitatory transmission; octopaminergic type II boutons (∼1 µm diameter) are found on a subset of muscles, and appear to have a primarily modulatory function (Johansen et al., 1989; Keshishian et al., 1996). Growth at all arbors is activity dependent (Budnik et al., 1990) and a number of molecules involved in regulating this morphological plasticity have been identified (Budnik et al., 1996).

We became intrigued about the possible function of integrins in morphological synaptic plasticity in this system due to the isolation of Volado, an α integrin subunit required for Drosophila memory (Grotewiel et al., 1998). Additionally, expression of one subunit, αPS2, at the NMJ had been briefly reported (Fernandes et al., 1996). To assay synaptic integrin expression, we probed larval preparations with specific antibodies against αPS1, αPS2, βPS and βν integrin subunits (Fig. 1A-C). The αPS1, αPS2 and βPS subunits were clearly detectable; only the βν subunit did not appear to be present in the NMJ (not shown). PS integrin expression was restricted to type I boutons in a halo surrounding an unlabeled center, suggesting primarily postsynaptic localization (Fig. 1A-C). In double-labeled confocal sections, αPS and βPS integrin staining overlapped anti-DLG (discs-large), a primarily postsynaptic marker, although the PS integrin expression appeared to be less uniform than the anti-DLG staining (Figs 2A, 3A,D). αPS1 integrin staining was consistently fainter and less extensive than αPS2 and βPS integrin staining (Figs 1C, 3D). In addition to the intense βPS localization at the NMJ, fainter staining was observed throughout the muscle. We interpret this to be sarcomeric expression of βPS dimerized with αPS2, as previously reported (Volk et al., 1990).

In order to localize PS integrins expression more precisely, mature larval NMJs were probed with anti-PS integrin antibodies and embedded for ultrastructural analysis (Fig. 1D-J). Antibody staining against αPS1, αPS2 and βPS integrin subunits, but not the control staining (Fig. 1D, F), all showed expression throughout the subsynaptic reticulum (SSR), a system of postsynaptic foldings of the muscle membrane that surrounds the presynaptic boutons in type I arbors (Budnik et al., 1996; Fig. 1D-J; Table 1). These results clearly confirm that all three PS integrins occur at least postsynaptically. In additional experiments, pre-embedding-anti-βPS staining could be achieved in the absence of detergent, thus preserving the cell membranes. In these specimens, unopposed presynaptic membranes often revealed staining deposits suggesting that at least βPS integrin is expressed in the presynaptic membrane in addition to the postsynaptic SSR (white arrowhead in Fig. 1E′; Table 1).

Developmentally, all three PS integrins become detectable at the NMJ in late 1^st or early 2^nd instar larvae, although others have reported earlier expression in the CNS (Hoang and Chiba, 1998). We were unable to detect expression in embryos or hatching 1^st instar larvae in the peripheral nerves or NMJ, although the expected expression at the muscle attachment sites was clearly visible for all three subunits (not shown). Thus, PS integrin expression becomes concentrated at NMJs only after they have been differentiated and stabilized during embryogenesis. This late onset of expression coincides with the appearance of the SSR, the period when different morphological classes of NMJs first appear and the terminals first become subject to measurable activity-dependent morphological plasticity (Budnik, 1996; Keshishian et al., 1996).

We next addressed potential functions of PS integrins at the Drosophila NMJ, focusing primarily on the βPS subunit. Because null alleles of myospheroid (mys), the gene encoding the βPS integrin subunit, cause embryonic lethality due to muscle detachment (Leptin et al., 1989), we characterized two viable, hypomorphic alleles, mys^b9 and mys^ts1. These alleles encode and express structurally normal proteins as assayed on western blots, but show specific regulatory defects (Bunch et al., 1992; D. Brower, personal communication) and mys^b9 shows incomplete late larval lethality at 29°C. Experiments were carried out at both 29°C and 25°C, but temperature did not appear to affect synaptic expression of the βPS protein or any of the observed NMJ phenotypes (see below).

Homozygous animals from both mutant strains were tested for alterations in expression of the βPS integrin subunit at the NMJ in wandering 3^rd instar larvae (Table 1). βPS expression at both 29°C and 25°C was intense and reliably observed in all type I NMJs of normal larvae (Fig. 2A, n=97 terminals, 15 animals), but detected in just 50% of NMJs in mys^b9 animals (n=106, 17 animals, Fig. 2B) and in only 20% of NMJs in mys^ts1 larvae (n=136 terminals, 18 animals; Fig. 2C). Detectable expression ranged from nearly wild type to nearly background in both mutants. In both alleles, reduction in synaptic βPS expression followed a dorsoventral gradient with dorsal NMJs consistently less likely to stain above background (ventral muscles 6/7, ∼70%; dorsolateral muscle 4, ∼30%). However, the mechanistic cause for the reduction of synaptic protein levels in these alleles might be different, since extrasynaptic muscle expression appeared comparable to wild type in mys^ts1 mutant larvae, but was reproducibly increased in mys^b9 (Fig. 2B′). We conclude, therefore, that mys^b9 causes an inappropriate distribution of the βPS protein, moderately reducing the frequency of detectable synaptic βPS expression (reduced 50%) while consistently upregulating extrasynaptic muscle expression, whereas mys^ts1 strongly and specifically reduces the frequency of detectable βPS expression at the NMJ terminal (reduced 80%) with no detectable corresponding increase elsewhere in the muscle.

It is generally accepted that α and β integrins must heterodimerize to be transported to the plasma membrane (Cheresh and Spiro, 1987). Since the βPS subunit is known to form heterodimers with the αPS1 and αPS2 subunits in other Drosophila tissues (Bogaert et al., 1987; Leptin et al., 1989), we tested whether reduced levels of βPS affect expression of αPS integrins at the NMJ (Table 1). Anti-αPS1 and anti-αPS2 immunostaining was detectable at all mutant NMJs, but was moderately reduced at mys^b9 mutant NMJs and strongly reduced at mys^ts1 mutant NMJs (Fig. 3). This observation strongly suggests that βPS dimerizes with both αPS1 and αPS2 at the NMJ. However, in both mys mutant alleles, αPS expression persists at a level higher than predicted if βPS was the sole β subunit. Therefore, we suggest that a novel β subunit may be present at the NMJ (Fig. 3; see Discussion).

The postembryonic onset and strictly localized synaptic expression of the PS integrins suggests that these proteins may play a specialized role in synaptic structural or functional modulation. We first asked whether the mys mutant alleles altered the NMJ synaptic architecture. We assayed the morphology of NMJs in abdominal segment A2 innervating muscle 4 (a simple terminal, i.e. having, on average, only two nerve inputs and low orders of branching), muscle 6 (a more complex terminal also having two inputs, but many more boutons and much more complex branching) and muscle 12 (the most complex terminal, i.e. having, on average, three or more inputs, and five or more primary branches with increased branching complexity). We quantified NMJ morphology on muscle 4 and muscle 12 by counting branches and boutons in A2. Due to the complexity of branching, and the compaction of all mutant muscle 6 NMJs, we counted boutons but did not attempt to quantify branching on muscle 6. A bouton was defined as a terminal varicosity expressing synaptic markers (e.g. vesicle-associated CSP) and an arboreal branch was defined as a terminal process consisting of at least two boutons. Bouton quantification was restricted to type I boutons, since type II terminals do not appear perturbed under these mutant conditions.

Using the synaptic vesicle marker anti-CSP to visualize NMJs (Fig. 4), we observed that arboreal branches of all homozygous mys^b9 mutant NMJs were dramatically shortened and synaptic boutons were both smaller and more tightly spaced than in wild-type animals, greatly reducing the synaptic terminal area (Fig. 4). In addition, mutant terminals contained abundant clusters of ‘mini-boutons’; very small (<1 µm) boutons that are only infrequently observed in wild-type terminals (Fig. 4B-B″, arrowheads in E). Labeling with the SSR marker anti-DLG revealed the mini-boutons to be surrounded by SSR as is characteristic of type I boutons (Fig. 4E, arrowhead), and we therefore counted them as such; however, this inclusion undoubtedly artificially inflated the bouton quantification (see below), leading to an underestimate of the phenotype severity. In addition to these aberrant mini-boutons, even the ‘mature’ (i.e. >1 µm) type I boutons are significantly smaller in mys^b9 than wild type (compare Fig. 4A ‴with B ‴).

On both muscle 4 and muscle 12, mys^b9 mutant terminals also showed a significant decrease in all orders of terminal branches, resulting in a very highly significant decrease in the total number of arboreal branches (Fig. 5A,B; n≥17 hemisegments per strain). As well, type I bouton numbers are very significantly reduced at NMJs on muscles 4, 6 and 12 (Fig. 5C). Thus, the undergrowth phenotype in mys^b9 mutant larvae results in a compaction and simplification of the synaptic architecture, and reduction in size and number of type I boutons. This reduction is not related to a decrease in muscle size, as all mutant genotypes showed no significant difference from wild type in muscle area. The average area of muscle 4 and muscle 12, respectively, were 1526±7.36 µm² and 846±5.30 µm² in OR controls, 1466±5.52 µm² and 796±3.88 µm² in mys^b9 and 1564±7.88 µm² and 844±4.26 µm² in mys^ts1. These values do not significantly differ between genotypes (muscle 4, OR/mys^b9, P=0.0944, OR/mys^ts1 P=0.7057; muscle 12, OR/mys^b9, P=0.5871, OR/mys^ts1 P=0.8746). To confirm that these phenotypes were specific to the mys mutation, we also analyzed mys^b9/mys^XG43 heterozygous animals (mys^XG43 is a protein null; Fig. 5). NMJs in these animals also have small, tightly clustered boutons surrounded by numerous mini-boutons. Both bouton numbers and numbers of branches tend to increase slightly towards wild type; the significance of this increase is discussed below (see Discussion).

Homozygous mys^ts1 mutant NMJs were significantly larger and more complex than normal, a phenotype opposite of mys^b9 mutant NMJs (compare in Fig. 4). This striking overgrowth phenotype was similar in animals raised at 25°C and 29°C, and more obvious at the normally simpler muscle 4 terminal than at the more complex muscle 12 terminal. Although the number of primary NMJ branches (1°) on muscle 4 was normal, all higher order branching (>2°) was very significantly increased resulting in a highly significant increase in total branching (Figs 4C, 5A). This trend is less obvious on muscle 12, perhaps because this NMJ is already more complex in the wild type (Fig. 5B). As well, type I boutons in mys^ts1 mutant NMJs were significantly increased on all scored muscles, in close agreement with the branching defects (Fig. 5C). Thus, the overgrowth phenotype in mys^ts1 mutant larvae results in an elaboration of the synaptic architecture and an increase in type I boutons. However, when mys^ts1/mys^XG43 heterozygous animals raised at 29°C were analyzed, although the branching phenotype was entirely consistent, these animals showed no significant increase in bouton numbers, indicating that the differentiation of boutons in these enlarged arbors may be a multigenic trait, influenced by the genetic background. Alternatively, changing the protein expression by placing a complex regulatory mutant over a null may change expression in unexpected ways, significantly affecting the observed phenotype (see Discussion).

In order to confirm the specificity of the phenotypes observed in both mutant strains, preparations were identified by genotype in a blind study. Out of 34 preparations, 32 were correctly identified. As complex morphological phenotypes tend to vary around a mean overlapping wild type, this indicates a highly reliable identification of the mutant phenotypes and confirms the qualitative severity of these NMJ growth defects in distinct mys regulatory mutants.

Our results with the mys mutant alleles suggest that the level and/or distribution of PS integrin in the synaptic terminal is important for controlling NMJ sprouting and bouton formation. We tested this hypothesis by overexpressing the mys gene in wild-type animals and assaying effects on NMJ morphology. We used larvae carrying a heat-inducible hsp70-mys construct that has been shown to express βPS protein at a high enough level to rescue lethal mutant phenotypes (Brabant and Brower, 1993). Using the same induction-protocol as Brabant and colleagues (37°C heat shock every 7 hours from hatching to scoring) and quantifying βPS expression at the NMJ from confocal images using the NIH Image program, we observed a significant increase in βPS protein expression at the NMJ and throughout the muscle consistent with the observed phenotype.

Since the heat-inducible hsp70 promoter is not cell or tissue specific, the βPS levels increased ubiquitously in all larval tissues. This overexpression caused larval and pupal lethality. Larvae that survived to wandering 3^rd instar often undergo premature histolysis, complicating analyses of NMJ morphology. However, at a qualitative level, NMJs of heat-shocked larvae were clearly structurally disordered, with reduced branches and smaller, abnormally shaped boutons (Fig. 4D). The simple muscle 4 terminals showed no significant quantifiable alterations in the number of synaptic branches or type I boutons (Fig. 6A,C), perhaps owing to the starting simplicity of this junction. In contrast, the complex muscle 12 terminals showed a significant decrease in the number of total synaptic branches, resulting primarily from a loss of higher order branches (≥3°; Fig. 6B) and a highly significant decrease in the number of type I boutons (Fig. 6C). Wild-type controls subjected to the same protocol were qualitatively and quantitatively normal (not shown). Thus, ubiquitous overexpression of βPS causes an undergrowth phenotype similar to that observed at mys^b9 mutant NMJs, as well as abnormal bouton morphology. We note that both mys^b9 and the hs-construct cause an increase in βPS expression in extrasynaptic regions of the muscle.

To clarify the results observed with ubiquitous βPS overexpression, we next induced overexpression with GAL4 insertions with expression patterns specific to either nerve or muscle tissue (Fig. 6). When βPS expression was induced in all neural tissue throughout development using elavGAL4 (Flybase, 1998), branching and bouton numbers were again significantly reduced in all NMJs. However, a very intriguing observation was that bouton morphology was normal and there was no increase in mini-boutons. A similar result was seen using two muscle-specific GAL4s, GAL4-α and GAL4-F11 (Fig. 6), which begin muscle expression during early 2^nd instar. In both cases, bouton number and branching defects at all NMJs were consistent with the results observed with hsβPS – significantly decreased at muscles 6 and 12, with branching, but not bouton numbers reduced at muscle 4 (Fig. 6). Additionally, muscle-specific βPS overexpression caused a moderate increase in mini-boutons (not shown). Thus, overexpression of βPS in neural and/or muscle tissue results in reductions in both branching and bouton differentiation at NMJs on muscles 6 and 12, and in a loss of branching, but not boutons, on muscle 4. Early muscle βPS overexpression results in an increase in mini-boutons.

In summary, our results show that regulated levels of PS integrin expression in synaptic and extrasynaptic regions of the NMJ synaptic arbor are required as promoting and inhibiting growth regulators of morphological growth and bouton differentiation. The expression of βPS in the membranes of both synaptic partners plays a key role in the regulation of growth. One important conclusion is that elevated extrasynaptic βPS may serve to inhibit growth and arrest nascent boutons in a developmentally immature state.

As errors in pathfinding have been reported in animals lacking α integrin subunits (Hoang and Chiba, 1998), 3^rd instar larvae with altered integrin expression at the NMJ were examined for aberrations that might be the result of either errors in pathfinding or in synapse remodeling or maturation (Gorczyca et al., 1999). For this analysis, we chose to quantitatively assay the complex, muti-innervated muscle 12 NMJ. Three types of gross patterning abnormalities, designated as backbranched, pulled away or multiple insertions, were noted and quantified in a blind study (Fig. 7). The first of these phenotypes, backbranched, consists of at least one arboreal branch leaving muscle 12 and branching back onto muscle 13 from a site away from the point of nerve entry onto muscle 12 (Fig. 7D). This architecture was very rare in normal animals (<5%) but abnormally occurred at high levels in all mys mutant animals; the error is most common in mys^b9 and mys^b9/mys^XG43 (both >30%; Fig. 7E). The second phenotype, pulled away, is characterized by boutons suspended in the space between muscles 12/13, and defasciculation, or branching, beginning before the nerve has actually reached the point of insertion on the muscle (Fig. 7B). This phenotype was again rare in controls (<10%) but common in all genotypes with severely reduced βPS expression; mys^ts1/mys^XG43 mutant larvae showed a 50% incidence of this defect (Fig. 7E). Finally, the third phenotype, multiple insertions, is characterized by two or more separate and distinct nerve entry points on muscle 12 (Fig. 7C). In these junctions, both nerves could frequently be traced back to a point of separation above muscle 13, though occasionally the separation would occur after the nerve had left muscle 13. The appearance of multiple nerve entry points was rare in controls (<10%) and was not significantly increased in mys^b9 and mys^b9/mys^XG43 mutants. However, multiple nerve entry points were a common consequence of severe βPS loss in both mys^ts1 and mys^ts1/mys^XG43 animals (Fig. 7E). We conclude that alterations in βPS expression also exert effects on the pattern of innervation and synaptic specificity during NMJ development.

All three PS integrins are clearly localized to the postsynaptic SSR and their postembryonic expression correlates with the time during which SSR develops (Budnik, 1996). Therefore, we examined the ultrastructure of the larval SSR at type I NMJs in wild type and mys^b9/mys^XG43, a severe hypomorphic combination that rarely survives to adulthood (Fig. 8). Qualitatively, SSR morphology appeared unaffected (Fig. 8A,B), and quantitative measurements (see Methods) revealed no significant deviation from wild type: the SSR mean cross-section area was 5.0±3.1 µm² in wild-type and 4.7±2.9 µm² in mutant larvae (P=0.9; see also SSR/bouton ratio in Fig. 8C). Therefore, we conclude that SSR structural integrity is maintained despite strong reduction in βPS integrin expression.

Our immunological results suggested that βPS integrin is also present presynaptically; however, the mys^b9/mys^XG43 mutant boutons appeared ultrastructurally normal (Fig. 8A,B). The mys synapses showed all features described for wild type (T-bars, regular material in the synaptic cleft, omega figures of the presynaptic membrane below the T-bar; see also Prokop, 1999). Likewise, the presynaptic membranes were normally attached to the postsynaptic membrane, indicating that gross membrane adhesion was unaffected. Synaptic vesicle density in mys mutant boutons was also indistinguishable from wild type (wild type 291±59 and mys^b9/mys^XG43 mutant larvae 286±43 vesicles per µm²; P=0.7; Fig. 8C). The lack of ultrastructural defects at mys^b9/mys^XG43 mutant NMJs is in agreement with the lack of ultrastructural defects at synapses of mys null mutant embryos (Prokop et al., 1998).

To assay whether altered mutant synaptic morphology is accompanied by altered functional synaptic transmission, we recorded nerve-evoked excitatory junctional currents (EJCs) at mys mutant NMJs. EJCs were recorded in physiological (1.5 mM Ca²⁺-containing) saline in muscle 4 and muscle 6 (Fig. 9A). Visible muscle contractions were evoked in both wild-type and mutant larvae by single suprathreshold nerve stimulations (3-10 V, 0.4-0.5 mseconds). At the muscle 4 NMJ, the majority (6 of 10) of mys^b9 larvae (reduced bouton number and branching) displayed EJC amplitudes in the normal range, while a subset (4 of 10) had unusually small EJC amplitudes (range: 96 to 111 nA; Fig. 9B) even after increasing the stimulation intensity (>20 V; not shown). Overall however, mys^b9 mean EJC amplitude (164±57 nA, n=10) was only slightly reduced compared to wild type (197±42 nA; Fig. 9B). In mys^b9/mys^XG43 larvae, EJC amplitudes were similarly unaffected (211±46 nA). Similarly, at the more complex muscle 6 NMJ, mys^b9 larvae (reduced bouton number and branching) had significantly decreased EJC amplitude (182±39 nA) compared to wild type (250±35 nA, Fig. 9B), although the mys^b9/mys^XG43 larvae appeared normal. By contrast, both mys^ts1 (increased bouton number and branching) and mys^ts1/ mys^XG43 larvae (increased branching) had significantly larger mean EJC amplitudes at muscle 4 than wild type or mys^b9 genotypes (Fig. 9B).

To determine whether the altered EJC amplitudes might reflect a change in quantal size, we recorded spontaneous miniature EJCs (mEJCs) in mys^b9 and mys^ts1 larvae (Fig. 9C). Neither mutant strain displayed altered mEJC frequency relative to wild type. Mean mEJC (quantal) amplitude was similarly unaltered in mys^ts1 NMJs, and very slightly increased (15%; P<0.001) in mys^b9, compared to wild type. These results suggest that that alterations in MEJC frequency or amplitude cannot explain any of the changes observed in evoked neurotransmission.

In summary, the observed alterations in functional synaptic transmission in mys mutants are generally consistent with alterations in mutant terminal morphology. Specifically, EJC amplitudes are increased only in alleles (mysts1, mysts1/mysXG43) displaying the most marked terminal overgrowth, and EJC amplitudes are decreased only in the allele (mys^b9) displaying the most marked terminal undergrowth. However, the functional changes recorded under these mutant conditions are relatively moderate (25-50%), and are not consistently observed at different classes of NMJs in the same mutant allele. We conclude, therefore, that integrins can modulate synaptic function in line with terminal morphology, but that this modulation is unlikely to have a profound impact on the physiology of the organism.

We have shown that three integrin subunits are expressed at the Drosophila NMJ; two alpha subunits, αPS1 and αPS2, are expressed at least in the postsynaptic SSR membrane, and a beta subunit, βPS, is located both presynaptically and postsynaptically. Thus, at the NMJ, αPS1 and αPS2 subunits are co-localized in a single cell type, in contrast to other Drosophila tissues (Brown, 1993). However, the localization of multiple subunits to a single tissue is common in vertebrates (Pinkstaff et al., 1999) and has been suggested by results on PS integrin function also in Drosophila wing development (Brower et al., 1995). All three subunits show a specific onset of synaptic expression in postembryonic stages, and are restricted to type I boutons. Downregulation of αPS subunits in mys mutant alleles strongly suggests that the βPS subunit heterodimerizes with both α subunits at the NMJ. In addition to the αPS subunits, a third α integrin subunit, α-Volado, which is expressed presynaptically at the NMJ (J. R. et al., unpublished data) is a potential partner for the βPS subunit in the neuronal membrane. Thus, similar to vertebrates, Drosophila exhibits several integrin isoforms at the NMJ, distributed on both faces of the synaptic cleft (Martin et al., 1996).

Both mys mutants described here exhibit reduced βPS amplitudes are increased only in alleles (mys^ts1, mys^ts1/mys^XG43) displaying the most marked terminal overgrowth, and EJC amplitudes are decreased only in the protein at the NMJ, frequently below detectable levels. Since both alleles have been sequenced (D. Brower, personal communication) and shown to have normal coding regions, we suggest that this reduction is due to mutations in the complex regulatory regions of the mys gene. Alternatively, the defect may be in the developmental regulation of distinct splice isoforms (Zusman et al., 1993). Since α integrin subunits are reported to leave the endoplasmic reticulum only as heterodimers (Cheresh and Spiro, 1987), we were surprised to see that reduction in αPS1 and αPS2 subunit expression was less severe than the reduction of the βPS subunit. This does not seem to be due to different sensitivities of the PS antibodies since, in our hands, βPS immunological staining is more consistent and robust than any of our αPS stainings (see also Hoang and Chiba, 1998). Therefore, either αPS subunits depend only partly on heterodimerization for their synaptic localization, or there might be an as yet undiscovered β integrin subunit at the Drosophila NMJ. This does not seem to be the Drosophila β subunit βν (Yee and Hynes, 1993), since it is not detectably expressed at the NMJ.

Our observations demonstrate that the level of PS integrins at postembryonic NMJs is a critical determinant of morphological sprouting, growth and bouton differentiation. First, moderate decrease in the detection of βPS expression at the NMJ accompanied by an increase in extrasynaptic muscle expression (in mys^b9) results in undergrowth and compaction of the terminal, loss of boutons, a decrease in overall bouton size and the proliferation of a new class of ‘mini-boutons’. Second, in contrast, severe reduction in the detection of βPS expression at the NMJ, unaccompanied by a detectable increase in extrasynaptic expression (in mys^ts1), results in an increase in sprouting and morphological growth to cause overgrowth and an increase in terminal complexity. Third, overexpression of βPS, generalized or nerve/muscle-specific, reduces branching complexity and bouton numbers. In agreement with the mys^b9 observations, early βPS overexpression in the muscle also increases the frequency of mini-boutons. Thus, the subcellular distribution of βPS in both the motor neuron and its muscle partner is an important determinant of morphological development at the postembryonic NMJ.

The identity of the mini-boutons is of particular interest. Mini-boutons are found very abundantly in mys^b9 and βPS-overexpressing terminals, but are also observed at wild-type NMJs at a low frequency (see Fig. 4A). Molecularly, these structures show the hallmarks of type I boutons but are many-fold smaller. An exciting recent study (Zito et al., 1999) using confocal imagery of live synapses has suggested that these structures may represent immature type I boutons and/or new terminal branches. Our molecular labeling defines these structures as nascent type I boutons. It appears, therefore, that decreased βPS integrin expression triggers the differentiation process for these bouton precursors, but that some subsequent condition required for maturation is not being met. Our observations suggest that proliferation of these immature type I boutons is correlated with extrasynaptic expression of βPS. Thus, we suggest that bouton formation is developmentally arrested when presented with the incorrect expression of βPS in the target membrane, leading to a failure of terminal bouton maturation.

In addition to the terminal defects, changes in βPS expression also alter distinct aspects of synapse specificity in the neuromusculature. In mys mutants, arboreal branches alter normal developmental paths and branch inappropriately on to non-target muscles. Moreover, the nerve often defasciculates and branching initiates from sites prior to the normal nerve entry point, and boutons differentiate in the absence of the appropriate postsynaptic target. Finally, multiple, spatially separated insertions can occur on muscles which normally have only a single, unified nerve entry point. These defects indicate that integrins may be involved in the mechanisms that identify muscle targets, and specify the particular morphology, synapse type and function of the synapse on each muscle. Recently, a potential transcription factor, mod(mdg4), has been implicated in similar synaptic defects, as well as in ultrastructural defects not noted in integrin mutants (Gorczyca et al., 1999). It has been suggested that this transcription factor may regulate genes involved in the development of synapse identity. Potentially, integrins are part of the mechanisms involved in this development of identity.

The change in NMJ morphology in mys mutant larvae is accompanied by alterations in synaptic transmission strength. The overgrowth observed in mys^ts1 and mys^ts1/mys ^XG43 mutants is accompanied by a parallel increase in transmission amplitude at some NMJs, whereas a decrease in synaptic function accompanies the undergrowth observed in mys^b9 mutant larvae at other NMJs. This change in function is clearly in parallel to the change in NMJ morphology, but does not occur in every NMJ in every genotype. This may be consistent with βPS integrin being involved in NMJ specificity and playing different roles at different NMJs.

It is presently unclear whether the phenotypes observed in different genotypes reflects alterations in the level/distribution of βPS or interaction with other, more subtle mutations in specific genetic backgrounds. Specifically, when animals carrying either mys^b9 or mys^ts1 over mys^XG43, a null allele, were examined, we usually, but not always, observed changes in the severity of the various synaptic phenotypes. The usual case was to modify the extent of the observed phenotypes, which remained quantitatively significant although, in some cases, the phenotype disappeared entirely. There are two potential explanations for this observation. First, only a portion of the phenotype observed in the homozygous animals may be due solely to the integrin mutation, with the remainder arising from interaction with modifiers in the genetic background. Second, since both mutants are regulatory and the phenotype is most likely due not to a change in the protein but to a change in the relative distribution of the protein, making the mutant heterozygous over a protein null changes the protein distribution to a new state with novel phenotypic consequences. At this point, we are unable to distinguish between these two possibilities.

In Drosophila, a number of mutations cause morphological changes in NMJ structural architecture. In the majority of these mutations, synaptic transmission strength is affected in parallel with NMJ size (Table 2; Budnik et al., 1996; Keshishian et al., 1996). This coupled alteration has been shown for hyperactive NMJs such as those in Shaker ether-a-go-go (Sh eag; both genes encode K⁺ channel subunits) double mutants, mutants with affected second messenger pathways such as dunce (dnc, a cAMP-specific phosphodiesterase) and CaM kinase II, and for mutants with affected levels of the adhesion molecule Fasciclin I (Fas I). (Table 2; Zhong and Shanley, 1995; reviewed in Budnik et al., 1996; Keshishian et al., 1996). Like the above genes and regulatory components, PS integrins alter both growth and strength of transmission at the NMJs in parallel, although the degree of this coupled alteration is less striking than in these other examples.

In contrast, it has been elegantly shown that growth and synaptic strength can be genetically separated and singly regulated through independent molecular pathways (Davis et al., 1996); the adhesion molecule Fasciclin II serves as a growth regulator and the transcription factor CREB controls synaptic transmission strength. The dual activation of both pathways is required to alter NMJ structure in parallel with function (Davis et al., 1996; Table 2). A decrease in Fas II is not in itself sufficient to alter synaptic function, which requires a concurrent reduction in CREB (Davis et al., 1996). This division between growth and function is more blurred in the case of βPS mutants. On the one hand, it appears that mys mutations by themselves can alter function and growth in parallel. However, it is also clear that alteration in βPS expression alone is not sufficient to alter function at all junctions, and that functional phenotypes may be sensitive to genetic background. Integrins may interact with distinct sets of molecules at different junctions, affecting function in different ways.

Integrin transmembrane receptors link the extracellular matrix and the intracellular cytoskeleton, and might therefore serve either as simple adhesion molecules or as more complex signaling mediators between outer and inner cell surface (Hynes, 1992). The adhesion and signaling functions of the Drosophila PS integrins have been shown to be separable and to be differentially required in different tissues (Martin-Bermudo and Brown, 1999). Since the cell adhesion molecule Fas I can regulate both NMJ size and function (Zhong and Shanley, 1995), the PS integrins might similarly function as relatively simple cell adhesion molecules at the NMJ. If so, they do not appear required for maintained intercellular adhesion, since the synaptic cleft looks normal in ultrastructural analyses of mys mutant NMJs (Fig. 6; and Prokop et al., 1998). Integrins are more likely to modify adhesion, e.g. by interacting with other adhesion molecules, such as the fasciclins, in cis (Porter and Hogg, 1998). Such a modulatory role might explain both the synaptic specificity of the functional effect and the dosage dependence of PS integrins for growth control shown here, as well as the inability of the several adhesion molecules (Fas I, Fas II, integrins) found at the NMJ to compensate for one another.

Alternatively, the PS integrins could serve as transmembrane signaling molecules at the larval NMJ, independently initiating signaling pathways. One potential link of PS integrins to the regulation of NMJ growth and strength is through the CaM kinase II pathway, since reduction in CaM kinase II function results in NMJ structural defects which are strikingly similar to those documented here in mys mutant larvae (Table 2; Wang et al., 1994). As well, CaM kinase II is known to function in ‘cross-talk’, or co-regulation, between different integrin heterodimers in vertebrate macrophages (Blystone et al., 1999; Bouvard and Block, 1998). Currently, the only known targets for CaM kinase II in the terminal are Eag, a K⁺ channel subunit, and Dlg, involved in localizing synaptic proteins, which may explain the alteration in transmission strength following CaM kinase II perturbation experiments (Griffith et al., 1994; Koh et al., 1999). However, potentially, CaM kinase II also targets one or more PS integrin heterodimers, activating signal cascades, mediating cross-talk, and influencing both structure and function. It seems clear that we will need to consider the integrins in light of their signal transduction features as well as their adhesion features and explore mechanisms that may link morphological and functional synapse specificity.

We thank D. Brower, N. Brown, L. Mart ín-Bermudo, E. Martín-Blanco, and R. Hynes for kindly providing fly stocks and antibodies. We are especially grateful to D. Brower for the unpublished mys^b9 line, N. Brown for extensive assistance and insightful discussions, M. Condic for comments on this manuscript and to M. Bate in whose laboratory some of this work was initiated. We thank Keri Swearingen and Amber Bradshaw for extensive technical assistance. A.P. was funded by a Human Capital and Mobility Fellowship (European Community) and a Fellowship from the Lloyd’s of London Tercentenery Foundation. K. B. and laboratory were funded by a Searle Scholarship and NSF CAREER grant.