Development and plasticity of the Drosophila larval neuromuscular junction

A. A brief overview of Drosophila NMJ development

Motor neurons are individually specified, and are generated in lineages deriving from at least 10 different neuroblasts.^13,14 Their muscle targets, which are also individually specified, are produced by cell fusion events. During stages 13-15 of embryonic development, motor neurons extend their axons into the musculature. Motor axons leave the CNS in three pathways: the segmental (SN) and intersegmental (ISN) nerve roots and the transverse nerve (TN). In the periphery, the SN and ISN split into five nerve pathways, designated as the SNa (innervates lateral muscles), SNc (innervates ventral muscles), ISN (innervates dorsal muscles), ISNb (innervates ventrolateral muscles (VLMs)), and ISNd (innervates other ventral muscles).¹⁵ Each motor axon follows a genetically determined pathway to a specific muscle fiber or group of fibers.¹⁶ These are shown in both an immunohistological composite (ISN root-derived branches only, Figure 1C) and as a schematic in Figure 1D.

After an axonal growth cone makes contact with its target muscle, postsynaptic GluRs and Discs large (Dlg), the Drosophila ortholog of the mammalian PSD-95 postsynaptic scaffolding protein, begin to cluster at the contact site.^17,18 The growth cone then differentiates into a presynaptic terminal. By the end of embryonic development, functional NMJs, each containing a few synaptic boutons, have formed on each muscle fiber (Figure 3C, D). Boutons are oval-shaped structures that house synapses. Boutons contain multiple active zones (neurotransmitter release sites), and each of these is apposed to a GluR cluster. The presynaptic bouton at larval NMJs eventually becomes surrounded by an infolded membranous structure called the subsynaptic reticulum (SSR), which contains neurotransmitter receptors, scaffolding proteins, and postsynaptic signaling complexes.

Early neural development is often characterized by an initial overproduction of synaptic connections, followed by a period of selective elimination of improper processes. This phenomenon was first observed in vertebrates¹⁹ and has been studied extensively at the visual system and NMJ. In the visual system, the refinement of the connections is necessary for the formation of the retinotopic maps in the mammalian brain. Relay of visual information from the retina to the primary visual cortex in the brain occurs through the lateral geniculate nucleus (LGN) located in the thalamus. Initially, axon terminals of retinal ganglion cells (RGCs) from the two eyes form ectopic connections and overlap within the different layers of the LGN. Later on, these connections are refined to form specific eye layers. This segregation of RGC inputs involves retraction from incorrect target layers and synapse formation in the correct layer.^20,21 At the vertebrate NMJ, multiple motor neurons initially innervate the same muscle fiber. As development progresses, all but one of the motor neurons are eliminated.²² Activity is critical for this refinement: altering the activity of the motor neurons results in the more active neuron being maintained and stabilized.²³

Synaptic refinement also occurs at the Drosophila NMJ. However, this refinement is most similar to the process that happens in the vertebrate visual system as opposed to the vertebrate NMJ. In early embryonic development, motor neurons form ectopic contacts on non-target muscles. These misplaced synapses are then eliminated in late-stage embryos by an activity-dependent process.^24-27 An additional form of refinement occurs after embryogenesis at the level of synaptic gain control once the motor neurons have reached their appropriate muscle targets. Here, the NMJ arbor must grow in order to maintain the proper synaptic drive that is needed due to the dramatic increase in muscle fiber size. From hatching of the embryo to the late third instar, the surface area of each muscle fiber increases by up to 100-fold (Figures 1A, B and 2A). During this growth period, boutons are continuously being added (and some are eliminated), and these processes result in the number of boutons and the number of active zones per bouton both increasing by up to 10-fold.^28,29 The final increase in the number of active zones by up to 100-fold matches the increase in muscle surface area (Figure 1B). Another round of synapse elimination occurs during metamorphosis.³⁰

In addition to the structural changes that result because of expansion in muscle size, Drosophila NMJs also undergo plastic changes in response to short-term perturbations of neuronal and muscle activity. Some of these involve structural alterations in the NMJ, and will be reviewed here. In others, such as facilitation and homeostatic compensation, electrophysiological changes that alter transmitter release and/or postsynaptic responses are observed. In many of these cases, these changes are not associated with major alterations in NMJ structure, and thus are not discussed in this review.

B. Patterned growth of the larval neuromuscular junction

The region of contact between the motor neuron and the muscle is the NMJ. The presynaptic terminals of Drosophila larval NMJs are organized into branched arbors that are composed of chains of synaptic boutons. There are 3 types of boutons: Types I, II and III (Figure 3A). These differ in size and shape, the neurotransmitter that is released, the amount of SSR that surrounds them, and the subunit composition of the glutamate receptors with which they are associated (Figure 3B). In this review, we consider only Type I boutons, which are glutamatergic and can be divided into two classes: 1b (large) and 1s (small). Type II and type III boutons are modulatory and use other neurotransmitters. In a third instar larva, a typical NMJ has approximately 20-50 Type I boutons on each muscle, with each individual bouton housing about 10 active zones (Figures 3C, D). Most studies examine NMJs on muscles 6/7 (this NMJ has twice as many boutons because it innervates two muscles), muscle 4, or muscle 12, in segments A2-A5.¹⁶

The structure of a larval NMJ is stereotypic and shows a similar arborization pattern for a specific muscle in different abdominal segments. Numerous studies have examined development of this complex structure from a 1^st larval instar to a 3^rd instar using fixed larval preparations. However, live imaging of NMJ development has allowed the study of the mechanisms involved in bouton addition and branch formation.³¹ Live imaging was first done by using a chimeric transmembrane green fluorescent protein that was targeted to the SSR. This construct allowed visualization of postsynaptic structures that outline Type Ib synaptic boutons, through the cuticle of live larvae. The same NMJ could be viewed multiple times during development, from first instar through third instar. These studies showed that location of new bouton formation was either between pre-existing boutons or at the end of a branch of the NMJ arbor. The new boutons arose by asymmetric budding of a mature (parent) bouton (similar to cell division in yeast), by symmetric division of a pre-exisiting bouton, or by de novo formation of a bouton from the axonal membrane (Figure 2B).

During NMJ development, many transient structures are either stabilized or retracted during formation of the complex terminal arbor. Since the above study examined synaptic boutons indirectly through visualization of a postsynaptic marker protein that surrounds these boutons, nascent presynaptic structures could not be observed. These include synaptopods, presynaptic debris, and ghost boutons. These transient structures are seen at normal NMJs during development and seem to be remnants of the synaptic refinement process. However, under various conditions, such as acute stimulation of motor neurons, these structures are stabilized and not properly eliminated.³² Synaptopods are highly dynamic presynaptic filopodial extensions that can only be visualized in live larval preparations.^33-36

C. Molecular mechanisms involved in NMJ growth

The mechanisms that regulate the different stages of bouton formation, development and maintenance are not fully understood. Our current knowledge of the molecules involved is based on genetic and biochemical analyses of mutants that have morphological NMJ phenotypes. The purpose of this section is to review a few of the genes involved in bouton growth and classify them based on their mutant phenotypes. The ultimate goal of investigators would be to understand the molecular mechanisms involved in the life history of a bouton from birth through maturity, as well as those that are required for branch formation during development of the terminal arbor. The genes discussed below function either cell autonomously on the presynaptic side (in motor neurons) or in trans-synaptic pathways involved in signaling from muscles to the neurons. Trans-synaptic signaling pathways are also discussed in a separate section below. Table 1 is a partial list of genes that are implicated in NMJ growth, which includes many genes in addition to those explicitly discussed in the text.

For the purposes of organization, we group the genes that play a role in the development of the larval NMJ into four categories. Each of these categories includes genes that encode proteins that function in a variety of different pathways. The first category consists of genes whose products promote NMJ growth. These are defined as those for which loss-of-function (LOF) mutants have smaller terminal arbors. In the second category are genes whose products inhibit NMJ growth, and LOF mutants for these genes have expanded terminal arbors. The third category discusses genes involved in neuronal activity that alter NMJ growth. The fourth category encompasses genes that regulate the formation and maintenance of boutons and do not fall into the other 3 groups. Disruption of these genes produces boutons that are arrested at various stages of development. For each of these categories, only a few genes that fall into these groups are discussed. In the last part of this section we describe how many of these genes may work in parallel to affect the same downstream effectors that regulate NMJ growth.

D. Trans-synaptic signaling pathways

Two of the most extensively studied trans-synaptic pathways that regulate development of synaptic arbors at Drosophila NMJs are the Wnt pathway and the BMP pathway. Excellent recent reviews exist for these pathways, so we do not discuss them here.^35,94-98 The Neurexin-Neuroligin trans-synaptic pathway has also recently been reviewed.⁹⁹ Below we discuss two less well-known trans-synaptic pathways.

A. miRNAs

miRNAs are 21-25 nucleotide non-coding RNAs that regulate gene expression by binding to target mRNAs and recruiting a repressor complex known as the RNA-induced silencing complex (RISC), which includes Dicer, Argonaute proteins, and other components.^113,114 RISC not only functions in the biogenesis of miRNAs and siRNAs (silencing RNAs), but also is required for their activity. The degree of complementarity between the miRNA and its target mRNA determines the mode of regulation: near-perfect complementarity results in cleavage of the duplex, whereas partial complementarity leads to translational repression.

miRNAs are known to be important regulators of neural development and function in a variety of systems.^115-117 However, the only miRNA whose individual function at the NMJ has been characterized thus far is miR-8. This miRNA is also implicated in neurodegeneration,¹¹⁸ Wnt signaling,¹¹⁹ and innate immune homeostasis.¹²⁰ miR-8 was found to be a regulator of NMJ growth in a forward genetic screen. miR-8 activity at the NMJ was examined using a deletion of miR-8 and a modified ‘microRNA sponge’ system, which can inactivate specific miRNAs.¹²¹ Knocking out miR-8 function presynaptically had no effect. However, postsynaptic knockout resulted in a decrease in the number of synaptic boutons and branches. The 3′UTR of the enabled (ena) mRNA has one predicted miR-8 binding site, and if miR-8 indeed binds to and represses ena mRNA translation, Ena protein levels should increase when miR-8 activity is inhibited. This was in fact observed: postsynaptic expression of the miR-8 sponge resulted in elevated levels of Ena. It was also shown that postsynaptic overexpression of Ena can phenocopy the loss of miR-8.¹²² These results provide evidence for a role of miRNA-mediated translational repression in regulating synaptic growth at the NMJ. Some important questions remain, however. First, are ena mRNA and/or miR-8 localized to the SSR? Second, the genetic interaction between miR-8 and ena does not necessarily indicate that miR-8 directly controls Ena expression. It will be important to determine whether miR-8 binds to the ena 3′UTR.

Although miR-8 is the only miRNA that has been shown to regulate NMJ growth thus far, Dicer and miR-284a control GluRIIA and GluRIIB protein levels at the NMJ³⁴ (Figure 5). GluR mutations can affect bouton numbers,^123-125 so miRNA-mediated effects on translation of GluR mRNAs may have an impact on NMJ growth during development.

The roles of miRNAs at the NMJ have also been explored using mutations that affect the RISC complex. Argonaute 2 (Ago2) is expressed presynaptically at the NMJ and has been shown to be a positive regulator of bouton growth. Ago 2 mutants have decreases in bouton number and in the number of arbor branches.¹²⁶ Ago2 functions predominantly in the siRNA pathway rather than in miRNA processing, although alternate miRNA biogenesis pathways may require Ago2.¹²⁷

Two other Drosophila miRNAs have been demonstrated to regulate dendritic outgrowth and to repress translation of proteins that are important for NMJ growth. It would be interesting to examine whether these miRNAs also have functions at the NMJ. In Drosophila embryos and larvae, GFP driven by the miR-124a promoter is expressed at high levels in the ventral nerve cord and motor neurons and at lower levels in dendritic arborization (DA) neurons in the peripheral nervous system (PNS).¹²⁸ Overexpression of miR-124a in DA neurons caused a reduction in the number of dendritic ends. miR-184 functions in ovaries to allow differentiation of the germline stem cells by reducing expression of the Decapentaplegic receptor, Saxophone (Sax).¹²⁹ sax mutants also have an NMJ phenotype characterized by reduced numbers of boutons and decreased synaptic strength.¹³⁰ It is not known whether miR-184 contributes to this sax phenotype.

B. Fragile X mental retardation protein

Fragile X syndrome is one of the most common forms of mental retardation in humans, affecting an estimated 1 in 4000 males and 1 in 8000 females. The disease is caused by mutations in a single gene, FMR1 (for fragile X mental retardation 1).^131-133 FMR1 protein (FMRP) is widely expressed in fetal and adult tissues with enhanced expression in the brain and testes where major symptoms are manifested.¹³⁴ Loss of FMRP results in delayed dendritic spine maturation in both FXS patients and Fmr1 knockout mice.^135-137 FMRP is a translational repressor ^138,139 that interacts with sequences in the 3′UTRs of its target mRNAs. Flies have only one ortholog of FMRP, called dFMR1, which regulates synaptic structure and physiology.¹⁴⁰

At the Drosophila NMJ, dFMR1 is expressed presynaptically in the motor neurons and postsynaptically in the muscles. Analysis of dfmr1 mutants revealed synaptic overgrowth characterized by an increase in the number of boutons, an increase in branching, and increased synaptic strength. Satellite boutons were also present in greater numbers than in wild-type.¹⁴¹ Overexpression of dFMR1 presynaptically produces a phenotype with fewer boutons that are significantly larger than normal. One of the neuronal targets of FMRP is Futsch, the MAP-1b ortholog.^142,143 Futsch mRNA levels are upregulated in dfmr1 mutants. dFMR1 repression of Futsch expression is important for NMJ development, because futsch mutations suppress the dfmr1 NMJ phenotype.¹⁴⁰

Like miRNAs, dFMR1 controls expression of the subunits of the postsynaptic ionotropic glutamate receptor.¹⁴⁴ NMJ GluRs can be divided into two classes: the larger amplitude, slower acting A class and the smaller amplitude, faster-acting B class. Each functional GluR consists of four subunits, three of which are shared among all receptors (GluRIIC, GluRIID, GluRIIE). The fourth subunit is either GluRIIA (A class) or GluRIIB (B class).^110,145 In dfmr1 mutants, GluRIIA is increased and GluRIIB is decreased (Figure 6). Postsynaptic overexpression of dFMR1 caused a decrease in both GluRIIA and GluRIIB levels. Because GluRIIA mRNA has been localized to the postsynaptic region of the muscle fiber, it is possible that this repression of the GluRIIA subunit by dFmr1 occurs locally at synapses.¹¹¹ The contribution of dFMR1-regulated GluR expression to bouton growth has not been examined, although it is well documented that GluR levels affect NMJ growth during development.^123-125

As described above, miR-124a may be likely to have roles at the NMJ, due to its high expression in motor neurons. miR-124a levels are regulated by dFmr1.¹²⁸ Thus, other than direct repression of target mRNAs, dFmr1 may also affect neuronal development at the NMJ by controlling levels of miR-124a. Another potential target of dFMR1 is Dlg, the PSD-95 ortholog. In mice, FMRP interacts with the 3′UTR of PSD-95 mRNA to regulate its stability.¹⁴⁶ Since Dlg plays a major role in synaptic structure and function at the larval NMJ,^6,18 dFMR1 could also regulate bouton growth by repressing translation of dlg mRNA. Finally, interactions between FMRP and the miRNA pathway may further contribute to the roles that both of these systems play in synaptic growth. These interactions are discussed in more detail below.

C. Translational repression by Pumilio and Nanos

The Drosophila Pumilio (Pum) protein is a member of the PUF RNA-binding protein family. Maternal Pum functions primarily as a translational repressor in embryonic patterning and germ cell proliferation and migration.^147,148 Pum recognizes sites called Nanos response elements (NREs) in its target mRNAs, including those encoding Hunchback, Cyclin B, and Para.^149-151 Most NREs are in 3′ UTRs. In most cases, Nanos (Nos) functions as a corepressor necessary for Pum-mediated translational repression. However, Pum and Nos also have functions that are separate from those of their partners.^152-154

Zygotic Pum also functions later in development at the larval NMJ. Pum is postsynaptically localized at the NMJ, and is also present in neuronal cell bodies.¹⁵⁵ Pum has distinct roles on the two sides of the synapse. In pum mutants, type 1b boutons are much larger and their numbers are decreased, whereas the number of 1s boutons is increased. The 1b bouton phenotype can be fully rescued by neuronal expression of full-length Pum in the pum mutant background. Postsynaptic Pum expression in pum mutants has no effect on the 1b bouton phenotype, but it rescues the increase in 1s bouton numbers.¹⁵⁵

The idea that local postsynaptic translation occurs at the NMJ emerged from the findings that puncta (‘aggregates’) of the translation factors eIF-4E and PABP appear at NMJ boutons after larval motor activity is induced by moving larvae from slurry (liquid) to solid food for a period of a few hours.¹¹¹ Larvae can remain relatively stationary while ingesting liquid food, but must actively burrow through solid food. GluRIIA levels are also increased by this protocol. It was speculated that the purpose of local postsynaptic translation at the NMJ is to allow rapid changes in synaptic strength and facilitate the growth of new boutons in response to increases in larval motor activity. These synaptic alterations might be able to occur more quickly if they are implemented through translation of mRNAs that are already localized to the postsynaptic SSR.

Since Pum is a translational repressor¹⁵⁶ and is postsynaptically localized, Menon et al.¹⁵⁵ reasoned that it might control the levels of postsynaptic eIF-4E, which is limiting for translation in many systems. Indeed, it was observed that eIF-4E levels at the NMJ are very high in pum mutants (up to 12-fold higher than in wild-type), and that these levels are unchanged by increases in larval motor activity. eIF-4E is encoded by a single essential gene, so eIF-4E protein is also present within the cytoplasm of the muscle fiber. However, cytoplasmic eIF-4E levels are unchanged in pum mutants, indicating that translational repression by Pum only occurs at the synaptic sites where Pum Is localized. Pum binds selectively to the 3′ UTR of eIF-4E mRNA, suggesting that it is a direct target. GluRIIA levels are also greatly increased in pum mutants, and Pum binds selectively to the 3′ UTR of GluRIIA mRNA as well.¹⁵² These results are consistent with a model in which Pum normally represses translation of synaptically localized eIF-4E and GluRIIA mRNAs in larvae that are not moving vigorously. When larval motor activity increases, Pum (or a Pum cofactor) is partially inactivated. This would cause the levels of eIF-4E, GluRIIA, and other direct Pum targets to increase rapidly, since these proteins would be translated from pre-existing mRNAs that had been translationally repressed by Pum. When larvae are forced to move they require more transmission at the NMJ, and this could be facilitated by eIF-4E induction, since eIF-4E might be limiting for translation of all postsynaptically localized mRNAs. The induction of GluRIIA, which produces receptor complexes that conduct more current, would also increase the ability of the NMJs to effectively depolarize the muscles. The mechanisms by which Pum might be inactivated after induction of larval motor activity are unknown, but there is evidence that Pum’s postsynaptic functions are regulated by an aggregation-prone sequence within its unstructured N-terminal region.¹⁵⁷

The regulation of eIF-4E and GluRIIA by Pum is part of a more complex circuit of translational regulation that operates at the NMJ. During early development, Pum and Nos work together to repress Hb and other targets. However, they work in opposition to each other at the NMJ. Pum binds to the 3′ UTR of nos mRNA, and Nos levels are increased in pum mutants. Nos represses expression of the alternate GluR subunit GluRIIB. Also, GluRIIA and GluRIIB compete with each other for occupancy in synaptic receptor clusters. Thus, when Pum levels are reduced, Nos is increased, leading to downregulation of GluRIIB, which amplifies the elevation of GluRIIA produced by loss of repression of its mRNA by Pum. Conversely, if Pum is increased, GluRIIA and Nos are both repressed, leading to increased expression of GluRIIB at the expense of GluRIIA¹⁵² (Figure 6). The mechanisms by which Nos represses GluRIIB without the involvement of Pum are unknown.

Pum is also involved in synaptic growth and plasticity in other types of neurons, both in Drosophila and in vertebrate systems. Pum can bind to the 3′UTR of dlg mRNA. In the mushroom bodies of adult Drosophila, overexpression of Pum reduces the levels of Dlg and causes a defect in the elaboration of axonal projections.¹⁵⁸ However, Dlg levels at the NMJ are not affected by Pum. Hypomorphic pum mutants have been reported to have learning and memory defects.¹⁵⁹ Pum and Nos also affect dendritic arborization in Drosophila sensory neurons.¹⁶⁰ In dissociated mammalian hippocampal neurons, Pum is localized to granules in dendrites.¹⁶¹ Knockdown of Pum by siRNA causes increases in dendritic arborization, while Pum overexpression reduces the size of the dendritic arbor.¹⁶² Finally, a novel function of Pum in controlling translational repression by miRNAs was recently reported¹⁶³ and this is discussed below.

D. Orb2/CPEB2

CPEB proteins can be divided into two subfamilies. The first CPEB subfamily functions mainly during oogenesis and early embryonic development ¹⁶⁴ and includes the Drosophila CPEB ortholog Orb. CPEBs recognize specific sequences in the 3′UTRs of their target mRNAs and control their translation. Orb is required for establishing anteroposterior and dorsoventral axes in early development by translationally activating oskar and gurken mRNAs, respectively.^165-169 The second CPEB subfamily includes vertebrate CPEB2-4 and Drosophila Orb2. This subfamily is more broadly expressed and has roles outside of the germline.^170,171

Although no direct evidence yet links Orb2 to synaptic growth at the NMJ, it is reasonable to think that it may have a role there. Drosphila Orb2 is widely expressed in the nervous system from embryonic to adult stages. Specific localization of Orb2 at synaptic sites was observed in the CNS, suggesting that it might be involved in synaptic translation.¹⁷² A study aimed at identifying mRNA targets of Orb2 identified a variety of genes involved in synaptic growth and stability at the NMJ, including neuroligin, still life, and aPKC.¹⁷³ It is unknown, however, if Orb2 regulates translation of these mRNAs in the neuromuscular system. Finally, Orb2 was identified in a screen for proteins likely to function in the dFmr1 pathway, suggesting that that Orb2 might regulate dFmr1-mediated synaptic growth.¹⁷⁴

E. Interplay among translational regulation pathways

In addition to the separate action of each of these translational regulatory mechanisms on its respective targets, evidence suggests that these systems can regulate one another and/or work in tandem to control the same target mRNAs. The miRNA pathway and mammalian FMRP are intimately linked.^175,176 In Drosophila, orb mRNA translation is activated by Orb protein, forming a positive feedback loop. dFMR1 also binds orb mRNA and inhibits its translation, thus keeping the positive feedback loop in check.¹⁷⁷ The 3′UTR of the mammalian tumor suppressor p27 mRNA has binding sites for Pum and for two miRNAs, miR-221 and miR-222. However, in order for the miRNAs to efficiently repress p27 translation, Pum must first bind and induce a conformational change in the mRNA.¹⁶³ A similar observation was made for translational regulation of the E2F3 oncogene.¹⁷⁸

Collectively, these interactions suggest that a complex series of interconnected regulatory mechanisms control the translation of mRNAs that encode key regulators of synaptic growth and function at the NMJ. However, we still lack an overall understanding of how these mechanisms work together to ensure that mRNAs encoding synaptic regulators are translated at the correct times and places. We need to determine which mechanisms function in the postsynaptic SSR, and which act in the muscle and neuronal cytoplasm. Are all of the necessary components for translation present in the SSR? This is not yet known. It is also unknown whether all of these mechanisms function during the same developmental stages. Finally, it has not yet been directly demonstrated that local synaptic translation actually occurs at the NMJ. It would be valuable to develop an optical method to detect and localize synaptic translational events in wild-type and mutant larvae.

The genes described in this review encode proteins that control synaptic bouton growth through many different mechanisms. These include cytoskeletal dynamics, protein degradation, cell adhesion, and neuronal activity. NMJ growth and development is further fine-tuned at the level of protein synthesis by translational regulators that include miRNAs, FMRP and Pum. The study of the larval NMJ is likely to continue to generate exciting new findings. It will also be of interest to examine the development and maintenance of the adult neuromuscular system, which is still poorly understood. We have highlighted molecular mechanisms employed at the Drosophila NMJ that are similar to those used at glutamatergic synapses in the vertebrate nervous system. Many vertebrate synaptic proteins have orthologs that are used for the development and function of the fly NMJ. Because of this, researchers can productively use forward genetic screens in Drosophila to find new synaptic components that are likely to be important for development and/or function of mammalian excitatory synapses. Insights gained from studies of the fly NMJ should provide information relevant to the development and function of synapses in many other systems