The wiring of the nervous system, from initial axon sprouting to the formation of specific synaptic connections, represents one of the most dramatic and precise examples of directed cellular outgrowth. Developing axons navigate sometimes tortuous routes as they seek out the appropriate target cells. Once in their target area, interactions between axons and their potential targets are extremely dynamic, attempts are made to identify appropriate postsynaptic partners, and initial synaptic contacts are established 
,[and reviewed in 3]
. A next critical step in the formation of functional neural circuits is the remodeling of initial patterns of connectivity. To facilitate the elaboration and refinement of developing neural circuits synaptic partners often remain highly responsive to their environment and add or eliminate synaptic connections 
, frequently in an activity-dependent fashion, presumably to fine-tune connectivity to specific activity patterns.
After the axons have found their partners, two distinct mechanisms can drive the developmental reorganization of synaptic connectivity: intercellular competition between cells for common targets (reviewed in 
), and the addition/elimination of synapses within a single arbor in response to the physiological demands of the signaling unit 
. The former mechanism dictates the circuit “wiring diagram” by defining precisely which subsets of cells will communicate through synaptic contacts; while the latter, in contrast, modulates the strength of connectivity between specific pre- and postsynaptic cells after circuits are assembled.
Early in nervous system development an excessive number of axonal projections and synaptic connections are initially established. What then ensues is cell–cell competition between neurons innervating the same target for limiting target-derived cues or sites of innervation during synaptogenesis. Appropriate synaptic contacts are then strengthened and exuberant processes are destabilized and eliminated through activity-dependent mechanisms 
. For example, at the mammalian neuromuscular junction (NMJ) muscles are initially innervated by more than one motor input. However, through a process of intercellular competition for motor endplates, all but one motor input are eliminated, with the “losers” retracting wholesale from the motor endplate 
. Likewise, at the retinotectal projection in frogs, retinal axons initially establish a rough topographic map with substantial overlap between branches. However, these local synaptic terminals ultimately compete for target space and through activity-dependent modulation of synapse stabilization the spatial map of synaptic inputs is ultimately refined to a highly selective subset of inputs 
In the intercellular competition model the elimination of exuberant inputs (the “losers”) can entail large-scale elimination of axon branches, and perhaps smaller scale pruning of individual synaptic contacts. During axon and synaptic pruning in mammals and Drosophila
entire axon branches are destabilized, degenerate, and are then cleared from the central nervous system by engulfing cell types (reviewed in 
). Similarly, recent work has shown that excessive motorneuron inputs at the mammalian NMJ also become destabilized, detach from the motor endplate, and undergo axosome shedding. In this process local Schwann cells processively engulf motorneuron terminals in a distal to proximal direction and constitute the force that drives retraction bulbs toward the parent arbor during input elimination 
. Ultimately, this mechanism results in a reduction of the total number of cells supplying synaptic input to the target cell.
In the second and mechanistically distinct mode of synapse remodeling, individual synaptic contacts are added or removed from a single arbor to strengthen or weaken synaptic input to the target cell. Such changes are generally elicited by changes in the target size or neural activity. For example, Drosophila
motorneurons have established synaptic contacts with specific embryonic muscle cells by the end of embryogenesis 
. At subsequent larval stages individual arbors, along with the target muscle itself, grow in size ~100-fold 
. This coordinate increase in muscle size and synaptic contacts at motorneuron terminals serves to increase synaptic input from the motorneuron as needed to drive activation of the expanding muscle fiber. Similar mechanisms appear in place to modulate the balance of neural input versus target cell size in mammals: at the mammalian adult bulbocavernous muscle, testosterone manipulation lead to increases or decreases in muscle size, and these changes were accompanied by respective expansion or shrinkage of the postsynaptic region of the NMJ, respectively 
Here we explore the in vivo dynamics of synaptic expansion in motorneuron arbors at the Drosophila
NMJ. We show in live preparations that the addition of new synapses during normal synaptic growth entails a large amount of shedding of presynaptic membranes in the form of small debris and a subpopulation of undifferentiated synaptic boutons (ghost boutons) that failed to mature. This process is distinct from intercellular competition, as none of the motorneuron terminals are eliminated. Rather, this mechanism appears to regulate the final size of the terminal arbor. We find that the formation of presynaptic debris (this report) and ghost boutons 
are modulated by neural activity, as acute stimulation of motor inputs leads to increased appearance of these structures. Intriguingly, presynaptic debris and the subpopulation of ghost boutons that become detached from the parent arbor appear to be actively cleared from the NMJ as they disappear over developmental time. We show that glia dynamically invade the NMJ and phagocytose presynaptically shed debris, and that ghost boutons are engulfed or degraded primarily by muscle cells. Loss of phagocytic function in glia or muscle cells through manipulating the Draper signaling pathway (a key engulfment signaling pathway) results in an accumulation of presynaptic debris or ghost boutons at the NMJ and a severe reduction in NMJ expansion, indicating that continuous clearance of shed presynaptic debris and/or ghost boutons is essential for normal synaptic growth. Thus glia and muscles work together to sculpt connectivity at developing NMJ arbors, clearing multiple types of shed presynaptic structures that are inhibitory to the formation of new synaptic boutons.