Cellular mechanisms of dendrite pruning in Drosophila: insights from in vivo time-lapse of remodeling dendritic arborizing sensory neurons

The nervous system is generated by progressive developmental phenomena,such as cell division and growth, and regressive phenomena, such as cell death and pruning (Cowan et al.,1984). One aspect of morphogenesis that is more pressing for neurons than other cell types is the elimination of extensive processes. Removal of such neuritic processes takes place in cells where the soma has undergone developmental cell death (Perry et al., 1983) and also in viable cells removing exuberant or inaccurate projections (Luo and O'Leary,2005).

Selective branch removal, or pruning, is found throughout the vertebrate nervous system. Two of the best examples are the pruning of the sub-cortical axonal projections of Layer 5 neurons of the cortex(O'Leary and Koester, 1993)and the pruning of motoneurons at neuromuscular junctions(Keller-Peck et al., 2001). Both underscore the importance of this phenomenon in shaping neural circuitry,but represent two extremes of pruning. Layer 5 cortical neurons remove large identifiable lengths of axon from animal to animal; this has been termed`stereotyped' pruning (Bagri et al.,2003). Pruning of motor axons in the vertebrate peripheral nervous system, by contrast, is usually small scale (i.e. local) and stochastic in nature.

The modes by which branches are eliminated also appear to vary. When motor axon branches lose their contacts, they lift off the muscle, form a`retraction bulb' and retreat backwards to the parent branch(Keller-Peck et al., 2001). In contrast, chick retinal axons that `overshoot' their targets within the optic tectum appear to correct errors by local degeneration of the axon(Nakamura and O'Leary, 1989). At present, we know little about the mechanisms that underlie either type of pruning, and how different they truly are.

Axonal and dendritic pruning is especially pronounced in insects that undergo complete metamorphosis. These insects build two distinct bodies: a larval form for feeding and growing, and an adult form for reproduction and dispersal. To transition between these two forms, the nervous system undergoes dramatic changes which include the differentiation of adult-specific neurons,the death of some larval neurons and remodeling of others(Levine and Weeks, 1996; Truman, 1996). Studies of central neurons in Drosophila have shown that pruning of axons and dendrites can be blocked by interfering with either ecdysone signaling(Schubiger et al., 1998; Schubiger et al., 2003; Lee et al., 2000), TGFβsignaling (Zheng et al., 2003)or the ubiquitin proteasome system (Watts et al., 2003). Alongside the requirement for such intrinsic factors, evidence is accumulating that extrinsic factors are also important for pruning in some neurons. Glial processes infiltrate the neuropil just prior to pruning of mushroom body γ neurons, and afterwards contain axonal debris (Watts et al.,2004). If glia are prevented from penetrating the neuropil, axon pruning is blocked (Awasaki and Ito,2004).

The pruning of neuritic arbors is also seen in peripheral neurons in Drosophila. During metamorphosis, the majority of larval sensory neurons die but a small number persist to become functional adult neurons(Williams and Shepherd, 1999). Among those that survive are dendritic arborizing (da) sensory neurons(Usui-Ishihara et al., 2000; Williams and Shepherd, 2002). The large, complex dendritic arbors of these neurons are completely removed during early metamorphosis, before the cells elaborate their adult-specific arbors (Smith and Shepherd,1996; Williams and Truman,2004).

Here we describe how da neuron dendrites are deconstructed during early metamorphosis and the contributions of intrinsic and extrinsic factors to this process. We find that microtubule cytoskeleton remodeling precedes the severing of proximal dendrites, and that this intrinsic mechanism can be blocked by the expression of a dominant negative ecdysone receptor. Our data also suggest that two other cell populations may participate in the deconstruction of the arbor. Phagocytic blood cells attack and sever intact branches, and the epidermis, the substrate over which these cells arborize,remodels whilst the neurons are pruning.

For live imaging, C161-GAL4(Smith and Shepherd, 1996) and ppk1.9-GAL4 drivers (Ainsley et al., 2003) were used to express either mCD8::GFP(Lee and Luo, 1999) or tubulin::GFP fusion proteins (Grieder et al., 2000) specifically in da neurons.

To disrupt ecdysone signaling, females of the genotype UAS-EcR.B1-ΔC655.W650ATP1-9(Cherbas et al., 2003) were crossed with UAS-CD8::GFP; C161-GAL4,UAS-CD8::GFP/TM6b. To block cell death, UAS-p35 females were crossed to UAS-CD8::GFP; C161-GAL4,UAS-CD8::GFP/TM6b. To image the epidermis, we used a genomic rescue construct of histone H2A fused to GFP(Clarkson and Saint, 1999).

Individual animals were collected at pupariation and maintained at 25°C. Staging was denoted as hours after puparium formation (h APF).

Dissection and immunohistochemistry were performed as described by Truman et al. (Truman et al., 2004). Primary antibodies included Rat anti-mCD8 (1:1000; Caltag Laboratories,Burlingame, CA, USA), mAb 22C10 [1:500; Developmental Studies Hybridoma Bank(DSHB), Iowa City, IA, USA] and anti-Cut (F2) (1:20, DSHB). Secondary antibodies included; 488 Alexa Fluor anti-rabbit IgG (1:500; Molecular Probes,Eugene, OR, USA) and Texas Red donkey anti-mouse IgG (1:500; Jackson ImmunoResearch Laboratories, West Grove, PA, USA).

Acidic organelles were labeled with LysoTracker Red (Molecular Probes). A 1 nl bolus of 50% LysoTracker DND-99 was injected directly into prepupa 10 h APF using a pico-spritzer and a glass microneedle.

Confocal images were taken using a BioRad (Hercules, CA, USA) Radiance 2000 system equipped with a krypton–argon laser. Z-stacks were collected at 1.5 μm intervals (40×). For time-lapse imaging, individual prepupae or pupae were mounted in an imaging chamber(Williams and Truman, 2004)and data acquired with the same Radiance system using a Ti:Sapphire laser(Spectra-Physics, Fremont, CA, USA) set at 905 nm. Time-lapse frames consisted of stacks of ∼25 sections at 1.5 μm intervals, acquired every 10 minutes. Stacks and movies were assembled in ImageJ, adjusted for brightness and contrast using Photoshop (Adobe Systems, San Jose, CA, USA).

The dorsal neurons labeled by C161-GAL4 were individually identified by dendritic morphology and levels of Cut protein expressed(Grueber et al., 2002; Grueber et al., 2003)(Fig. 1A). Two da neurons, ddaD and ddaE, have simple dendritic arbors and show no measurable Cut expression. The dendritic arbor of ddaD projects anteriorly whereas that of ddaE projects posteriorly. Two other da neurons show high levels of Cut protein and have spike-like protrusions on their branches, identifying them as ddaF and ddaA. A fifth da neuron, ddaB, has low levels of Cut and a simpler arbor. A sixth neuron expresses GFP at low levels. Using transgenes containing regulatory regions from the gene pickpocket (ppk), we identified it as ddaC.

The five dorsal da neurons strongly labeled by C161-GAL4 show two different fates: ddaA, ddaB and ddaF undergo cell death, whereas ddaD and ddaE survive and remodel. Neither group shows any obvious change at 0 h APF (data not shown). By 6 h APF the arbors of ddaF, ddaB and ddaA have largely disappeared, although some fragments and blebs of GFP remain(Fig. 1B). The corpses of the cell bodies of these neurons can be seen in close proximity to the nerve.

The dendrites of ddaD and ddaE are still larval in their number and extent at 6 h APF. By 10 h APF (Fig. 1C), they have started to prune, with arbor loss more advanced in ddaE than ddaD. Filopodia are also found on the branches of both neurons. By 16 h APF, many of the higher order branches are lost, and detached branches appear to be separated from the main arbor(Fig. 1D). Filopodia are still found along the branches as the proximal regions begin to thin. By 24 h APF,the only remaining dorsal neurons that express the C161-GAL4 driver are ddaD, ddaE and dbd. Debris derived from pruned dendrites remain, and the cell bodies have filopodia-rich growth cones(Fig. 1E). Axons remain intact throughout and show no change in caliber.

To understand the cellular mechanisms that bring about these dramatic changes, we used in vivo time-lapse microscopy. Fig. 2A shows frames from Movie 1 (see supplementary material) focusing on the distal arbors of 3 da neurons. The neurons imaged are ddaF and ddaB, both of which die, and ddaD, which survives and remodels.

At 0 h APF, the dendritic arbors are indistinguishable from their larval form. By 2 h APF, the dorsally projecting branch of ddaF develops constrictions and swellings which become more obvious `beads' by 3 h APF. Between 3 and 4 h APF, the branch severs at multiple sites, leaving blebs of GFP, which move away randomly. The other branches of ddaF follow the same sequence.

Between 3 and 4 h APF, the dendrites of ddaB also begin to generate constrictions and beads. Just after 5 h APF its branches sever, leaving only the arbor of ddaD intact by 6 h. The dying cells showed no filopodia. Beading appears to propagate through the arbor as a proximal to distal wave. Fig. 2B illustrates this wave in the dorsally projecting arbor of ddaF. At 1.5 h a distinct bead forms in the proximal region whereas the most distal part is intact. By 3 h the whole branch appears beaded, and by 4 h these beads move in random directions.

To visualize the microtubule cytoskeleton in these neurons, we used a tubulin GFP fusion protein, tub::GFP. At 0 h APF, the arbors have a uniform distribution of tub::GFP (data not shown). By 3 h APF, ddaF shows a redistribution of tub::GFP (Fig. 2C), with some regions losing tub::GFP while others accumulate it into beads. In contrast, the arbor of ddaD at 3 h APF still shows a uniform distribution of tub::GFP.

When p35 was expressed with C161-GAL4, cell death was blocked in ddaA, ddaF and ddaB (Fig. 2D). The distal dendrites of ddaF and ddaB did not bead and undergo degeneration by 5 h APF (Fig. 2D). At 18 h APF the cell bodies of these `rescued' da neurons were still evident, but their arbors had been largely removed (Fig. 2E) and abundant filopodia extend from all cells.

To obtain a greater understanding of dendritic pruning in remodeling neurons, we made time-lapse movies of ddaD and ddaE. Fig. 3A shows selected frames from a movie between 14 and 24 h APF (Movie 2 in supplementary material). At 14 h 10 min APF <50% of the arbors of ddaD and ddaE remain, with filopodia found throughout. Severed branches, along with smaller fragments and blebs of GFP are seen close by. By 17 h 10 min APF more branches are severed. The primary branch of ddaE is also thinner than at 14 h 10 min APF. By 19 h APF the major branches of both neurons have been severed from their cell bodies. Between 19 and 24 h APF the separated branches decrease in length, thin and form swellings.

Movies of ddaD and ddaE taken during this period (n=8) reveal that branch severing occurs at both distal and proximal sites within the arbor. Fig. 3B shows selected frames from Movie 2 (in supplementary material), focusing on the proximal arbor. Large and small swellings are found along the primary branches of both ddaD and ddaE throughout the thinning phase and move in both proximal and distal directions. Another feature coincident with the thinning is the appearance of filopodia, which extend and retract in the thinning zone (17 h 00 min). Between 17 h 00 min and 17 h 50 min the caliber of the primary branch of ddaE becomes noticeably thinner and changes from straight to undulating. Just prior to 18 h the primary branch of ddaE is severed. The primary branch of ddaD continues to thin and severs at 18 h 30 min.

Fig. 3C shows a ddaD branch undergoing a `distal' severing event. At 14 h 10 min the arbor appears stable,although there are swellings similar to those described above for proximal severing. Between 15 h 10 min and 15 h 20 min the branch severs, but shows no obvious thinning prior to this event.

Severed branches undergo fragmentation as illustrated in Fig. 3D. At 17 h 50 min the segment begins to thin and generate swellings. At 18 h 50 min the branch is beaded and is fragmenting into blebs. By 20 h most debris has disappeared.

In addition to severing and fragmentation, the branches of ddaD and ddaE can also undergo distal to proximal retraction. Fig. 3E shows a movie sequence in which retraction occurs in the proximal region of ddaD. At 16 h 30 min the primary branch of ddaD possesses a number of filopodia. Between 16 h 30 min and 20 h 00 min the branch retreats retrogradely towards the cell body. We saw retraction events in the distal regions of the arbor (branch labeled with asterisk in Fig. 3C retracts after severing event). While branches are retracting distally they may be severed at more proximal sites. Interestingly, severed branches can retract at both ends before undergoing fragmentation.

Because branch thinning precedes proximal severing, we examined the cytoskeleton in this region (Fig. 4A) during pruning using 22C10, a monoclonal antibody against the microtubule-associated protein Futsch(Hummel et al., 2000). The proximal branches of ddaC in third instar larvae have a uniform caliber(Fig. 4B) and even staining with 22C10 (Fig. 4B′). Following the onset of metamorphosis, the proximal branches thin and generate swellings (Fig. 4C). Futsch is redistributed showing clear gaps in staining where branches are constricted,and high levels in the beads (Fig. 4C′). The emergence of filopodia on these branches indicates that the actin cytoskeleton is also being locally remodeled.

To examine the dynamics of the microtubule cytoskeleton, we made time-lapse movies of tub::GFP in ddaC (Movie 3 in supplementary material; Fig. 4D). At 0 h APF proximal arbors show uniform distribution of tub::GFP. By 2 h APF tubulin is redistributed and non-uniform. By 4 h APF, there is significant loss of tub::GFP in some regions while it is aggregated in large beads in other regions. These data show that a local loss of the microtubule cytoskeleton is associated with regions that undergo dendritic thinning.

The pruning of da neuron dendrites occurs in the broader context of the dismantling of a larval body through the actions of the steroid hormone ecdysone and its metabolites. To establish the relative contributions of cell-autonomous versus non cell-autonomous mechanisms, we blocked ecdysone signaling in the neurons using a dominant negative version of the ecdysone receptor (EcR^DN).

When EcR^DN is expressed in ddaA, ddaF and ddaB, cell death is blocked and the dendrites of these cells remain intact(Fig. 5B). Similarly ddaD and ddaE fail to prune properly; no severed branches or cellular debris are evident, and the proximal arbor does not `thin'(Fig. 5B). Instead the distal tips of the dendritic arbors of these neurons show bulb-like swellings,similar to retraction bulbs (Fig. 5E,F).

Time-lapse movies of EcR^DN-expressing ddaD and ddaE (Movie 4 in supplementary material) reveal how branches are removed(Fig. 5G). At 16 h APF the arbor shows features described above with prominent retraction bulbs forming on the distal tips of dendrites. By 24 h 10 min APF the branches have shortened and the retraction bulbs have increased in size. The subsequent frames show a slow but steady progression of retraction towards the cell body. One branch in ddaE (Fig. 5G, 25 h 30 min) was seen to sever during retraction. This was the only severing event seen in five movies and it was not preceded by thinning. Thinning was never seen in the proximal region of the dendritic arbor, and even at 64 h APF EcR^DN-expressing neurons still possess vestiges of larval branches and show no sign of adult outgrowth (data not shown).

Although the expression of EcR^DN suppressed branch severing and local degeneration, the branches show a retrograde retraction. This retraction may be due either to an ecdysone-induced cellular program that is not blocked by EcR^DN or to the neurons' response to changes in the epidermis,i.e. the substrate on which they arborize. Consequently, we examined the changes taking place in the epidermis during early metamorphosis.

The dendrites of ddaD and ddaE arborize over a discrete region of the dorsal larval epidermis. In the early pupa the abdominal epidermis consists of two populations of cells, large polyploid larval cells and small diploid cells of the imaginal histoblast nests (Fig. 6A). To visualize how the epidermal target remodels relative to the pruning neuron, we used a histone GFP fusion line. Movie 5 (in supplementary material) shows the cells of the dorsal histoblast nests beginning to spread dorsally. Fig. 6 shows how a larval epidermal cell is consumed by phagocytic blood cells. At 15 h 00 min the large nucleus of the larval cell is intact and the diploid nuclei of phagocytic blood cells are close by. By 17 h 20 min the cell has been consumed by the phagocyte. The remnants of the larval cell (i.e. nuclear GFP) are within the phagocyte and allow it to be tracked. Fig. 6C summarizes the fate of 106 larval epidermal cells imaged in Movie 5. The 17 cells (i.e. 16% of population) labeled with orange dots die by 25 h APF and 70% of those cells are found distant from the histoblast migration front. Movie 5 also shows that after a larval cell is removed, the epidermis reorganizes to compensate for the gap. Hence, during the time that the da neurons are pruning back their dendrites, the underlying epidermis is in a state of flux as larval cells are being removed throughout the domain of innervation. This removal of larval cells may play a role in directing the behavior of distal dendrites during this period.

Besides their role in removing dead and dying cells during metamorphosis,blood cells appear to be intimately involved in pruning dendrites. Migratory cells labeled with GFP were found in most of the pupae with GFP-expressing da neurons (`ph' in Fig. 3A; 24 h 00 min and Movie 2). Fig. 7Ashows a faintly labeled phagocyte entering the field from the top right. By 19 h 40 min a severed branch from ddaE shows beading and the phagocyte moving into close proximity. At 19 h 50 min the branch fragments, and by 22 h 00 min has disappeared and the GFP has been transferred to the phagocyte through engulfment of labeled dendrite fragments. In Fig. 7B such a phagocyte is co-labeled with the acidotropic marker LysoTracker. These cells are not glia,since they did not label with repo-GAL4 but instead express hemolectin-GAL4, a phagocytic blood marker (data not shown).

Phagocytes also appear to play an active role in severing intact dendritic branches. Such attacks can occur at the distal tips of branches or at proximal sites near the cell body. Fig. 7C shows selected frames from a time-lapse sequence where the distal part of the primary branch of ddaE is being attacked by phagocytes. At 16 h 00 min a phagocyte enters the field from the top right and comes into contact with the distal tip of ddaE. By 17 h 00 min the distal part of the branch is gone and the phagocyte appears to be labeled with more CD8::GFP. At 17 h 00 min another phagocyte moves from the left field. This phagocyte attacks the end of the primary branch as it retracts.

Phagocytes can also attack proximal regions of the dendritic arbor. Fig. 7D shows a time-lapse sequence (movie 6) from 15 h 40 min APF focusing on interactions between a phagocyte and the proximal region of ddaE. Between 15 h 40 min and 17 h 30 min the phagocyte moves close to the pruning arbor. Between 18 h 10 min and 18 h 20 min the branch is severed where the phagocyte is located. A small bleb of GFP derived from the pruning branch appears to be internalized by the phagocyte at severing (18 h 30 min).

We found the behavior of phagocytes to differ if they encountered neurons expressing EcR^DN. Fig. 8A shows a typical phagocyte from a wild-type animal(C161-GAL4>CD8::GFP). The path of the phagocyte shows it tracking around a proximal branch of ddaD, which is undergoing fragmentation, then after severing the primary branch it moves to the right and attacks the retracting branch of ddaE. In contrast a phagocyte found near the proximal arbor of neurons expressing EcR^DN(Fig. 8B) moves slowly across the field, showing little pseudopod activity, and fails to attack the proximal arbor. The phagocyte on the left in Fig. 8B follows a retracting distal tip, through the 8h 20 min of the movie, and removes pieces of the retraction bulb as the branch retracts (also see Movie 4 in supplementary material).

The abdominal sensory system of the adult fly is a mosaic consisting of postembryonic neurons derived from the imaginal histoblast nests and genital disk (Hartenstein and Posakony,1989; Taylor,1989), and a subset of larval sensory neurons of embryonic origin(Williams and Shepherd, 1999; Williams and Truman, 2004). At the onset of metamorphosis, larval sensory neurons either die or survive and remodel. Among the dorsal neurons of abdominal segments 2-5, we found that ddaA, ddaF and ddaB die whilst ddaC, ddaD and ddaE survive.

Programmed cell death is a near ubiquitous phenomenon in the vertebrate nervous system and essential for proper morphogenesis(Burek and Oppenheim, 1996),Surprisingly, most studies have focused on the cell body and paid little attention to the fate of neuritic processes, even though the axons of most projection neurons have reached their targets before the onset of cell death(Perry et al., 1983).

We used time-lapse imaging to observe the dendrites of dying neurons in intact animals. The distal branches of the ddaF develop swellings at 2 h APF that become distinct beads, which accumulate tubulin. Between 3 and 4 h APF branches undergo severing at multiple sites where they have thinned. This beading propagates through the arbor in a proximal to distal direction with proximal changes occurring 20-30 minutes ahead of the distal ones(Fig. 2C). This same progression is seen in ddaB but with a lag of 1 hour. A similar degeneration is found in the peripheral arbors of Rohon-Beard neurons, which also thin and fragment when dying (Reyes et al.,2004).

The destruction of peripheral processes has been most widely studied following acute local trauma, where the axon branch distal to the lesion undergoes a stereotyped degeneration by fragmentation, termed Wallerian degeneration (Waller, 1850; Beirowski et al., 2005). There are similarities between the degeneration of dendrites in our study and the axon pathology seen in Wallerian degeneration. Firstly, in vitro models of Wallerian degeneration reveal a proximal to distal spread of beading in the separated distal neurites (Sievers et al.,2003). Secondly, the microtubule cytoskeleton is disrupted following transection (Zhai et al.,2003) and the beaded morphology and redistribution of tubulin we see in the da neuron dendrites resembles that seen in neurons treated with microtubule depolymerizing drugs (He et al., 2003).

We found that the expression of EcR^DN in neurons that were fated to die blocked their death and prevented the local degeneration seen between 3 and 6 h APF (data not shown). This prompted us to ask whether the degeneration normally observed is due to cell death directly or is the result of a hormone-induced microtubule disassembly program running in parallel with cell death. When we block cell death by expressing p35, the dendrites showed no signs of degeneration by 5 h APF (Fig. 2D), yet by 18 h APF their arbor was largely removed(Fig. 2E). Thus, with p35, the dendrites of these doomed cells prune at the same time and in the same way as ddaD and ddaE (data not shown), suggesting that under normal conditions, death directly causes the dendrite degeneration observed in ddaA, ddaF and ddaB at 3-6 h APF. Similar observations made after killing da neurons by laser ablation support this proposition(Williams and Shepherd,2002).

Pruning of the dendrites of ddaD and ddaE starts between 6 and 10 h APF and is largely complete by 24 h APF (Fig. 1). Our time-lapse movies reveal that dendrites undergo deconstruction by two different cellular mechanisms: local degeneration(Watts et al., 2003) and branch retraction.

During local degeneration, branches are detached from the main arbor and undergo fragmentation (Fig. 3A,E). Watts et al. (Watts et al., 2003) found that axons and dendrites of mushroom body γneurons undergo local degeneration: between 4 and 6 h APF the processes undergo blebbing, and by 8 h APF most of the dendritic arbor has been removed. By 18 h APF the axons have fragmented, and during that time there was no evidence to suggest that γ neurons undergo retraction. Similarly, larval projection neurons in the olfactory lobe undergo local degeneration(Marin et al., 2005).

A key step in local degeneration appears to be severing, which we were able to visualize with time-lapse movies. Severing happens in one of two ways,depending on position within the arbor. In proximal regions severing is preceded by thinning (Fig. 3B). Once a branch has thinned, it severs and the stump retracts, while the separated arbor undergoes fragmentation. We found that beads adjacent to thinning regions contain abundant Futsch and tub::GFP, whereas these are lacking in the thinned regions, suggesting that the bulk of the microtubule cytoskeleton is lost in these focal regions. This supports observations by Watts et al. (Watts et al.,2003) that myc-marked tubulin disappears from pruning γneurons before the axon is lost. Similarly, Bishop et al.(Bishop et al., 2004) showed that just prior to shedding `axosomes', the proximal region of a retracting axon becomes devoid of organized microtubules. When neurons express EcR^DN, the caliber of the proximal branches does not change,suggesting that a redistribution of the microtubule cytoskeleton fails and proximal severing is subsequently blocked.

The second type of severing occurs at more distal sites within the arbor(Fig. 3C). Here there is no thinning and only occasional beading; thus it is possible that different mechanisms are responsible for severing at distal sites. Nevertheless, distal severing is also suppressed in neurons that express EcR^DN,suggesting that a destabilization is also important here.

After severing, both proximal and distally detached branches undergo fragmentation. The branches thin whilst beads form along their length(Fig. 3D) and break at multiple sites, generating GFP blebs that are removed by phagocytes (see below). Although most of the events observed in branch fragmentation during pruning appear similar to those seen in the arbors of dying da neurons, an exception is the appearance of filopodia. When we have simultaneously imaged the microtubule cytoskeleton and the membrane in pruning neurons, we often find filopodia are coincident with the areas of microtubule disorganization(Fig. 4C). Bray et al.(Bray et al., 1978) found lateral filopodia when they applied colcemid to chick neurons in culture. Filopodia do not appear on the arbors of the doomed cells as the cells are undergoing programmed cell death. However, when cell death is blocked by expression of p35 these cells later show abundant filopodia as their dendrites are removed (Fig. 2E). This suggests a fundamental difference between the dendrite removal observed during the pruning of ddaD and ddaE, and that seen in cells that undergo programmed cell death.

The primary branches of ddaD and ddaE can be retracted as shown in Fig. 3E. These retraction events occurred in 6/17 neurons imaged and the branches of such neurons often possessed many filopodia. When retraction events happen, local degeneration may still occur in secondary branches and on other primary branches nearby. It is somewhat problematic to apportion what percentage of pruning is due to retraction and what is due to severing and fragmentation, as they both appear to happen simultaneously, e.g. where a branch is in the process of distal retraction and suddenly is severed at a more proximal site.

The dendrites of neurons expressing EcR^DN form bulb-like structures on their distal tips and produce few if any filopodia. Local degeneration is blocked in these neurons, but they do eventually remove their dendrites by retraction. A number of scenarios could explain this retraction phenotype. It is possible that EcR^DN does not block all ecdysone signaling and so the cell intrinsic ecdysone pruning program has not been entirely eliminated. It is also possible, but unlikely, that the EcR^DN results in a non-specific `neomorphic' as the specificity of this reagent has been demonstrated by Cherbas et al.(Cherbas et al., 2003). The other possibility is that EcR^DN completely blocks ecdysone signaling and that another parallel intrinsic signaling pathway plays a role in dendrite retraction. It is most likely that EcR^DN completely blocks ecdysone signaling and that the phenotype we see reveals extrinsic factors that are important for dendrite pruning under wild-type conditions. The neuron-specific expression of EcR^DN means that ecdysone signaling is only disrupted in the neuron and that the local environment can undergo its normal hormone induced changes.

A dorsal region of the epidermis is the target for the dendrites of ddaD and ddaE. Two potential interactions could be taking place between the target and pruning arbor. The epidermis could be sending an instructive signal causing the arbor to prune or there could be a passive loss of epidermal contacts as the larval epidermis is replaced. To explore the latter possibility we made movies at the time when ddaD and ddaE are pruning. It has been widely held that the larval epidermal cells are only removed by the migrating front of adult epidermal cells derived from the histoblast nest(Madhavan and Madhavan, 1980; Fristrom and Fristrom, 1993). To our surprise we found that during early metamorphosis larval cells are removed at sites distant from the histoblast migration front(Fig. 6B). The epidermal cells are removed by phagocytic blood cells and the cells that are removed do not show obvious signs of apoptosis before being contacted by the phagocyte. This suggests that the larval cells are killed by a phagocytosis-induced cell death as described in C. elegans(Reddien et al., 2001) and Drosophila (Mergliano and Minden,2003). Importantly, when a larval cell is removed, other cells move to compensate for its loss. Thus the epidermis becomes a dynamic substrate during prepupal and early pupal development and this movement may result in shearing events in distal dendritic branches and could explain the phenomenon of distal severing without thinning. Likewise, the disruption of neuron-epidermal contacts may contribute to the retraction of primary branches seen in an extreme form when neurons express the EcR^DN.

In flies with GFP-expressing neurons we found that phagocytic blood cells also became labeled, which in turn allowed us to visualize their behavior in early pupae. Our time-lapse movies reveal that phagocytes are intimately involved in the fragmentation and engulfment of severed dendritic branches. The sequence in Fig. 7A shows a faintly labeled phagocyte coming into close proximity with a severed and beaded branch. After contact the branch fragments and the phagocyte consumes the debris, thereby becoming strongly labeled with GFP. The degenerating branches of dying da neurons share the same fate, as we see GFP blebs move off in random directions following fragmentation because of their movement in phagocytes. In the vertebrate nervous system, cells with phagocytic capacity play a scavenger role, clearing up the cellular debris that results from Wallerian degeneration and normal developmental cell death(Hirata and Kawabuchi, 2002; Mallat et al., 2005).

During early metamorphosis, glial processes infiltrate the mushroom body neuropil when axons are undergoing local degeneration(Awasaki and Ito, 2004; Watts et al., 2004). Ultrastructure studies show that remnants of labeled neurons appear in these glia (Watts et al., 2004). Awasaki and Ito (Awasaki and Ito,2004) also suggest that glia play an active role in the destruction of the axons, since inhibiting endocytosis in the glia stopped infiltration and blocked axon pruning. Our time-lapse data also reveal that in the periphery phagocytic blood cells also attack intact dendritic branches. As shown in Fig. 7B,C, phagocytes both attack the distal tips of retracting branches and sever branches at proximal positions. We do not observe the phenomenon of proximal severing by phagocytes in every movie, which suggests that either severing can result from other forces such as the shearing action of the reorganizing epidermis, or we are not seeing all the phagocyte/neuron interactions because only a fraction of phagocytes are labeled. Nevertheless our data shows that phagocytic cells have an active role in the pruning of peripheral neurons.

During the clearance of apoptotic cells, phagocytes are known to recognize specific `eat me' signals on the surface of a dying cell(Savill and Fadok, 2000). Are the phagocytes recognizing similar `bite me' signals on the arbors of the pruning da neurons? Our observations suggest that they are targeting specific regions of branches that show signs of destabilization. Targeting of destabilized regions is especially obvious from the behavior of phagocytes encountering neurons expressing EcR^DN. These phagocytes ignore the stable proximal regions of the arbor and instead only attack the distal tips(Fig. 8).

Taken together, our data show that the pruning of the da neuron dendrites during metamorphosis is achieved by local degeneration and branch retraction. We propose that these phenomena are controlled by both intrinsic and extrinsic cellular mechanisms (Fig. 9). The initiation of pruning by the steroid hormone ecdysone results in the destabilization of the microtubule cytoskeleton within the dendritic arbor. This loss of microtubules results in the severing of the branches, either by attacking phagocytes or possibly from shearing forces produced in the remodeling epidermis. Severed dendritic branches undergo a distinctive`beading', similar to redistribution of the microtubule cytoskeleton found in the arbors of dying cells. Phagocytes are intimately involved in the fragmentation and consumption of severed branches, which leads to a rapid clearance of cellular debris.

We expect that the mechanisms described here are not restricted to the dendrites of Drosophila sensory neurons undergoing remodeling during metamorphosis. It will be very interesting to learn which mechanisms used by developing cells to remove their neuritic processes are employed by cells following trauma or the onset of disease.

We would like to thank Wayne Johnson, Lucy Cherbas, Allan Spradling and the Bloomington for fly stocks, the DSHB for monoclonal antibodies, Sara Lennox for technical assistance, Margrit Schubiger, Janet Altman and Lisa Marin for reading the manuscript and Sanjay Sane for helpful discussions. Research was supported by NIH grant NS13079.