An ESCRT module is required for neuron pruning

Neural circuits are refined by both functional and structural changes. Structural remodeling by large-scale pruning occurs where relatively long neuronal branches are cut away from their parent neuron and removed by local degeneration. Until now, the molecular mechanisms executing such branch severing events have remained poorly understood. Here, we reveal a role for the Endosomal Sorting Complex Required for Transport (ESCRT) machinery during neuronal remodeling. Our data show that a specific ESCRT pruning module, including members of the ESCRT-I and ESCRT-III complexes, but not ESCRT-0 or ESCRT-II, are required for the neurite scission event during pruning. Furthermore we show that this ESCRT module requires a direct, in vivo, interaction between Shrub/CHMP4B and the accessory protein Myopic/HD-PTP.

Neural circuits are refined by both functional and structural changes. Structural remodeling by large-scale pruning occurs where relatively long neuronal branches are cut away from their parent neuron and removed by local degeneration. Until now, the molecular mechanisms executing such branch severing events have remained poorly understood. Here, we reveal a role for the Endosomal Sorting Complex Required for Transport (ESCRT) machinery during neuronal remodeling. Our data show that a specific ESCRT pruning module, including members of the ESCRT-I and ESCRT-III complexes, but not ESCRT-0 or ESCRT-II, are required for the neurite scission event during pruning. Furthermore we show that this ESCRT module requires a direct, in vivo, interaction between Shrub/CHMP4B and the accessory protein Myopic/HD-PTP. N euronal remodeling is a fundamental mechanism that is essential for building a functional nervous system 1 . During development individual neurons grow exuberantly, often generating many more branches than required. A significant number of aberrant or redundant branches are pruned back by branch-specific local degeneration events during a period of circuit maturation 2,3 .
In recent years, studies in Drosophila have contributed greatly to our understanding of the cellular and molecular mechanisms of large-scale pruning. These studies have revealed that TGFb and Ecdysone signaling pathways initiate pruning and that the Ubiquitin-Proteasome System, caspases, hdc and calcium signaling events are required, along with Kat60L, IKK and Mical for pruning to proceed to completion 4 . All of these activities lead up to the point at which a branch is cut away from the parent neuron. How this final scission event is controlled and what cuts the nerve cell membrane is currently unknown.
To address this question we looked at how similar membrane bending and scission events occur within cells. The Endosomal Sorting Complex Required for Transport (ESCRT) machinery has been shown to execute membrane-cutting events during intraluminal vesicular formation, viral budding, membrane repair and cytokinetic abscission [5][6][7][8][9] . The ESCRT proteins were first identified in yeast as class E vacuolar sorting (Vps) proteins and they are functionally conserved from yeast to human 10 . The ESCRTs form four heteromeric protein complexes named ESCRT-0, -I, -II, and -III involved in cargo recognition and sorting as well as membrane bending and cutting. Each of these processes requires the sequential assembly of a specific ESCRT module, composed of proteins from different ESCRT complexes, as well as associated adaptor proteins 11 . For example, during final stages of cytokinesis, the ESCRT-I component TSG101 and the Bro-domain adaptor protein Alix, are recruited to the midbody, where they, in turn, recruit the ESCRT-III CHMP4B, a key player in the scission machinery 12 .
Here we explore and identify a requirement for the ESCRTs during the developmental pruning of axons and dendrites in Drosophila.

Results
A subset of ESCRT proteins are required for pruning. We performed a candidate RNAi screen to knock down multiple representatives from each of the four ESCRT complexes (-0, -I, -II and -III) and ESCRT related proteins in the class IV dendritic arborizing (da) sensory neuron, ddaC ( Fig. 1a and b) 13 . These neurons were specifically imaged with the pickpocket-GAL4 driver (ppk1.9-GAL4) at 18 h After Puparium Formation (APF) to assess pruning (Fig. 1b0). We tested twenty-five RNAi lines targeting fifteen of the 18 genes forming the ESCRT machinery in Drosophila (Fig. 1c).
The downregulation of five ESCRT related genes, TSG101, Shrub, Vps4, mop and UBPY, by expressing 9 different RNAi lines, lead to severe disruptions in pruning with dendrites still being attached to the cell body by 18 h APF, resulting from a lack of branch severing ( Fig. 1c and 2b, g, i, k and l). These experiments demonstrate the involvement of the ESCRT machinery during pruning in a cell autonomous manner.  The loss of ESCRT-0 subunit Hrs led to minor delays in pruning ( Fig. 1c) and none of the three subunits of ESCRT-II (Vps22, 25 and 36) displayed a pruning defect ( Fig. 1e-g). These data suggested the interesting possibility that only ESCRT proteins belonging to ESCRT-I and -III are required for pruning.
As these experiments were obtained using RNAi based tools, we attempted to confirm these results using a modified MARCM clonal analysis 14 that allows us to visualize two control neurons alongside the homozygous mutant neurons in the same animal. This MARCM clone analysis allows the cell autonomous requirement of the ESCRT genes to be unequivocally tested. We generated MARCM clones mutant for Hrs, Vps22, Vps25 as well as double mutant for both Hrs and Stam (removing both ESCRT-0 subunits) ( Fig. 1h-k). All the homozygous mutant neurons remodeled like wild type neurons, revealing that these ESCRT-0 and ESCRT-II subunits are not required for pruning. These homozygous mutant clones show normal regrowth following pruning, generating numerous filopodia on their emerging growth cones.
Alongside this, MARCM clones against Vps28 (Fig. 2m) confirmed that ESCRT-I is necessary for pruning, as the strong severing defect observed with TSG101 RNAi had suggested ( Fig. 1c and 2b). For the ESCRT-III, we focused on Shrub the CHMP4B homolog in flies 15 . Shrub is a key component of the scission machinery and forms the oligomeric helical filament that bends membranes together to allow cutting to take place 16 . MARCM analysis of neurons homozygous mutant for Shrub reveal strong defects in severing, with dendrites still attached to the soma by 18 h APF (Fig. 2n).
Together multiple RNAi knockdown and MARCM clone data with null and strong alleles show that ESCRTs are required, in a cell autonomous manner, for pruning. The requirement of a specific subset of ESCRT including components of ESCRT-I, -III and Vps4 suggest the existence of an ESCRT pruning module.
Myopic accessory protein is required for pruning. The ESCRT genes are expressed ubiquitously and the activity of their proteins needs to be tightly regulated to allow them to undertake different roles within the cell. One aspect of this regulation is their modular assembly, i.e. the context-specific generation of discrete combinations of ESCRT proteins and their accessory proteins 7,12 . The ESCRT-accessory proteins play a key role in recruiting and stabilizing ESCRT subunits. During cytokinesis the ESCRT-III complex is recruited to the midbody by the Bro-domain ESCRT-accessory protein Alix [17][18][19] . In our screen none of the RNAi reagents targeting Alix lead to defects in pruning (Fig. 2j). To explore this further we generated single cell mutant MARCM clones homozygous for Alix LL05494 and also found no disruption in pruning (n 5 8)   (Fig. 2o). These data suggest that another accessory protein is required for ESCRT-III mediated severing in pruning neurons. Myopic (mop), the homolog of the human gene HD-PTP is another Bro-domain protein found in the fly genome. Mop is known to be involved in ESCRT-mediated receptor sorting 20,21 . When we knocked down the expression of mop with two independent RNAi lines, we observed a robust block of dendrite severing in ddaC ( Fig. 1c and 2k) revealing it to be a key ESCRT-accessory protein required for pruning. Shrub function during pruning requires a direct interaction with the accessory protein Mop. Since shrub and mop are required for pruning, we next investigated how these two proteins interact with each other and whether a direct interaction between them is necessary for their function. Previous work identified that HD-PTP (the human homologue of mop) physically interacts with CHMP4B (the homologue of shrub) through the Bro1 domain of HD-PTP and that this interaction can be disrupted by introducing the point mutations L202D and I206D 20 . We confirmed that this interaction is conserved between the Drosophila Shrub and Mop using a yeast-two-hybrid assay (Fig. 3a) and that the disruption of the conserved amino acid I201D and V205D within the Bro1 domain of Mop (equivalent to the human L202 and I206) is sufficient to inhibit this interaction. To test if this same interaction is relevant in vivo and required for branch severing, we generated transgenic flies expressing wild-type or mutant forms of mop (I201D and V205D) at similar levels ( Fig. 3i-m). Each was used in turn to test whether they could rescue pruning in a mop-RNAi background (Fig. 3f). We found wild-type Mop was able to fully rescue the RNAi severing phenotype in ddaC (Fig. 3g) but the Bro-domain mutant Mop could not (Fig. 3h). Together, these results show that branch severing requires a physical interaction between Mop and Shrub, which is dependent on the Bro domain of Mop.
ESCRT function is required for pruning of other classes of neurons. To determine if ESCRT function during pruning is also conserved in other type of neurons, we focused on Shrub and Mop. We found that shrub is required for dendrite severing in class I da sensory neurons ( Fig. 4a and b) 22 and also for severing the axons of the mushroom body c-neurons within the central nervous system (Fig. 4d and e) 23 . Mop-RNAi recapitulates the strong severing defect seen with shrub loss of function revealing a new role for mop in branch severing ( Fig. 4c and f). We conclude that Shrub and mop are key players in pruning and that ESCRT function is required, in a cell autonomous manner, for pruning in different classes of neurons in both the peripheral and the central nervous system of Drosophila.
ESCRT pruning defects are not due to early developmental disruption. Previous work showed that shrub loss of function disrupts the development of da neuron arborizations and leads to the eponymous bushy 'shrub' phenotype 15 . To determine if these early developmental defects in arbor growth could be responsible for the observed disruptions during pruning we utilized a GFP tagged Shrub (shrub::GFP) that behaves as a dominant negative 15 . This construct provided us with the opportunity to measure the level of Shrub expression through monitoring the GFP. When shrub::GFP is expressed continuously in ddaC neurons we observe both a 'shrub' branching phenotype at 0 h APF (Fig. 5a) as well as disruption in pruning at 18 h APF (Fig. 5b). If we delay the onset of expression shrub::GFP to late wL3 stage, the dendritic arborization is wild-type at 0 h APF ( Fig. 5c) but we still find strong severing defects, n 5 15 (Fig. 5d). Alongside this, we find that when mop-RNAi is expressed throughout development, the dendritic arborization of ddaC does not show a shrub-like branching phenotype at 0 h APF (Fig. 5e), but does have a robust severing defect at 18 h APF (Fig. 5f). Taken together, these data show that disruptions in pruning can be decoupled from the early developmental defects of ESCRT loss of function.
To determine whether the initiation of pruning takes place on schedule we also looked at the expression of the Ecdysone Receptor (EcR), and Sox14 24,25 , two key genes that are upstream components in the pruning cascade. We found that ddaC neurons expressing mop-RNAi had same onset and same levels of expression of both transcription factors as adjacent wild-type da neurons at 0 h APF ( Fig. 5g  and h).
Taken together these data show that the gating of pruning proceeds normally and that the severing defects are the result of a requirement for the ESCRTs function during pruning rather than from some earlier developmental sequela.  In wild-type neurons we found ubiquitinated proteins are rapidly degraded and are almost undetectable (Fig. 6a) 26,27 . We then analyzed the accumulation of ubiquitinated proteins in ddaC neurons where we had down regulated specific ESCRT subunits using RNAi and also compared their pruning. For all the subunits tested, we found a robust accumulation of ubiquitinated proteins revealing ESCRT dependent defects in MVB formation (Fig. 6b-e). With RNAi against the ESCRT-0 subunit Hrs we observed that ubiquitinated proteins accumulate in the cell body but show no severing defects at 18 h APF and only a very mild delay in the clearance of severed branches ( Fig. 6b and b9). Similarly, when we knock down Vps25, we found the accumulation of ubiquitinated protein and no pruning defect ( Fig. 6c and c9). In contrast, shrub and mop RNAi lead to an accumulation of ubiquitinated proteins in the ddaC neurons, as expected from their requirement in MVB biogenesis and also show robust disruptions in severing with many dendritic branches still attached at 18 h APF (Fig. 6d-e9). These data suggest that the MVB compartment may be involved in pruning but disrupting its function is not sufficient to cause the severing defects observed with shrub and mop knockdown. These data confirm the importance of the ESCRT machinery in controlling MVB biogenesis in neurons but also appear to decouple MVB cargo processing and branch severing. This raises the possibility that the ESCRTs may play an additional role in pruning that is independent of MVB biogenesis.
Could the ESCRT machinery play a local role within dendrites during pruning? Live imaging studies of pruning in ddaC neurons show that the proximal branches remodel their cytoskeleton, become thin and then sever 22 . A closer analysis of the severing defect shows that a number of clones homozygous mutant for Shrub ( Fig. 7a-b) had a very thin membrane tether connecting the distal dendrites to the cell body ( Fig. 7a9-b9). This change in the calibre is in stark contrast to neurons in which EcR or Sox14 function has been disrupted by RNAi, where the diameter of the proximal dendrites remain similar to those of larval neurons (Fig. 7c-d9). Following these observations, we wondered whether the ESCRT machinery could be acting locally to cut the dendritic membranes and that these very thin tethers were unresolved scission events.
To further explore the idea of a local action, we examined the localization of Shrub in ddaC neurons at 6 h APF while pruning is occurring. We found Shrub present in the cell body, as would be expected for its role in the MVB biogenesis, but also found it within the proximal dendrites ( Fig. 7e-e9). The distribution of Shrub within the proximal dendrites was not homogenous, but localized to varicosities and in small puncta within the thinner parts of the dendrites, in the region where the severing events will occur (Fig. 7e-e9). We also analyzed the localization of shrub::GFP in ddaC neurons during pruning ( Fig. 7g-g9) and found it mirrored the antibody staining.
From the interaction study, we hypothesized that the accessory protein Mop should be spatially and temporally localized with Shrub. Using a Flag-tagged version of Mop 21 , we observed a robust co-localization of Mop with Shrub::GFP during the early phases of ddaC pruning (Fig. 7f-f9) consistent with Mop playing a local action to recruit/stabilize Shrub and the ESCRT pruning module. In contrast to this, we found that Vps36::GFP, a member of ESCRT-II, required for MVB formation but not for pruning, was localized only within the cell body and not in the proximal dendrites (Fig. 7h-h9). Altogether, these data provide support for a local action of an ESCRT pruning module, requiring Shrub and Mop.

Discussion
The normal functioning of the nervous system requires appropriate matching between signaling and receiving cells. This 'matching' of network components is achieved by progressive developmental phenomena, such as cell division and cell growth, and regressive phenomena, such as programmed cell death and pruning. Large-scale pruning, where relatively long neuron branches are removed, happens not by a distal to proximal retraction event but by a local degeneration. How such highly ordered, branch-specific autodestruction events are orchestrated still remains unclear.
The proteins of the ESCRT family are highly conserved from yeast to human 28 with a striking functional conservation of the cellular roles they play in multicellular organisms i.e. MVB formation and cytokinesis 9,28 . Here, we describe a RNAi screen to study the involvement of the ESCRT machinery during pruning. Functional knockdown within neurons of many members of this large family showed clearance or severing defects. As partial loss of function or off target effects can be an issue when using RNAi tools we used mulitiple RNAi reagents and also followed up candidates by generating MARCM clones with previously characterized null or strong alleles of the ESCRT proteins.
RNAi reagents against Hrs were not sufficient to allow us to conclude whether it has a role during pruning. The absence of any phenotype in MARCM clones with Hrs D28 , a strong allele, alone or in combination with the null allele stam 2L-2896 confirmed that none of the subunits of the ESCRT-0 are required for pruning 29,30 . Likewise, RNAi against ESCRT-II subunits Vps22, 25 were confirmed by MARCM clonal analysis, suggest that these subunits are not involved in pruning. Combined with the absence of phenotype in Vps36 RNAi, the third subunit of the ESCRT-II, these data indicate that ESCRT-II is not required for pruning. On the other hand, MARCM analysis of Vps28 and Shrub confirmed the cell autonomous requirement of ESCRT-I and -III, for pruning.
Together, we analyzed the loss of function of 17 ESCRT and ESCRT accessory proteins on neuron pruning. Our data reveal a novel combination of the ESCRT components that forms an 'ESCRT pruning module' necessary for branch severing. This module is composed of components of ESCRT-I, ESCRT-I together with Vps4 and the accessory proteins Mop and Ubpy. The module does not appear to require ESCRT -0 and -II function.
The loss of function of ESCRT genes led to both clearance and severing defects. The clearance defects we observed in this system could be the result of some disruption to the engulfment machinery or due to a delay in severing branches. Nevertheless, as the RNAi and MARCM clones are generated specifically in the neurons we decided to focus on the severing process. More specifically, we focused our attention on Shrub and Mop, as Shrub is a key component of the ESCRT-III scission machinery which forms a helical oligomeric filament that brings together membranes to allow scission to take place. The formation of the Shrub filament requires the assembly of different combinations of ESCRT complexes depending on the particular cellular process in question e.g. membrane repair, MVB, cytokinetic abscission and viral budding. The ESCRT-accessory proteins play a key role in recruiting and/or stabilizing these specific complexes. Our RNAi screen reveals that Mop is likely to be an important accessory protein required for pruning. We found that the in vivo physical interaction seen between the human ESCRT-III protein CHMP4B and the accessory protein HD-PTP 20 is conserved between the Drosophila homologues, Shrub and Mop. Furthermore, our analyses of other da sensory neurons and c-neurons of the mushroom body reveal that Shrub and Mop are required for both dendrite and axon pruning. Taken together these data suggest that the 'ESCRT pruning module' may be part of a universal machinery used for the structural remodeling of neurites.
A key question raised by the work is whether the disruptions to pruning we observe are due to a pleiotropic dysregulation of development or to a more specific event during pruning. The analysis of shrub::GFP and mop RNAi reveal that the pruning defect seen with ESCRT loss of function are independent from the earlier developmental disruption they can induce. Furthermore, the expression of EcR and Sox14, two markers of pruning initiation, were normal in mop RNAi neurons. These results reveal a role for the ESCRTs in ddaC neurons during metamorphosis.
How is a dysregulation of the ESCRT machinery actually disrupting pruning? Two recent papers forward the hypothesis that the ESCRT dependent disruption of pruning is through its role in MVB biogenesis and the processing of cell surface receptors 31,32 . Zhang et al showed that Vps28, Vps32 and Vps4 are required for pruning in ddaC 31 . These three ESCRT subunits belong to what we call the ESCRT pruning module, thus corroborating our findings. Our analyses show that these subunits are important in MVB formation and processing ubiquitinated proteins. Interestingly, we observed that other subunits of the ESCRT-0 and -II machinery, Hrs or Vps25, also affect MVB formation and ddaCs ability to process ubiquitinated proteins. We noticed different patterns of accumulation of the ubiquitin staining depending on the ESCRT complex analyzed. These variations could be due to the stage of MVB formation at which these subunits act i.e. Hrs at the beginning, Vps32 and Vps4 for the later steps. Importantly, we find that disruption in ESCRT-0 and -II do not lead to severing defects. This could be due to redundancy of the components, with one being able to compensate for the other, although we find severing occurs normally in the double mutant for Hrs/Stam. Another possible caveat could be that there is a perdurance of the wild-type protein within the single-cell mutant clones i.e. sufficient wild-type protein is transferred across into the homozygous mutant cell at the point of clone induction. This scenario also seems unlikely as 4 days have elapsed between the recom-bination event and branch severing, during which time the cell has grown in size by at least two orders of magnitude. Thus, our results point toward a decoupling between MVB formation and severing, suggesting that even though the ESCRTs may be involved in pruning through MVB biogenesis they may also be required through some other mode of action.
It has not escaped our attention that there is a striking topological similarity between neurite severing and the events taking place at cytokinetic abscission. It could be that the ESCRT pruning module is required locally to physically cut the neurites. We find that Shrub and Mop are co-localized within the proximal dendrites. This 'local hypothesis' could also explain the gossamer-thin tethers linking the distal dendrites to the cell body we observed in the ESCRT loss of function, a phenotype that we have never seen before in our studies of pruning. This structure is reminiscent of an unresolved intercellular bridge as seen when the ESCRTs is disrupted during cytokinetic abscission. Furthermore, there is a parallel between cytokinetic abscission and pruning in that both ESCRT-I and -III are involved but not ESCRT-II.
We found it appealing that the ESCRT machinery could be deployed for a very 'neuron-specific' function, like pruning, as until now most studies have focused on common cellular phenomena found in all cell types (Fig. 8). Hopefully improvements in imaging techniques along with genome engineering will soon enable us to observe the dynamics and localization of shrub within the dendrites during pruning to explicitly test this 'local hypothesis'. Future work exploring the detailed composition and assembly of this ESCRT 'pruning module' will provide important insights into the cellular events leading up to pruning.

Methods
Plasmid Construction. Shrub and the Bro domain of Mop (aa 1-389) were amplified by PCR from cDNA (pUAST-mop a gift from Jessica Treissman 21 ) using primers directed to the 59 and 39 ends of the coding sequence. Primers contained NotI sites to allow the insertion of the PCR product into pGBKT7 (Clontech) and pVP16 33 for the yeast two-hybrid assays. The point mutations I201D/V205D was introduced by PCR based site-directed mutagenesis. cDNA were then cloned into the phiC31 vector pUASTattB to allow directed transgenesis at two independent AttP sites, attP40 on 2L and VK00027 on 3R. The table summarises the deployment of ESCRT components and highlights that specific combinations of ESCRT proteins assemble depending on which membrane-cutting event is required [5][6][7][8][9] . Tick indicates that the specific ESCRT complex is involved, cross that it is not required and ? indicates that at present its role is not known. Staging of animals. Individual animals were collected at pupariation and maintained at 25uC in a Petri dish with moist filter paper. Staging was denoted as hours after puparium formation or APF.
Imaging, image analysis and quantification. The body walls of third-instar larvae, white pre-pupae, and pupae (until 12 h APF) were imaged directly using a Zeiss LSM 510 confocal microscope. After 12 h APF, pupae were peeled out of the pupal case. Between 15-40 optical sections at 1 mm intervals were taken for each neuron, and assembled in NIH ImageJ (http://rsb.info.nih.gov/ij/). Maximum projections generated. Images were adjusted only for brightness and contrast using Adobe Photoshop (Adobe Systems).