Sec71 functions as a GEF for the small GTPase Arf1 to govern dendrite pruning of Drosophila sensory neurons

In the developing nervous systems, neurons often extend excessive neurites and form superfluous connections at early stages. Subsequent removal of those exuberant or inappropriate neurites without causing the death of parental neurons, a process known as pruning, is crucial for the refinement of neural circuits at late developmental stages (Luo and O'Leary, 2005; Riccomagno and Kolodkin, 2015; Schuldiner and Yaron, 2015). Neuronal pruning is a conserved process widely occurring in both vertebrates and invertebrates. In vertebrates, many neurons in the neocortex, neuromuscular system and hippocampal dentate gyrus prune their unwanted neurites to control the proper wiring of the nervous systems (Bagri et al., 2003; O'Leary and Koester, 1993; Tapia et al., 2012). In invertebrates, such as Drosophila, the nervous systems undergo drastic remodeling during metamorphosis, a transition stage from a larva to an adult fly (Kanamori et al., 2015a; Truman, 1990; Yu and Schuldiner, 2014). In the central nervous system (CNS), mushroom body (MB) γ neurons prune their dorsal and medial axon branches as well as entire dendrites (Lee et al., 1999). In the peripheral nervous system (PNS), some dorsal dendritic arborization (da) neurons, ddaC, ddaD and ddaE, selectively eliminate their larval dendrites without affecting their axons (Kuo et al., 2005; Williams and Truman, 2005), whereas ddaF neurons are apoptotic during early metamorphosis (Williams and Truman, 2005). The pruning event involves both local degeneration and retraction (Luo and O'Leary, 2005), resembling neurodegeneration associated with brain injury and neurodegenerative diseases. Thus, a complete understanding of cellular and molecular mechanisms of developmental pruning would shed some light on pathological neurodegeneration following neurological diseases and injury.

In Drosophila, ddaC sensory neurons have emerged as an attractive model system with which to elucidate the molecular and cellular mechanisms of dendrite-specific pruning during early metamorphosis. In response to the steroid-molting hormone 20-hydroxyecdysone (ecdysone) at the late larval stage, ddaC neurons sever the proximal region of their dendrites and subsequently undergo rapid fragmentation of the severed dendrites, as well as dendritic clearance via phagocytosis (Fig. 1A) (Kuo et al., 2005; Williams and Truman, 2005). It has been well documented that the Ecdysone Receptor and its co-receptor Ultraspiracle are required to activate the expression of several downstream targets to initiate dendrite pruning (Yu and Schuldiner, 2014). We and others have recently reported the identification of endocytic pathways that are crucial for dendrite pruning (Kanamori et al., 2015b; Zhang et al., 2014). Rab5/Avalanche and ESCRT complexes, the components of the endocytic pathways, are required for downregulation of the L1-type cell adhesion molecule (L1-CAM) Neuroglian (Nrg) (Zhang et al., 2014). Nrg is drastically redistributed to endosomes and its protein levels are strongly downregulated prior to pruning, suggesting that massive Nrg endocytosis promotes dendrite pruning (Zhang et al., 2014). It is conceivable that Nrg endocytosis might be triggered by secreted ligands/signals through the secretory pathway. However, it is completely unknown whether the secretory pathway, an opposite route of the endocytic pathway, also plays a role in dendrite pruning of ddaC neurons.

The primary sites of the secretory pathway consist of the endoplasmic reticulum (ER), the Golgi apparatus and the trans-Golgi network in eukaryotic cells (Lippincott-Schwartz et al., 2000). Newly synthesized membrane proteins and lipids exit from the ER, pass through the Golgi complexes and are delivered to the plasma membrane via the post-Golgi exocytosis (Pfenninger, 2009). In developing neurons, the continuous addition of membrane proteins and lipids via the secretory pathway plays a key role in the outgrowth and elongation of dendrites and axons (Horton and Ehlers, 2004). Disruption of the ER-to-Golgi transport leads to the inhibition of dendritic or axonal growth in Drosophila sensory neurons and rodent hippocampal neurons (Aridor and Fish, 2009; Ye et al., 2007). In an attempt to isolate novel players of dendrite pruning, we carried out a large-scale RNA interference (RNAi) screen and identified ADP-ribosylation factor 1(Arf1), also known as ADP-ribosylation factor at 79F (Arf79F), as an important player for dendrite pruning in ddaC sensory neurons. Arf1 is a small GTPase and belongs to the Class I Arf family (D'Souza-Schorey and Chavrier, 2006; Gillingham and Munro, 2007). It has been reported that Arf1 can recruit COPI coat proteins on cis-Golgi and clathrin adaptor proteins, such as AP-1, AP-3 and GGAs, on trans-Golgi in a GTP-dependent manner, thereby facilitating vesicle formation and trafficking (Cherfils, 2014). Studies from yeast and mammals indicate that Arf1 is activated by two conserved families of guanine nucleotide exchange factors (GEFs), including the Gea1/GBF1 family on cis-Golgi and the Sec7/BIG family on trans-Golgi (Gillingham and Munro, 2007). Arf1 cycles between GDP- and GTP-bound forms, and both the GTP- and GDP-locked forms can interfere with its functions and disrupt secretory trafficking (Dascher and Balch, 1994). In mammalian hippocampal neurons, overexpression of the GTP-locked form of Arf1 (Arf1^Q71L), which abolishes Arf1 activity, inhibits dendrite growth (Horton et al., 2005). The mammalian Arf1GEF, BIG2, is required for vesicle trafficking and mutations in human BIG2 (ARFGEF2 – HGNC) lead to autosomal recessive periventricular heterotopia with microcephaly (ARPHM), a brain disorder characterized by defective neural proliferation and migration (Ferland et al., 2009; Sheen et al., 2004). Thus, various studies have documented that the secretory pathway plays a crucial role in neurite growth and extension in developing neurons. However, very little is known about its role in regulating neurite pruning, a developmental degenerative process.

Here, we report the identification of Arf1 as an important player of dendrite pruning in ddaC sensory neurons. The cycling of Arf1 between GDP-bound and GTP-bound forms is essential for dendrite pruning. We also further identify Sec71 as a GEF for Arf1 in Drosophila. We show that Sec71, like Arf1, is cell-autonomously required for dendrite pruning of ddaC/D/E sensory neurons but not for apoptosis of ddaF neurons during early metamorphosis. Arf1 and Sec71 colocalize on Golgi apparatus and regulate secretory vesicle biogenesis in ddaC neurons. Furthermore, we show that Sec71/Arf1-dependent secretory pathway acts upstream of Rab5-dependent endocytosis and facilitates the internalization and downregulation of the cell-adhesion molecule Nrg prior to dendrite pruning. Thus, our study demonstrates a novel and important role of Arf1/Sec71-mediated secretory pathway in promoting developmental pruning via the regulation of Nrg endocytosis.

To identify novel players of dendrite pruning in ddaC sensory neurons, we conducted a genome-wide RNAi screen by crossing the class IV da neuron driver ppk-Gal4 (Grueber et al., 2003) and UAS-Dicer2 with a large collection of RNAi lines from Vienna and Bloomington Drosophila stock centers (Kirilly et al., 2009). From this screen, we isolated two distinct RNAi lines, v23082 (#1) and v103572 (#2), corresponding to Arf1. In Drosophila, Arf1 was reported to regulate planar cell polarity, blood cell homeostasis and lamellipodium formation (Carvajal-Gonzalez et al., 2015; Humphreys et al., 2012; Khadilkar et al., 2014). The expression of these Arf1 RNAi transgenes, via ppk-Gal4, led to consistent dendrite pruning defects in the vast majority of ddaC neurons at 16 h after puparium formation (APF) (90% and 100%, respectively; Fig. 1C; Fig. S1A). On average, 3.2 primary and secondary dendrites remained attached to Arf1 RNAi ddaC neurons at 16 h APF (Fig. 1J). Those larval dendrites were eventually pruned away at 32 h APF (Fig. S1D). By contrast, all larval dendrites were completely eliminated in the control neurons expressing an irrelevant RNAi line at 16 h APF (Fig. 1B). Development of larval dendrite arbors was also impaired in Arf1 RNAi ddaC neurons, as the numbers of primary and secondary dendrites at the white prepupal (WP) stage were reduced; moreover, elaboration of high-order dendrites was also severely affected (Fig. 1C,J). To validate the RNAi knockdown phenotypes, we generated a synthetic RNAi-resistant transgene that bears minimal homology to the Arf1 dsRNA sequence (#1) but encodes the wild-type Arf1 protein. Both dendrite pruning and morphology defects observed in Arf1 RNAi (#1) ddaC neurons were fully rescued by co-expression of this RNAi-resistant Arf1 transgene (Fig. 1D,J), confirming the specificity of the Arf1 RNAi effect.

Arf1 is a small GTPase that is required for secretory membrane biogenesis and transport (D'Souza-Schorey and Chavrier, 2006). It can cycle between an inactive GDP-bound form and an active GTP-bound form (Gillingham and Munro, 2007). We expressed Arf1^T31N, a GDP-locked form of Arf1, in ddaC neurons. Consistently, the expression of Arf1^T31N, via two copies of ppk-Gal4 driver, also resulted in a dendrite pruning defect at 16 h APF (Fig. 1E,J). The mild pruning defect is probably due to an initial dendrite development defect, as few WP dendrites were present in the vicinity of Arf1^T31N ddaC neurons (Fig. 1E,J). The cytological location of the Arf1 gene near the centromere precludes our MARCM clonal analysis. To circumvent the requirement of Arf1 for initial dendrite development, we induced Arf1^T31N expression at the middle third instar larval stage (90-96 h after egg laying, AEL) using the gene-switch system. Full penetrance of dendrite pruning defect was observed in the Arf1^T31N-expressing ddaC neurons derived from animals fed with RU486 for 6 h (100%; Fig. 1G), in contrast to no pruning defect observed in non-fed animals (Fig. 1F). Approximately 9.4 primary and secondary dendrites persisted with the attachment to the soma of Arf1^T31N-induced ddaC neurons at 16 h APF (Fig. 1J), indicating a severe dendrite pruning defect. Moreover, pulse expression of Arf1^T31N did not impair initial ddaC dendritic development and morphology, as shown at the WP stage (Fig. 1G,J). These data highlight that the Arf1^T31N-associated dendrite pruning defect is not a secondary consequence of its initial dendrite arborization defect. Thus, Arf1 is required to promote dendrite pruning in ddaC sensory neurons.

Both the GTP- and GDP-locked forms of Arf1 can interfere with its functions in the secretory pathway (Dascher and Balch, 1994). Indeed, like that of Arf1^T31N, pulse expression of Arf1^Q71L, a GTP-locked form of Arf1, also led to strong dendrite severing defects, with the persistence of 10.6 primary and secondary dendrites at 16 h APF in RU486-fed animals (100%, Fig. 1H). However, no obvious pruning defect phenotype was observed in ddaC neurons expressing wild-type Arf1 protein (Arf1^WT) at 16 h APF with two copies of the continuously expressing driver ppk-Gal4 (Fig. S1A) or the gene-switch system (Fig. 1I,J). These results strongly support the notion that the cycling of Arf1 between GDP-bound and GTP-bound forms is essential for normal progression of dendrite pruning. In addition to ddaC neurons, ddaD/E sensory neurons also completely eliminated their dendrites by 19 h APF (Fig. S1B). Arf1^T31N-expressing mutant ddaD/E neurons retained some of their larval dendrites attached to their soma (83%, Fig. S1B). Wild-type ddaF neurons are apoptotic during early metamorphosis (Williams and Truman, 2005). Interestingly, ddaF neurons expressing Arf1^T31N or Arf1 RNAi constructs were eliminated (Fig. S1B), similar to wild-type ddaF neurons, suggesting that Arf1 is dispensable for ddaF apoptosis. Collectively, Arf1 plays an important role in regulating dendrite pruning, rather than apoptosis, of sensory neurons; the cycling of GDP-Arf1 and GTP-Arf1 is essential for normal dendrite pruning.

To assess subcellular localization of Arf1 in Drosophila, we generated an antibody against Arf1 and performed the immunostaining with the anti-Arf1 antibody in ddaC sensory neurons. The anti-Arf1 antibody is specific, as its signals were completely eliminated by both Arf1 RNAi knockdowns (Fig. S2A). We observed many Arf1-positive punctate structures in wild-type ddaC soma and these discrete puncta were colocalized with galactosyltransferase-GFP (GalT-GFP), a trans-Golgi marker (Fig. 2A). Arf1 punctate structures were partially colocalized and juxtaposed with the cis-Golgi marker GM130 (Fig. 2B). Likewise, Arf1 was also colocalized with another Golgi protein: Lava Lamp (Lva) (Fig. S2B). Arf1-positive puncta were also localized next to the ER exit site marker Sec31-mCherry (Fig. 2C). However, Arf1 did not overlap with the recycling endosomal marker Rab4-mRFP, the early endosomal marker GFP-Rab5 and the mitochondrial marker Mito-GFP (Fig. S2C-E).

We then determined whether Arf1 regulates the integrity of the Golgi apparatus in ddaC neurons. We observed that GalT-GFP punctate structures were completely disrupted, as GalT-GFP signals became diffused in the soma of either Arf1^T31N gene-switch (Fig. 2D) or Arf1 RNAi (Fig. S2F) mutant ddaC neurons, compared with those bright puncta in wild type (Fig. 2A′). Similarly, GalT-GFP-positive (Fig. S2H) or ManII-VENUS (Fig. 2H′) signals were also diffused in the dendrites of Arf1 RNAi ddaC neurons, compared with the controls (Fig. S2G, Fig. 2G′, respectively). The GM130 marker was also strongly reduced in size and number in Arf1^T31N neurons (Fig. 2E). These data indicate that the integrity of Golgi apparatus is disrupted in Arf1 mutant neurons. By contrast, the Sec31-mCherry marker was still distributed as many discrete punctate structures, although the numbers of their respective puncta were slightly reduced (Fig. 2F). These data indicate that Arf1 regulates the integrity of the Golgi apparatus. Thus, Arf1 is predominantly localized on the Golgi compartments and is important for the integrity of the Golgi structures in ddaC neurons.

We then took advantage of RNAi and biochemical approaches to identify an important Arf1GEF that is also involved in dendrite pruning. Arf1 can be activated from a GDP-bound state to GTP-bound state by Sec7 domain-containing GEFs (Gillingham and Munro, 2007). We first examined the potential requirements of six Sec7 domain-containing GEFs from the Drosophila genome using RNAi. Among them, we isolated two independent RNAi transgenes, BL32366 (#1) and v100300 (#2), which both target Sec71. Sec71 is a GEF with a catalytic Sec7 domain, which is more closely related to the BIG1/BIG2 family (Christis and Munro, 2012). Mutations in BIG2, a human counterpart of Sec71, lead to defective vesicle trafficking, impaired cell adhesion and ARPHM disease (Sheen et al., 2004). Notably, RNAi-mediated knockdown of Sec71 led to complete penetrance of dendrite pruning defects in all ddaC neurons with 5.7 and 4.7 primary and secondary dendrites attached to their soma at 16 h APF (Fig. 3B,D,G). Those larval dendrites were eventually pruned away at 32 h APF (Fig. S1D). By contrast, all larval dendrites were completely pruned in the control neurons expressing an irrelevant RNAi line at 16 h APF (Fig. 3A,G). Similar to Arf1 mutants, Sec71 RNAi ddaC neurons exhibited simplified dendrite arbors at larval and WP stages (WP: Fig. 3B,D,G; wL3: Fig. S3E). Co-expression of an RNAi-resistant Sec71 transgene fully rescued both dendrite pruning and morphology defects associated with Sec71 RNAi (#1) ddaC neurons (Fig. 3C,G), confirming that the RNAi transgene specifically targets Sec71.

Sec7 domain-containing ArfGEFs activate Arf proteins from the GDP-bound state to the GTP-bound state. The glutamic acid-to-lysine substitution in the conserved Sec7 domain behaves as a catalytically inactive form, as shown for the corresponding mutation in the human ArfGEF GBF1 (E794K) (Garcia-Mata et al., 2003). As described below, the E794K change indeed abolished the GEF activity of the Sec71 protein in a GEF assay (Fig. 4E,F), thus behaving as a dominant-negative form (hereafter referred to as Sec71^DN). To determine a specific requirement of Sec71 function for dendrite pruning, we used the gene-switch system to pulse induce the expression of Sec71^DN at the middle third instar larval (90-96 h AEL) stage, which did not affect initial dendrite development and morphology in ddaC neurons at the WP stage (Fig. 3F). Strong dendrite severing defects were observed in all Sec71^DN-expressing ddaC neurons derived from RU486-induced animals and an average of 9.7 primary and secondary dendrites remained unpruned at 16 h APF (Fig. 3F,G), in contrast to no larval dendrites being present in non-induced animals (Fig. 3E,G). Some larval dendrites in Sec71^DN-expressing ddaC neurons remained attached by 24 h APF (data not shown). Continuous expression of Sec71^DN via ppk-Gal4 also caused simplified dendrite arbors in ddaC neurons (data not shown). Thus, the gene-switch experiments strongly suggest that the Sec71^DN dendrite pruning defect is not a secondary effect of its initial dendrite arborization defect. Thus, Sec71 regulates dendrite pruning of ddaC neurons, which is separable from its role in initial dendrite growth.

To further verify the functions of Sec71 in dendrite pruning, we generated a strong mutant allele, Sec71^EX11, from the mobilization of the P-element insertion Sec71^GS16990 (Fig. S3A). MARCM clones of Sec71^EX11 consistently exhibited simple dendrite arbors and dendrite pruning defects (Fig. S3B). Reintroduction of the Sec71 protein into Sec71^EX11 mutant ddaC neurons fully restored the complex dendrite arbors and rescued the pruning defects (Fig. S3B). In addition to ddaC neurons, ddaD/E neurons expressing Sec71^DN also exhibited dendrite pruning defects, as their larval dendrites were attached to their soma by 19 h APF (50%, Fig. S3C). Sec71^DN-expressing ddaF neurons were eliminated by 16 h APF (Fig. S3C), similar to wild-type neurons. Other known ArfGEFs, namely Steppke and Arf6GEF [encoded by the loner (siz – FlyBase) gene], are dispensable for ddaC dendrite pruning (Fig. S3D).

To examine the possibility that Sec71 functions as an Arf1GEF, we first determined the physical interaction between Arf1 and Sec71. We overexpressed the full-length Sec71 protein in S2 cells and performed GST pull-down experiments using various GST fusion proteins containing wild-type Arf1 (Arf1^WT), a GTP-bound form (Arf1^Q71L) or a GDP-bound form (Arf1^T31N). GEF proteins preferentially bind to the nucleotide-free form or the GDP-bound form, but not the GTP-bound form. Importantly, Sec71 possessed a strong binding affinity for Arf1^WT and GDP-Arf1 but not for GTP-Arf1 (Fig. 4A). To confirm these interactions, we conducted co-immunoprecipitation experiments for Sec71 and Arf1. Similarly, we observed that Sec71 interacted strongly with Arf1^WT and GDP-Arf1, but weakly with GTP-Arf1 (Fig. 4B). Moreover, the interaction between Sec71 and wild-type Arf1 was attenuated in the presence of excess γ-GTP, a non-hydrolyzable GTP analog (Fig. S4A). Thus, Sec71 preferentially interacts with the nucleotide-free form or the GDP-bound form of Arf1.

To further substantiate that Sec71 is an Arf1GEF, we investigated whether Sec71 is able to specifically catalyze the guanine nucleotide exchange of Arf1. We monitored the quantum yield of a non-hydrolyzable and fluorescent analogue of GTP, which increases upon binding to GTPase. His-Arf1 was incubated with different GST proteins, and the levels of the guanine nucleotide exchange of Arf1 were measured via the increase of fluorescence intensity. Sec71 tagged with GST resulted in rapid kinetics of fluorescence increase (Fig. 4D), compared with the GST protein control (Fig. 4C). The inverse of time constant, 1/τ value, reflects the intrinsic exchange activity of an GEF. The exchange activity of GST-Sec71 on Arf1 was significantly higher than that of the GST control (Fig. 4F). This exchange activity was abolished by the catalytically incompetent form of Sec71 (GST-Sec71^EK) (Fig. 4E,F). As a control, we did not observe any exchange activity of GST-Sec71 on another small GTPase Arl1 (Fig. S4B). Thus, Sec71 possesses specific GEF activity toward Arf1.

We further provided genetic evidence to strengthen the notion that Arf1 functions downstream of Sec71 to regulate dendrite pruning. We had observed that the expression of Arf1^Q71L, like Arf1^T31N, caused a block of dendrite pruning, whereas the expression of wild-type Arf1 (Arf1^WT) did not (Fig. 1G,H,I). The GTP-locked form Arf1^Q71L was unable to recue sec71 RNAi knockdown in terms of its dendrite pruning defects (data not shown) because excess GTP-Arf1, induced by Arf1^QL, is not functional. We therefore overexpressed Arf1^WT in Sec71 RNAi ddaC neurons to investigate its potential rescue effect on dendrite pruning. Importantly, the expression of wild-type Arf1 in Sec71 RNAi mutant ddaC (Fig. 4H,I) fully rescued the dendrite pruning defects, compared with the control UAS transgene (Fig. 4G,I). Moreover, overexpression of Arf1 also rescued the initial dendrite morphology defects at the WP stage (Fig. 4H,I). These data suggest that elevated level of Arf1 may enhance basal levels of activation/inactivation cycling, which is sufficient to override the requirement of the Arf1GEF Sec71 for dendrite pruning. As a control, the expression of another small GTPase Arf6^WT was unable to suppress pruning defects in Sec71 RNAi ddaC neurons (Fig. S4C). Collectively, Sec71 is an important GEF for Arf1 and facilitates the Arf1 cycling between GDP-bound and GTP-bound forms during dendrite pruning.

We next assessed the subcellular localization of Sec71 and its colocalization with Arf1. An antibody against the N-terminal epitope of Sec71 (see Fig. 6A) was generated and the specificity was confirmed in Sec71 RNAi and Sec71^Ex11 mutant neurons (Fig. S5A). Using this anti-Sec71 antibody, we observed that endogenous Sec71 was distributed on punctate structures and colocalized with the trans-Golgi marker GalT-GFP (Fig. 5A-A″). The Sec71 signals were also colocalized with Arf1 (Fig. 5B-B″), partially with the cis-Golgi marker GM130 (Fig. 5C-C″) and adjacent to the ER exit marker Sec31-mCherry (Fig. 5D-D″). Likewise, Sec71 is also important for the integrity of the Golgi apparatus in ddaC neurons. In Sec71^DN or Sec71 RNAi-expressing ddaC neurons, the trans-Golgi marker GalT-GFP was completely diffused in their somas (Fig. 5E; Fig. S2F) compared with those in the wild type (Fig. 5A′). The cis-Golgi compartments, indicated by GM130 signals, were also strongly reduced in size and number in Sec71^DN-expressing ddaC neurons (Fig. 5F). The ER exit marker Sec31-mCherry remained largely similar in Sec71^DN-expressing ddaC neurons (Fig. 5G). These data suggest that the Golgi apparatus, particularly trans-Golgi, is severely disrupted in Sec71 mutant neurons, similar to that in Arf1 mutant neurons. Thus, Sec71, like Arf1, is localized on Golgi and is also required for the integrity of the Golgi compartment in ddaC neurons.

To determine the dependency of Arf1 and Sec71 localizations, we checked their endogenous proteins in either Sec71 or Arf1 mutants. In Sec71 RNAi (Fig. 5H′) and Sec71^DN gene-switch (Fig. 5H″) mutant ddaCs, Golgi localization of Arf1 was almost completely disrupted compared with wild-type neurons (Fig. 5H). Likewise, Sec71-positive puncta were largely absent in Arf1 RNAi (Fig. 5I′) or Arf1^T31N gene-switch (Fig. 5I″) ddaC neurons compared with wild-type ddaC neurons (Fig. 5I). Occasionally, a couple of Sec71-labelled puncta were observed in Arf1 RNAi ddaC (Fig. 5I′). Hence, Arf1 and Sec71 are mutually dependent on one another for their localizations on Golgi compartments. We therefore cannot exclude the possibility that their localization interdependence is due to the disruption of Golgi integrity caused by loss of either gene.

To investigate whether Arf1 and Sec71 regulate the formation of secretory vesicles in ddaC neurons, we made use of a GFP-tagged version of Sec15 (Sec15-GFP), a component of the exocyst complex, to monitor the distribution of secretory vesicles in sensory neurons. Sec15-GFP has been previously used to visualize a subset of secretory vesicles in neurons (Jafar-Nejad et al., 2005). In wild-type ddaC neurons, Sec15-GFP was distributed as discrete punctate structures in the somas (Fig. S5B), and was also enriched at dendritic branch points along major dendrites at the WP stage (data not shown). Consistent with our observations, a previous study also reported Sec15 distribution in larval sensory neurons (Peng et al., 2015). Some Sec15-positive puncta were localized adjacent to or partially colocalized with the Arf1-positive Golgi apparatus (Fig. S5B). Notably, Sec15-GFP-positive puncta were completely absent in all Arf1 RNAi (Fig. S5C) or Sec71 RNAi (Fig. S5C) ddaC neurons. As a control, the ER exit marker Sec31-mCherry remained present in either Arf1^T31N (Fig. 2F) or Sec71^DN (Fig. 5G) mutant neurons. Thus, Arf1 and Sec71 appear to regulate biogenesis of secretory vesicles in ddaC sensory neurons.

To examine whether Arf1 and Sec71 regulate protein transport to the cell surface, we conducted a trafficking assay by detecting the levels of extracellular mCD8 epitope in a detergent-free condition. Via the gene-switch system, we pulse induced the expression of mCD8-GFP (murine CD8 fused to GFP) at the middle L3 stage and measured the levels of newly synthesized mCD8-GFP protein at the WP stage in ddaC neurons. In wild-type WP ddaC neurons, the extracellular mCD8 epitope was robustly detectable on the dendrites using the anti-mCD8 antibody in the absence of the detergent (Fig. S5D). We observed no or strongly reduced mCD8 signal on the surface of the dendrites in either Arf1^T31N (Fig. S5D) or Sec71^DN (Fig. S5D) ddaC neurons in the detergent-free condition. We found that overall GFP fluorescence representing both surface and internal pools of the protein in the soma of Sec71^DN and Arf1^T31N ddaC neurons was similar to that in wild-type neurons. In the trafficking assays, mCD8 signals were absent on the soma, axons and proximal regions of dendrites of ddaC neurons, presumably because these structures are tightly wrapped by glia and the antibody was not able to penetrate into the wrapped parts without the presence of detergent. Taken together, these observations show that Arf1 and Sec71 colocalize on Golgi and regulate protein secretion in sensory neurons.

Sec71 contains DCB (dimerization and cyclophilin binding) and HUS (homology upstream of Sec7) domains in its N-terminal region, a Sec7 domain in the middle region and HDS1-4 (homology downstream of Sec7) domains in its C-terminal region (Fig. 6A). We found that the Sec7 domain is essential for the Sec71 function in regulating dendrite pruning (Fig. 3F). To elucidate whether other domains of Sec71 are also required for its function, we generated a series of RNAi-resistant transgenes expressing various truncated proteins and examined their abilities to rescue sec71 RNAi phenotypes (Fig. S6A). Although the expression of full-length Sec71 fully rescued the Sec71 RNAi phenotypes in terms of dendrite arborization and pruning (Fig. 6C,G), the expression of DCB-deleted Sec71 protein was unable to significantly rescue the pruning defect in sec71 RNAi ddaC neurons at 16 h APF (Fig. 6D,G) but partially restored the dendrite morphology (Fig. 6D,G). Sec71^ΔDCB did not appear to localize on Golgi (Fig. S6B), suggesting that the DCB domain of Sec71 is essential for its Golgi localization as well as for its function in regulating dendrite pruning. We observed that HDS2-4 domains are not important for Sec71 function, as the Sec71 variants deleting either HDS2-4 (Fig. 6E,G) or HDS3-4 domains (Fig. 6F,G), like the full-length protein, completely rescued both dendrite pruning and arborization defects. Apart from Sec71^ΔDCB, all the other Sec71 variants were able to localize on Golgi (Fig. S6B). Thus, the DCB domain of Sec71 is important for proper function and Golgi localization of Sec71 during dendrite pruning.

We have previously reported that Rab5/ESCRT-dependent endocytic pathways downregulate the L1-CAM Nrg to promote dendrite pruning during early metamorphosis. In wild-type ddaC neurons, Nrg is drastically redistributed to early endosomes at 6 h APF prior to dendrite pruning and concomitantly its protein levels are also strongly reduced in the somas, dendrites and axons (Zhang et al., 2014). In Rab5^DN ddaC neurons, Nrg protein levels were significantly increased in the soma, dendrites and axons (Fig. 7B,E) (Zhang et al., 2014) compared with wild-type neurons (Fig. 7A,E). We therefore examined whether the Arf1/Sec71-dependent secretory pathway regulates the protein levels of Nrg at 6 h APF. Surprisingly, Nrg protein levels were significantly elevated in the dendrites, axons or soma of Arf1^T31N or Sec71^DN gene-switch ddaC neurons (Fig. 7C-E), similar to those in Rab5^DN neurons. Nrg protein levels in Arf1^T31N or Sec71^DN gene-switch ddaC neurons were comparable with those in Rab5^DN-expressing neurons (Fig. 7B-E). Moreover, Arf1 and Sec71 also regulate endosomal localization of Nrg prior to dendrite pruning. In wild-type neurons, Nrg protein mainly colocalized with the early endosomal marker GFP-2xFYVE at 6 h APF (Fig. S7A). By contrast, in Arf1^T31N or Sec71^DN gene-switch ddaC neurons, Nrg did not colocalize with those GFP-2xFYVE puncta (Fig. S7A), suggesting a block of Nrg endocytosis. Thus, Arf1 and Sec71, like Rab5, are required to facilitate Nrg endocytosis and downregulation prior to dendrite pruning.

To examine whether Arf1 and Sec71 regulate dendrite pruning upstream of Nrg function, we knocked down Nrg, via two independent Nrg RNAi transgenes, in Arf1 RNAi- or Sec71 RNAi-expressing ddaC neurons. Importantly, the expression of two independent Nrg RNAi transgenes, both of which efficiently knocked down Nrg protein (Zhang et al., 2014), dramatically suppressed the dendrite severing defects in Arf1 RNAi (Fig. 7G-I) or Sec71 RNAi (Fig. 7K-M) ddaC neurons, further supporting the notion that, similar to Rab5, Arf1 and Sec71 facilitate Nrg endocytosis and degradation to promote dendrite pruning. There are at least two possibilities to explain how Arf1 and Sec71 regulate Nrg endocytosis: they act in a pathway that is either upstream of or parallel to Rab5 to facilitate endocytosis. To distinguish these two possibilities, we created Rab5 and Arf1/Sec71 double mutants and examined their epistasis. Pulse expression of Rab5^DN at the late larval stage via the gene-switch system led to pronounced accumulation of ubiquitylated protein deposits on two or three enlarged endosomes that were positively labelled with ubiquitin (Fig. S7B), the early endosomal marker Avl (Fig. S7C) and Nrg (Fig. S7C). These data suggest that Rab5 does not regulate the initial formation of early endosomes but is important for subsequent endosomal maturation as well as for Nrg downregulation (Zhang et al., 2014). By contrast, ubiquitin and Avl did not accumulate in Arf1^T31N or Sec71^DN-expressing ddaC neurons (Fig. S7B,C). Importantly, Arf1^T31N and Rab5^DN double mutants did not exhibit any ubiquitylated aggregates (Fig. S7B) and Avl-positive enlarged endosomes (Fig. S7C) in ddaC neurons, resembling Arf1^T31N single mutant neurons. Likewise, co-expression of Sec71^DN and Rab5^DN did not result in formation of ubiquitylated protein deposits (Fig. S7B) and Avl-positive endosomes (Fig. S7C), resembling Sec71^DN mutant neurons. Thus, these epistatic data support the first possibility that Arf1 and Sec71 act upstream of Rab5 to facilitate initial formation of endosomes and robust endocytosis prior to dendrite pruning.

The small G protein Arf1 regulates vesicular trafficking in eukaryotes and is activated on cis-Golgi by the Gea1/GBF1 family of Arf1GEF or on trans-Golgi by the Sec7/BIG1 family (Cherfils, 2014). It has been reported that Drosophila Arf1 regulates hematopoietic niche maintenance, blood cell precursor differentiation in vivo, planar cell polarity and lamellipodium formation in S2 cells (Carvajal-Gonzalez et al., 2015; Humphreys et al., 2012; Khadilkar et al., 2014). Arf1 and other Golgi proteins were reported to exhibit upregulation of their transcripts in axon pruning of MB γ neurons during the larval-pupal transition (Hoopfer et al., 2008). In this study, we report an important role for Arf1 in regulating dendrite pruning of ddaC sensory neurons. We found that Arf1 puncta overlap with the trans-Golgi marker GalT and partially with the cis-Golgi marker GM130, suggesting that Arf1 is primarily localized on trans-Golgi in ddaC sensory neurons. A specific GEF for Arf1 has not been identified in Drosophila. Our in vivo and in vitro data provide compelling evidence that Sec71 is a GEF for Arf1 in Drosophila. First, Sec71 was colocalized with Arf1 on Golgi and their localizations were inter-dependent in ddaC neurons. Second, Sec71 preferentially bound to the GDP-bound form of Arf1 instead of GTP-bound form. Third, in the GEF assays Sec71 accelerated the release of GDP from Arf1. Fourth, both Arf1 and Sec71 are required for dendrite pruning, as loss of Sec71 or Arf1 led to comparable pruning defects in ddaC sensory neurons. Finally, the expression of Arf1 fully restored WP dendrite morphology and rescued the pruning defects in Sec71 RNAi ddaC neurons. Thus, Sec71 is specifically required for the GDP-to-GTP exchange of Arf1.

Our structure-function analyses indicate that the DCB domain of Sec71 is important for its targeting on Golgi (Fig. S6). By contrast, another small GTPase, Arl1, was reported to bind to the N-terminal region of Sec71 (DCB and HUS1 domains) and to recruit Sec71 on trans-Golgi apparatus in Drosophila S2 cells; mammalian Arl1 is required for the recruitment of BIG1/2 (mammalian homologs of Sec71) on trans-Golgi (Christis and Munro, 2012). Unexpectedly, we found that mutant ddaC neurons derived from arl1¹, a null arl1 mutant (Torres et al., 2014), did not exhibit delocalization of Sec71 or dendrite pruning defects in ddaC neurons (Fig. S1C), suggesting an Arl1-independent mechanism regulating Sec71 localization and function in ddaC sensory neurons.

Extensive studies have attempted to understand roles of post-Golgi trafficking in outgrowth and elaboration of dendrites in growing neurons. Post-Golgi trafficking is polarized towards apical dendrites of rodent hippocampal neurons and selectively regulates the growth of dendrites (Horton and Ehlers, 2003; Horton et al., 2005). The dynamics of the Golgi outposts, mediated by the golgin Lava Lamp, dynein-dynactin complex and Leucine-rich repeat kinase (Lrrk), is important for dendrite growth in Drosophila class IV da neurons (Lin et al., 2015; Ye et al., 2007). Rab10, a small GTPase that mediates post-Golgi vesicle trafficking, regulates dendrite growth and branching of multi-dendritic sensory neurons in both C. elegans and Drosophila (Taylor et al., 2015; Zou et al., 2015).

In this study, we provide compelling evidence to demonstrate that post-Golgi trafficking plays a crucial role in proper dendrite pruning in sensory neurons. First, we identified that the key small GTPase Arf1, which is important for post-Golgi trafficking, regulates secretory vesicle biogenesis and dendrite pruning in sensory neurons during metamorphosis. Second, a Sec7-domain-containing protein Sec71 acts as a specific GEF for Arf1 and colocalizes with Arf1. Like Arf1, Sec71 is also an essential factor for regulating dendrite pruning. Third, given that both Arf1 and Sec71 also regulate dendrite growth and arborization in ddaC neurons, we further confirmed a crucial role for Arf1 and Sec71 in dendrite pruning using the gene-switch system. Pulse induction of Arf1^T31N or Sec71^DN at the middle third instar larval stage, when the complete larval dendrite arbors form in ddaC neurons, consistently caused much more severe dendrite pruning defects. These results highlight that the secretory pathway play separable roles in two distinct processes: dendrite growth and dendrite pruning. Arf1 was reported to regulate post-Golgi secretion by recruiting its downstream effectors, including the clathrin adaptors AP-1 and AP-3, and GGA (Cherfils, 2014). Post-Golgi trafficking pathways include the transport from Golgi to plasma membrane (exocyst complex mediated), from Golgi to early/sorting endosomes (AP-1 meditated), from Golgi to late endosomes (Golgi-localized gamma ear-containing Arf-binding protein or GGA mediated) as well as from Golgi to lysosomes (AP-3 mediated). It is conceivable that at least one of these post-Golgi trafficking routes is involved in dendrite pruning of sensory neurons.

We have previously reported that Rab5/ESCRT-dependent endocytic pathways facilitate dendrite pruning by downregulating the L1-CAM Nrg in ddaC neurons during metamorphosis (Zhang et al., 2014). In MB γ neurons, the JNK pathway promotes axon pruning by downregulating another adhesion molecule: Fasciclin II (Bornstein et al., 2015). These studies suggest a general mechanism whereby cell-adhesion molecules are internalized and downregulated to destabilize dendrites and/or axons during neurite pruning. However, the mechanism that triggers Nrg endocytosis is poorly understood. In this study, we demonstrate that Arf1/Sec71-mediated secretory pathway promotes endocytosis and downregulates Nrg prior to dendrite pruning. First, although Nrg levels were strongly reduced prior to dendrite pruning, loss of Arf1 or Sec71 led to elevated levels of Nrg protein in dendrites, axons and soma, comparable with Rab5 mutant neurons. Second, Nrg was no longer redistributed on FYVE-positive endosomes in Arf1 or Sec71 mutant ddaC neurons, suggesting a blockage of Nrg endocytosis. Third, whereas Rab5 mutant neurons exhibited robust ubiquitylated protein aggregates and enlarged endosomes, further removal of either Arf1 or Sec71 suppressed these Rab5 mutant phenotypes, suggesting that the secretory pathway acts upstream of Rab5 to positively regulates endocytosis. Finally, knockdown of Nrg significantly suppressed the dendrite pruning defect in Arf1 or Sec71 mutant neurons, supporting the notion that the secretory pathway promotes Nrg endocytosis and downregulation. Thus, the secretory pathway not only secretes the cell-adhesion molecules to the dendrite surface and stabilize dendrites (Taylor et al., 2015; Zou et al., 2015), but also unexpectedly promotes the internalization and turnover of the adhesion molecules (this study). It is conceivable that, in response to an ecdysone pulse, the secretory pathway might be required to specifically secrete an as yet unidentified ligand to trigger robust endocytosis of the L1-CAM Nrg and thereby lead to degeneration of larval dendrites. Further studies may continue to elucidate what ligand or secreted protein promotes Nrg endocytosis.

The fly stocks used in this study were maintained on standard food at 25°C. See supplementary Materials and Methods for further details of fly stocks used in this study.

We carried out MARCM clonal analysis, dendrite imaging and branch quantification as previously described (Kirilly et al., 2009). See supplementary Materials and Methods for further details.

Wild-type and mutant embryos were collected at 3 h intervals and were reared on standard food to the middle 3rd instar larval stage before being transferred to the standard culture medium containing 240 µg/ml mifepristone (Sigma Aldrich, M8046).

We crossed Sec71^GS16990 flies with a fly strain carrying the Δ2–3 transposase to induce imprecise excision as previously described (Kirilly et al., 2009). See supplementary Materials and Methods for further details.

The Arf1 and Sec71 full-length cDNA were PCR from EST LD24904 and LD29171 (DGRC) into Topo Entry and pDonor, respectively (Life Tech). The variants of Sec71 and Arf1 were generated by site mutagenesis (Agilent Tech). The cDNA fragment encoding the 1-100 amino acid region of rat ManII protein were used to generate ManII-Venus transgene (Bestgene). See supplementary Materials and Methods for further details.

The full-length Arf1 and the 60-345 amino acid fragment of Sec71 were used for their respective antibody generation. See supplementary Materials and Methods for further details.

Pupae or larvae were dissected in PBS and fixed with 4% formaldehyde for 20 min. The primary antibodies and Cy3-, Cy5- or fluorescein isothiocyanate (FITC)-conjugated secondary antibodies were used for immunohistochemistry. Mounting was performed in VectaShield mounting medium, and the samples were directly visualized by confocal microscopy. See supplementary Materials and Methods for further details.

Transfected S2 cells were homogenized with the lysis buffer. The supernatants were used for immunoprecipitation, followed by incubation with protein A/G beads (Pierce Chemical). Protein A/G beads were washed four times. Bound proteins were separated by SDS-PAGE and analyzed by western blotting with anti-Myc, anti-Flag HRP-conjugated antibody. See supplementary Materials and Methods for further details.

The fillets were incubated with rat monoclonal anti-CD8α (1:100, CALTAG Laboratories) in PBS and washed three times with PBS in the detergent-free condition. See supplementary Materials and Methods for further details.

We carried out the in vitro guanine nucleotide exchange assays as previously described (Mahajan et al., 2013). GST fusion proteins for the Sec7 domain of Sec71 (amino acids 554-790), its dominant negative form (E-K) and Sec7 domain of human BIG1 were expressed using the GST expression vector pGEB. 6×His-tagged Arf1^WT and 6×His-tagged Arl1^WT were expressed via the vector pET37b. See supplementary Materials and Methods for further details.

We quantified the immunolabelling intensities of Nrg at 6 h APF, as described previously (Zhang et al., 2014). One-way ANOVA and Bonferroni tests were used for comparisons between different conditions (*P<0.05, **P<0.01, ***P<0.001; n.s., not significant). See supplementary Materials and Methods for further details.

Live confocal images of ddaC neurons were shown at the WP stage and 16 h APF. Statistical significance was determined using either two-tailed Student's t-test (two samples) or one-way ANOVA and Bonferroni test (multiple samples) (*P<0.05, **P<0.01, ***P<0.001; n.s., not significant). See supplementary Materials and Methods for further details.

We thank H. Bellen, R. L. Borgne, E. Chen, J. Eissenberg, M. Gonzalez-Gaitan, T. Harris, S. Hayashi, Y. Jan, J. Kennison, A. Kolodkin, S. Luschnig, M. Metzstein, E. Olson, T. Schwarz, T. Uemura, the Bloomington Stock Center (BSC), DSHB (University of Iowa), the Kyoto Stock Center (Japan) and the VDRC (Austria) for generously providing antibodies and fly stocks. We thank members of the Yu laboratory for helpful assistance and discussion.