A Cullin1-Based SCF E3 Ubiquitin Ligase Targets the InR/PI3K/TOR Pathway to Regulate Neuronal Pruning

Pruning that selectively eliminates unnecessary axons/dendrites is crucial for sculpting the nervous system during development. During Drosophila metamorphosis, dendrite arborization neurons, ddaCs, selectively prune their larval dendrites in response to the steroid hormone ecdysone, whereas mushroom body γ neurons specifically eliminate their axon branches within dorsal and medial lobes. However, it is unknown which E3 ligase directs these two modes of pruning. Here, we identified a conserved SCF E3 ubiquitin ligase that plays a critical role in pruning of both ddaC dendrites and mushroom body γ axons. The SCF E3 ligase consists of four core components Cullin1/Roc1a/SkpA/Slimb and promotes ddaC dendrite pruning downstream of EcR-B1 and Sox14, but independently of Mical. Moreover, we demonstrate that the Cullin1-based E3 ligase facilitates ddaC dendrite pruning primarily through inactivation of the InR/PI3K/TOR pathway. We show that the F-box protein Slimb forms a complex with Akt, an activator of the InR/PI3K/TOR pathway, and promotes Akt ubiquitination. Activation of the InR/PI3K/TOR pathway is sufficient to inhibit ddaC dendrite pruning. Thus, our findings provide a novel link between the E3 ligase and the InR/PI3K/TOR pathway during dendrite pruning.


Introduction
The selective removal of unnecessary or exuberant neuronal processes without loss of neurons, referred to as pruning, is a central theme in the maturation of the nervous system during animal development [1]. Pruning occurs widely in a variety of neurons of invertebrates [2,3] and vertebrates [4]. In vertebrates, neurons normally extend exuberant branches to multiple targets, such as muscles or partner neurons, and prune away inappropriate or redundant branches to develop mature and functional connectivity [1,5]. One well-characterized example is pervasive synaptic branch removal in the mammalian neuromuscular system at birth [6]. In invertebrates, such as holometabolous insects Manduca and Drosophila, the nervous systems are extensively remodeled via pruning and apoptosis during metamorphosis, a transition stage between larval and adult forms [2,3]. In the Drosophila central nervous system (CNS), mushroom body (MB) c neurons, serotonergic neurons, and thoracic ventral neurons prune their larval dendrites and/or axons to form the adult neuronal circuits [7][8][9][10]. In contrast, in the peripheral nervous system (PNS), a subset of dorsal dendrite arborization (dda) sensory neurons, such as Class I (ddaD/E) and class IV (ddaC) neurons, selectively remove their larval dendrite arbors with their axons intact and subsequently regrow their adult-specific dendrites [11,12], whereas class II (ddaB) and class III (ddaA/F) neurons are eliminated via apoptosis during early metamorphosis [11]. Despite the wide occurrence and key roles of pruning in the maturing nervous systems, the molecular and cellular mechanisms underlying pruning remain poorly understood in both invertebrates and vertebrates.
Drosophila MB c and ddaC neurons have been emerging as appealing systems for unraveling mechanisms of axon pruning and dendrite pruning, respectively. These pruning processes occur in a stereotyped but context-specific manner: ddaC neurons sever their major larval dendrites at the proximal regions, followed by rapid fragmentation and hemocyte-dependent debris clearance ( Figure 1A) [11,12], whereas MB c neurons selectively prune their axon branches within the dorsal and medial lobes via local degeneration and glia-mediated engulfment ( Figure 2A) [7,[13][14][15]. Pruning of the Drosophila nervous system is regulated by the steroid molting hormone 20-hydroxyecdysone (ecdysone) and its nuclear receptor heterodimer (EcR-B1/Usp) [16]. Ecdysone binds to a nuclear receptor heterodimer consisting of EcR-B1 and its coreceptor Ultraspiracle (Usp) to trigger the onset of pruning by activating the downstream pruning programs, rather than the conventional ecdysone response genes Broad complex, E74, and E75 [7]. The ecdysone-induced pruning programs are activated via a multilayered regulatory process. First, in response to a late larval pulse of ecdysone, EcR-B1 expression is upregulated, depending on TGF-b signaling [17,18], the cohesin complex [19,20], and the Ftz-F1/Hr39 nuclear receptors [21]. Second, EcR-B1, together with two epigenetic factors, Brahma and CREB-binding protein, activates the expression of their common target gene sox14 [22]. Third, Sox14, a key transcription factor, in turn induces the expression of the cytoskeletal regulator Mical [23]. Interestingly, despite its elevated expression in ddaC and MB c neurons, Mical is only required for ddaC dendrite pruning but dispensable for MB axon pruning [23]. Aside from Mical, other molecules, such as caspases [24,25], Ik2 and Kat60L [26], are also specifically required for ddaC dendrite pruning. Therefore, it remains poorly understood what common pruning machinery directs two distinct modes of pruning in ddaC and MB c neurons.
The ubiquitin-proteasome system (UPS) is a potent cellular pathway regulating protein stability and homeostasis in eukaryotic cells [27]. Growing evidence indicates that the UPS pathway plays critical roles in neuronal development [28] and neurodisorders [29]. Pharmacological inhibition of the UPS activity leads to apparent retardation of injury-induced axon degeneration [30]. In the UPS pathway, the polypeptide ubiquitin is activated by a ubiquitin-activating enzyme E1; the activated ubiquitin is transferred to a ubiquitin-conjugating enzyme E2 and then, via a ubiquitin ligase E3, to protein substrates, destined for 26S proteasome-mediated degradation. Substrate specificity is conferred by different classes of E3 ligases, one of which contains RING (really interesting new gene) domain. An important RING E3 ligase is the SCF (Skp1-Cullin-F-box) E3 ligase composed of the adaptor Skp1, the scaffold protein Cullin, the RING-domain protein Roc1-Rbx1-Hrt1, as well as the F-box protein for substrate recognition [31]. The conserved SCF E3 ligase complex containing the RPM-1/Highwire RING protein negatively regulates the MAP kinase signaling or the receptor tyrosine kinase ALK to control synapse formation and homeostasis in C. elegans and Drosophila [32][33][34]. Another worm SEL-10 SCF E3 ligase, whose activity is spatially restricted, directs selective synapse pruning, presumably by locally disrupting as-yet-unidentified substrates or pathways [35]. In Drosophila, inactivation of the UPS pathway by removing the proteasome subunits, the E1 activating enzyme Uba1, or expressing the yeast ubiquitin protease UBP2 blocks both MB c axon pruning [13] and ddaC dendrite pruning [12]. Autoubiquitination and degradation of the caspase-antagonizing RING E3 ligase, Drosophila inhibitor of apoptosis protein 1 (DIAP1), allows for local activation of the Dronc caspase and thereby dendrite pruning of ddaC neurons [24]. However, DIAP1 (Wong and Yu, unpublished data) and Dronc [24] are not involved in axon pruning of MB c neurons, raising the intriguing questions of whether and which E3 ligase complex directs two distinct modes of pruning in ddaC and MB c neurons. Moreover, it is also of great interest to understand which downstream pathway is inactivated by the E3 ligase in order to facilitate the pruning process in ddaC neurons during early metamorphosis.
Here, we identified an E3 ubiquitin ligase complex that plays a crucial role in directing two distinct modes of pruning in ddaC and MB c neurons during early metamorphosis. In a genome-wide RNA interference (RNAi) screen, we isolated Cullin1 (Cul1), a core scaffold protein of the SCF E3 ligase, which governs pruning of ddaC dendrites and MB axons. Additional components of the Cul1-based SCF E3 ligase include the RING domain protein Roc1a, the adaptor protein SkpA, and the F-box protein Slimb, all of which are essential for ddaC dendrite pruning and MB c axon pruning. Cul1 acts together with Roc1a, SkpA, and Slimb to promote ddaC dendrite pruning downstream of EcR-B1 and Sox14, but independently of Mical. Furthermore, the Cul1-based SCF E3 ligase promotes ddaC dendrite pruning through negative regulation of the InR/PI3K/TOR signaling pathway but not its conventional target pathways, such as the Hedgehog (Hh) and Wingless (Wg) pathways [36]. We show that attenuation of the InR/PI3K/TOR pathway leads to strong suppression of dendrite pruning defects in mutant ddaC neurons deficient in the Cul1 SCF ligase, whereas activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. Moreover, the F-box protein Slimb forms a complex with Akt, an activator of the InR/ PI3K/TOR pathway, and promotes Akt ubiquitination. We demonstrate that the Cul1-based SCF E3 ligase complex facilitates ddaC dendrite pruning primarily through inactivation of the InR/ PI3K/TOR pathway. Finally, the SCF ligase and the InR/PI3K/ TOR pathway regulate dendrite pruning in ddaC neurons at least in part by promoting local caspase activation in the dendrites. Thus, this study provides the first example, to our knowledge, of a critical E3 ligase complex in directing two distinct modes of pruning and also provides mechanistic insights into how this evolutionarily conserved SCF E3 ligase attenuates a key signaling pathway to promote pruning during the maturation of the nervous system.

Author Summary
Neurons have the ability to engage in selective pruning that eliminates unnecessary axons/dendrites. This process is crucial for sculpting the nervous system during development. During Drosophila development, dendrite arborization sensory neurons (ddaCs) selectively prune their larval dendrites in response to the molting steroid hormone ecdysone, whereas mushroom body c neurons eliminate their axon branches. However, the underlying molecular mechanisms for both of these modes of pruning were not well understood. Here, we conduct a genomewide screen and identify a conserved E3 ubiquitin ligase that is critical for pruning both ddaC dendrites and mushroom body c axons. This ligase complex has four core components-Cullin1, Roc1a, SkpA, and Slimb-that promote ddaC dendrite pruning in response to ecdysone. We show that this ligase facilitates ddaC dendrite pruning through regulation of the InR/PI3K/TOR pathway. The substrate-recognition protein Slimb promotes ubiquitination of Akt, an activator of the InR/PI3K/TOR pathway. Akt ubiquitination leads to its degradation and inactivation of the InR/PI3K/TOR pathway, which is required for dendritic pruning. Consistent with this, ddaC dendrite pruning is inhibited when the InR/PI3K/TOR pathway is activated. Thus, we identify a link between the Cullin1-based E3 ligase and the InR/PI3K/TOR pathway in regulating dendrite pruning. This work represents the first link between neuronal pruning and the insulin signaling pathway, raising interesting questions about how metabolic states may influence the control of such developmental processes. Red arrowheads point to the ddaC somas, and purple arrowheads point to the ddaF somas that are labeled by two or three copies of ppk-Gal4 (Chr II). While the wild-type neurons eliminated dendrites (B, B9), ddaC neurons overexpressing two copies of cul1 RNAi #2 (C, C9), three copies of cul1 DN (D, D9), or cul1 EX MARCM (E, E9) ddaC clones exhibited severe dendrite pruning defects at 16 h APF. Overexpression of the full-length Cul1 protein in the cul1 EX background was able to fully rescue the pruning defect (F, F9). (G) Quantification of the average number of primary and secondary dendrites attached to the soma of the wild-type and mutant ddaC neurons at WP stage and 16 h APF. The number of samples (n) in each group is shown on the bars. Error bars represent S.E.M. ***p,0.001, n.s., not significant. Dorsal is up in all images. The scale bars are 50 mm. See genotypes in Text S1. doi:10.1371/journal.pbio.1001657.g001

Results
Cullin-1, a Core Component of the SCF E3 Ligase, Governs Pruning of ddaC Dendrites We previously reported high efficacy and specificity of inducible RNAi knockdown in ddaC neurons during the larval-pupal transition [23]. It prompted us to carry out an unbiased genome-wide RNAi screen searching for novel players of dendrite pruning (Wong and Yu, unpublished data). In this large-scale screen, we isolated two independent RNAi transgenes, v108558 (#1) and v42445 (#2), corresponding to cullin-1(cul1, also known as lin19) [37]. RNAi knockdown of Cul1, via one copy of the ppk-Gal4 driver that can target Gal4 expression in the class IV da neurons, led to dendrite pruning defects in ddaC neurons at 16 h after puparium formation (APF) (60%, n = 25 and 100%, n = 17, respectively; Figure S1A). The expression of cul1 RNAi transgene (#2) with two copies of ppk-Gal4 drivers caused severe defects with approximately 10.2 primary and secondary dendrites attached at 16 h APF (100%, n = 21; Figure 1C9 and 1G). Larval dendrites of cul1 RNAi ddaC neurons were eventually removed by 48 h APF (n = 13, unpublished data), probably due to extensive migration and death of the abdominal epidermis on which the dendrites arborize [11]. Likewise, knockdown with other cul1 RNAi lines, v33406 and v33407, also resulted in similar pruning defects at 16 h APF (unpublished data). In contrast, all dendrites were pruned completely in the wild-type ddaC neurons (n = 15; Figure 1B9 and 1G). The Cullin proteins serve as scaffold proteins of the SCF E3 ligase and interact with Roc1-Rbx1-Hrt1. We next generated transgenic flies expressing the dominant-negative form of Cul1 (Cul1 DN ) lacking its C-terminal putative Roc1-binding domain and neddylation site [38,39]. The expression of Cul1 DN also resulted in consistent dendrite pruning defects at 16 h APF with an average of 4.5 primary and secondary dendrites (n = 20, 100%; Figure 1D9 and 1G). Development of major dendrite branches appears not to be affected in cul1 RNAi and Cul1 DNexpressing ddaC neurons, as judged by the number of their respective primary and secondary dendrites at the white prepupal (WP) stage ( Figure 1C-D and 1G). Importantly, cul1 RNAi (#2) ddaC neurons under one copy of ppk-Gal4 exhibited normal larval dendrite morphology, as shown by the number of dendrite termini at the wandering 3 rd instar (wL3) stage ( Figure S2A), while these mutant neurons showed notable dendrite pruning defects at 16 h APF ( Figure S1A). Further, we made use of the RU486-inducible Gene-Switch system [40] to drive the expression of Cul1 DN from the early 3 rd instar (eL3) stage when larval dendrite morphogenesis was largely completed. RU486 treatment did not affect WP dendrite morphology in Cul1 DN -expressing ddaC neurons ( Figure  S2B) or dendrite pruning in the control ddaC neurons (n = 35; Figure S2B). However, we consistently observed dendrite pruning defects in 67% of Cul1 DN -expressing ddaC neurons at 16 h APF (n = 30; Figure S2B), contrasting with the noninduced controls (0%, n = 12, Figure S2B). The Cul1 DN pruning defect in the Gene- Wild-type MB c neurons projected their axons into the dorsal and medial lobe of the brain at wL3. (C-F) cul1 EX (C), roc1a G1 (D), skpA 1 (E), and slimb 2 (F) neuroblast clones that were labeled with 201Y-Gal4 projected their dorsal and medial axons normally but exhibited proliferation defects at wL3, as the axon branches were less dense than the wild-type control. (B9) Wild-type MB c neurons pruned their dorsal and medial axon branches by 24 h APF, leaving intact a/b axons (GFP negative and FasII positive). (C9-F9) cul1 EX (C9), roc1a G1 (D9), skpA 1 (E9), and slimb 2 (F9) MB neuroblast clones displayed notable axon pruning defects at 24 h APF. The right panels in (C9-F9) show that many unpruned dorsal c axons persisted outside the main a lobes and were co-labeled by GFP and FasII (white arrowheads) in single confocal sections of the dorsal lobe. (G) At adulthood, MB c neurons regenerated their medial branches but not the dorsal branches. Blue arrowheads point to a axon branches. (H) cul1 EX MB neuroblast clones retained many GFP-positive unpruned larval c axons (white arrowheads) outside the FasII-positive main a lobe at the adult stage. However, the GFP-positive homozygous cul1 EX mutant a axon branches were absent in this main a lobe (H), suggesting a proliferation defect. Please note that the FasII-positive a lobe is derived from the other heterozygous MB neuroblasts. Dorsal is up in all images. The scale bar is 50 mm. See genotypes in Text S1. doi:10.1371/journal.pbio.1001657.g002 Switch system was not as strong as those induced by ppk-Gal4 driver (three copies) ( Figure 1D9), due to weaker expression of the elav-GS-Gal4 driver (one copy). Thus, cul1 plays a specific role in dendrite pruning in ddaC neurons.
Taken together, Cul1 is identified as the core component of the SCF E3 ligase that regulates pruning and apoptosis of sensory neurons during early metamorphosis.
The RING-Domain Protein Roc1a, But Not Roc1b, Is Required for ddaC Dendrite Pruning We next attempted to identify other components of the Cul1based SCF E3 ligase complex that are required for ddaC dendrite pruning. The Cullin proteins, via their C-terminal regions, bind to the RING-domain protein Roc1-Rbx1-Hrt1, which recruits an ubiquitin conjugating enzyme E2 [31]. In Drosophila, two closely related Roc1 proteins, Roc1a and Roc1b, were reported to play differential roles during development [42,43]. We observed a strong dendrite pruning defect in roc1a mutants, but not in roc1b mutants, suggesting a specific requirement of Roc1a for pruning. MARCM analysis of roc1a G1 , a null allele [42], revealed a severe cul1-like pruning defect in all ddaC neurons: about 10.3 primary and secondary dendrites were attached at 16 h APF (n = 9; Figure 3B9 and 3F). Furthermore, RNAi knockdown of Roc1a, via independent v106315 (#1) and v32398 (#2) lines, also resulted in pruning defects (n = 22 and n = 24, respectively; Figure S3). Loss of roc1a function also led to severe defects in ddaD/E pruning (n = 2; Figure S4A) and ddaF apoptosis (n = 6; Figure S4B). In contrast, roc1b null mutants showed no defects in dendrite pruning of ddaC (n = 9; Figure S5) and ddaD/E neurons (n = 8; unpublished data) or ddaF apoptosis (n = 8; unpublished data). Consistent with the preferential binding between Roc1b and Cul3 [43], MARCM ddaC clones for cul3 gft2 , a null cul3 allele, did not exhibit apparent pruning defects (n = 14; Figure S5). These results underscore the specific roles of Cul1 and Roc1a in governing dendrite pruning.
Thus, Roc1a, like Cul1, plays an essential role in dendrite pruning of ddaC sensory neurons.

The Adaptor Protein SkpA and the F-Box Protein Slimb Govern ddaC Dendrite Pruning
The N-terminal region of the Cullin proteins interacts with the adaptor subunit Skp1, and via it with an F-box protein to recruit protein substrates in proximity to the E2 enzyme, thereby promoting substrate ubiquitination [31]. We first examined the potential involvement of SkpA, a Drosophila homologue of Skp1 [44], in ddaC dendrite pruning. Knockdown of SkpA with three independent RNAi lines, BL28974 (#1), v32790 (#2), and v107815 (#3), resulted in consistent dendrite pruning defects in most of ddaC neurons ( Figure 3C9 and 3F, and unpublished data). The expression of v32790 (#2) with two copies of ppk-Gal4 driver resulted in moderate ddaC dendrite pruning defects with full penetrance: an average of 6.9 primary and secondary dendrites remained attached (n = 21, 100%; Figure 3D9 and 3F). Attenuation of skpA function also inhibited ddaD/E pruning (n = 13; Figure S4A) and ddaF apoptosis (n = 15; Figure S4B). No MARCM clones for skpA 1 , a skpA null allele, could be recovered in ddaC neurons, presumably due to its essential functions during cell division [44]. Hence, SkpA, like Cul1 and Roc1a, appears to govern ddaC dendrite pruning.
Alternatively, we cannot exclude the possible existence of other Fbox proteins involved in ddaC dendrite pruning and ddaF apoptosis.
Thus, SkpA and Slimb appear to play important roles in regulating ddaC/D/E dendrite pruning and ddaF apoptosis.

The Cul1-Based SCF E3 Ligase Complex Regulates Axon Pruning of MB c Neurons
We next examined the potential requirements of Cul1, Roc1a, SkpA, and Slimb for MB axon pruning. In wild type, MB c neurons selectively eliminated their larval axon branches by 24 h APF (n = 5; Figure 2A and 2B-B9) and regenerated the medial branches in the adulthood (n = 6; Figure 2G). Importantly, cul1 EX mutant MB c neurons retained many larval axons, which were labeled by 201Y-Gal4-driven mCD8GFP expression (n = 19, 100%; Figure 2C9). These unpruned larval axons colabeled by FasII were located outside the major FasII-positive a/b lobes at 24 h APF (arrowheads in the right panels of Figure 2C9) and persisted in the adult brains (n = 5, 100%; Figure 2H). In addition to the pruning defect, we also observed a neuroblast-proliferation defect, as adult cul1 MB clones lacked late-born a/b neurons ( Figure 2H). Moreover, the cul1 EX axon pruning defects were fully rescued by reintroducing the Cul1 protein ( Figure S6A). Thus, Cul1, a core component of the SCF E3 ligase, also plays a critical role in regulating axon pruning of MB c neurons. The MB axon pruning and proliferation defects in cul1 EX mutant resemble those in mutants of uba1, a single E1 gene in fly, as reported previously [13]. Interestingly, roc1a G1 MARCM analyses also revealed a notable axon pruning defect in MB c neurons. Unpruned axon branches, positively labeled by 201Y-Gal4-driven mCD8GFP expression, remained at 24 h APF (n = 13, 62%; Figure 2D9). The axon pruning defects were rescued by expression of the full-length Roc1a protein (n = 11; Figure S6A). In contrast, roc1b exhibited normal pruning of MB c axons at 24 h APF ( Figure S6B). We were able to generate skpA 1 MARCM clones in MB c neurons, all of which exhibited strong axon pruning defects at 24 h APF (n = 8, 100%; Figure 2E9). Likewise, MB axon pruning was also inhibited at 24 h APF in slimb 2 (78%, n = 14; Figure 2F9) and slimb 8 (57%, n = 14; unpublished data) MARCM clones. Finally, dendrite pruning of MB c neurons failed to occur in cul1 EX (n = 9, 100%) or skpA 1 (n = 4, 100%) MB neuroblast clones at 24 h APF ( Figure S6C). Taken together, Cul1, Roc1a, SkpA, and Slimb govern both ddaC dendrite pruning and MB axon pruning, presumably as a Cul1-based SCF E3 ligase complex.

The Cul1-Based SCF E3 Ligase Complex Acts Downstream of EcR-B1 and Sox14
We next demonstrated the physical association among Cul1, Roc1a, SkpA, and Slimb. To this end, we first performed coimmunoprecipitation (co-IP) experiments in nontreated and ecdysone-treated S2 cells transfected with Myc-tagged Slimb. Endogenous unmodified Cul1 and Neddylated-Cul1 (Nedd8-Cul1 [37]), together with endogenous SkpA, were detected specifically in the immune complex when Myc-Slimb was immunoprecipitated from nontreated or ecdysone-treated protein extracts using an anti-Myc antibody ( Figure 4A), suggesting that Cul1, SkpA, and Slimb form a protein complex, independently of ecdysone. Moreover, Roc1a was specifically co-immunoprecipitated using the protein extracts of S2 cells co-transfected with Flag-tagged Roc1a and Myc-Slimb ( Figure 4B), also supported by a previous report that Roc1a preferentially associates with Cul1 in embryos [43]. We next performed coIP experiments using larval brain extracts expressing Myc-Slimb. Slimb was able to specifically co- To investigate whether the Cul1 SCF E3 ligase expression is upregulated during the larval-pupal transition, we utilized Laser Capture Microdissection (LCM) technique to microdissect MB c neurons from wild-type, EcR DN , and sox14 brains, subject to total RNA extraction and quantitative real-time PCR experiments (Q-PCR). Importantly, mRNA levels of cul1, like mical, were significantly upregulated from eL3, WP to 6 h APF ( Figure 4D). Upregulation of cul1 transcription at 6 h APF, like mical, was strongly inhibited by EcR DN expression or loss of sox14 function (50.4% and 48.9% reduction, respectively; Figure 4E), suggesting that cul1 upregulation is dependent on EcR-B1 and Sox14. Therefore, ecdysone signaling appears to regulate the abundance, but not assembly, of the Cul1 SCF E3 ligase complex. mRNA levels of the E1 gene uba1 were also increased upon pupal formation ( Figure 4D), in agreement with a previous microarray analysis showing that the UPS genes including cul1 and uba1 were upregulated in MB c neurons during the larval-pupal transition [46]. Moreover, knockdown of cul1 did not enhance the dendrite pruning defects in sox14 null mutant ddaC neurons (n = 41; Figure  S8). We also made MARCM ddaC clones for sox14 D13 and cul1 EX double mutant and observed an average of 15.2 primary and secondary dendrites attached to the double mutant ddaC neurons (n = 5; Figure 5G9 and 5J), similar to either cul1 EX (15.8; Figure 1E9 and 1G) or sox14 D13 (14.5, n = 6; Figure 5F9 and 5J) null MARCM neurons, supporting a linear relationship between sox14 and cul1.
Thus, cul1 appears to act downstream of sox14 during dendrite pruning.
Taken together, Cul1, Roc1a, SkpA, and Slimb function as the components of the SCF E3 ligase complex, among which the expression of cul1 is dependent on EcR-B1 and sox14 during dendrite pruning.

The Cul1-Based SCF E3 Ligase Complex Governs Dendrite Pruning in a Mical-Independent Manner
Given that the EcR-B1/Sox14/Mical pathway governs ddaC dendrite pruning [23], we next investigated how the SCF E3 ligase complex may integrate into this pathway. We first examined the protein levels of Mical, Sox14, and EcR-B1 in various SCF mutant ddaC neurons. Mical protein levels remained largely unchanged in the cul1 RNAi ddaC neurons (n = 12; Figure 5B and 5E) or cul1 EX MARCM (n = 8; unpublished data) at the WP stage, compared to those in the wild-type neurons (n = 14; Figure 5A and 5E). Likewise, the Mical levels were not affected in roc1a G1 MARCM (n = 7; Figure 5C and 5E), skpA RNAi (n = 12; Figure 5D and 5E), as well as slimb 8 MARCM (n = 13; unpublished data) ddaC neurons. Using a mical-lacZ reporter that drives upregulation of the LacZ expression under a mical enhancer (Y. Gu and F. Yu, unpublished data), we detected similar LacZ expression in wildtype, cul1 RNAi, and skpA RNAi WP ddaC neurons ( Figure S7A). These data indicate that the Cul1 SCF E3 ligase is dispensable for regulation of Mical transcription/expression in ddaC neurons. Moreover, the expression levels of EcR-B1 and Sox14 were unchanged in WP cul1 EX mutant ddaC neurons, compared with the wild-type controls ( Figure S7B). Likewise, the rest of the SCF components are not important for the expression of EcR-B1 and Sox14, as their protein levels were not affected in roc1a G1 , skpA RNAi, slimb 8 , or nedd8 AN015 mutant ddaC neurons ( Figure S7B). Thus, these data are consistent with the conclusion that the SCF complex functions downstream of EcR-B1 or sox14 during dendrite pruning. Given that Mical expression is unaffected by the mutants of the SCF components, we hypothesized that the Cul1 SCF E3 complex might act in parallel with Mical during dendrite pruning. If this is true, we would expect enhancement of dendrite defects of mical null mutants by compromising the SCF components. Indeed, RNAi knockdown of Cul1 in mical mutant ddaC neurons resulted in a drastic pruning defect with the persistence of 10.1 primary and secondary dendrites at 16 h APF (n = 22; Figure S8), whereas the mical null ddaC neurons retained 5.3 major dendrites (n = 26; Figure S8). Likewise, knockdown of Roc1a (n = 36; Figure S8), SkpA, or Slimb (n = 22 and 30, respectively; unpublished data) significantly enhanced the mical null mutant phenotypes. Knockdown of cul1 or roc1a in mical mutants exhibited normal elaboration of major dendrites at WP stage ( Figure S9). Moreover, MARCM clones of mical 15256 and slimb 8 double null mutant also exhibited a significant enhancement of dendrite pruning defects ( Figure 5K). An average of 9.7 primary and secondary dendrites were attached to the double mutant ddaC neurons (n = 10; Figure 5I9 and 5K), compared to either mical 15256 (3.3, n = 14; Figure 5H9 and 5K) or slimb 8 (4.4, n = 16; Figure 3E9 and 3F) null MARCM neurons. Thus, these data suggest that the Cul1 SCF E3 ligase facilitates ddaC dendrite pruning in parallel to Mical.
Taken together, our data suggest that the Cul1-based SCF E3 ligase complex acts downstream of EcR-B1/Sox14 and governs dendrite pruning in parallel with Mical.
The Cul1-Based SCF Ligase Complex Attenuates the Insulin Pathway to Promote ddaC Dendrite Pruning During tissue growth and pattern formation, Cul1, Roc1a, and Slimb negatively regulate Hh or Wg pathways by specifically degrading their respective effectors, Ci or Armadillo [45]. We hypothesized that if the Cul1-based SCF complex also attenuates the Hh and/or Wg pathways during ddaC pruning, the dendrite pruning defects associated with loss of the SCF ligase complex would be attributable to hyperactivation of either or both pathways. However, three lines of evidence indicate that the Cul1-based SCF complex acts independently of Hh and Wg pathways during dendrite pruning. First, inhibition of either pathway did not suppress cul1 RNAi dendrite pruning defects. Expression of the Hh repressors, Ci Cell (n = 29; Figure 6B and 6I) or the Patched receptor (unpublished data), did not suppress the pruning defects associated with cul1 RNAi, compared to the nonfunctional Mical N-ter control (n = 30; Figure 6A and 6I) that was unable to rescue the dendrite pruning defect in mical mutant ddaC neurons and its expression alone did not interfere with ddaC dendrite pruning [23]. Likewise, expression of the Wg inhibitors, Sgg S9A (n = 19; Figure 6C and 6I) or the truncated form of Dishevelled (unpublished data), did not suppress the cul1 RNAi effects. Second, the expression of the nondegradable Hh activator Ci U or the Wg activator Arm S10 did not affect normal ddaC dendrite pruning ( Figure S10). Finally, forced expression of the Hh activators, Ci U (n = 16; Figure S11) or the truncated form of Smoothened (D661-818; unpublished data), did not enhance the Cul1 DN pruning defects, compared to the Mical N-ter control (n = 25; Figure S11). Similarly, the expression of the Wg activator Arm S10 did not affect the pruning defects in the Cul1 DNexpressing ddaC neurons (n = 22; Figure S11).
To further investigate which signaling pathway might be attenuated by the Cul1-based SCF complex during ddaC pruning, we performed a candidate-based screen to systematically examine other important signaling pathways including Notch, Insulin, JNK, JAK/STAT, Hippo, EGFR, PVR, and Dpp. We expressed the dominant-negative repressors of these pathways, such as Notch DN , InR DN , Bsk DN , Dome DCYT , Yorkie S168A , EGFR DN , PVR DN , and Tkv DN , in cul1 RNAi ddaC neurons. From this screen, the Insulin pathway was identified as a potential target pathway that is negatively regulated by the Cul1 SCF E3 ligase complex. Notably, the expression of a dominant negative form of the Insulin Receptor (InR DN ), via one copy of ppk-Gal4 driver, dramatically suppressed the dendrite pruning defects in the cul1 RNAi-expressing ddaC neurons (n = 43; Figure 6D). On average, only 1.2 primary and secondary dendrites remained attached to these InR DN , cul1 RNAi double ddaC neurons, contrasting with 7.8 major dendrites observed in the mical N-ter , cul1 RNAi control (n = 30; Figure 6A and Figure 6I). In contrast, the expression of Notch DN (n = 24; Figure 6E and 6I) or other repressors did not influence the cul1 RNAi effects on ddaC dendrite pruning ( Figure  S12A). Expression of either of these dominant-negative transgenes did not affect normal ddaC pruning ( Figure S10, unpublished data). The numbers of primary and secondary dendrites were primarily unchanged at the WP stage in various double-mutant combinations, except an apparent reduction in the Sgg S9A , cul1 RNAi combination ( Figure S12B).
To verify the specific effect of InR, we further conducted the genetic enhancement experiments in which InR CA or Notch CA , known to activate Insulin or Notch pathways, respectively, was expressed in cul1 DN ddaC neurons. The expression of InR CA (n = 28; Figure S11) but not Notch CA (n = 18; Figure S11) resulted in a significant enhancement of the cul1 DN -associated pruning defects with full penetrance. Moreover, the InR DN expression significantly mitigated either cul1 DN (n = 22, Figure S13) or skpA RNAi (n = 15; Figure 6G and 6J) effects on dendrite pruning. Conversely, the expression of InR CA significantly enhanced the ddaC dendrite pruning defects caused by skpA RNAi knockdown (n = 19; Figure 6H and 6J). The numbers of primary and secondary dendrites remained similar at the WP stage in these double-mutant combinations, compared with single mutants ( Figure S12B).
Collectively, these genetic suppression/enhancement results indicate that the Cul1-based SCF complex promotes ddaC dendrite pruning primarily through attenuation of the Insulin pathway.

The PI3K/TOR Pathway Is Inhibited by the Cul1-Based SCF Complex During ddaC Dendrite Pruning
InR functions through the PI3K/TOR signaling pathway to mediate protein translation, metabolism, and ribosome biogenesis [47]. To examine whether the PI3K/TOR pathway could also be inhibited by the Cul1-based SCF complex during pruning, we performed genetic suppression assays by inactivating the PI3K/ TOR pathway in cul1 RNAi or cul1 DN ddaC neurons. Importantly, the expression of the dominant-negative form of PI3K (PI3K DN ) [48] or the Phosphatase and tensin homologue (PTEN) [49], both known to inactivate the PI3K pathway, like InR DN , drastically mitigated the dendrite pruning phenotypes in cul1 RNAi ddaC neurons (n = 32 and 39, respectively; Figure 7B, 7C, and 7K).  Figure S13). Thus, these data suggest that Cul1 also attenuates the PI3K pathway during ddaC pruning.
The PI3K pathway is interconnected with the Target of Rapamycin (TOR) pathway via TSC1 and TSC2, which act as negative regulators of the TOR pathway [47]. We next ascertained whether the TOR signaling pathway is also attenuated by the Cul1 SCF complex during ddaC dendrite pruning. Coexpression of TSC1 and TSC2, like PTEN, significantly suppressed the pruning defects in cul1 RNAi (n = 25; Figure 7E and 7K) or cul1 DN (n = 30; Figure S13) ddaC neurons. The truncated protein TOR TED lacking its toxic effector domain behaves as the dominant-negative form to disrupt the TOR signaling [50]. The TOR TED expression also strongly suppressed the dendrite pruning defects in cul1 RNAi (n = 47; Figure 7F and 7K) or cul1 DN (n = 24; Figure S13) ddaC neurons, respectively. The protein kinase TOR regulates protein synthesis via phosphorylation of the p70 ribosomal protein S6 kinase (S6K) [47] and the eukaryotic translation initiation factor 4E binding protein (4E-BP) [51]. S6K KQ , a catalytically inactive version, and 4E-BP(AA), a nonphosphorylated version, are able to repress protein translation and the TOR pathway [51,52]. The expression of 4E-BP(AA) rescued the dendrite pruning defects in almost all cul1 RNAi (n = 55; Figure 7G and 7K) or Cul1 DN -expressing (n = 32; Figure   S13) ddaC neurons. Likewise, the expression of S6K KQ also dramatically mitigated the dendrite pruning defects caused by cul1 RNAi knockdown (n = 42, Figure 7H and 7K) or cul1 DN expression (n = 18; Figure S13). Moreover, the expression of InR DN (n = 8; Figure 7I9) or 4E-BP(AA) (n = 8; Figure 7J9) also strongly suppressed the dendrite pruning defects in cul1 EX MARCM ddaC neurons, as approximately 4.9 and 4.3 primary and secondary dendrites remained attached at 16 h APF, respectively, compared to 15.8 major dendrites remaining in cul1 EX mutant neurons ( Figure 7L). To further confirm these genetic suppressions, we attenuated the InR/PI3K/TOR pathway by feeding the 3 rd instar larvae with Rapamycin, a pharmacological inhibitor of TOR. Rapamycin treatment did not affect the onset of puparium formation/adult eclosion, larval dendrite development ( Figure  S14A) or wild-type ddaC dendrite pruning ( Figure S14B). However, Rapamycin treatment significantly suppressed the ddaC dendrite pruning defects in cul1 RNAi ddaC neurons (n = 58), but not in mical RNAi neurons (n = 48, Figure S14B). Thus, these data further support the conclusion that the InR/PI3K/TOR pathway is inactivated by the Cul1-based SCF complex during ddaC dendrite pruning.
The following lines of evidence indicate the specificity of these suppression effects. First, attenuation of the InR/PI3K/TOR pathway did not affect normal elaboration of major dendrites in cul1 RNAi or cul1 DN mutant ddaC neurons, as the numbers of their primary and secondary WP dendrites were essentially unchanged despite the simplified terminal branches ( Figure S15A and S15B). Inactivation of the InR/PI3K/TOR pathway alone did not affect normal ddaC dendrite pruning ( Figure S10 and unpublished data). More importantly, the severing of major dendrites, a hallmark feature of dendrite pruning, occurred, similar to the wild type, in these suppression experiments (empty arrowheads, Figure S16A), suggesting that inactivation of the InR/PI3K/TOR pathway restores the severing of proximal dendrites from the cul1 RNAi ddaC neurons. Furthermore, the expression of InR DN , PI3K DN , PTEN, TOR TED , S6K KQ , or 4E-BP(AA) was not able to suppress the dendrite pruning defects associated with mical RNAi (unpublished data) or mical 15256 mutant ddaC neurons ( Figure S16B), supporting their specific genetic interactions with the SCF components.
Taken together, the InR/PI3K/TOR pathway is specifically inhibited by the Cul1-based SCF E3 complex in order to promote ddaC dendrite pruning during early metamorphosis.

Slimb Forms a Protein Complex with Akt and Promotes Akt Ubiquitination
Since inactivation of the InR/PI3K/TOR pathway suppresses the dendrite pruning defects in ddaC neurons lacking the Cul1 E3 ligase activity, we next assessed whether compromised E3 ligase function causes hyperactivation of the InR/PI3K/TOR pathway. To this end, we examined the expression and activity of Akt, a positive regulator of the InR/PI3K/TOR pathway, in ddaC neurons and 6 h APF brain lysates. Using an anti-Akt antibody, a weak localization of endogenous Akt was detected in ddaC somas ( Figure 8A), but not in dendrites and axons (unpublished data). Endogenous Akt was significantly upregulated in the somas of cul1 RNAi ddaC neurons at the WP stage (2.8 folds, n = 9; Figure 8B and 8D), compared to the wild-type somas (n = 8, Figure 8A and 8D). Akt signals were abolished in akt RNAi ddaC neurons ( Figure 8C and 8D). Since overexpressed Akt could be observed weakly in axons and dendrites in addition to its robust localization in ddaC somas ( Figure S17), we co-expressed Akt with cul1 RNAi or the control RNAi to examine the Akt protein levels throughout the neurons. Consistently, cul1 RNAi knockdown also caused a significant increase in Akt protein levels in ddaC somas (2 folds, n = 16), dendrites (2.2 folds, n = 12), and axons (1.9 folds, n = 11; Figure 8F-F0 and 8G), compared to the RNAi controls ( Figure 8E-E0 and 8G). As a control, the expression levels of GFP were the same or similar throughout cul1 RNAi or the control RNAi ddaC neurons (unpublished data). These data suggest that SCF-dependent Akt degradation is not restricted to dendrites of ddaC neurons. Consistently, the Cul1 SCF E3 complex appears to be localized throughout ddaC neurons as indicated by uniform distribution of exogenously expressed SkpA-RFP in the dendrites, axons, and somas (n = 7; Figure S17). Moreover, the protein levels of Akt were significantly increased in the cul1 RNAi brain lysates (the upper panel, Figure 8H) where Cul1 proteins were knocked down via the cul1 RNAi line (#1) using a pan-neuronal driver elav-Gal4 ( Figure  S18A). Concomitantly, Akt activity was also substantially increased, as judged by an increase in active and phosphorylated Akt levels (the middle panel, Figure 8H). Thus, the Cul1-based SCF E3 ligase negatively regulates Akt expression and activity in ddaC neurons and prepupal brains. We then investigated whether attenuation of Akt suppresses the dendrite pruning defects in cul1 RNAi ddaC neurons. Interestingly, knockdown of Akt with two independent RNAi lines (#1: BL31701 or #2: BL33615) potently suppressed the dendrite pruning defects in cul1 RNAi ddaC neurons (n = 26, Figure 7D; unpublished data). Reduction of akt function resulted in an average of 0.2 primary/ secondary dendrites connected to the cul1 RNAi ddaC neurons ( Figure 7K), compared to 8.1 in the Mical N-ter control ( Figure 7A and 7K). These biochemical and genetic data indicate that compromised SCF E3 ligase function causes hyperactivation of the InR/PI3K/TOR pathway.
To further examine a potential link between the Cul1 SCF E3 ligase and the InR/PI3K/TOR pathway, we assessed the physical interaction between Akt and the F-box protein Slimb. The C-terminal region of Slimb contains seven WD40 domains that are responsible for binding to its substrates and targeting them for ubiquitination. Interestingly, Akt was co-immunoprecipitated with Slimb using the protein extracts of S2 cells cotransfected with Myc-Akt and Flag-Slimb ( Figure 8I). We confirmed this interaction in postmitotic neurons, as Slimb was specifically co-immunoprecipitated with Akt in the prepual brain extracts expressing Myc-Slimb and Akt ( Figure 8J). Furthermore, Akt specifically interacted with Slimb, as Akt was not pulled down by either the truncated Slimb protein lacking its WD40 domains (Slimb DWD40 , Figure 8K) or another F-box protein Ago ( Figure  S18B). We then investigated whether Slimb can mediate ubiquitination of Akt. Notably, Slimb expression strongly increased the amount of polyubiquitinated Akt (lane 7, Figure 8L), compared to the control (lane 6, Figure 8L). In contrast, the expression of Slimb DWD40 failed to facilitate Akt ubiquitination (lane 8, Figure 8L), suggesting that the WD40 domains are responsible for Slimb-mediated ubiquitination of Akt. Thus, Slimb associates with Akt and targets Akt for ubiquitination.

Activation of the InR/PI3K/TOR Pathway Alone Is Sufficient to Inhibit ddaC Dendrite Pruning
We have demonstrated that inactivation of the InR/PI3K/ TOR pathway by the Cul1-based SCF E3 ligase complex is required to facilitate ddaC dendrite pruning during early metamorphosis. To further substantiate it, we examined whether constitutive activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit normal progression of ddaC dendrite pruning. Notably, the expression of InR CA , or PI3K CA , both known to constitutively activate the InR/PI3K pathway [48,53,54], via two copies of ppk-Gal4 driver, caused consistent dendrite pruning defects in the vast majority of ddaC neurons (InR CA , n = 25, 92%; and PI3K CA , n = 23, 91%; Figure 9B9 and 9C9). On average, 5.2 (InR CA ) and 4.3 (PI3K CA ) primary/ secondary dendrites retained the attachment to their respective ddaC neurons at 16 h APF ( Figure 9I). The dendrite pruning defect caused by InR CA expression was fully suppressed by akt RNAi knockdown (n = 20; Figure S19D), suggesting that expression of InR CA likely activates residual Akt to inhibit dendrite pruning. Supportively, using two PTEN null/strong alleles, PTEN C494 and PTEN 1 , we observed similar dendrite pruning defects in ddaC neurons (89%, n = 9 and 48%, n = 21, respectively; Figures 9D9 and S19B). Loss of PTEN function also inhibited ddaD/E dendrite pruning, but not ddaF apoptosis or MB axon pruning ( Figure S19B and S19C, unpublished data). Moreover, the expression of the small GTPase Rheb (Ras homologue enriched in brain) or the constitutively active form of S6K (S6K STDETE ), activators of the TOR pathway [47,52], led to dendrite pruning defects in the majority of ddaC neurons (77%, n = 25 and 63%, n = 24, respectively; Figure S19A). Activation of the InR/PI3K/TOR pathway appears not to affect the morphology and numbers of their major WP dendrites ( Figures 9A-D and S19A). Further, we co-expressed InR CA , PI3K CA , Rheb, or S6K STDETE with Cul1 DN (Figure S20) or roc1a RNAi (unpublished data) in ddaC neurons, resulting in more severe dendrite pruning defects. Similar to the Cul1 SCF E3 ligase, the InR/PI3K/TOR pathway regulates ddaC dendrite pruning in a Mical-independent manner. First, the protein levels of Mical, Sox14, or EcR-B1 were unaffected upon activation of the InR/PI3K/TOR pathway, via InR CA /PI3K CA expression or loss of PTEN function ( Figure  S21A-B). The upregulation of the Mical expression during the larval-pupal transition was also unaffected upon inactivation of the InR/PI3K/TOR pathway (unpublished data). Second, inactivation of the InR/PI3K/TOR pathway, via expression of InR DN , PI3K DN , PTEN, TOR TED , S6K KQ , or 4E-BP(AA), did not suppress the dendrite pruning defects in mical null mutant ddaCs ( Figure S16B). Finally, activation of the InR/PI3K/TOR pathway, via expression of InR CA (n = 27, Figure 9F), PI3K (n = 21, Figure 9G), or Rheb (n = 20, Figure 9H), significantly enhanced the mical null mutant pruning phenotypes ( Figure 9J).
Local caspase activation in dendrites was shown to be required for elimination of dendrites in ddaC neurons [24,25]. We therefore assessed whether SCF and InR/PI3K/TOR govern dendrite pruning through local caspase activation. Using the genetically encoded caspase reporter CD8::PARP::Venus [25], we observed no or negligible caspase activity in dendrites of cul1 RNAi (n = 9) or InR CA -expressing (n = 6) ddaC neurons at 6 h APF, in contrast to strong caspase activity in wild-type ddaC dendrites (n = 6; Figure  S22). Therefore, the SCF ligase and the InR/PI3K/TOR pathway regulate dendrite-specific pruning in ddaC neurons at least in part by promoting local caspase activation in the dendrites.
In summary, our data indicate that activation of the InR/ PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning in a Mical-independent manner. Thus, we demonstrate that during early metamorphosis, the Cul1-based SCF E3 ligase complex facilitates ddaC dendrite pruning primarily through inactivation of the InR/PI3K/TOR pathway.

Discussion
Previous studies showed that the UPS activity plays an intrinsic and essential role in governing both modes of pruning in ddaC [12] and MB c neurons [13]. However, little is known about the E3 ubiquitin ligase that is able to direct two distinct modes of neuronal pruning in Drosophila. Moreover, it is also unknown which downstream pathway is inactivated by the E3 ligase in order for ddaC neurons to prune their dendrites. Here, we report the identification of the Cul1-based SCF E3 ligase complex that plays a critical role in both modes of pruning of ddaC and MB c neurons. In a genome-wide RNAi screen, we first isolated Cul1, a core scaffold protein of the SCF E3 ligase, which is required for ddaC dendrite pruning and MB axon pruning during early metamorphosis. We further identified the other components of the E3 ligase complex including the RING domain protein Roc1a, the adaptor protein SkpA, as well as the F-box protein Slimb. These molecules, like Cul1, are all required for pruning of ddaC and MB c neurons. We show that the Cul1-based SCF E3 ligase acts downstream of EcR-B1/Sox14 and promotes ddaC dendrite pruning in a Micalindependent manner. Moreover, via a candidate-based screen, we observed that during ddaC dendrite pruning, the Cul1 E3 ligase negatively regulates the InR/PI3K/TOR signaling pathway but not other major developmental pathways examined. We demonstrate that inactivation of the InR/PI3K/TOR pathway leads to strong suppression of dendrite pruning defects in ddaC neurons deficient in the Cul1 E3 ligase activity, whereas activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. Thus, the Cul1-based SCF E3 ligase promotes dendrite pruning in ddaC neurons primarily through inactivation of the InR/PI3K/TOR pathway ( Figure 9K).

The Cul1-Based SCF E3 Ubiquitin Ligase Regulates Pruning of ddaC Dendrites and MB Axons in Drosophila
Previous studies indicated that UPS activity is cell-autonomously required for both MB axon pruning and ddaC dendrite pruning in Drosophila [12,13]. First, the expression of a yeast ubiquitin protease, which eliminates ubiquitin from substrates and inhibits UPS-mediated degradation [55], causes severe pruning defects in MB c and ddaC neurons [12,13]. Second, loss of the E1 enzyme Uba1 leads to strong pruning defects in both types of neurons. Third, removal of the proteasome subunits, Mov34 or Rpn6, also results in uba1-like pruning defects in MB and ddaC neurons. The selectivity of the UPS machinery for pruning is mainly conferred by a specific E3 ligase. Given the existence of a large number of E3 ligases in the Drosophila genome, it is challenging to identify the specific one that directs these two modes of neuronal pruning. It was reported that DIAP1, an RING E3 ligase, antagonizes the Dronc caspase activity and inhibits ddaC dendrite severing [24]. A model had been proposed that autoubiquitination and degradation of DIAP1 allows for the activation of the Dronc caspase in ddaC dendrites and thereby pruning of ddaC neurons. However, recent studies reported that all ddaC neurons sever their dendrites normally in DIAP1 gain-of-function mutants [26] and overexpression condition [25]. Nevertheless, Dronc caspase and DIAP1 appear not to be important for axon pruning of MB c neurons, as analyses of Dronc MARCM clones [24], expression of the caspase inhibitor p35 [13], or gain of DIAP1 function (Wong and Yu, unpublished data) revealed no axon pruning defects. Thus, the question of which specific E3 ligase directs both modes of pruning remains open for a long time. In this study, we demonstrate that the Cul1-based SCF E3 ligase complex plays key roles in regulating both ddaC dendrite pruning and MB axon pruning during Drosophila metamorphosis.
Several lines of evidence indicate that the Cul1-based SCF E3 ligase complex is the specific one that governs both ddaC dendrite pruning and MB axon pruning. First, while Cul1 and its binding partner Roc1a are critical for pruning of ddaC and MB neurons, Cul3 and Roc1b that preferentially bind each other are not important. Second, the Cul4-based E3 ligase, which regulates TSC2 protein stability and TSC1/2 complex turnover in Drosophila [56], is dispensable for ddaC dendrite pruning. MARCM analyses of the null allele cul-4 11L exhibited no ddaC dendrite pruning defect (n = 5; Figure S5). Consistently, overexpression of TSC1/TSC2 complex did not inhibit ddaC pruning (unpublished data). Third, among 31 fly F-box proteins, Slimb was identified, which when knocked down via RNAi, resulted in apparent dendrite pruning defects in ddaC neurons. Furthermore, Slimb appears to be important for both ddaC dendrite pruning and MB c axon pruning. Another F-box protein, Archipelago (Ago), is not required for ddaC pruning, in contrast to the essential role of its worm homologue SEL10 in synapse elimination [35]. MARCM analyses with ago 3 , a null allele [57], revealed no pruning defect in ddaC neurons (n = 3; Figure S5), supporting the selectivity of the E3 ligases for pruning. Finally, Nutcracker, the Fbox protein of an SCF ubiquitin ligase (E3) required for caspase activation during sperm differentiation [58], appears not to be essential for ddaC pruning (n = 9, unpublished data).

The Cul1-Based SCF E3 Ligase Acts in Parallel to Mical to Promote ddaC Dendrite Pruning
We previously reported that a transcriptional hierarchy, consisting of EcR-B1, Sox14, and Mical, is commonly induced in both ddaC neurons and MB c neurons, and essential for ddaC dendrite pruning [23]. How is the Cul1 E3 ligase integrated into this linear pathway? We propose that the Cul1-based SCF E3 ligase likely acts downstream of the transcriptional activators EcR-B1 and Sox14, however, in parallel to Mical, during ddaC dendrite pruning (see the model in Figure 9K). First, previous microarray analyses showed that the UPS genes, involved in all steps of the UPS pathway including cul1 and skpA, appear to be upregulated in remodeling MB c neurons during the larval-pupal transition [46]. Upregulation of these genes is abolished by the expression of EcR DN [46]. Second, our Q-PCR data further verify that cul1 is a downstream effector of EcR-B1 and sox14, suggesting ecdysone signaling regulates the abundance, but not assembly, of the Cul1 SCF E3 ligase complex. Third, ddaC neurons devoid of the Cul1 E3 ligase retain strong expression of Sox14. Reduction of cul1 function does not enhance sox14 mutant phenotype, supporting that the Cul1 E3 ligase acts downstream of sox14 during ddaC pruning. Finally, the Cul1-based E3 ligase is dispensable for the transcription and translation of mical in ddaC neurons, suggesting that it likely exerts its roles in a Micalindependent pathway. This is further supported by the fact that reduction of the E3 ligase function in mical null mutant ddaC neurons results in additive effects on dendrite pruning. Taken together, our data support the model in which the Cul1-based SCF E3 ligase acts in parallel with Mical to govern ddaC dendrite pruning, downstream of EcR-B1 and Sox14 ( Figure 9K).

The Cul1-Based E3 Ligase Inactivates the InR/PI3K/TOR Pathway During ddaC Dendrite Pruning
During pattern formation, the Cul1-based SCF E3 ligase degrades Ci or Arm to negatively regulate Hh or Wg pathways, respectively [45]. Importantly, our data demonstrate that the InR/ PI3K/TOR pathway, rather than Wg or Hh pathways, is inactivated by the Cul1-based E3 ligase in order for ddaC neurons to prune their dendrites. First, we show that inactivation of the InR/PI3K/TOR pathway suppresses the dendrite pruning defects in ddaC neurons lacking the Cul1 E3 ligase activity. Second, compromised cul1 function causes enhanced expression/activity of Akt, a positive regulator of the InR/PI3K/TOR pathway. These data suggest that compromised E3 ligase function causes activation of the InR/PI3K/TOR pathway. Consistently, a reversal of the UPS pathway by a deubiquitinating enzyme also leads to enhancement of the insulin signaling in mammals [59]. Finally, activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. In contrast, activation of the InR/ PI3K/TOR pathway by loss of PTEN function (n = 11) or InR CA expression (n = 7) is dispensable for MB axon pruning ( Figure  S19C). Given the distinct architecture and morphology of ddaC dendrites and MB c axons, it is conceivable that the SCF E3 ligase regulates differential target pathways during these two types of pruning. Like the Cul1 E3 ligase, the InR/PI3K/TOR pathway acts downstream of EcR-B1/Sox14 and regulates dendrite pruning in parallel to Mical (see the model in Figure 9K).
It has been reported that ecdysone can inhibit the release of insulin-like peptides from insulin-producing-cells during the larvalpupal transition, systemically inactivate the InR/PI3K/TOR pathway, and thereby terminate larval growth [60]. We show that during dendrite pruning, ecdysone signaling increases abundance of the Cul1-based SCF complex, which in turn inactivates the InR/PI3K/TOR pathway ( Figure 9K). These regulations with different mechanisms together would ultimately ensure lower levels of InR/PI3K/TOR activity/function in ddaC neurons, thereby resulting in dendrite pruning in these neurons during early metamorphosis. How does the Cul1-based SCF E3 ligase inactivate the InR/PI3K/TOR pathway during ddaC pruning? This negative regulation may occur through direct degradation of positive regulators of the pathway, such as IRS (fly Chico), Akt, and TOR. Several lines of evidence indicate that Akt is a good candidate as a substrate of the Cul1 SCF E3 ligase. First, the F-box protein Slimb, a component of the Cul1 SCF E3 ligase required for substrate recognition, specifically interacts with Akt. Second, Slimb promotes ubiquitination of Akt in a WD40dependent manner. Third, Akt levels are significantly increased in cul1 RNAi mutant ddaC neurons and brains. Finally, knockdown of Akt strongly suppresses the cul1 RNAi dendrite pruning defects. Since the InR/PI3K/TOR pathway and Akt are important for ddaC dendrite pruning but not for MB axon pruning, it is conceivable that the targets of the Cul1 SCF E3 ligase are divergent in these two modes of neuronal pruning. Interestingly, the SCF-Fbxo40 ligase targets the mammalian IRS for degradation and limits the insulin/InR signaling in skeletal muscle [61], whereas the RING E3 ligase ZNRF1 ubiquitinates Akt to promote Wallerian degeneration of injured dorsal root ganglia neurons, a pruning-like process [62]. Finally, our results also show that the Cul1-based SCF E3 ligase and the InR/PI3K/TOR pathway regulate local caspase activation in ddaC dendrites during pruning. Attenuation of the insulin pathway can downregulate IAP, an inhibitor of caspases, during the differentiation of embryonic chicken lens epithelial cells [63]. It will be of great interest to determine whether global attenuation of InR/PI3K/ TOR pathway can similarly downregulate DIAP1, a Drosophila homologue of IAP, to locally activate caspases during dendrite pruning.
In summary, we demonstrate that the Cul1-based SCF E3 ligase plays crucial roles in directing two distinct modes of pruning in ddaC and MB c neurons in Drosophila. We further show that the Cul1 SCF E3 ligase inactivates the InR/PI3K/TOR pathway, a key signaling pathway, in order for ddaC neurons to prune their unnecessary larval dendrites during metamorphosis. This study provides a novel link between the SCF E3 ligase and the InR/ PI3K/TOR pathway in regulating neuronal pruning. Thus, we open up new avenues for further studies of the E3 ligase in the remodeling and maturation of the developing nervous system, as well as their implications in the pathogenesis of many neurodegenerative diseases.

UAS-cul1 DN and UAS-skpA-RFP Transgenes
The GATEWAY pTW vector containing a fragment of the cul1 cDNA (encoding aa 1-532; Cul1 DN ) or GATEWAY pTRW vector containing a fragment of the skpA cDNA were constructed and several transgenic lines were established by the Bestgene Inc.

Mical Antibody Production
The cDNA fragment corresponding to the last 359 aa of Mical was amplified by PCR and verified by DNA sequencing. The product was expressed using the GST expression vector (pGEX 4T-1, Pharmacia) and the purified protein was used to immunize guinea pigs to generate antibodies against Mical. The specificity of the guinea pig anti-Mical antibody was verified using mical mutants.

MARCM Analysis and Dendrite Imaging/Quantification
We carried out MARCM analysis, dendrite imaging, and quantification as previously described [23].

Quantification of ddaC Dendrites
Live confocal images of ddaC neurons expressing UAS-mCD8-GFP driven by ppk-GAL4 were shown at w3L, WP, and 16 h APF. The average number of primary and secondary dendrites attached to soma or total dendritic termini was counted from wild-type and mutant ddaC neurons. The number of samples (n) in each group is shown on the bars. Error represents S.E.M. Dorsal is up in all images. The strength of the pruning phenotypes was divided into three levels: mild (less than 5 primary and secondary dendrites), moderate (5-10 primary and secondary dendrites), and strong/ severe (.10 primary and secondary dendrites).

Dissection of Brains, Visualization of Mushroom Body Neurons, and MARCM Mosaic Analysis
Larval, pupal, and adult brains were dissected in PBS and fixed in 4% formaldehyde for 15 min. Brains were washed in PBS+1%Triton X for 3 times for 10 min each. For the clonal analysis study (MARCM), embryos were collected at 6 h interval. The clones were induced in the first instar lavae by applying a 1 h heat shock at 38uC. The antibody to FASII (1D4) was obtained from the Developmental Studies Hybridoma Bank and used at 1:50. Rabbit anti-GFP antibody was obtained from Invitrogen and used at 1:1,000. 201Y-Gal4 labels postmitotic c neurons and a small subset of late-born a/b neurons. The Anti-FasII (1D4) antibody labels a/b neurons strongly and c neurons weakly. The severity of the axon pruning phenotypes was divided into three levels: weak, strong, and complete, according to a previous study [21].

Quantification of Immunolabeling
To quantify the immunolabeling intensities, cell nuclei (EcR-B1/Sox14 immunostaining) or whole cell body (Mical immunostaining) were drawn on the appropriate fluorescent channel based on the GFP channel relative cellular localization in ImageJ software. After subtracting the background (Rolling Ball Radius = 30) on the entire image of that channel, we measured the mean grey value in the marked areas in ddaC and ddaE on the same images and calculated their ratios. The ratios were normalized to corresponding average control values and subjected to statistical t test for comparison between different conditions (*p,0.05, **p,0.01, ***p,0.001, n.s., not significant). Graphs display the average values of ddaC/ddaE ratios and the standard error of means (S.E.M). n is shown on the bars. Insets show the ddaC neurons labeled by ppk-GAL4 driven UAS-mCD8-GFP expression. Dorsal is up in all images.
To quantify the immunolabeling intensities of Akt, soma/ dendrite/axon regions were drawn on the appropriate fluorescent channel based on the GFP channel relative cellular localization in ImageJ software. After subtracting the background, we measured the mean grey value of Akt in the marked areas. The values were normalized to corresponding average control values and subjected to statistical t test for comparison between different conditions. Graphs display the average values of normalized Akt expression and the standard error of means (S.E.M). n is shown on the bars. Insets show the soma/dendrite/axon labeled by ppk-GAL4 driven UAS-mCD8-GFP expression (*p,0.05, **p,0.01, ***p,0.001, n.s., not significant).

Rapamycin Treatment
To avoid any developmental delay, wild-type or mutant embryos were collected at 6 h intervals and were reared on standard food to the 3 rd instar stage (96 h after egg laying, AEL) before being transferred to the standard culture medium containing 2 mM of Rapamycin dissolved in ethanol (Sigma Aldrich R0395). Larvae were fed in Rapamycin food for approximately 8 h before cessation of feeding. The onset of puparium formation and adult eclosion was not affected by Rapamycin treatment. wL3 larvae were used to quantify the total dendrite termini.

RU486/Mifepristone Treatment for elav-GeneSwitch System
To avoid any developmental delay, wild-type or mutant embryos were collected at 6 h intervals and were reared on standard food to the e3L stage before being transferred to the standard culture medium containing 240 mg/ml mifepristone dissolved in ethanol (Sigma Aldrich M8046). The onset of puparium formation and adult eclosion was not affected by RU486 treatment.

RNA Isolation and Reverse Transcription
Isolation of RNA from MB c neurons was accomplished using Laser Capture Microdissection (LCM). Ten-micrometer frozen sections were cut from larval or pupal brains. MB c neurons labeled by 201Y-GAL4.mCD8-GFP were microdissected using the Zeiss PALM microbeam microdissection system. Each capture consisted of ,80 cell bodies, and 30 captures were pooled to obtain each replicate. Total RNA was extracted using the PicoPure RNA isolation kit from Arcturus and subject to reverse transcription using Oligo dT and the SuperScript III First-Stand Synthesis SuperMix (Invitrogen). The genomic DNA was digested by RNase-free DNase (Qiagen).

Quantitative Real-Time PCR (Q-PCR)
Independent experiments were conducted in triplicates using Maxima SYBR Green/ROX qPCR Master Mix (Fermentas) and 7900HT Fast Real-time PCR system (Applied Biosystems) according to the manufacturer's recommendations. rp49 (CG7939) was used as an internal control gene. Results were normalized to the controls indicated. Error bars represent standard error of mean (SEM) for four experimental repeats (n = 4).
Primers listed in 59-to-39 sequence: Expression level of ppk-Gal4 (inserted on Chr II) is much stronger than that of ppk-Gal4 (on Chr III). Thus, the phenotype of cul1 RNAi #2 in Figure 1C9 is almost double compared to that in Figure S1A due to the presence of two copies of ppk-Gal4 (Chr II). Please note that the ddaF neuron is occasionally labeled by one copy of ppk-Gal4 (Chr II) but always by two copies of the driver.  Figure S22 The SCF ligase and the InR/PI3K/TOR pathway regulate dendrite pruning in ddaC neurons at least in part by promoting local caspase activation in the dendrites. Confocal images of ddaC neurons expressing the caspase reporter construct CD8::PARP::VENUS at 6 h APF. While cleaved PARP marked in white bracket was readily detected in the wild-type ddaC neurons at 6 h APF, ddaC neurons expressing cul1 RNAi or InR CA failed to activate the caspase activity at 6 h APF. White arrowhead points to ddaC somas. The scale bar is 20 mm. See genotypes in Text S1. (TIF)

Supporting Information
Text S1 List of fly strains. Various genotypes were used in the main and supplementary figures. (DOCX)