JNK signaling coordinates with ecdysone signaling to promote pruning of Drosophila sensory neuron dendrites

During the development of the nervous system, neurons often form exuberant axon branches and dendritic arbors and then selectively prune those that make incorrect connections. One major form of developmental pruning is stereotyped large-scale pruning, in which a significant proportion of specific primary axons/dendrites and their branches are eliminated. Such large-scale pruning happens widely during the development of the nervous system, such as the development of subcortical projections and formation of retinotopic maps, and is crucial for the formation of precise neuronal connections (Luo and O'Leary, 2005). However, molecular mechanisms that control large-scale pruning are not fully understood.

In Drosophila, large-scale axon/dendrite pruning has been well characterized in mushroom body (MB) γ neurons in the brain and class IV dendritic arborization (c4da) sensory neurons in the peripheral nervous system (Yu and Schuldiner, 2014; Kanamori et al., 2015). These neurons undergo drastic remodeling during metamorphosis. MB γ neurons prune both their larval axons and dendrites, and then re-grow their axons and dendrites in adult-specific patterns (Lee et al., 1999), whereas c4da neurons only prune and re-grow their dendrites (Kuo et al., 2005). Studies in these two types of neuron have begun to reveal cellular processes and molecular mechanisms underlying axon/dendrite pruning. At the cellular level, both axon and dendrite pruning follow a similar series of cellular events, including breakdown of the microtubule and actin cytoskeleton, blebbing and fragmentation of axons/dendrites, and clearance of axon/dendrite fragments by glial or epidermal cells through phagocytosis (Watts et al., 2003; Lee et al., 2009; Awasaki et al., 2006; Han et al., 2014; Williams and Truman, 2005). Recent studies also reveal that pruning of MB axons and da neuron dendrites involves endocytosis (Zhang et al., 2014; Issman-Zecharya and Schuldiner, 2014) and a transient increase in calcium influx (Kanamori et al., 2013). At the molecular level, both axon and dendrite pruning require ecdysone receptor B1 (EcRB1) and its co-receptor Ultraspiracle (Usp) (Lee et al., 2000; Kuo et al., 2005). In the MB, TGF-β signaling together with the immunoglobulin protein Plum activates the expression of EcRB1 at early pupal stages (Zheng et al., 2003; Yu et al., 2013). EcRB1 promotes axon and dendrite pruning in part by activating Sox14 (Kirilly et al., 2009). In addition, the ubiquitin-proteosome system and caspases play crucial roles in both axon and dendrite pruning, and a few ubiquitylating enzymes and components of the proteasome involved in axon and/or dendrite pruning have been identified (Kuo et al., 2005; Watts et al., 2003; Kuo et al., 2006; Rumpf et al., 2011; Williams et al., 2006; Wong et al., 2013). Besides their shared mechanisms, MB axon pruning and da neuron dendrite pruning also employ distinct mechanisms. For example, Ik2 kinase, the cytoskeleton regulators Katanin-60 and Mical, and the ubiquitylating enzyme UbcD1 (Eff) are required for da neuron dendrite pruning but not for MB axon pruning (Lee et al., 2009; Kirilly et al., 2009; Kuo et al., 2006; Watts et al., 2003). However, despite the significant progress made over the past two decades, our understanding about the molecular mechanisms regulating axon/dendrite pruning is far from complete.

The JNK signaling pathway plays diverse roles in both the development of the nervous system and neurological diseases. Studies in JNK knockout mice have shown that JNK is required for brain morphogenesis, apoptosis, axon outgrowth/guidance and dendrite morphogenesis during the development. In addition, JNK also triggers neuronal death in various neurodegenerative diseases (reviewed by Coffey, 2014). JNK performs these functions by either functioning in the nucleus or in the cytoplasm. In the nucleus, JNK regulates gene transcription by phosphorylating substrates such as Jun or histone modifiers (Crocker et al., 2001; Tiwari et al., 2011). In the cytoplasm, JNK phosphorylates a variety of substrates, including microtubule-associate proteins, superior cervical ganglion 10 protein (SCG10), and postsynaptic density protein 95 (PSD95) that are involved in neural development or degeneration (Bjorkblom et al., 2005; Shin et al., 2012; Kim et al., 2007). However, the function of JNK signaling in large-scale dendrite pruning has not been investigated previously. In this study, we found that JNK coordinates with EcRB1 signaling to promote dendrite pruning and identified Fos as a novel downstream target of EcRB1 in c4da neurons. Furthermore, we discovered that Wnt5 signaling is also possibly required for dendrite pruning of c4da neurons and genetically interacts with JNK signaling.

In each hemi-segment of Drosophila larvae, there are three c4da neurons, ddaC, v'ada and vdaB, from dorsal to ventral. All c4da neurons undergo similar dendrite pruning processes during early metamorphosis, although dendrite pruning of v'ada and vdaB neurons is slightly later than that of ddaCs. Furthermore, vdaB neurons do not survive to adulthood (Kuo et al., 2005). Therefore, we focused on ddaC neurons for phenotypic analyses of dendrite pruning. During dendrite pruning, major dendritic branches, including primary and secondary branches of c4da neurons, are first severed from the soma at 5-10 h after puparium formation (h APF) (Fig. 1A-B). Severed dendrites are then cleared through degeneration and phagocytosis (Han et al., 2014; Williams and Truman, 2005). By 16-18 h APF, all dendrites are removed, leaving only the soma and axons (Fig. 1C) (Kuo et al., 2005).

To test whether the JNK signaling pathway is involved in dendrite pruning of c4da neurons, we examined how loss of Drosophila JNK, which is encoded by Basket (Bsk), would affect dendrite pruning of c4da neurons. We first inhibited JNK activity in c4da neurons using ppk-Gal4 (Grueber et al., 2007) to drive the expression of a dominant-negative form of JNK, JNK^DN (Adachi-Yamada et al., 1999). We found that expressing JNK^DN inhibited dendrite pruning in a dose-dependent manner. When one copy of UAS-JNK^DN was expressed, about 11% of ddaC neurons still had primary dendritic branches attached to the soma and 24% of ddaCs had severed dendritic fragments at 16-18 h APF (Fig. 1D,E,H). When two copies of UAS-JNK^DN were expressed, the percentages of ddaC neurons with attached dendritic branches or severed dendritic fragments were further increased to about 70% and 30%, respectively (Fig. 1H). In contrast, wild-type ddaC neurons only occasionally had some severed dendritic fragments but no attached dendritic branches at similar stages (Fig. 1C,H). The retention of attached dendritic branches or severed dendritic fragments indicates that dendrite severing was blocked or delayed when JNK was inhibited by its dominant-negative form. The dendrite-pruning defects resulting from the expression of JNK^DN is unlikely to be a secondary effect due to dendrite morphogenesis defects for the following two reasons. First, no obvious dendrite morphogenesis defects were observed in ddaC neurons expressing UAS-JNK^DN or homozygous mutant for bsk² at 3rd instar larval stages (Fig. S1A-C). Second, similar dendrite-pruning defects were still observed when we used tub-GAL80^ts to restrict the expression of UAS-JNK^DN to 1 day before puparium formation (Fig. S1D).

In order to confirm that dendrite-pruning defects resulting from the expression of JNK^DN were indeed caused by the loss of JNK function, we next examined whether knocking down JNK would lead to similar dendrite-pruning defects in c4da neurons. To maximize the RNAi knockdown efficiency, we raised animals at 29°C after larval hatching and examined dendrite pruning at 11-13 h APF, when dendrites are largely pruned in wild-type animals raised at the same temperature. We found that 8% of JNK knockdown ddaC neurons still had unsevered dendrites and 12% had severed dendritic fragments at 11-13 h APF, whereas in the wild type, severed dendritic fragments were only observed in fewer than 4% of ddaC neurons at the same stage (Fig. 1F,H).

To prove that the dendrite-pruning defects resulting from the expression of UAS-JNK RNAi were not off-target effects, we performed the following two experiments. First, we tried to rescue the knockdown phenotypes by expressing UAS-JNK. Indeed, expressing UAS-JNK significantly rescued the dendrite-pruning defects caused by JNK knockdown. Only about 1% of ddaC neurons had attached dendrites and 10% had severed dendritic fragments at 11-13 h APF when UAS-JNK was co-expressed with UAS-JNK RNAi (Fig. 1G,H). The rescue is unlikely due to dilution of the GAL4 effect because similar dendrite-severing defects were still observed in about 9% of ddaC neurons (13 out of 140 ddaC neurons examined) when one additional copy of UAS-mCD8-GFP was included while knocking down JNK by RNAi (Fig. S1G). Second, we examined whether removing one wild-type copy of the bsk gene would enhance the RNAi knockdown phenotypes. To this end, we performed the knockdown experiments at 25°C to reduce the efficiency of JNK knockdown. When JNK was knocked down in the wild-type background at 25°C, ∼3% of ddaC neurons had unsevered dendrites and 8% had severed dendritic fragments at 16-18 h APF. However, when JNK was knocked down in a bsk² heterozygous mutant background, the percentage of ddaC neurons with attached dendrites was increased to about 7% and the other 14% had severed fragments. The rescue of the knockdown phenotypes by UAS-JNK and the enhancement of the phenotypes in a bsk² heterozygous mutant background indicate that the dendrite-pruning defects resulting from the RNAi knockdown were not off-target effects. Consistent with these results, we observed unsevered dendrites in one out of 10 bsk² mutant ddaC clones and two out of 23 bsk^flp147E mutant ddaC clones (Fig. S1E-F). Taken together, the dendrite-pruning defects observed in ddaC neurons expressing JNK^DN or JNK RNAi and bsk mutant clones demonstrate that JNK is indeed required for dendrite severing of c4da neurons.

Local disassembly of microtubule structures in the dendrites is one of the earliest visible signs during dendrite pruning and occurs before thinning and severing of membrane dendrites (Fig. 2A-C′) (Williams and Truman, 2005). Previous studies have shown that loss of the microtubule-severing protein Katanin p60-like 1 (Kat-60L1) or the Tau kinase PAR-1 blocks dendrite pruning (Herzmann et al., 2017; Lee et al., 2009). In addition, JNK is known to regulate microtubule dynamics by phosphorylating microtubule-associated proteins, including Tau (Bjorkblom et al., 2005; Buée-Scherrer and Goedert, 2002). In order to understand how JNK regulates dendrite pruning, therefore, we next examined how microtubule structures in the dendrites might be affected when JNK was inhibited by JNK^DN. We found that microtubule structures remained intact in unsevered dendrites even at 11 h APF when JNK^DN was expressed in c4da neurons (Fig. 2D-F′), whereas in the wild-type c4da neurons, microtubules have already broken down around 5 h APF (Fig. 2A-C′). These results indicate that JNK is required for disassembly of microtubules during dendrite pruning in c4da neurons.

Dendrite pruning of c4da neurons occurs only at early pupal stages. Therefore, we wondered whether JNK is only activated at early pupal stages and whether JNK acts through the canonical nuclear pathway in c4da neurons. To address these issues, we examined the expression of the JNK reporter puckered (puc)-lacZ in c4da neurons from larval to early pupal stages. puc encodes a JNK phosphatase and is a transcriptional target of JNK signaling (Martin-Blanco et al., 1998). Activation of JNK signaling upregulates puc expression, which in turn provides negative feedback to JNK by dephosphorylation (Martin-Blanco et al., 1998). puc-lacZ lines (puc^A251.1F3 or puc^E69) are lacZ enhancer trap lines (Dobens et al., 2001) that have been widely used as reporters of nuclear JNK activity (Neisch et al., 2010; Willsey et al., 2016). Using the puc-lacZ (puc^A251.1F3) as a reporter, we observed that the lacZ expression could be detected in c4da neurons at both larval and early pupal stages, and its expression levels remained similar (Fig. 3A-D). These data suggest that JNK is constantly activated in c4da neurons from larval and pupal stages, and its activity does not show obvious dynamic changes. Furthermore, the expression of puc-lacZ also indicates that JNK likely acts through the canonical nuclear pathway.

If JNK is activated at both larval and pupal stages, then why does JNK promote dendrite pruning only at early pupal stages? Could it be that JNK downstream effectors are only expressed/activated at early pupal stages? JNK controls various cellular processes by regulating the activity of a number of substrates through phosphorylation. As puc-lacZ is expressed in c4da neurons, it is possible that JNK acts through the canonical pathway in the nucleus to phosphorylate and activate the AP-1 complex, which is composed of Drosophila Jun and Drosophila Fos. Studies in both vertebrates and invertebrates show that Fos expression is dynamically regulated by various stimuli (Greenberg and Ziff, 1984; Weiser et al., 1993; Rusak et al., 1990; Morgan et al., 1987). Therefore we wondered whether the expression of Drosophila Fos, which is encoded by kayak (kay), in ddaC neurons shows dynamic changes during the transition from larval to early pupal stages. Immunostaining of Fos in ddaC neurons showed that this is indeed the case. At the 3rd instar larval stages, no obvious Fos expression could be detected (Fig. 4A,E). At the white pupal stage (0 h APF), we observed relatively low levels of Fos expression (Fig. 4B,E). The expression of Fos further increased at 2-3 h APF (Fig. 4C,E). These data demonstrate that Fos expression in ddaC neurons is specifically activated at early pupal stages.

Given that ecdysone signaling plays a key role in regulating axon/dendrite pruning in Drosophila (Lee et al., 2000; Kuo et al., 2005), we then asked whether the expression of fos was activated by ecdysone signaling. To address this issue, we inhibited ecdysone signaling in c4da neurons by expressing a dominant-negative form of ecdysone receptor B1 (EcRB1^DN) (Cherbas et al., 2003). Expressing EcRB1^DN completely inhibits dendrite pruning of c4da neurons (data not shown) (Kuo et al., 2005). Interestingly, expressing EcRB1^DN abolished the expression of Fos in ddaC neurons at early pupal stages (Fig. 3D-E). These results suggest that the expression of Fos in c4da neurons is activated by ecdysone signaling.

Previous studies have shown that EcRB1 regulates dendrite pruning by activating the expression of the transcription factor Sox14 (Kirilly et al., 2009). Therefore, we wondered whether there might be a cross-regulatory relationship between Fos and Sox14 for their expression in c4da neurons. To address this issue, we examined how loss of Sox14 would affect the expression of Fos and vice versa. We expressed UAS-sox14 RNAi to knockdown Sox14 or expressed a dominant-negative form of Fos (Fos^DN) to inhibit the activity of Fos in c4da neurons. Knocking down Sox14 by RNAi completely inhibited dendrite pruning like the expression of EcRB1^DN (Fig. S2E-F,H) as reported previously (Kirilly et al., 2009). However, Fos expression in c4da neurons at 3 h APF remained comparable with that in wild-type animals when Sox14 was knocked down by RNAi (Fig. S2A-B′). Likewise, expressing Fos^DN did not lead to obvious changes of the expression of Sox14 in c4da neurons at 3 h APF either (Fig. S2C-D′). These data suggest that, although the expression of both Fos and Sox14 is activated by EcRB1, Fos and Sox14 do not depend on each other for their expression and likely act in two independent pathways downstream of EcRB1. In support of this notion, expressing UAS-Fos could not rescue the dendrite-pruning defects resulting from Sox14 knockdown (Fig. S2G,H).

We next wanted to determine whether Jun and/or Fos function downstream of JNK to promote dendrite pruning of ddaC neurons. We first examined how loss of Jun, encoded by Jra (Jun-related antigen), or Fos would affect dendrite pruning of ddaC neurons. For the loss-of-function phenotypic analyses of Jun, we either generated jun mutant clones or knocked down Jun using RNAi. We found that over 60% of ddaC neurons homozygous mutant for jun¹, a nonsense mutant allele that produces a truncated protein lacking the basic region and leucine zipper (Kockel et al., 1997), still had unsevered dendrites at 16-18 h APF (Fig. 5A,B,F). Similarly, knockdown of Jun led to retention of unsevered dendrites in about 40% of ddaC neurons (Fig. 5C,F). The retention of unsevered dendrites in jun¹ mutant and Jun knockdown ddaC neurons demonstrates that Jun is required for dendrite pruning of c4da neurons.

Next, we tried to determine whether Fos is similarly required for dendrite pruning of c4da neurons. To this end, we inhibited Fos activity by expressing a dominant-negative mutant form of Fos (Fos^DN), which is a truncated version of Fos carrying the bZIP domain only (Fos^bZIP) (Eresh et al., 1997). The isolated bZIP domain can still dimerize with Jun but lacks transcriptional activation activity. Fos^DN has been widely used for inhibiting Fos function (Collins et al., 2006; West et al., 2015). Our results showed that inhibiting Fos activity by Fos^DN led to dendrite-severing defects in ddaC neurons in a dose-dependent manner. When one copy of UAS-d-Fos^DN was expressed, ∼23% of ddaC neurons had unsevered dendrites and other 10% of ddaCs had severed dendritic fragments at 16-18 h APF (Fig. 5D,F). When two copies of UAS-Fos^DN were expressed, the percentage of ddaC neurons with attached dendrites was increased to nearly 50% and another 20% had severed dendritic fragments (Fig. 5E,F). Dendrite-pruning defects resulting from the expression of UAS-d-Fos^DN suggest that, like Jun, Fos is also required for dendrite pruning of c4da neurons. Taken together, the dendrite-pruning defects observed in jun mutant and Fos^DN-expressing ddaC neurons suggest that JNK promotes dendrite pruning likely by regulating the activity of the AP1 complex in C4da neurons.

JNK signaling is involved in many cellular processes and can be activated by a variety of extracellular stimuli (Behrens et al., 2000; Leppä and Bohmann, 1999). We next asked what extracellular stimuli activated JNK signaling during dendrite pruning. We investigated some candidates that could potentially activate JNK signaling by examining dendrite-pruning defects in ddaC neurons expressing UAS-RNAi transgenes or homozygous mutant for candidate genes. These candidates include TNFα (Eiger) and its receptor (Wengen) as well as TNF receptor-associated factors, receptors for growth factors (including EGFR, VEGFR and TGFβR) and Wnt5 (Table S1). We found that knockdown of the Activin receptor Baboon (Babo) (Fig. S3A-B) and mutations of wnt5 (Fig. 6A-C,G) led to dendrite-pruning defects in c4da neurons. It has been demonstrated previously that Babo regulates axon pruning by functioning through Smad to activate EcRB1 expression in MB neurons (Zheng et al., 2003). Consistently, knockdown of Smad also resulted in dendrite-pruning defects in ddaC neurons (Fig. S3C), suggesting that Babo regulates dendrite pruning of c4da neurons similarly by activating Smad in ddaC neurons. Therefore, we focused on Wnt5 in our studies.

Wnt5 is not essential for viability, making it possible to examine dendrite pruning in wnt5 homozygous mutant animals. We used two wnt5 mutant alleles, wnt5⁴⁰⁰ and wnt5^D7, both of which are loss-of-function alleles generated by imprecise excision from the P-element insertion line P{GT1} wnt5^BG00642 (Fradkin et al., 2004; Yoshikawa et al., 2003). We found that ∼7-10% of ddaC neurons in wnt5⁴⁰⁰ or wnt5^D7 mutant animals still had unsevered dendrites and another 10-12% had severed dendritic fragments at 16-18 h APF, when wild-type c4da neurons have already pruned their dendrites (Fig. 6A-C,G). Similar dendrite-pruning defects were also observed in wnt5⁴⁰⁰/wnt5^D7 transheterozygous mutant animals (Fig. 6D,G), indicating that the dendrite-pruning defects in wnt5 mutants were not caused by background mutations. These results demonstrate that Wnt5 is required for dendrite pruning in c4da neurons.

To determine whether Wnt5 regulates dendrite pruning by activating JNK signaling in c4da neurons, we then performed the following three experiments. First, we carried out dominant genetic interaction tests to determine whether Wnt5 and JNK act in the same pathway. To this end, we examined whether removing one wild-type copy of wnt5 would enhance the dendrite-pruning defects resulting from the expression of one copy of UAS-JNK^DN in c4da neurons. We reasoned that if Wnt5 activates JNK in c4da neurons, reducing the expression of Wnt5 would lead to reduced JNK activity and enhance the dendrite-pruning defects resulting from the expression of UAS-JNK^DN. Indeed, dendrite-pruning defects induced by one copy of UAS-JNK^DN were significantly enhanced by the reduction of Wnt5 expression. The percentage of c4da neurons with attached dendrites at 16-18 h APF increased from 12% to ∼25% when one copy of UAS-JNK^DN was expressed in animals heterozygous mutant for wnt5^D7 or wnt5⁴⁰⁰, whereas wnt5^D7 or wnt5⁴⁰⁰ heterozygous mutants did not show obvious dendrite-pruning defects at the same stage (Fig. 6H). The enhancement of JNK^DN-induced dendrite-pruning defects appears to be specific for wnt5 because no similar enhancement was observed in animals heterozygous mutant for wntD^KO1, wntD^KO2, wg^1-8, wnt2^I and wnt4^EMS23 (Fig. 6H and data not shown). These data suggest that Wnt5 and JNK genetically interact and possibly function in the same pathway to promote dendrite pruning.

Second, we examined whether the dendrite-pruning defects in wnt5^D7 mutants could be partially rescued by enhancing JNK signaling. We enhanced JNK activity by reducing the expression of Puc, which provides inhibitory feedback on JNK (Martin-Blanco et al., 1998). Interestingly, the dendrite-pruning defects in wnt5^D7 mutant animals were significantly rescued by removing one wild-type copy of puc. The percentage of wnt5^D7 mutant ddaC neurons with attached dendrites was reduced from 10% to only 1.5% at 16-18 h APF in puc^A251.1F3/+ heterozygous mutant background, whereas puc^A251.1F3/+ heterozygotes did not show any dendrite-pruning defects (Fig. 6E-G). The partial rescue of the dendrite-pruning defects in wnt5^D7 mutants by increasing JNK activity and the enhancement of the JNK^DN-induced dendrite-pruning defects by removing one wild-type copy of wnt5 strongly suggest that Wnt5 likely regulates dendrite pruning of c4da neurons by activating JNK signaling.

Last, we examined JNK activity in wnt5 mutants using puc-lacZ as a reporter to see whether JNK activity might be reduced in wnt5 mutants. However, we did not observe dramatic reduction in puc-lacZ expression in wnt5^D7 hemizygous mutant animals at 0 h APF compared with that in wild-type animals (Fig. S4A-C), indicating that Wnt5 might not be solely responsible for the activation of JNK signaling in c4da neurons.

In Drosophila, Wnt5 binds to the RYK (related-to-tyrosine-kinase) receptor Derailed (Drl) to regulate axon guidance and salivary gland development (Yoshikawa et al., 2003; Harris and Beckendorf, 2007). A recent study also demonstrated that both Drl and a closely related RYK receptor, Derailed-2 (Drl-2), have a partially redundant function in c4da neurons to mediate Wnt5 signaling to specify the dendritic territory of adult c4da neurons (Yasunaga et al., 2015). To investigate whether Wnt5 similarly functions through Drl and/or Drl-2 to regulate dendrite pruning of c4da neurons, we first examined whether Drl was expressed in larval c4da neurons. Antibody staining revealed that Drl was indeed expressed in c4da neurons at both the 3rd instar larval stage and early pupal stages (Fig. 7A-B′). Consistently, UAS-mCD8-GFP driven by drl-GAL4, a GAL4 enhancer trap line in drl that reflects the endogenous expression pattern of Drl (Moreau-Fauvarque et al., 1998), was also specifically expressed in c4da neurons on the larval body surface but not in other peripheral sensory neurons (Fig. 7C). The expression pattern of drl-GAL4 in larval c4da neurons is consistent with its expression in adult c4da neurons (Yasunaga et al., 2015). Therefore, Drl is endogenously expressed in c4da neurons throughout development and could potentially mediate Wnt5 signaling.

We next investigated whether Drl and Drl-2 are required for dendrite pruning of c4da neurons. We used three viable mutant alleles, drl², drl^P1 and drl^Red2 (Dura et al., 1995; Bonkowsky et al., 1999), for examining the involvement of Drl in dendrite pruning. We observed consistent weak dendrite-pruning defects in animal homozygous mutant for drl², drl^P1 or drl^Red2. About 6-11% of ddaC neurons still had attached dendrites and the other 8-22% had severed dendritic fragments in these mutant animals at 16-18 h APF (Fig. 7D-G,I). To determine whether the pruning defects in drl mutants were indeed caused by loss of Drl rather than background mutations and whether Drl function cell autonomously in c4da neurons, we next examined whether similar dendrite-pruning defects could still be observed in drl²/drl^P1 transheterozygous mutants and whether expressing Drl in c4da neurons could rescue the pruning defects. We found that dendrite-pruning defects still existed but were much weaker in drl²/drl^P1 transheterozygous mutants. Only 2% of ddaC neurons that still had unsevered dendrites and 5% had severed fragments in drl²/drl^P1 animals at 16-18 h APF (Fig. 7H,L). However, expressing UAS-drl-myc in c4da neurons could not rescue the dendrite-pruning defects in drl^P1 or drl^Red2 mutants (Fig. 7L). Given that similar dendrite-pruning defects were observed in pupae homozygous mutant for three independent drl alleles, as well as in drl²/drl^P1 transheterozygotes, the pruning defects are likely caused by the loss of Drl. However, failure of the rescue of the pruning defect by expressing UAS-drl-myc indicates that Drl may actually function non-cell-autonomously to regulate dendrite pruning in spite of the expression of Drl in c4da neurons. Alternatively, the pruning defects observed in drl mutants may result from other subtle developmental defects that were not visible at the gross morphology level.

For assessing the potential role of Drl-2 in dendrite pruning, we examined dendrite pruning in drl2^E124 (Inaki et al., 2007) and drl2^DG38805 (Bellen et al., 2004) homozygous mutant animals at 16-18 h APF. Our results showed that, in drl-2^E124 mutants, ∼14% of ddaC neurons still had attached dendrites and 27% had severed dendritic fragments at 16-18 h APF (Fig. 7I,L). In drl-2^DG38805 mutants, about 12% of ddaC neurons still had attached dendrites and many higher-order branches were also retained in these neurons at 16-18 h APF. In addition, over 70% of drl-2^DG38805 mutant ddaC neurons had a large amount of severed dendritic fragments at 16-18 h APF (Fig. 7J,L), similar to wild-type ddaC neurons at around 10 h APF (Fig. 1B), indicating that severing of primary dendrites had been significantly delayed in most drl-2^DG38805 mutant ddaC neurons. Similarly, in drl-2^E124/drl-2^DG38805 transheterozygotes, about 8% and 12% of ddaC neurons still had unsevered dendrites or severed dendritic fragments, respectively, at 16-18 h APF. However, the dendrite-pruning defects in drl-2 mutants were only observed in animals with the 3rd chromosome homozygous but not heterozygous for UAS-mCD8-GFP, ppk-GAL4, although animals homozygous for UAS-mCD8-GFP, ppk-GAL4 alone on the 3rd chromosome did not have any obvious dendrite-pruning defects in c4da neurons. These results suggest that Drl-2 could also be potentially involved in dendrite pruning in C4da neurons, but it may not play a major role given that the pruning defects were only observed in specific genetic backgrounds.

In this study, we found that JNK signaling is required for dendrite pruning of c4da neurons. Our data suggest that JNK regulates dendrite pruning likely by acting through its canonical pathway to promote microtubule disassembly. Remarkably, the temporal control of dendrite pruning by JNK at early pupal stages seems to be achieved not through changes in JNK activity but rather by the availability of its substrate Fos. We show that the expression of Fos is specifically activated at early pupal stages by EcRB1, a major regulator of axon/dendrite pruning (Kuo et al., 2005; Lee et al., 2000). Thus, our work not only identified a novel signaling pathway involved in dendrite pruning and a new downstream effector of EcRB1 in c4da neurons, but also revealed that JNK and ecdysone signaling coordinate to regulate dendrite pruning (Fig. S5).

Dendrite pruning shares many common cellular and molecular mechanisms with pathological axon degeneration as well as developmental axon pruning (Yu and Schuldiner, 2014; Luo and O'Leary, 2005). It has been well documented that JNK and other MAPKs are central to axon degeneration resulting from injury or neurodegenerative diseases (reviewed by Geden and Deshmukh, 2016; Coffey, 2014). A recent study reported that JNK is also required for developmental pruning of mushroom body axons in developing Drosophila brains (Bornstein et al., 2015). However, mechanisms of JNK-mediated dendrite pruning of c4da neurons are likely different from those of JNK-mediated developmental axon pruning or injury-induced axon degeneration. In injury-induced axon degeneration or developmental axon pruning, JNK mainly functions locally in the axon to promote axon degeneration instead of acting through the canonical pathway to regulate gene expression by activating the AP-1 complex in the nucleus. In injury-induced axon degeneration, JNK acts locally in the axon to promote the degradation of the axon survival factors NMNAT2 and SCG10, and the subsequent depletion of NAD⁺ and ATP in severed axons (Walker et al., 2017; Shin et al., 2012; Yang et al., 2015). In axon pruning of MB neurons, JNK destabilizes the cell-adhesion molecules Fasciclin (FasII) in axons (Bornstein et al., 2015). The nuclear reporter of JNK, puc-lacZ, is not expressed in MB neurons, nor does loss of Jra or Kay function lead to axon-pruning defects in MB neurons (Bornstein et al., 2015). In contrast, in c4da neurons, our results demonstrate that JNK likely functions through the canonical pathway to promote dendrite pruning. First, puc-lacZ is expressed in the nuclei of c4da neurons. Second, loss of either Jun or Fos results in dendrite-pruning defects. It will be interesting to identify target genes that are activated by the AP-1 complex during dendrite pruning of c4da neurons, which may provide mechanistic details of developmental dendrite pruning. Interestingly, even though JNK is required for dendrite pruning of da neurons and axon pruning of MB neurons, loss of JNK does not block dendrite pruning in MB neurons (Bornstein et al., 2015). Therefore, even dendrites in different neuronal types could employ distinct mechanisms for regulating their development pruning.

Although JNK is required for the dendrite pruning of da neurons, our work shows that the temporal control of the JNK-mediated dendrite pruning is achieved not through dynamic changes of the JNK activity but rather through activation of Fos expression specifically at early pupal stages. The expression of the JNK reporter puc-lacZ in c4da neurons remains constant from larval to early pupal stages, whereas the expression of Fos can only be detected at early pupal stages. Therefore, JNK may perform different functions in c4da neurons at different developmental stages, depending on the availability of distinct substrates. It has been well documented that the expression of Fos is dynamically regulated and can be induced by various physiological and pathological stimuli, such as growth factors, neuronal activity and injury (Greenberg and Ziff, 1984; Morgan et al., 1987; Rusak et al., 1990; Weiser et al., 1993). In this study, we found that expression of Fos in c4da neurons was activated by EcRB1 at early pupal stages. Therefore, JNK signaling coordinates with ecdysone signaling to regulate dendrite pruning of da neurons. Ecdysone signaling activates the expression of Fos either directly or indirectly, whereas JNK regulates the activity of the AP-1 complex composed of Fos and Jun through phosphorylation.

In addition to Fos, a previous study has shown that EcRB1 also activates the expression of Sox14 to regulate dendrite pruning of c4da neurons. Sox14 in turn activates the expression of the cytoskeleton regulator Mical (Kirilly et al., 2009). However, Fos and Sox14 likely function in parallel instead of linearly because our results show that the expression of Fos and Sox14 does not seem to depend on each other, nor could expression of UAS-Fos rescue the dendrite-pruning defects resulting from Sox14 knockdown. Therefore, EcRB1 promotes dendrite pruning possibly by activating the expression of multiple target genes, including Sox14, Fos and possibly other unknown targets.

How is JNK activated in c4da neurons? JNKs are components of the canonical MAPK pathway and can be activated by a variety of stimuli, such as cytokines, growth factors and cellular stress (Leppä and Bohmann, 1999; Behrens et al., 2000). In this study, we showed that loss of Wnt5 led to similar dendrite-pruning defects that could be significantly rescued by reducing the expression of JNK inhibitor Puc, whereas removing one wild-type copy of wnt5 significantly enhanced the dendrite-pruning defects caused by the expression of JNK^DN. These genetic interaction data suggest that Wnt5 regulates dendrite pruning possibly by activating JNK signaling in c4da neurons, but Wnt5 might be only partially responsible for the activation of JNK in c4da neurons because there was no dramatic reduction in the expression of the JNK reporter puc-lacZ in wnt5^D7 hemizygous mutants. Other unknown signals might also be involved in the activation of JNK in c4da neurons. Alternatively, Wnt5 and JNK signaling could act independently in parallel to regulate dendrite pruning of c4da neurons. However, we do not rule out the possibility that the dendrite-pruning defects observed in wnt5 as well as drl mutants (see discussion below) might be secondary effects due to other subtle developmental defects that are not obviously visible at the gross morphological level. It will be interesting in the future to investigate whether other unknown signals that activate JNK signaling regulate dendrite pruning in c4da neurons.

Studies in various model systems have shown that Wnt5 can bind to different receptors, including Frizzled, Ror2, CD146 and RYK, to activate JNK (Kumawat and Gosens, 2016; Fradkin et al., 2010; Nishita et al., 2010). A recent article reported that, in adult c4da neurons, Wnt5 acts through Drl and Drl2 to restrict the growth of dendrites within specific dendritic territories (Yasunaga et al., 2015). We show in this study that there are consistent dendrite-pruning defects in different drl or drl-2 mutant alleles, suggesting that Drl and Drl-2 may also be involved in dendrite pruning in c4da neurons. However, the failure of UAS-drl-myc to rescue the pruning defects observed in drl mutants indicates that the dendrite-pruning defects in drl mutants are likely non-cell-autonomous, or may be secondary to other subtle developmental defects in drl mutants. The fact that the dendrite-pruning defects in drl-2 mutants were only observed in a genetic background with homozygous UAS-mCD8-GFP, ppk-GAL4 on the 3rd chromosome suggests that Drl-2 may only have a minor role in mediating dendrite pruning. Therefore, further investigations are needed in the future to determine which receptor(s) mediate(s) Wnt5 function in dendrite pruning if Wnt5 indeed activates JNK signaling in c4da neurons to promote dendrite pruning.

For dominant-negative inhibition, the following transgenic lines were used: UAS-JNK^DN [6409, Bloomington Drosophila Stock Center (BDSC)], UAS-d-Fos^DN (also named UAS-Fra.Fbz, BDSC 7214 and BDSC 7215), and UAS-EcRB1^DN (BDSC 6872). The following lines were used for RNAi knockdown: UAS-djun RNAi [fly stocks of the National Institute of Genetics (NIG), Japan, 2275R-1], UAS-sox14 RNAi (BDSC 34794), UAS-bsk RNAi [BDSC 31323, targeting to a 483 bp sequence (from +1028 bps to +1051 bps downstream of the transcription start site) within the coding region], UAS-babo RNAi [Vienna Drosophila Resource Center (VDRC), Austria, 853] and UAS-dSmad2 RNAi (NIG 2262-R-1). bsk², FRT^40A/Cyo, bsk^flp147E, FRT^40A/Cyo, FRT^G13, jun¹/Cyo, wnt5^D7, wnt5⁴⁰⁰, drl², drl^P1, drl^Red2, drl2^E124 and drl-2^DG38805 were used for mutant phenotypic analyses. elav-GAL4, hs-FLPase, UAS-mCD8-GFP; tubP-GAL80, FRT^40A/Cyo and elav-GAL4, hs-FLPase, UAS-mCD8-GFP; FRT^G13, tubP-GAL80/Cyo fly lines were used for generating bsk and jun mutant clones of c4da neurons with the mosaic analysis with repressible cell marker (MARCM) technology (Lee et al., 1999). ppk-GAL4 lines on the 2nd or 3rd chromosome (BDSC 32078 and BDSC 32079) were used to drive the expression of transgenes in c4da neurons.

MARCM clones of c4da neurons were induced by 1 h heat shock at 38°C at 3 h after egg laying. Animals were then raised at 25°C until 3rd instar larval stage for identifying larvae with c4da neuron clones with live confocal imaging. Larvae with c4da neuron clones were further kept at 25°C until 16-18 h APF for examining dendrite-pruning defects. For Jun RNAi knockdown and dominant-negative inhibition of JNK and Fos, animals were also maintained at 25°C and dendrite-pruning defects were examined at 16-18 h APF. RNAi knockdown of JNK was performed at 25°C or 29°C, and dendrite pruning was examined at either 16-18 h APF or 11-13 h APF, respectively, as indicated in the text. UAS-Dcr2 (BGSC #24650) was co-expressed in ddaC neurons to boost RNAi knockdown efficiency.

For examining the expression of puc-lacZ, Fos, Sox14 or Drl in c4da neurons, animals at 3rd instar larval or early pupal stages were dissected, fixed and stained with the following primary antibodies: rat anti-mCD8 (Life Technologies, 1:100, 14-0081-82), rabbit anti-galactosidase (MP Biomedicals, 1:1000, 55976), rabbit anti-Fos (FabGennix International, 1:250, DFOS-101AP), mouse anti-Sox14 (Kirilly et al., 2009) (a gift from Dr. Fenwei Yu, Temasek Life Sciences Laboratory, National University of Singapore, 1:200) and rat anti-Drl (Lundgren et al., 1995) (a gift from Dr. J. B. Thomas, Molecular Neurobiology Laboratory, Salk Institute for Biological Studies, La Jolla, CA, USA, 1:200). Secondary antibodies conjugated to Cy2, Cy3, Cy5 or DyLight 647 (Jackson ImmunoResearch; 715-165-151, 711-165-152, 711-225-152, 712-225-153 and 712-175-153) were used at 1:100, 1:500, or 1:500, respectively. C4da neurons were imaged either in live animals or in stained larval or pupal fillets using a Leica Sp2 or Zeiss LSM780 confocal microscope and processed with Adobe Photoshop. Staining intensities of β-gal and Fos in the nuclei of c4da neurons were measured using Adobe Photoshop. Two-tailed Student's t-test was used for statistical analyses.

We thank Drs T. Adachi-Yamada, A. Singh, A. DiAntonio, J. B. Thomas, D. Bhomann, F. Yu, H. Ying, C. Hama, the Bloomington Drosophila Stock Center and Fly Stocks of National Institute of Genetics for fly stocks and antibodies; Sriharsha Gowtham and Xiaobing Deng for technical support; members of the Jab lab and the Zhu lab for thoughtful discussion and comments; and Dr M. Connell for proofreading.