Drosophila GSK3β promotes microtubule disassembly and dendrite pruning in sensory neurons

Neurons can remodel their axons and dendrites during animal development. This characteristic enables them to adapt to physiological and environmental changes. During a specific period of development, axons and dendrites undergo pruning. Exuberant or incorrect neuronal projections or neurites, which form at earlier stages, can be selectively removed at later stages (Luo and O'Leary, 2005; Riccomagno and Kolodkin, 2015; Schuldiner and Yaron, 2015). Both insufficient and excessive pruning are correlated with human neurological disorders, such as autism-spectrum disorders and schizophrenia (Sellgren et al., 2019; Tang et al., 2014). Thus, a full understanding of the mechanisms underlying neuronal pruning might provide some insights into the pathogenesis of neurological disorders.

Holometabolous insects such as Drosophila experience complete metamorphosis (Truman and Riddiford, 2019). Their central (CNS) and peripheral (PNS) nervous systems are required to restructure accordingly as the body transitions from larval to adult stages. Dendritic arborization (da) neurons are a group of PNS neurons that undergo extensive remodeling during metamorphosis. Among the four classes of dorsal da neurons, ddaA/F (class III) and ddaB (class II) neurons are eliminated via apoptosis, whereas ddaC (class IV, also known as C4da) and ddaD/E (class I) prune their entire dendrite arbors without a concurrent loss of their axons (Kuo et al., 2005; Shimono et al., 2009; Williams and Truman, 2005). In ddaC neurons, the pruning process is initiated by the release of the steroid hormone ecdysone at the late larval stage (Truman, 1990). Upon ecdysone release, it binds to the ecdysone receptor (EcR) and its co-receptor Ultraspiracle (Usp) to activate downstream signaling and thereby promote dendrite pruning (Kuo et al., 2005; Williams and Truman, 2005). EcR and Usp are cell-autonomously required for the expression of the transcription factor Sox14, which in turn upregulates the F-actin disassembly enzyme Mical (Kirilly et al., 2009). Apart from this pathway, ecdysone signaling also enhances the activity of the ubiquitin proteasomal system (Chew et al., 2021; Wong et al., 2013), caspase, calcium and calpain-mediated fragmentation of the dendritic membrane (Kanamori et al., 2013; Williams et al., 2006), as well as the degradation of cell-adhesion molecules, such as Neuroglian and Ppk26 via the endolysosomal pathway (Krämer et al., 2019; Zhang et al., 2014). These successive events result in a stereotypic progression of ddaC dendrite pruning, beginning with the appearance of varicosities along the proximal dendrites 2-3 h after puparium formation (h APF). Disassembly of the cytoskeletons leads to a severing event at the proximal regions of the dendrites around 6 h APF, followed by dendrite fragmentation and debris phagocytosis by the surrounding epidermal cells (Han et al., 2014; Williams and Truman, 2005) (Fig. 1A). A loss of microtubules occurs around 6 h APF in the proximal dendrites, a process known as microtubule disassembly, before dendritic membrane breaks down (Lee et al., 2009). Local breakdown of microtubules requires regulators of both microtubule polarity (Herzmann et al., 2018; Rui et al., 2020; Tang et al., 2020; Wang et al., 2019) and disassembly (Bu et al., 2021; Herzmann et al., 2017; Lee et al., 2009). Par-1 kinase, a well-known regulator of cell polarity (Benton and St Johnston, 2003) and cytoskeletal integrity (Drewes et al., 1997), promotes microtubule breakdown and dendrite pruning likely via phosphorylating its substrate Tau (Herzmann et al., 2017).

In an effort to identify other genes or pathways involved in dendrite pruning, we isolated Shaggy (Sgg), the sole fly homologue of mammalian Glycogen synthase kinase 3β (GSK3β), which is required for dendrite pruning in ddaC neurons. GSK3β/Sgg kinases are expressed abundantly in the brain (Trostnikov et al., 2019; Woodgett, 1990) and have been well documented to play important roles in governing neuronal growth, polarization and migration (Hur and Zhou, 2010), as well as in the regulation of axonal transport (Banerjee et al., 2021; Kim et al., 2011; Mudher et al., 2004). Unlike most kinases, GSK3β/Sgg kinases are active by default, unless subjected to inhibitory regulation through the competitive phosphorylation of specific residues (Cohen and Goedert, 2004). Their effects are therefore wide-ranging with more than 100 substrates bearing various phosphorylation sites for GSK3β/Sgg (Patel and Woodgett, 2017; Sutherland, 2011). Many of them are microtubule-associated proteins (MAPs) such as MAP1B/Futsch (Gögel et al., 2006; Goold et al., 1999), through which GSK-3β/Sgg modulates cytoskeletal dynamics and consequently synaptic formation/morphology (Cuesto et al., 2015; Franco et al., 2004). More intriguingly, GSK3β/Sgg not only directly phosphorylates MAP2/Tau and affects its microtubule-binding capabilities (Wagner et al., 1996), but also cooperates with Par-1 kinase in the regulation of Tau function and toxicity (Nishimura et al., 2004). However, despite their inherent connections to multiple MAPs and their overall importance in neuronal development, a potential role for GSK-3β/Sgg in developmental neurite pruning has so far not been reported in either fly or mammals.

In this study, we identify Sgg in an RNA interference (RNAi) screen and report its important role in dendrite pruning of ddaC sensory neurons. Sgg functions as a negative regulator of microtubules to disassemble microtubules and increase turnover rates. We show that the pruning defects associated with sgg knockdown are due to enhanced microtubule stability. Moreover, Sgg acts synergistically with Par-1 kinase to accelerate microtubule disassembly during dendrite pruning. Thus, Sgg and Par-1 kinases might phosphorylate a common MAP(s) to promote microtubule disassembly and thereby dendrite pruning.

In a large-scale RNAi screen searching for new players of dendrite pruning, we used the class IV da-specific driver pickpocket (ppk)-Gal4 to knock down gene function in ddaC neurons (Kirilly et al., 2009, 2011). Among those positive hits isolated, two independent RNAi lines, 1 (BDSC 35364) and 2 (BDSC 38293), target the same gene sgg. Knockdown of sgg (RNAi 1 or 2), using two copies of ppk-Gal4 driver, led to consistent dendrite severing defects in ddaC neurons by 16 h APF (1, 68%; 2, 66%) (Fig. 1C,D,H,I). In contrast, wild-type ddaC neurons pruned away all their larval dendrites at this time point (Fig. 1B,H,I). Sgg, a Drosophila homolog of mammalian GSK3β kinase, is highly conserved in animals and regulates various signaling pathways and developmental processes (Patel and Woodgett, 2017; Ruel et al., 1993). To further verify the sgg function in dendrite pruning, we generated mutant ddaC clones using the hypomorphic allele sgg^M1-1 (Peifer et al., 1994). sgg^M1-1 mutant clones exhibited similar but weaker dendrite pruning defects at 16 h APF (Fig. 1E,H,I; wild-type clones, Fig. 1H,I). To examine the requirement of the kinase activity of Sgg, we continuously overexpressed either of two kinase-inactive forms, Sgg^A81T and Sgg^Y214F (Bourouis, 2002), using the ppk-Gal4 driver. Overexpression of Sgg^A81T or Sgg^Y214F also caused dendrite pruning defects in ddaC neurons (Fig. S1A), attesting its important role in pruning. Moreover, we temporally induced the expression of Sgg^A81T from the early 3rd instar larval stage via the gene-switch system. After RU486 induction, Sgg^A81T-overexpressing ddaC neurons exhibited notable dendrite severing defects in 77% of ddaC neurons examined (Fig. 1G-I), in contrast to no or negligible pruning defects in non-induced neurons (Fig. 1F,H,I). Thus, these temporal gene-switch data suggest a requirement for Sgg at the prepupal stage when dendrite pruning starts to take place.

We next examined the expression of Sgg in ddaC neurons by using the sgg protein trap line sgg::GFP, in which a GFP exon is spliced into the endogenous sgg gene (Morin et al., 2001). sgg::GFP adults are viable and fertile, indicating that Sgg::GFP protein is functional. Using the sgg::GFP line, we observed the expression of Sgg in ddaC and other neighboring da neurons (Fig. S1C). Its expression did not fluctuate from the early 3rd instar larval (eL3) stage to 6 h APF (Fig. S1C), indicating that sgg expression is not regulated by ecdysone signaling. To assess whether Sgg can regulate ecdysone signaling during the larval-to-pupal transition, we examined EcR and Mical expression upon sgg knockdown. The expression levels of EcR and Mical were unaltered in sgg RNAi ddaC neurons at the white prepupal (WP) stage (Fig. S1D), showing that Sgg is not required for activation of ecdysone signaling.

Like class IV neurons, other classes of da neurons also undergo remodeling within a similar timespan. Wild-type dendrites of class I da (ddaD/E) neurons were completely pruned by 19 h APF (Fig. S2A,C), whereas class III da (ddaA/F) neurons were removed via apoptosis by 16 h APF (Fig. S2D). We observed that although Sgg knockdown led to dendrite retention in 47% of class I ddaD neurons (Fig. S2B,C), it did not prevent the apoptosis of class III ddaF neurons (Fig. S2E). Taken together, the serine-threonine kinase Sgg is cell-autonomously required for dendrite pruning in da neurons, but not for apoptotic events during their remodeling.

Sgg suppresses Hedgehog (Hh) or Wingless (Wg) pathways by promoting the degradation of their respective effectors, Cubitus interruptus or Armadillo, during pattern formation or tissue growth (Jia et al., 2002; Price and Kalderon, 2002). Given that in our previous study both Hh and Wg pathways were shown to be dispensable for dendrite pruning (Wong et al., 2013), Sgg is unlikely to suppress the Hh or Wg pathways in ddaC neurons. Microtubule regulators, such as Par-1, kinesins, Patronin, Mini spindles, Protein phosphatase 2A, Efa6 and Stathmin (Stai), have recently been reported to regulate dendritic microtubule disassembly/orientation and dendrite pruning in ddaC neurons (Bu et al., 2021; Herzmann et al., 2018, 2017; Rui et al., 2020; Tang et al., 2020; Wang et al., 2019). As GSK3β/Sgg suppresses microtubule polymerization via phosphorylation of MAPs, including MAP1B/Futsch and MAP2/Tau (Gögel et al., 2006; Goold et al., 1999; Wagner et al., 1996), we therefore explored its potential roles in microtubule polymerization in the dendrites of ddaC neurons. To this end, we first made use of the EB1-GFP marker to label microtubule plus ends and assess in vivo polymerization/orientation of dendritic microtubules. In mammalian neurons, axonal microtubules are aligned with their plus-ends distal (plus-end out), whereas dendritic microtubule orientation is mixed (Akhmanova and Steinmetz, 2015; Baas and Lin, 2011). In Drosophila da sensory neurons, the vast majority of dendritic EB1-GFP comets move retrogradely towards the soma, indicative of a predominantly minus-end-out microtubule orientation in major dendrites. In contrast, all axonal EB1-GFP comets migrate away from the soma in an anterograde manner, signifying a plus-end-out microtubule orientation in axons (Stone et al., 2008). Indeed, in the control ddaC neurons, almost all the dendritic EB1-GFP comets moved towards the soma with the average track length of 5 µm within 3 min (Fig. 2A,G,I). The average speed and number of EB1-GFP comets were 0.08 µm/s (Fig. 2A,H) and 5 comets per 30 µm dendrites (Fig. 2A,J), respectively. However, in sgg RNAi (1 and 2) neurons, the track length and speed of EB1 comets were increased significantly (Fig. 2B,C,G,H), whereas the orientation and number of the comets remained unaltered (Fig. 2B,C,I,J). Thus, these data suggest that Sgg negatively regulates the polymerization of dendritic microtubules in ddaC neurons but is dispensable for their minus-end-out orientation.

To assess whether Sgg is sufficient to accelerate dendrite pruning and inhibit microtubule polymerization, we continuously overexpressed either the wild-type form Sgg or the constitutively active form Sgg^S9A from the embryonic to pupal stages. Overexpression of Sgg or Sgg^S9A did not promote precocious dendrite pruning at 7 h APF (Fig. S1B), suggesting that active Sgg is not sufficient to accelerate dendrite pruning. Unexpectedly, overexpression of Sgg^S9A, but not Sgg, caused dendrite pruning defects in 43% of ddaC neurons at 16 h APF (Fig. S1B). In contrast to those in the control neurons (Fig. 2D,G,J), the EB1-GFP track length was significantly decreased from 5.8 µm to 4.6 µm upon Sgg overexpression (Fig. 2E,G), concomitantly with a significant increase in average numbers from 5.8 to 7.7 per 30 µm dendrites in Sgg-overexpressing neurons (Fig. 2E,J). The decrease in EB1-GFP track length is associated with the increase in the comet number, suggesting that, upon Sgg overexpression, reduced microtubule polymerization might be compensated for by enhanced microtubule nucleation. Interestingly, overexpression of active Sgg (Sgg^S9A) led to inhibition of microtubule polymerization in ddaC neurons. Sgg overexpression did not affect the speed and direction of EB1-GFP movement in the dendrites (Fig. 2E,H,I). Likewise, overexpression of Sgg^S9A also led to a significant reduction in comet track lengths (Fig. 2F,G). Interestingly, the dendritic microtubule orientation was impaired in such neurons, as the percentage of the anterogradely moving EB1-GFP comets was significantly increased from 2% to 12% upon Sgg^S9A overexpression (Fig. 2F,I; control, Fig. 2D,I). Given a high correlation between dendrite pruning and microtubule orientation, the pruning defects associated with Sgg^S9A overexpression (Fig. S1B) might be due to impaired microtubule orientation in the dendrites. Overexpression of Sgg^S9E, which potentially mimics a partially inactive state of the kinase, caused weak dendrite pruning defects in ddaC neurons (Fig. S3A), in contrast to sgg RNAi knockdown. Moreover, we observed significant reductions in microtubule levels and EB1-GFP comet length, as well as a significant increase in EB1-GFP comet number in the dendrites of Sgg^S9E-overexpressing ddaC neurons (Fig. S3B,C). These Sgg^S9E phenotypes differ from the sgg RNAi phenotypes but resemble overexpression of wild-type Sgg. Thus, these data indicate that the S9E mutation is unlikely to impair endogenous Sgg function in ddaC neurons. Similar to our finding in ddaC neurons, a previous study also shows that S9E mutation does not disrupt Sgg function; instead, it resembles wild-type Sgg protein and S9A mutation during wing formation (Bourouis, 2002).

To further confirm that overexpression of Sgg^S9A impairs dendritic microtubule orientation, we also examined the distribution of the dendritic and axonal microtubule markers Nod-lacZ and Kin-lacZ, respectively. Nod-lacZ is the chimera in which the motor domain from the minus-end-directed kinesin Nod and the coiled-coil domain of kinesin 1 are fused to the N-terminus of β-Galactosidase. Nod-lacZ is often used to detect microtubule minus ends in fly neurons (Clark et al., 1997). As in control neurons, sgg RNAi neurons showed normal distribution of Nod-lacZ in the dendrites (Fig. S4A). Kin-lacZ is a plus-end microtubule marker consisting of the motor and coiled-coil domains of kinesin 1 fused to β-Galactosidase (Clark et al., 1997). In sgg RNAi neurons, Kin-lacZ was localized to the axons but not in the dendrites (Fig. S4B), similar to that in control neurons (Fig. S4B). However, overexpression of Sgg^S9A resulted in a significant reduction in dendritic Nod-lacZ levels (Fig. S4A) and the mislocalization of Kin-lacZ into the dendrites (Fig. S4B), confirming that Sgg^S9A overexpression has impaired the typical minus-end-out orientation of the dendrites. Collectively, Sgg inhibits microtubule polymerization in the dendrites of ddaC neurons, while hyperactive Sgg disturbs the dendritic microtubule orientation.

Given that Sgg negatively regulates microtubule polymerization, we further examined whether it promotes microtubule turnover before dendrite pruning. To do so, we took advantage of a chimeric protein that contains a tandem photoconvertible EOS dimer and α-tubulin (tdEOS::α-tubulin) (Barlan et al., 2013; Lu et al., 2013; Tao et al., 2016). As the ability of microtubule sliding is lost in mature neurons (Lu et al., 2013, 2015), we made use of the photoconversion approaches to measure microtubule turnover in mature ddaC neurons from the 3rd instar larval stage (Tao et al., 2016). After tdEOS::α-tubulin was photoconverted from green to red in a segment of proximal dendrites, we measured the remaining intensity of the converted signals after a 30 min recovery. In the dendrites of sgg RNAi (1 and 2) ddaC neurons, higher amounts of the photoconverted tdEOS::α-tubulin remained in their dendrites at WP (Fig. 3B-D) and wL3 (Fig. S5B-D) stages, when compared with those in control neurons (Fig. 3A,D, Fig. S5A,D), suggesting that dendritic microtubule turnover slows down in the absence of Sgg. As microtubule sliding does not occur in mature neurons (Lu et al., 2013), decreased turnover rates in the absence of Sgg suggest that microtubule disassembly was inhibited. Conversely, Sgg or Sgg^S9A overexpression significantly accelerated the decay of the photoconverted tdEOS::α-tubulin in the dendrites at both WP (Fig. 3F-H) and wL3 (Fig. S5F-H) stages, compared with the controls (Fig. 3E,H, Fig. S5E,H). Thus, Sgg promotes microtubule turnover, possibly via microtubule depolymerization.

To further examine whether Sgg is required for microtubule levels in the dendrites, we measured the levels of microtubules via the MAPs Tau and Futsch (22C10) (Heidary and Fortini, 2001; Roos et al., 2000) or of stable microtubules via acetylated α-tubulin (K40) (Prokop, 2022). Futsch and Tau are the Drosophila homologues of mammalian MAP1B and MAP2, respectively (Heidary and Fortini, 2001; Roos et al., 2000). In contrast to control neurons, sgg RNAi neurons exhibited significant increases in microtubules in their proximal and distal dendrites at the wL3 stage (Fig. 4A,B, Fig. S6A), as shown by both Futsch and Tau staining. Knockdown of Sgg also resulted in elevated levels of acetylated α-tubulin (K40) (Fig. S6B), which indicates stable microtubules (Janke and Bulinski, 2011; Prokop, 2022). On the contrary, Sgg^S9A overexpression led to mild but significant reductions in Futsch and Tau levels in the dendrites (Fig. 4C,D), compared with those in control neurons (Fig. 4C,D). These data indicate that Sgg negatively regulates microtubule mass in the dendrites of ddaC neurons. To further confirm that Sgg is required for the disassembly of dendritic microtubules at the onset of dendrite pruning, we measured the levels of Futsch-positive microtubules after sgg RNAi knockdown was induced by RU486 treatment from the eL3 stage onwards. As a control, RU486-induced sgg RNAi knockdown exhibited dendrite severing defects in 58% of ddaC neurons (Fig. S6C). Compared with no alterations at the wL3 stage (Fig. 4E), Futsch levels were slightly elevated at the WP stage (Fig. S6D) but more significantly increased by 6 h APF in ddaC dendrites (Fig. 4F). Thus, this gene-switch result suggests that Sgg is required to facilitate the disassembly of dendritic microtubules at the prepupal stage and thereby promote dendrite pruning. It is possible that Sgg activity or function might be regulated by ecdysone signaling before the onset of dendrite pruning.

During dendrite pruning, local microtubule disassembly precedes membrane breakage, suggesting that this early cellular alteration may drive the severing of proximal dendrites (Lee et al., 2009; Watts et al., 2003; Williams and Truman, 2005). Given the fact that Sgg accelerates microtubule turnover and disassembly before pruning, the dendrite pruning defects in sgg RNAi and mutant neurons might be due to increased microtubule stability. To test this possibility, we made use of both pharmacological and genetic manipulations to depolymerize microtubules in sgg RNAi neurons and examined if the dendrite pruning defects were suppressed. We first impaired microtubule polymerization by treatment with the microtubule-depolymerizing drug colchicine. 72 h AEL larvae were fed with a low concentration of colchicine (5 μg/ml) for 2 days, which did not affect normal head eversion and survival to adulthood. Compared with those in the DMSO-treated controls, the length of EB1-GFP comets was significantly reduced; however, the number of EB1-GFP comets was increased in the dendrites of colchicine-treated sgg RNAi ddaC neurons (Fig. S7A). As a control, colchicine treatment did not interfere with the normal progression of dendrite pruning in wild-type ddaC neurons at both 6 h APF (Fig. S7B) and 16 h APF (Fig. 5B,I), similar to that in non-treated neurons (Fig. S7B, Fig. 5A,I). Compared with those in non-treated sgg RNAi animals (Fig. 5C,I), colchicine treatment largely rescued dendrite pruning defects in sgg RNAi neurons at 16 h APF (Fig. 5D,I).

Recently, we have reported that the microtubule regulators Efa6 and Stai inhibit microtubule polymerization in the dendrites of ddaC neurons (Bu et al., 2021). We found that the length, number and speed of EB1-GFP comets were drastically reduced in sgg RNAi neurons overexpressing Efa6 (Fig. S7A), compared with that in sgg RNAi controls (Fig. S7A). Like colchicine treatment, the overexpression of Efa6 significantly suppressed the sgg RNAi dendrite pruning phenotype (Fig. 5F,J), in comparison with the control (Fig. 5E,J). We then removed one copy of γTub23C to attenuate microtubule nucleation and thereby compromise microtubule polymerization in dendrites (Bu et al., 2021; Hertzler et al., 2020). Removal of one copy of γTub23C (γTub23C^A15-2/+) in sgg RNAi neurons led to a significant reduction in the number of dendritic EB1-GFP comets (Fig. S7A). Interestingly, the loss of a single γTub23C copy (γTub23C^A15-2/+), which did not cause precocious dendrite pruning at 6 h APF (Fig. S7C), led to a significant suppression in the dendrite pruning defects of sgg RNAi neurons at 16 h APF (Fig. 5H,J). These data indicate that reduced microtubule polymerization suppressed the dendrite pruning defects in sgg RNAi neurons. On the other hand, we stabilized the microtubules by removing either Efa6 or Stai (Belmont and Mitchison, 1996; Qu et al., 2019). Knockdown of Efa6 or Stai was previously shown to significantly increase microtubule mass in the dendrites of ddaC neurons (Bu et al., 2021). Accordingly, both Efa6 and Stai knockdowns significantly enhanced dendrite pruning defects in sgg RNAi neurons (Fig. S8A-H). Thus, these genetic and pharmacological data suggest that the dendrite pruning defects of sgg RNAi neurons are caused by increased microtubule polymerization/mass. Taken together, our data suggest that Sgg regulates dendrite pruning at least partly by promoting microtubule disassembly.

Mammalian Sgg and Par-1 kinases (GSK3β and MARKs, respectively) can phosphorylate Tau and other MAPs to release them from microtubules and thereby destabilize microtubules (Drewes et al., 1997; Wagner et al., 1996). In Drosophila, Sgg has been reported to act downstream of Par-1 to phosphorylate Tau and modulate its toxicity (Nishimura et al., 2004). As fly Par-1 can phosphorylate Tau and promote dendrite pruning (Herzmann et al., 2017), we further explored the possible relationship between sgg and par-1 during pruning. To this end, we conducted a series of genetic interaction analyses between sgg and par-1. Double knockdown of Sgg and Par-1 led to severe pruning defects with full penetrance (100%; Fig. 6C-E), compared with their respective controls (33% in sgg RNAi control, Fig. 6A; 37% in par-1 RNAi control, Fig. 6B). Such an enhancement was not observed in the concurrent knockdown of Sgg and the F-actin disassembly factor Mical (Fig. S9A,C-E), which instead only showed defects at a level similar to mical RNAi (Fig. S9B,D,E). These observations support a synergistic effect between Sgg and Par-1 in dendrite pruning. Importantly, overexpression of active Sgg^S9A and Par-1 also synergistically increased dendrite pruning defects (Fig. 6K,L). Sgg^S9A and Par-1 co-expression caused severing defects in 86% of ddaC neurons (Fig. 6H,K), compared with either Sgg^S9A (18%) or Par-1 (22%) overexpression alone (Fig. 6F,G,K). We then overexpressed Par-1^KD, a kinase-dead form of Par-1 (Nishimura et al., 2004), together with Sgg^S9A. Co-overexpression of Par-1^KD did not exacerbate the dendrite pruning defects in Sgg^S9A-overexpressing ddaC neurons (Fig. 6J-L). As a control, overexpression of Par-1^KD alone exhibited no defects (Fig. 6I,K,L). These data suggest a requirement for Par-1 kinase activity during dendrite pruning. Surprisingly, neither overexpression nor knockdown of fly Tau significantly affected pruning defects in sgg RNAi neurons at 16 h APF (Fig. S10A,B), in contrast to a previously reported genetic interaction between Tau and Par-1 (Herzmann et al., 2017). Likewise, overexpression or knockdown of Tau did not modify the Sgg^S9A phenotypes (Fig. S10C,D). Sgg was also reported to phosphorylate other MAPs, including Futsch (Gögel et al., 2006). Like Tau, neither knockdown nor overexpression of Futsch modified the dendrite pruning defects in sgg RNAi or Sgg^S9A-overexpressing neurons (Fig. S11A-D). We show here that Sgg enhances microtubule turnover rate and thereby downregulates Tau and Futsch levels in ddaC neurons (Figs 3 and 4). Taken together, our data suggest that Sgg acts synergistically with Par-1 to promote dendrite pruning, likely by downregulating multiple MAPs.

We next attempted to examine whether Sgg and Par-1 act synergistically to modulate microtubule levels in the dendrites of ddaC neurons. Double knockdown of sgg and par-1 led to a further increase in dendritic Tau levels (Fig. 7C,I), compared with sgg RNAi or par-1 RNAi knockdown alone (Fig. 7A,B,I). Conversely, Tau levels were further reduced when Sgg^S9A was co-expressed with Par-1^WT (Fig. 7F,J), compared with Tau levels in Sgg^S9A or Par-1^WT (Fig. 7D,E,J). It has been recently reported that Par-1^WT overexpression downregulates Tau levels in the dendrites of ddaC neurons (Bu et al., 2022). Overexpression of the kinase-dead form Par-1^KD failed to further downregulate Tau levels in Sgg^S9A-overexpressing neurons (Fig. 7H,J; Par-1^KD control, Fig. 7G,J), suggesting that the kinase activity of Par-1 is required for the downregulation of microtubules in ddaC neurons. Thus, Sgg and Par-1 act synergistically to destabilize microtubules during dendrite pruning. Given that overexpression of either Sgg^S9A or Par-1 impairs dendritic microtubule orientation (Fig. 2F,I) (Bu et al., 2022), we also assessed whether a concurrent overexpression of Par-1 could exacerbate such Sgg^S9A-associated defects. Interestingly, overexpression of Par-1 further reduced Nod-lacZ levels in the dendrites of Sgg^S9A-overexpressing neurons (Fig. S12A,B). We then tracked EB1-GFP comets in the case of Sgg^S9A and Par-1 co-overexpression (Fig. S12C,D). Likewise, the percentage of anterograde-moving comets was significantly increased in the ddaC dendrites of such neurons, in comparison with that in the Sgg^S9A control (Fig. S12C,D). This is in stark contrast to the lack of changes in both Nod-lacZ level and EB1-GFP directionality, where Sgg^S9A was instead co-expressed with Par-1^KD (Fig. S12A-D). Thus, these suggest that active Sgg, together with Par-1, can interfere microtubule orientation in the dendrites. Taken together, these data suggest a synergistical role of Sgg and Par-1 in regulating microtubule mass and dendrite pruning in ddaC neurons.

GSK3β/Sgg is a key kinase involved in developmental pathways such as the Hh and Wg signaling pathways (Jia et al., 2002; Price and Kalderon, 2002). Given its abundant expression in the nervous systems, Sgg is also crucial for neuronal development, particularly in the formation of synapses and neural connections (Franco et al., 2004; Kim et al., 2011). Here, we report a novel role for Sgg in regulating dendrite pruning of sensory ddaC neurons during neuronal maturation. We show that loss or suppression of sgg function, via two RNAi lines, a mutant line or three dominant negative lines, resulted in prominent pruning defects. Importantly, sgg depletion conferred an overall stabilizing effect on microtubules. Temporal gene-switch assays show that sgg function is required at the larva-to-pupa transition stage. Although the overexpression of wild-type Sgg had no notable effects, overexpression of the constitutively active Sgg^S9A caused defects in dendrite pruning as well as the dendritic microtubule orientation. Moreover, Sgg acts synergistically with Par-1 to regulate microtubule mass or orientation during dendrite pruning. Our results support the vital role of Sgg, in concert with Par-1, in regulating microtubules and therefore dendrite pruning in ddaC neurons.

In its function as a primarily inhibitory kinase, GSK3β/Sgg is involved in a multitude of processes, ranging from glycogen/energy metabolism to cellular growth/repair (Hur and Zhou, 2010). Importantly, GSK3β/Sgg profoundly modulates microtubule stability via its ability to phosphorylate many MAPs (Gögel et al., 2006; Hur and Zhou, 2010; Kumar et al., 2009; Wagner et al., 1996), thereby downregulating the binding capabilities of such MAPs on microtubules. Through its microtubule-destabilizing actions, GSK3β/Sgg helps to promote an intrinsic dynamism to the cytoskeletal network, permitting the reconfiguration of synaptic ends (Franco et al., 2004; Gögel et al., 2006). Here, we duly show with real-time EB1-GFP movies that the loss of sgg enhanced microtubule polymerization by increasing the length and speed of EB1-GFP comets in ddaC dendrites. Conversely, the overexpression of both wild-type and constitutively active Sgg inhibited microtubule polymerization as it decreased average comet lengths (Fig. 2). Moreover, photoconversion assays, which depict in vivo microtubule turnover and therefore disassembly rate, show that, whereas microtubule turnover was indeed decreased in the dendrites upon Sgg knockdown, it was increased upon excessive Sgg (Fig. 3 and Fig. S5). Consistently, Sgg knockdown and overexpression increased or reduced the mass of microtubules in the dendrites, respectively (Fig. 4). Thus, these results demonstrate a negative effect of Sgg on microtubule polymerization and stability. Furthermore, our pharmacological and genetic data strongly suggest that Sgg promotes dendrite pruning partly via microtubule disassembly (Fig. 5). First, treatment with the microtubule-depolymerizing drug colchicine almost fully rescued the dendrite pruning defects in sgg RNAi neurons. Second, although the overexpression of the microtubule-destabilizing factor Efa6 largely rescued the sgg RNAi defects, knockdown of Efa6 strongly exacerbated the sgg RNAi phenotype. Finally, we show that the removal of one copy of γTub23C also significantly suppressed the dendrite pruning defects of sgg RNAi neurons.

How does Sgg inhibit microtubule stability and promote microtubule disassembly during dendrite pruning? Sgg might promote microtubule disassembly by phosphorylating its downstream MAPs. GSK3β/Sgg has been documented to reduce microtubule stability by phosphorylating MAP2/Tau in both fly and mammals (Lovestone et al., 1996; Nishimura et al., 2004). Furthermore, the recruitment of autoactivated Sgg by a hyperactive form of LRRK2 leads to the hyperphosphorylation of axonal Tau and thereby microtubule fragmentation in ddaC neurons (Lin et al., 2010). In flies, Sgg is also known to limit the number of synaptic boutons or nerve terminals formed at neuromuscular junctions (NMJs) (Franco et al., 2004). More microtubule loops and boutons were observed in the NMJs of sgg mutants, which requires the MAP Futsch (Franco et al., 2004). Given its ability to phosphorylate Futsch (Gögel et al., 2006), Sgg might modulate the microtubule association of Futsch and thereby microtubule stability at the terminals of growing axons. Likewise, the mammalian Sgg counterpart GSK3β can phosphorylate MAP1B and enhance the dynamic instability of local microtubules, thus promoting axon growth of differentiating neurons (Goold et al., 1999; Trivedi et al., 2005). Additionally, GSK3β was found to negatively regulate another MAP, namely CRMP-2 (Fukata et al., 2002). Inactivation of GSK3β increases non-phosphorylated active CRMP-2, which in turn enhances microtubule assembly and thereby promotes axonal outgrowth during axon specification (Yoshimura et al., 2005). Given the existence of multiple substrates of GSK3β/Sgg, it is conceivable that Sgg might promote dendrite pruning by phosphorylating multiple MAPs. Consistent with this possibility, we found that knockdown and overexpression of either Tau or Futsch did not modify the dendrite pruning defects in either sgg RNAi or Sgg^S9A-overexpressing neurons (Figs S10, S11).

Under normal physiological conditions, GSK3β/Sgg exists as an active kinase (Cohen and Goedert, 2004). The phosphorylation of specific Sgg residues either increases (at Y214) or suppresses (at S9) its activity (Beurel et al., 2015). In accordance with sgg knockdown, attenuating Sgg activity by disabling the activation of phosphorylation at residue Y214 led to pruning defects (Fig. S1A). Interestingly, although excess wild-type Sgg bore no effects, the mutant form Sgg^S9A, which could not be phosphorylated at its S9 site and is therefore constitutively active, caused dendrite pruning defects (Fig. S1B). The Sgg^S9A defects are likely due to a disturbance in dendritic microtubule orientation (Fig. 2, Fig. S4), as ddaC dendrite pruning is highly correlated with proper maintenance of dendritic microtubule orientation (Herzmann et al., 2018; Rui et al., 2020; Tang et al., 2020; Wang et al., 2019). A preceding study has also shown that LRRK2-recuited autoactivated Sgg induces the mislocalization of the axonal MAP Tau into the dendrites of ddaC neurons (Lin et al., 2010), further supporting the notion that excess Sgg activity impairs microtubule polarity in neurons.

The orientation of a microtubule lattice is determined by the protruding subunit on either of its ends (Akhmanova and Steinmetz, 2015). β-Tubulin spearheads the plus ends, which are regulated by plus-end-tracking proteins (+TIPs), such as EB1, Adenomatous Polyposis Coli (APC) and CLASP2 (Akhmanova and Steinmetz, 2008; Etienne-Manneville, 2013). The connections between GSK3β/Sgg and each of these +TIPs have been well documented. ROS-mediated activation of GSK3β in cancer cells phosphorylated EB1 and caused its dissociation from microtubule plus-ends (Le Grand et al., 2014). The ability of the MAP APC to bind to and stabilize microtubules was decreased with increased GSK3β activity (Zumbrunn et al., 2001). Excessive GSK3β kinase activity induced CLASP2 dissociation from microtubules and prevented their enrichment at the microtubule plus-ends, thereby impairing axon outgrowth (Hur et al., 2011). Thus, it is tempting to assume that hyperactive Sgg causes the microtubule orientation defects in ddaC dendrites by impairing distribution or functions of +TIPs. Further studies are required to determine this possibility.

Many GSK3β/Sgg substrates are first phosphorylated by a priming kinase (Beurel et al., 2015). This priming phosphorylation event is a prerequisite for further recognition and effective phosphorylation by GSK3β/Sgg (ter Haar et al., 2001). The priming step often confers substrate specificity to GSK3β/Sgg, as distinct priming kinases dictate GSK3β/Sgg activity on distinct proteins (Sutherland, 2011). For example, the priming action of Cdk5 or DYRK2 on different CRMP members determines the spatial and temporal specificities of GSK3β phosphorylation (Cole et al., 2006). However, Par-1, but not Cdk5, was demonstrated to be the priming kinase that mediates the initial phosphorylation preceding the phosphorylation of Tau by Sgg (Nishimura et al., 2004). In this study, our genetic interaction data suggest a synergy between Sgg and Par-1 during dendrite pruning. We observed a cumulative effect on pruning defects with the simultaneous knockdown of sgg and par-1 (Fig. 6A-E). This additive effect correlated with higher Tau levels along the dendrites of such ddaC neurons (Fig. 7A-C,I), indicative of increased microtubule mass. However, the dendrite pruning defects were not significantly exacerbated when Tau was overexpressed in sgg RNAi (Fig. S10A). This is in stark contrast to a previous study reporting an enhancement in pruning defects when Tau was co-expressed with par-1 RNAi (Herzmann et al., 2017). Thus, our genetic interaction experiments suggest that Tau is more inherently tied to Par-1 than it is to Sgg, and that Par-1 might be sufficient to regulate Tau independently of Sgg. Accordingly, Drosophila tauopathy models have emphasized that the phosphorylation of Tau by Par-1, rather than by Sgg, is more pivotal for Tau toxicity (Chatterjee et al., 2009; Nishimura et al., 2004). Importantly, our observations also support the notion that Sgg promotes dendrite pruning via its effects on multiple MAPs, as we found that, apart from Tau, sgg RNAi also increased Futsch and acetylated α-tubulin levels (Fig. 4).

On the other hand, simultaneous overexpression of active Sgg^S9A and wild-type Par-1 produced an even more robust disruption on dendrite pruning, compared with the overexpression of either protein alone (Fig. 6F-H,K,L). Interestingly, Tau levels were also further reduced upon Sgg^S9A and Par-1 co-overexpression (Fig. 7D-F,J). More intriguingly, functional Par-1 was required for these additive effects, as Sgg^S9A co-expression with the kinase dead Par-1^KD phenocopied Sgg^S9A alone (Figs 6I-L; 7E,G,H,J). Nod-lacZ displacement and impaired EB1 directionality duly attributed the combined effects of Sgg^S9A and Par-1^WT to an exaggerated perturbation of microtubule orientation (Fig. S12). It is tempting to speculate that the exacerbated pruning defects observed in Sgg^S9A/Par-1-co-overexpressing neurons stem from overtly active Sgg and Par-1 converging on common MAPs, thereby increasing microtubule destabilization and incidences of mixed orientations in dendrites. Alternatively, it is equally feasible that a more fundamental basis underlies this relationship between Sgg and Par-1, as GSK3β could both activate and inactivate MARK/Par-1 via selective phosphorylation (Kosuga et al., 2005; Timm et al., 2008). It is therefore possible that in ddaC neurons, a concurrent expression of wild-type Par-1^WT and constitutively active Sgg^S9A enable the coupled activation of both kinases, subsequently magnifying their individual effects.

Fly stocks and crosses were kept in standard cornmeal media at 25°C. The third instar larvae or pupae at 0, 6/7 or 16 h APF (both male and female) were used in this study. The following fly stocks were used: ppk-Gal4 on Chromosome II and III, expressed strongly in C4da neurons from the late embryonic stages (Grueber et al., 2003), SOP-flp (#42) (Matsubara et al., 2011), UAS-Mical^Nterm (Terman et al., 2002), UAS-EB1-GFP (Stone et al., 2008), UAS-Kin-β-Gal (Clark et al., 1997), UAS-Par-1^WT, UAS-Par-1^KD (Wang et al., 2007), UAS-dTau-HA (Herzmann et al., 2017) and UAS-Efa6^FL (Bu et al., 2021).

The stocks obtained from Bloomington Drosophila Stock Center (BDSC) were: UAS-mCD8::GFP, UAS-Dicer2, FRT19A, tubP-Gal80, Gal4^4-77 (expressed weakly in C4da neurons from the late embryonic stages, BDSC 8737), UAS-Nod-β-Gal (BDSC 9912), γ-tub23C^A15-2 (a null allele, BDSC 7042), ppk-CD4-tdGFP (BDSC 35843), GSG2295-Gal4 (expressed in da neurons only in the presence of mifepristone/RU486; BDSC 40266), UASp-αTub84B-tdEOS (BDSC 51313 and 51314), sgg^M1-1 (a hypomorphic allele, BDSC 68162), Sgg::GFP (BDSC 50887), sgg RNAi 1 (BDSC 35364), sgg RNAi 2 (BDSC 38293), UAS-sgg (BDSC 5435), UAS-sgg^S9A (BDSC 5255), UAS-sgg^A81T (1, BDSC 5360; 2, BDSC 5359), UAS-sgg^Y214F (BDSC 6817), Ctrl RNAi (mCherry, BDSC 35785), par-1 RNAi (BDSC 32410), stai RNAi (BDSC 36902) and futsch^EP1419 (BDSC 10751).

The stocks obtained from Vienna Drosophila Resource Center (VDRC) were: ctrl RNAi (γ-tub37C, v25271), efa6 RNAi (v330083), tau RNAi (v25023), fustch RNAi (v6972) and mical RNAi (v46097). Genotypes of the fly strains shown in each figure are listed in the supplementary Materials and Methods.

MARCM or RNAi analysis of ddaC neurons was conducted as previously described (Kirilly et al., 2009). Crosses were maintained and grown at 25°C. For visualization of dendrite morphology, pre-pupae (0 h or 7 h APF) were cleaned and mounted onto a thin layer of Corning grease before imaging. In screening for pruning defects, white pre-pupae were gently collected and placed onto filter paper. After aging for 16 h, pupal cases were removed. The exposed pupae were mounted and viewed immediately. All mounting was carried out in 90% glycerol. GFP-positive neurons were visualized live, dorsal-side up, on the Leica SPE-II upright confocal microscope platform with a 40× oil lens. The ImageJ plugin Simple Neurite Tracer was used for dendrite length quantification. A severing defect is defined by the presence of proximal dendrites that remain attached to the soma at 16 h APF.

The C4da driver Gal4^4-77 or ppk-Gal4 was used to drive the expression of EB1-GFP to visualize its comets in ddaC neurons. For live imaging of growing microtubule plus-ends via EB1-GFP comet tracking, 72-96 h AEL larvae were raised on low-yeast paste food and quickly cleaned in distilled water and placed onto slides before the application of a thin coating of halocarbon oil. Larvae were pushed down with a cover slip, as flatly as possible but alive. Time-lapse movies were acquired on an Olympus FV3000 inverted confocal microscope platform with the 60× oil lens set to 3× zoom. Parameters were standardized to 125 frames of six z-step images, obtained at 1.45 s intervals for a total of 3 min. To avoid prolonged compression stress on the larvae, two or three neurons from A3-A4 segments were imaged per animal. Kymographs were generated by the ImageJ plug-in KymographBuilder. The average of EB1-GFP comet track length, comet speed, percentage of anterograde moving comets and number of comets from 30 µm proximal dendrites within 3 min were calculated in each neuron for quantification.

Wandering 3rd instar larvae or newly turned WP stage pupae expressing UASp-αTub84B-EOS were briefly cleaned in water and mounted with halocarbon oil. All images were acquired on the Olympus FV3000 inverted confocal microscope with a 60× oil lens at 3× zoom. A square of ∼7 µm across was set as the standard ROI. A segment of a dorsal proximal dendrite was photoconverted from green to red using a 405 nm laser. The GFP and RFP channels were imaged immediately after bleaching, and once again after a 30 min period of recovery. Fluorescence intensity (FI) of the converted region was obtained as the mean gray value of the 7 µm stretch of dendrite within the specified ROI, as quantified by the ImageJ software. FI was also taken from a 7 µm length of dendrite immediately neighboring the ROI. Turnover rates were determined from calculating the remaining FI according to the formula: (FI[converted]−FI[neighbouring])_30min/(FI[converted]−FI[neighbouring])_0min.

Parental flies crossed 24 h earlier were flipped into a fresh vial containing control food to begin embryo collection. Flipping was repeated at 12 h intervals. Late 2nd or early 3rd instar larvae were collected and equally separated onto either control food or food prepared with 5 µg/ml colchicine (Sigma-Aldrich, C9754) or 240 µg/ml RU486 (Sigma-Aldrich, C84371). WP stage pupae were collected 2 days after treatment for dendrite imaging. wL3 larvae were collected 1 day after treatment for EB1-GFP time-lapse imaging.

The following primary antibodies were used in this study: mouse anti-Futsch (1:50, DSHB, 22C10), mouse anti-Galactosidase (1:1000, Promega, Z3781), mouse anti-EcR-B1 (1:50; DSHB, AD4.4), guinea pig anti-Mical (1:500; Yu lab), mouse anti-Acetylated Tubulin (1:750, Sigma-Aldrich, T6793) and rabbit anti-Tau (1:1000, a gift from Daniel St Johnston, Gurdon Institute, University of Cambridge, UK). Secondary antibodies were applied in 1:500 dilutions and included Cy3-, 488- and 649-conjugated goat secondary antibodies (Jackson ImmunoResearch, 115-165-003, 106-165-003 and 123-605-021).

To stain for LacZ, GFP, EcR-B1 and Mical, wL3 or WP animals from control and experimental groups were dissected simultaneously in ice-cold PBS and fixed with 4% formaldehyde for 20 min. For the staining of microtubule-associated proteins, animals were dissected in ice-cold Ca²⁺-free HL3.1 saline and fixed with a freshly prepared solution of PHEM buffer, 4% formaldehyde, 0.25% glutaraldehyde and 0.1% Triton X-100 for 20 min, then quenched with 50 mM NH₄Cl for 5 min (Witte et al., 2008). All samples were subsequently washed with 0.5% Triton-X in PBS (PBST) for 45 min, then blocked with 5% BSA in 0.5% PBST for another 45 min. Fillets were mounted epidermal layer-side down directly onto a cover slip in VectaShield AntiFade Mounting Medium (Vector Laboratories, H-1000). Imaging on the inverted Olympus FV3000 confocal microscope (40× oil lens, 2.3× zoom) was conducted in such a way that the entire ddaC/ddaE projections of a single neuron should be visible.

Assessment of immunofluorescence was tailored to the imaged entity: for quantification of Sgg::GFP intensity, the entire ddaC cell body was manually isolated and measured after background subtraction (Rolling Ball Radius=30) by ImageJ. The intensity levels of Nod-lacZ/Tau were determined from mean grey values along two dorsally extending ddaC dendrite branches, measured against an internal control, i.e. HRP. The defective Nod-lacZ distribution is defined as reduced LacZ levels in the dendrites with concurrent increases in the soma of ddaC neurons. For Futsch and acetylated α-Tubulin, three regions were chosen: two from the dorsal dendrites of ddaC and one from ddaE, all of which were also compared against individual HRP. The 20 µm dendritic fragments (either a 30 µm or 100 µm radius away from the center of the soma) were chosen for the intensity measurement. Fluorescence intensity levels presented as results have been further normalized against the average of the control group.

Two-tailed Student's t-test was used in pairwise analyses. One-way ANOVA with a Bonferroni test to compare all possible pairs were applied in assessing two or more groups against a single control. The statistical significance was defined as ***P<0.001, **P<0.01, *P<0.05 (ns, not significant). Error bars in all graphs show standard error of the mean (s.e.m.). Numbers on the base of some graphs represent the number of neurons (n) assessed. Final data are compiled from two or three independent experiments. All quantitative data are included in Table S1.

We thank Y. Jan, D. Johnston, B. Lu, M. M. Rolls, T. Uemura, the Bloomington Stock Center (BSC), DSHB (University of Iowa), the Kyoto Stock Center (Japan) and the VDRC (Austria) for generously providing antibodies and fly stocks. We are grateful to M. M. Rolls for providing the technical advice on EB1-GFP imaging. We thank Si Yun Ng for some genetic analysis and other Yu lab members for helpful discussion.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.200844.reviewer-comments.pdf