Astrocyte-dependent local neurite pruning in Beat-Va neurons

During development, the nervous system is initially overpopulated by neurons that form an excessive number of synaptic connections. This neural circuitry is subsequently refined through the elimination of exuberant synapses, neurites, or entire neurons often in response to activity-dependent signaling mechanisms (Luo and O’Leary, 2005). While neuronal remodeling occurs in all complex metazoans and provides a mechanism for the optimization of neuronal numbers and circuit connectivity, aberrant neuronal remodeling is associated with neurological conditions like autism spectrum disorders, schizophrenia, and epilepsy (Feinberg, 1982; Selemon et al., 1995; Atz et al. 2007; Winchester et al., 2012; Sekar et al., 2016; Ishizuka et al., 2017; Neniskyte and Gross, 2017).

How much neuronal remodeling occurs across the mammalian nervous system remains unclear, but it is thought to be extensive and have profound effects on the final wiring diagram (Luo and O’Leary, 2005, Neukomm and Freeman, 2014; Riccomagno and Kolodkin, 2015). Axonal and synaptic connectivity can be radically altered through local pruning of axons or individual synaptic connections. Retinal ganglion cells (RGCs) initially project axons to the dorsal lateral geniculate nucleus (dLGN) of the thalamus and form an excessive number of synapses on target cells, but later these are refined and exuberant RGC axons and synapses are eliminated (Stevens et al., 2007; Schafer et al., 2012). Large-scale changes in circuit wiring can also be driven by selective cell death (Oppenheim, 1985; Hutchins and Barger, 1998). For instance, 40% of mouse GABAergic inhibitory neurons in the developing postnatal cortex are culled by a wave of caspase-driven cell death in the first 20 days of life (Southwell et al., 2012). Many of these changes occur long after the neurons have integrated into neural circuits and fired for weeks. Precisely how specific neurons are selected for elimination remains unclear.

Glial cells help sculpt developing neural circuits by participating in the selective elimination of neurites, synapses, or entire neurons. In the dLGN, C1q is believed to opsonize synapses that are destined for removal, and C1q-bound synapses are then recognized and phagocytosed by complement receptor 3–bearing microglia (Schafer et al., 2012). Astrocytes can also engulf pruned synapses via the MERTK and MEGF10 receptors (Chung et al., 2013), likely by also binding C1q (Iram et al., 2016; Lin et al., 2017). Disruption of C1q signaling leads to less robust RGC terminal segregation and the retention of excess functional synapses (Stevens et al., 2007), which argues that glia somehow play a role in promoting the final execution of synapse/axon elimination rather than simply passively clearing pruned neuronal debris. However, C1q is not required in all brain areas that undergo developmental neural refinement, arguing that a diversity of pruning mechanisms exist across the brain. In the mouse barrel cortex, innervating axons are extensively refined into segregated somatosensory maps (Petersen, 2019). In this context, however, refinement relies on fractalkine signaling between neurons and microglia, and C1q signaling is dispensable (Gunner et al., 2019). Finally, it has been clear for some time that phagocytic cells, including glia, can help drive cells toward the final execution of cell death. In developing C. elegans, cell death requires both cell-intrinsic initiation of apoptotic pathways and signals from the engulfing hypodermal cells, with engulfing hypodermal cells signaling to partially dead cells to fully execute apoptosis (Hoeppner et al., 2001; Reddien et al., 2001). While not widely explored in the mammalian brain, similar roles for microglia have been identified in promoting the final execution of cell death in populations of developing Purkinje neurons (Marín-Teva et al., 2004). In this way, neuronal remodeling appears to entail a complex conversation between neurons and glia.

Drosophila has served as an excellent system in which to explore the molecular basis of neuronal remodeling events in vivo (Lee et al., 1999, 2000; Watts et al., 2003; Williams and Truman, 2005; Choi, 2006, Kirilly et al., 2009; Zhang et al., 2014). At pupariation, a burst of the steroid hormone 20-hydroxyecdysone (ecdysone) drives the activation of neuronal remodeling programs across the nervous system, most of which are executed during the first 12–18 h of metamorphosis (Truman et al., 1994; Schubiger et al., 1998, 2003). Ecdysone binds to the Ecdysone Receptor (EcR) nuclear hormone receptor, with EcR signaling acting as a timing mechanism to coordinate the initiation of animal-wide metamorphic changes (Pinto-Teixeira et al., 2016; Koelle et al., 1991; Riddiford and Truman, 1993; Talbot et al., 1993; Thummel, 1996). Cell-autonomous blockade of EcR signaling appears to inhibit known neuronal cell death and local pruning events in the Drosophila nervous system (Schubiger et al., 1998, 2003; Kuo et al., 2005; Choi, 2006; Winbush and Weeks; 2011, Hara et al., 2013; Yamaguchi and Miura, 2015; Marchetti and Tavosanis, 2017). For instance, in the first 6–8 h of metamorphosis in the central nervous system (CNS), blocking EcR signaling in ventral nerve cord (VNC) Corazonin neurons suppressed apoptotic death (Choi, 2006; Lee et al., 2011, 2013, 2019; Wang et al., 2019). Likewise, the mushroom body (MB) γ neurons in the brain, which are involved in olfactory learning and memory, undergo stereotyped axonal and dendritic pruning (Heisenberg, 1998) that can be blocked by inhibiting EcR (Lee et al., 2000; Watts et al., 2003). To our knowledge, neuronal cell types that undergo developmental remodeling independently of EcR have not been identified.

Like their mammalian counterparts, Drosophila glial cells play a crucial role in neuronal remodeling. Neurons become competent for remodeling when the transforming growth factor-β (TGFβ) family member Myoglianin is released from glia, which activates transcription of EcR via TGFβ receptor signaling on target neurons (Awasaki et al., 2011; Yu et al., 2013; Hakim et al., 2014; Wang et al., 2019). After cell death or neurite/synapse degeneration has occurred, glial cells act as phagocytes to recognize and phagocytose neuronal debris through conserved signaling pathways like Draper/MEGF10 (MacDonald et al., 2006; Doherty et al., 2009; Hakim et al., 2014) or Fractalkine/Orion (Boulanger et al., 2021; Ji et al., 2023; Perron et al., 2023). Whether there are other, novel mechanisms by which glial cells coordinate neuronal remodeling during metamorphosis is an open question.

A growing body of evidence implies that the signaling pathways engaged to drive neuronal remodeling are diverse and context-dependent across the brain, and we lack a deep understanding of the molecular basis of neuronal remodeling in any context (Schafer et al., 2012; Yaniv and Schuldiner, 2016; Boulanger and Dura, 2022). In this study, we sought to identify new neuronal subtypes that remodel in the Drosophila pupal nervous system, especially those that use novel mechanisms to regulate remodeling. We characterized two populations of neurons labeled by the BeatVa-Gal4 driver—medial (Beat-Va_M) and lateral (Beat-Va_L) neurons—that undergo complete local neurite pruning or segment-specific apoptotic cell death, respectively. We show that local pruning in Beat-Va_M neurons can happen largely independently of EcR, but requires signaling from phagocytic astrocytes, while segment-specific cell death of Beat-Va_L neurons is downstream of EcR and governed by spatially restricted expression of the Hox gene Abd-B. Our work establishes Beat-Va neurons as a new model to explore neuronal remodeling in vivo and identifies new mechanisms for the regulation of developmental pruning and cell death in Drosophila.

We wished to identify neuronal cell types that undergo developmental remodeling but use genetic programs that differ from previously characterized neurons in Drosophila. We, therefore, screened ∼5,500 Gal4 lines on the open source Janelia FlyLight website (Pfeiffer et al., 2008) to identify Gal4 driver lines that were active in sparse populations of ventral nerve cord (VNC) neurons during the wandering third instar larval stage (wL3), the developmental stage that directly precedes metamorphosis (Fig. 1 A). From these, we selected 87 lines and validated their wL3 expression patterns by crossing each line to membrane-tethered GFP (UAS-mCD8::GFP) (Table S1). We chose 28 lines to study further and visualized their morphologies during metamorphosis. We examined them at wL3, 6 h after puparium formation (APF) and head eversion (HE, ∼12 h APF), the time point by which most known cell types have completed remodeling, and 18 h APF (Fig. 1 A and Table S2). Among all of these lines, we chose to focus on GMR92H04-Gal4, which appeared to label a combination of cells undergoing cell death or local pruning.

GMR92H04-GAL4 was constructed by fusing an enhancer element for the Beat-Va gene to the DSCP and Gal4 promoters (Jenett et al., 2012). We refer to this line as BeatVa-Gal4 and the neurons labeled as Beat-Va neurons. To define the segmental patterns of neurons labeled by the BeatVa-Gal4 driver, we used antibody staining for the transcription factors Even-skipped (Eve) and Engrailed (En), which act as convenient landmarks for the identification of individual cells in the VNC (Manoukian and Krause, 1992). We found that the BeatVa-Gal4 driver labeled one lateral neuron (referred to as Beat-Va_L) and one medial neuron (referred to as Beat-Va_M) per hemisegment in abdominal segments A3–7 (Fig. S1, A and B). This driver also weakly labeled neurons in A8 and A1–2 segments, as well as a handful of cells in the brain lobes (data not shown). The total number of cells labeled by this BeatVa-Gal4 driver decreased prior to HE in the VNC, suggesting that a subset of neurons underwent cell death during metamorphosis. In addition, the complexity of neurite projections in the synaptic neuropil decreased significantly, indicating that some of these neurons underwent neurite pruning (Fig. 1, B and B″).

To examine the morphology of these cells in segments A3–7, we used the MultiColorFlpOut (MCFO) approach. MCFO is based on the use of UAS-regulated expression of spaghetti monster GFP transgenes (smGFP) (Nern et al., 2015). Stochastic expression of one of four UAS-regulated versions of fluorescently dead smGFP, each with a unique epitope tag (OLLAS, V5, HA, or Flag), is activated within the population of Gal4 expressing neurons (Nern et al., 2015; Viswanathan et al., 2015). Individual neurons are then visualized by staining for each epitope (Fig. S1 C). By examining single cells, we found that at the third instar larval stages, Beat-Va_L cells crossed the midline, projected anteriorly, and terminated within the VNC. Anterior Beat-Va_L neurons in segments A3–4 persisted through HE (Fig. 1 C) while posterior lateral Beat-Va_L neurons in segments A5–7 underwent cell death by 6 h APF (Fig. 1 D and Fig. S1 D). In contrast, we found that Beat-Va_M neurons extended a dense network of fine projections through multiple segments in the synaptic neuropil, totaling >1,000 µm in length. These fine processes began fragmenting by 6 h APF and were cleared from the CNS by HE (Fig. 1, E–G). These observations revealed that BeatVa-Gal4 labels at least two cell types, one that undergoes apoptotic death and the other local neurite pruning (summarized Fig. 1 H).

In all other neuronal subtypes studied to date, both neuronal cell death and local neurite pruning during Drosophila metamorphosis depend on neuronal ecdysone signaling (Lee et al., 2000; Kuo et al., 2005; Choi, 2006; Winbush and Weeks, 2011, Yu et al., 2013; Wang et al., 2019; Zhu et al., 2019). To determine whether ecdysone signaling also drove Beat-Va_L cell death or Beat-Va_M remodeling through EcR, we used a UAS-EcR^DN construct to cell-autonomously block EcR signaling in all Beat-Va neurons (Fig. 2, A–C). While neurite remodeling in the Beat-Va_M neurons was not disrupted by cell-intrinsic blockade of EcR signaling, expression of EcR^DN suppressed Beat-Va_L apoptosis (Fig. 2, A–C and E), similar to other neuronal cell death events in Drosophila metamorphosis (Winbush and Weeks, 2011, Denton et al., 2010; Lee et al., 2011; Zirin et al., 2013). We next stained for activated caspases using an antibody that recognized a cleaved version of Dcp-1. We found that posterior Beat-Va_L cells (A5–7) became positive for activated caspases during early metamorphosis while Beat-Va_M neurons and cells in anterior segments that survived (A3–4) were not caspase positive (Fig. S2, A–D). We then expressed P35, a baculovirus caspase inhibitor that blocks many caspase-dependent cell death events (Clem et al., 1991), using BeatVa-Gal4. We found that P35 had no obvious effect on Beat-Va_M remodeling, but strongly blocked Beat-Va_L cell death through HE in segments A5–6 and more weakly blocked death in A7 (Fig. 2, D and E) (perhaps due to driver strength variability), arguing posterior Beat-Va_L apoptosis is driven through caspase activation (Fig. 2, D and E; and Fig. S2).

We speculated that positional identity and segment-specific differences in survival could be regulated by differential expression of homeobox (Hox) genes. The boundary between the anterior and posterior lateral cells (A4/5) is defined by the Abdominal-B (Abd-B) Hox gene in developing embryos (Delorenzi and Bienz, 1990) and during embryonic development, Abd-B can either promote or inhibit apoptotic cell death of neuronal progenitors depending on the developmental context (Monedero Cobeta et al., 2017; Bakshi et al., 2020; Clarembaux-Badell et al., 2022). We used antibodies to determine the larval expression pattern of Abd-B and found that surviving A3–4 Beat-Va_L cells did not express Abd-B while Beat-Va_L cells in segments A5–7 strongly expressed Abd-B at wL3 and for as long as Beat-Va_L cells survived during metamorphosis (Fig. 2, F and G). We knocked down Abd-B expression specifically in the Beat-Va neurons by driving a UAS-Abd-B^RNAi construct with BeatVa-Gal4. This led to a strong suppression of caspase activation in posterior Beat-Va_L neurons (detected by cleaved Dcp-1 staining) through 6 h APF (Fig. S3, A–C) and partially suppressed lateral cell death compared with controls at HE (Fig. 2, H–J and L). This incomplete but strong blockade of cell death is likely due to a partial knockdown effect by the RNAi construct targeting Abd-B as staining for Abd-B revealed partial rather than complete knockdown of protein levels in some cells. We note that Beat-Va_L cells retained expression of EcR-B1 even when Abd-B was depleted from these cells (Fig. S3 D), arguing that the suppression of cell death could not be explained by changes in EcR expression. When we performed the reciprocal experiment and expressed Abd-B in all Beat-Va_L neurons by driving UAS-Abd-B with BeatVa-Gal4 (Fig. S3 E), we found that A3–A4 lateral cells underwent cell death by HE (Fig. 2, K and L), indicating that expression of Abd-B in Beat-Va_L neurons is sufficient to induce cell death during metamorphosis. Finally, we note that Beat-Va_M neurons also expressed Abd-B, but manipulations had no effect on cell survival or neurite pruning, indicating Abd-B promotes caspase activation in some but not all neurons during metamorphosis (Fig. 2, H–K; and Fig. S3, F and G).

It was surprising that EcR manipulations had no obvious effect on Beat-Va_M neurite pruning, as signaling through EcR-B1 drives most neuronal remodeling during Drosophila metamorphosis. One possibility was that Beat-VaM neurons do not express EcR-B1 and are therefore unaffected by EcR manipulations; however, we found EcR-B1 was indeed expressed in Beat-Va_M neurons in segments A3–7 (Fig. S3 H). To visualize individual cells more precisely in the context of EcR signaling manipulations, we generated single-cell clones in Beat-Va_M neurons and quantified neuronal complexity in wL3 stages and HE using the MCFO approach. At wL3 stages, we found no differences in Beat-Va_M neuron morphology and neurite complexity when we compared controls with EcR^DN expressing cells, regardless of the segment (A3–7), arguing that EcR^DN expression throughout larval stages does not alter Beat-Va_M neuron development (Fig. 3, A–F). We also confirmed that expression of EcR^DN did not block local pruning of Beat-Va_M neurites by HE (Fig. 3, A–F), as measured by total neurite length or total number of branch points (Fig. 3, E and F).

We sought to explore the possibility that EcR^DN could be failing to block pruning due to insufficiently high levels of expression in Beat-Va_M neurons. We first stained for the EcR^DN protein and observed robust expression of EcR^DN in all Beat-Va_M neurons at 6 h APF when endogenous EcR levels are low (Fig. S3 I). It was also possible that EcR^DN might have sufficient inherent activity to induce pruning in Beat-Va_M neurons (Cherbas et al., 2003), so we devised alternative strategies to block EcR signaling. Because EcR mutants are lethal at late embryonic or early larval stages and the EcR locus is proximal to FRT sites used for mosaic analysis, we used CRISPR/Cas9 technology to selectively eliminate genes required for EcR signaling in Beat-Va neurons (Meltzer et al., 2019). Briefly, we used lines that ubiquitously express guide RNAs (gRNAs) to EcR, its obligate heterodimer, ultraspiracle (usp) (Yao et al., 1993), or plum, an IgSF molecule that activates EcR transcription (Yu et al., 2013), and then expressed UAS-Cas9 selectively in Beat-Va neurons with Beat-Va-Gal4. Targeting EcR and plum using gRNAs/Cas9 led to a significant reduction in EcR expression as expected, while targeting usp did not change EcR expression (Fig. S3, J and K). When we examined the refinement of Beat-Va_M neurons at HE, we observed only minor preservation or regrowth of small projections in usp, EcR, and plum gRNA backgrounds (Fig. 3, G–N). We observed a similar small degree of neurite preservation when Cas9-mediated knockout of Rpn6 was induced in Beat-Va_M neurons (Fig. 3, O and P), suggesting that EcR-dependent pruning in these cells occurs through the ubiquitin-proteasome system (UPS), as shown previously in MB γ neurons (Watts et al., 2003; Hoopfer et al., 2008). Although caspase activation can drive neurite pruning in other populations of neurons, expression of P35 in Beat-Va_M neurons did not block their pruning (Fig. 2 D). Moreover, cleaved Dcp-1, an activated caspase, was not present either in Beat-Va_M cell bodies from wL3 to 6 h APF—the time at which Beat-Va_M neurons remodel (Fig. S2, A–D). We wondered if, alternatively, Beat-Va_M neurons might use axon injury signaling to prune neurites. To test this, we used a Gal4 to express Wallerian degeneration slow (WldS), which potently blocks axon degeneration after injury (Mack et al., 2001), and this had no effect on Beat-Va_M neuron pruning (Fig. 3, Q and R). Together these data suggest that Beat-Va_M neurons activate local pruning in a manner that does not depend on caspases or injury signaling and that the minor role we can detect for EcR occurs through the UPS pathway.

Finally, to examine single-cell morphology in a strong EcR depletion condition we used MCFO in combination with RNAi to concurrently knock down BaboA and plum, two transmembrane proteins that work together to induce EcR transcription, using UAS-driven RNAi constructs in BeatVa-Gal4 animals (Fig. 4, A–D) (Zheng et al., 2003; Yu et al., 2013; Wang et al., 2019). This strategy caused a strong depletion of EcR protein as determined by antibody stains (Fig. 4 E), but we found Beat-Va_M neurons still successfully pruned their neurites (Fig. 4, F–K). These data, coupled with our observations using EcR^DN and gRNAs/Cas9, support the notion that neurite pruning in Beat-Va_M neurons occurs largely independently of neuronal EcR signaling.

Recent work has shown that neuronal activity plays a role in MB γ neuron remodeling, with silencing or activating neuronal activity disrupting pruning (Mayseless et al., 2023). To determine whether neuronal activity might play a similar role in Beat-Va_M neurons, we first inhibited activity by expressing the human version of Kir2.1, an inwardly rectifying K+ channel that causes hyperpolarization, using UAS-Kir2.1 (Baines et al., 2001). We found that Beat-Va_M neuronal morphology and pruning were normal at the wL3 and HE stage (Fig. 5, A–D). To explore the possibility that increasing neuronal activity might disrupt pruning, we expressed TrpA1, a temperature-sensitive cation channel that is inactive at 25°C but active at 29°C (Hamada et al., 2008) in Beat-Va_M neurons. We shifted them to 29°C during the first 12 h of metamorphosis, similar to the manipulation that alters MB γ neuron remodeling (Mayseless et al., 2023), or more chronically by raising the animals at 29°C for 48 h prior to HE. Neither manipulation had any effect on Beat-Va_M neuron pruning (Fig. 5, E–L) Taken together, these data argue against the notion that neuronal activity plays a significant role in Beat-Va_M local pruning.

Beat-Va_M neurite fragmentation and clearance occur during the EcR-dependent transformation of larval astrocytes into phagocytic glia (Tasdemir-Yilmaz and Freeman, 2014; Yunsik et al., 2023, Preprint). To determine what role astrocytes might play in driving Beat-Va_M neuron remodeling, we generated a LexAop-EcR^DN line which we could drive with alrm-LexA to selectively block the transformation of astrocytes into phagocytes while allowing all other EcR-mediated neuronal signaling events to proceed normally. We first confirmed that our LexA/LexAop constructs efficiently blocked astrocyte transformation into phagocytes at HE. Indeed, expression of this LexAop-EcR^DN construct in astrocytes resulted in astrocytes retaining their bushy larval morphology, much like what we observe with the established, UAS-driven version of EcR^DN (Fig. S4, A–J) (Hakim et al., 2014; Tasdemir-Yilmaz and Freeman, 2014). We then used the MCFO approach to examine Beat-Va_M neuron architecture at wL3, 6 h APF, and HE after blocking EcR signaling selectively in astrocytes. Strikingly, we found that blockade of EcR signaling in astrocytes strongly suppressed local pruning of Beat-Va_M neurons to HE, where we observed significantly less neurite remodeling in Beat-Va_M neurons (Fig. 6, A–F).

Given that inhibition of EcR signaling in Beat-Va_M neurons had a small effect on local pruning, we speculated that the remodeling of Beat-Va_M neurons might be driven by a combination of EcR signaling in Beat-Va_M neurons and astrocytes synergistically. To test this directly, we blocked both astrocyte and neuronal ecdysone signaling by driving a UAS-EcR^DN simultaneously in both astrocytes and Beat-Va neurons (using alrm-Gal4 and BeatVa-Gal4). We found that the combination of blocking astrocyte transformation and blocking neuronal ecdysone signaling led to a complete inhibition of local pruning in Beat-Va_M neurons (Fig. 6, G–L; and Fig. S4, K–N). Together, these data, coupled with those above, indicate that local Beat-Va_M neurite pruning occurs in response to EcR-dependent signaling in both Beat-Va_M neurons and astrocytes, with astrocytic EcR signaling playing the primary role. Finally, we note that at wL3 stages before local pruning of Beat-Va_M neurons, neurites in animals expressing astrocytic EcR^DN were less elaborate compared with controls, with a small decrease in total neurite length and number of branch points (Fig. 6, E, F, K, and L). This observation argues that EcR-dependent signaling in astrocytes is important for a small fraction of Beat-Va_M neurite growth during embryonic or larval stages.

In previously studied Drosophila neuronal subtypes, pan-glial release of the TGFβ molecule Myo, which signals through the two TGFβ receptors BaboA and Plum, results in upregulation of EcR in neurons so they are competent to remodel in response to ecdysone (Awasaki et al., 2011; Yu et al., 2013). RNAi knockdown of Myo in all glia using repo-Gal4 suppresses the remodeling of these neurons in the Drosophila CNS (Awasaki et al., 2011). Although our data suggested that EcR signaling plays only a minor role in Beat-Va_M remodeling, we wanted to explore the possibility that astrocytic Myo might be the factor regulating Beat-Va_M neuron pruning, perhaps through an EcR-independent mechanism. We first generated a BeatVa-LexA line and drove LexAop-CD4-tdGFP, a membrane-targeted GFP (Poe et al., 2017), and co-expressed it with UAS-Myo^RNAi in glia. We also confirmed the fidelity of the BeatVa-LexA line by comparing its overlap with BeatVa-Gal4, UAS-mCD8::cherry (Fig. S5). When we drove the expression of Myo^RNAi in all glia, we observed a strong suppression of Beat-Va_M neuron local pruning compared to controls (Fig. 7, A–D). However, when we drove Myo^RNAi only in astrocytes or only in cortex glia, it did not affect Beat-Va_M neuron local pruning (Fig. 7, E and F). Thus, blockade of Myo signaling from all glia, but not astrocytes alone, can block Beat-Va_M neuron neuronal remodeling.

Given our data demonstrating the importance of astrocytic EcR signaling for Beat-Va_M pruning, we explored the possibility that pan-glial Myo knockdown could disrupt EcR activation in astrocytes. Previous work on MB pruning has shown that Myo is released in a redundant manner from all glia to drive the upregulation of neuronal EcR (Awasaki et al., 2011). We found that astrocytes failed to transform into phagocytes in a background where Myo^RNAi was expressed in all glia but did so normally if we drove Myo^RNAi only in astrocytes (Fig. 8, A–I). Furthermore, we found that EcR-B1 staining was absent from astrocytes when Myo^RNAi was driven in all glia but normal when driven in astrocytes alone (Fig. 8, J–N). We interpret these data to mean that other glia (likely cortex and ensheathing glia) redundantly provide astrocytes with Myo to activate EcR expression, which then drives astrocyte transformation into phagocytes and, in turn, Beat-Va_M neuron local pruning. Since Myo knockdown in astrocytes alone did not block Beat-Va_M neuron local pruning, we speculate that other astrocyte-derived factors regulate the activation of Beat-Va_M neuron local pruning.

Neuronal immunoglobulin superfamily (IgSF) molecules have previously been shown to regulate neurite remodeling (Awasaki et al., 2011; Yu et al., 2013; Bornstein et al., 2021). We reasoned they, or other surface molecules, could encode the astrocyte cue or cues that instruct Beat-Va_M neurons to prune. We therefore screened 104 genes encoding for IgSF and other cell surface molecules using RNAi to knock each gene down in astrocytes (using GMR25H07-Gal4) and visualized Beat-Va_M neurites (using Beat-LexA, LexAop-GFP) along with astrocyte transformation into phagocytes (Fig. 9). With respect to pruning phenotypes, we found three phenotypes: (1) those that resulted in intact Beat-Va_M neurites at HE (CG3164, Dpr2, DIP-ζ, Fipi, Sema2a, and Dpr16) (Fig. 9, B–G); (2) those where Beat-Va_M neurites fragmented but neuronal debris was not cleared by HE (CG15744, Babo, ImpL2, Kek5, Htl, Beat-IIIb, and Kek4) (Fig. 9, H–N); and (3) those that gave an intermediate phenotype (Fas3, Kek6, Vein, Wrapper, Dpr17, Dpr8, Dpr9, and Dscam2) (Fig. 9, O–V). When we assayed astrocyte morphology at wL3, 4 h APF, and HE in RNAi backgrounds for CG3164, Dpr2, DIP-ζ, and Fipi, we found that astrocytes looked relatively healthy at wL3—as evaluated by their bushy processes infiltrating the neuropil—but at 4 h APF all astrocytes had failed to transform into phagocytes (assessed by a lack of large, clustered, vesicles) (Fig. 10, A–E″). In contrast, when we assayed astrocyte morphology in hits with defects in clearance of neuronal debris (CG15744, Babo, ImpL2, Kek5, Htl, Beat-IIIb, and Kek4), we found that astrocytes were highly vesiculated by 4 h APF, a hallmark of successful transformation into phagocytes (Fig. 10, F–H″). Collectively, these findings highlight a striking correlation between astrocyte phagocytic transformation and neurite fragmentation: if astrocytes did not transform into phagocytes, Beat-Va_M neurons remained largely intact. These data reveal a tight correlation between astrocyte transformation into phagocytes and fragmentation of Beat-Va_M neurons during local pruning and imply that astrocytes have a more active role in shaping this neural circuit than simply clearing pruned neuronal debris.

A growing collection of mechanistic studies on neuronal remodeling suggests that a diversity of molecular pathways are deployed across the nervous system to accomplish remodeling in different contexts (Schafer et al., 2012; Neukomm and Freeman, 2014; Yaniv and Schuldiner, 2016; Boulanger and Dura, 2022). The relatively small number of neuronal populations in which remodeling has been studied, compared with the total number that likely exhibit remodeling across the CNS of complex metazoans, may have limited the depth of our knowledge about the diversity of signaling pathways that can drive cell death, and neurite or synaptic pruning. In this study, we identified the BeatVa-Gal4–expressing neurons and show they remodel using mechanisms not previously reported in other heavily studied CNS cell types: MB neurons (Lee et al., 2000; Yu et al., 2013; Lai et al., 2016) and ventral Corazonin neurons (Choi, 2006; Lee et al., 2013; Wang et al., 2019). Beat-Va_M neurons exhibit EcR-independent, astrocyte-activated local neurite pruning, while Beat-Va_L neurons die in a Hox gene-mediated, segmentally restricted pattern. The unique set of mechanisms that we found to underly Beat-Va neuron remodeling—among the first new neuronal subtypes whose remodeling we have characterized in detail—also supports the notion that a diversity of undiscovered remodeling mechanisms likely exist across the brain.

Beat-Va_L neurons die in a caspase and steroid hormone-dependent fashion, similar to other neuronal cell types that undergo cell death at metamorphosis (Choi, 2006; Lee et al., 2013). However, we found that the Hox gene Abd-B controls the segment-specific patterns of Beat-Va_L neuron cell death, with the three Abd-B–positive posterior Beat-Va_L cells undergoing cell death, while the anterior two Abd-B–negative cells survive. Knocking down Abd-B in Beat-Va neurons blocked caspase activation and cell death in the posterior cells, while overexpressing Abd-B in Beat-Va neurons drove cell death in the two, normally surviving, anterior Beat-Va_L cells. Notably, neither overexpression nor knockdown of Abd-B had any noticeable effect on Beat-Va_M neuron local pruning. We propose that Abd-B confers positional identity that leads to cell death in the appropriate Beat-Va_L cells.

Beat-Va_M neurons undergo massive local neurite pruning during the first 12 h of metamorphosis, which we visualized with single-cell precision. Surprisingly, unlike all other neuronal cell types studied in Drosophila, we found that Beat-Va_M neuron local pruning was not strongly suppressed by cell-autonomous blockade of EcR signaling. Rather, we found that blocking astrocyte developmental progression (into a phagocytic phenotype) using EcR^DN reduced Beat-Va_M remodeling by ∼50%, and when combined with simultaneous blockade of EcR signaling in Beat-Va_M neurons, local neurite pruning was almost entirely blocked. We proposed that extrinsic cues from phagocytic astrocytes and neuron-intrinsic EcR-mediated events converge to control Beat-Va_M remodeling. Beat-Va_M neurons should therefore provide an excellent system in which to genetically dissect astrocyte to neuron signaling pathways that promote local pruning.

Why a role for neuron-autonomous EcR signaling is more fully revealed when astrocytes are also blocked from transformation is an open question. A recent study showed that neuronal ecdysone signaling allows ddaC neurons to become competent for pruning by driving microtubule rearrangement, with subsequent physical force caused by tissue movement in the body wall during metamorphosis driving the severing and fragmentation of dendrites (Krämer et al., 2023). It is possible that neuronal ecdysone signaling may make Beat-Va_M neurons competent for remodeling through similar mechanisms, but the execution of fragmentation requires a secondary mechanical event. Astrocytes could, for example, provide the physical force that severs small neurites during metamorphosis as they transform and become phagocytic. We certainly observed a tight correlation between astrocyte transformation and Beat-Va_M neurite fragmentation. We noted that the fragmentation we observed when EcR signaling was blocked in Beat-Va_M neurons occurred primarily in smaller-diameter Beat-Va_M neurites. Perhaps the architectural integrity of large versus small diameter of a neuronal process makes fine neurite processes more susceptible to remodeling, as larger processes are more likely to be microtubule rich while smaller processes are more likely actin-based.

Do astrocytes release molecular cues onto neurites as they transform into phagocytes that promote the progression of neurite remodeling? The glial-secreted TGFβ ligand Myo is known to activate the expression of EcR in neurons, thereby establishing their competence to prune in response to ecdysone (Awasaki et al., 2011; Yu et al., 2013; Hakim et al., 2014; Wang et al., 2019). We found that depletion of Myo from astrocytes did not block Beat-Va_M neuron local pruning nor astrocyte transformation but pan-glial Myo depletion led to decreased EcR expression in astrocytes and a failure of astrocytes to transform. We proposed that Myo released redundantly by the cortex or ensheathing glia promotes EcR expression in astrocytes during early metamorphosis, which can then activate the phagocytic (or remodeling) program in astrocytes. To begin to explore the nature of astrocytic cues that might drive Beat-Va_M remodeling, we performed a targeted forward genetic screen for astrocyte molecules required for Beat-Va_M pruning. Although our screen explored roles for only transmembrane IgSF molecules on astrocytes, the overall results were informative. We found a tight correlation between astrocyte transformation and fragmentation of Beat-Va_M neurites—Beat-Va_M neurites only fragmented if astrocytes transformed into phagocytes. If these two processes were genetically separable, we might have expected to find a genetic manipulation where astrocytes transformed into phagocytes while neurites remained intact. Perhaps perturbation of other transmembrane or secreted molecules in the future will result in such a phenotype? Alternatively, it is possible that similar to peripheral dendrites, astrocyte phagocytic transformation enables them to physically break down CNS neurites as a normal part of local pruning.

See Table S4 for information on reagents used, including Drosophila lines, antibodies, and software.

Drosophila melanogaster was reared using standard laboratory conditions. All experiments were conducted at 25°C unless otherwise explicitly noted. A complete list of all fly strains used in this study can be found in the key resource table.

Dissection and immunostaining of larval and pupal fly brains were performed according to an adapted FlyLight protocol (Jenett et al., 2012). Larval brains were dissected in cold PBST (phosphate buffered saline; Invitrogen) with 0.1% Triton X-100 and fixed at 4% formaldehyde (Electron Microscopy Sciences) for 20 min at room temperature. Fixed brains were washed three times with PBS while nutating at room temperature for 10–15 min per wash. Primary and secondary antibodies were diluted in PBST (0.1% Triton X-100) and incubated with samples at 4°C for 24–48 h. Washes after antibody incubations were performed at room temperature with PBS for 3 × 15 min. Samples were mounted in VECTASHIELD antifade mounting medium (Vector Laboratories) and stored at 4°C until imaging.

Imaging was performed using a Zeiss LSM 880 with Airyscan. Confocal z-stacks were acquired using the optimal z-interval and 0.05–0.09 µm/pixel resolution with a 40×/1.3 Plan-Apochromat oil objective. Images were-Airyscan processed with Zeiss’ ZEN Black software, image tiles stitched with a 6–8% overlap when necessary, and converted into IMARIS format for 3D analysis. 3D rendering was performed with IMARIS; and single image or z-stack projections were analyzed in ImageJ. All imaging was done at 25°C.

To quantify individual neuronal morphologies, we used the Surface module in IMARIS 10 (Bitplane) with the Filament module. The module was trained on example images through iterative machine learning and then the algorithm was applied to all images. The first 100 µm of each neuron after it crossed the midline was analyzed for both total filament length and number of branch points. To determine if a neurite was intact or fragmented, individual filaments were then examined with the micrograph overlayed on the rendering. If a break in a neurite was observed (an area of any length without neurite labeling), the filament was discarded.

All statistical analyses were carried out using GraphPad Prism 9; statistical details, P values, and sample sizes are indicated in the figure legends. Any comparison between two parametric data sets that only had one independent variable was conducted with a Student’s t test. Any comparison between >2 parametric data sets that only had one independent variable was conducted with a one-way ANOVA with the Sidak test for multiple comparisons. Any comparison between >2 non-parametric data sets that only had one independent variable was conducted with a one-way ANOVA with the Kruskal-Wallis test for multiple comparisons. Any comparison between >2 parametric data sets that had two independent variables was conducted with a two-way ANOVA with the Sidak test for multiple comparisons. Parametric data distribution was assumed to be normal but this was not formally tested. For all graphs accompanied by imaging data, the data point that corresponds to the image shown is colored.

BeatVa-LexA lines were made with Gateway cloning. The 3,471 base pair enhancer region was cloned with primer sequences provided on the Janelia Flylight page. They were cloned into the pBPnlsLexAGADflUw vector (#26232; Addgene) and verified through Sanger sequencing (Genewiz). Injections were site directed using the y¹w^67c23; P[CaryP attP2] line (BL8622), y¹w^67c23P[CaryP attP18] (BL 32107) line and y¹w^67c23; P[CaryP attP40] line. The attp18 insertion was used in all the above experiments. The line was verified by comparing GFP expression generation from the Beat-LexA line driving a LexAop-GFP to a BeatVa-Gal4 line driving a UAS-mCD8::cherry.

The LexAop-EcR^DN construct was created by amplifying DNA from flies containing UAS-EcR.B1DC655.W650A (Bloomington Drosophila Stock Center). The amplification was carried out using the EcR^DN Kozak AgeI F primer (5′-CCC​CCA​ACC​GGT​CAA​AAC​ATG​AAG​CGG​CGC​TGG​TCG​AAC​A-3′) and EcR^DN-W650AXbaI R primer (5′-CCC​AAT​CTA​GAC​TAG​ATG​GCA​TGA​ACG​TCG​GCG​A-3′). The resulting PCR product was digested with AgeI and XbaI (NEB) and then ligated into the pattB-13xLexAop2EGFP plasmid (Coutinho-Budd et al., 2017), with GFP removed via AgeI and XbaI digestion. The final construct, pattB-13xLexAop2-EcR^W650A, was generated through ligation using T4 ligase (NEB), and the transformation was carried out in DH5α cells (Zymo). The construct’s sequence was verified through Sanger sequencing (Genewiz), and it was utilized for attp154 landing site-specific integration on the second chromosome (Bestgene).

Fig. S1 shows how we established the segmental characterization of Beat-Va neurons. Fig. S2 shows the segmental breakdown of Beat-Va_L and Beat-Va_M caspase activation. Fig. S3 contains supporting data about Abd-B and EcR-B1 localization in wildtype and knockdown conditions. Fig. S4 shows astrocyte transformation in wildtype conditions and when EcR signaling is blocked specifically in astrocytes; additionally, it shows micrographs of Beat-Va_M neurons in these conditions. Fig. S5 verifies the overlap between the BeatVa-Gal4 line and the BeatVa-LexA line. Table S1 is the in vivo evaluation of Drosophila lines from the Janelia Flylight collection at wL3, while Table S2 evaluates a subset of these lines at multiple time points. Table S3 describes the phenotypes from our targeted screen to drive astrocyte-mediated Beat-Va_M neuronal remodeling. Table S4 lists our reagents, including Drosophila lines, antibodies, software and algorithms, and cloning reagents. The source and RRID identifier for each reagent are also listed.

Raw data underlying this study is available from the corresponding author upon request.

We would like to acknowledge all members of the Freeman lab for thoughtful discussions and suggestions throughout this project. We acknowledge the fly community for the generous sharing of reagents.

This study was supported by National Science Foundation Graduate Research Fellowship Program Fellowship DGE-1448072 (to K.S. Lehmann), Damon Runyon Fellowship DRG 2329-18 (to Y. Kang), National Institutes of Health (NIH) grant K99 NS126642-01 (to Y. Kang), and NIH/National Institute of Neurological Disorders and Stroke RO1s R37NS053538 and R01NS124146 (to M.R. Freeman).

Author contributions: K.S. Lehmann: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Validation, Visualization, Writing - original draft, Writing - review & editing, M.T. Hupp: Data curation, Investigation, Visualization, Writing - review & editing, L. Abalde-Atristain: Data curation, Formal analysis, Investigation, Validation, Writing - review & editing, A. Jefferson: Data curation, Investigation, Validation, Writing - review & editing, Y.-C. Cheng: Resources, A.E. Sheehan: Investigation, Resources, Y. Kang: Conceptualization, Data curation, Investigation, Methodology, Project administration, Supervision, Validation, Visualization, Writing - review & editing, M.R. Freeman: Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing - original draft, Writing - review & editing.