Tracheal tube fusion in Drosophila involves release of extracellular vesicles from multivesicular bodies

Tubular organs, such as the vertebrate vasculature or the Drosophila tracheae, develop from separate units that fuse to form tubular networks. Tracheal tube fusion in Drosophila is mediated by specialized ‘fusion’ cells (FCs; Samakovlis et al., 1996), which transform into lumenized toroids to connect adjacent tubes. The fusion of luminal membranes inside FCs depends on the Munc13-4 (also known as UNC13D) ortholog Staccato (Stac; Caviglia et al., 2016), which localizes to lysosome-related organelles (LROs) exclusively in FCs. The formation of Stac-positive LROs (referred to hereafter as Stac-LROs) depends on the GTPase Arl3 (also known as Dnd in Drosophila; Jiang et al., 2007; Caviglia et al., 2016), which is specifically expressed in FCs and has been proposed to target the exocytosis machinery to apical membranes (Jiang et al., 2007; Kakihara et al., 2008). However, how LROs mediate tube fusion remains unclear. Munc13-4, together with the GTPase Rab27A, promotes SNARE-mediated membrane fusion of LROs in mammalian hematopoietic cells and melanocytes (Alzahofi et al., 2020; Boswell et al., 2012), and regulates multivesicular body (MVB) maturation and release of exosomes in cultured cancer cells (Messenger et al., 2018). Exosomes originate from exocytosis of MVBs, while ectosomes, another class of membranous extracellular vesicles (EVs), bud directly from the plasma membrane (Kalluri and LeBleu, 2020). EVs are thought to play important roles in intercellular communication during development, homeostasis and disease. However, despite many studies of EVs in cultured cells, evidence of EVs and their functions in vivo has remained rare (Corrigan et al., 2014; Scott et al., 2021; Tsai et al., 2019; Verweij et al., 2018). We report here the discovery of EVs in the developing tracheae of Drosophila embryos. A subset of these EVs contains Stac, is associated with tracheal tube fusion events, and depends on Arl3 function. Rab27 and Rab35 act in a partially redundant fashion downstream of Arl3 to promote formation of Stac-positive MVBs (Stac-MVBs) and tube fusion We propose that Stac-positive EVs (Stac-EVs) originate from the fusion of multivesicular LROs with the luminal plasma membrane during anastomosis formation.

We observed that EGFP–Stac and mCherry–Stac were distributed throughout the cytoplasm of embryonic tracheal cells, only accumulating at LROs in FCs (Fig. 1A; Caviglia et al., 2016). Surprisingly, we also noticed EGFP–Stac- or mCherry–Stac-positive puncta inside the tracheal lumen of stage 15 embryos (Fig. 1A–C; Movie 1). Extracellular EGFP–Stac puncta were submicrometer-sized (0.60±0.12 µm diameter based on confocal microscopy, mean±s.d.; n=19 EVs; Fig. 1I) and were surrounded by luminal material containing the secreted protein Vermiform–mRFP (Verm–mRFP; Fig. 1A). EGFP–Stac puncta were detectable in the lumen of the dorsal trunk (DT) tube in 40% of the embryos (Fig. 1H; n=50 embryos). Stac puncta were labeled by palmitoylated mKate2 (palm–mKate2; Fig. 1B) and by CD63–GFP, a GFP-tagged vertebrate tetraspanin that localizes to limiting and internal membranes of MVBs and is released on exosomes by mammalian as well as Drosophila cells (Fig. 1C; Escola et al., 1998; Panáková et al., 2005). These findings suggest that the luminal Stac puncta are membranous EVs. We observed a larger pool of CD63–GFP-positive EVs, only a fraction of which were also positive for mCherry–Stac (2%, n=89 EVs in 11 embryos; Fig. 1C). CD63–GFP-positive EVs were found in all embryos analyzed (n=14; Fig. 1C,H) and resembled EGFP–Stac-positive EVs in size (0.63±0.25 µm diameter; n=79 EVs in 11 embryos; Fig. 1I). Thus, the embryonic tracheal lumen contains distinct pools of EVs, a subset of which accumulates Stac (Stac-EVs).

Luminal EVs were largely immobile over the course of 4 h (Fig. S1A,B; Movies 2,3), suggesting that EVs are trapped in the matrix that fills the embryonic tracheal lumen (Dong et al., 2014; Tonning et al., 2006), and they disappeared during luminal liquid clearance (Fig. S1C and Movie 3). Interestingly, Stac-EVs showed a non-uniform distribution along the tracheal lumen and were frequently found near tube fusion sites at tracheal metamere boundaries (marked by adjacent transverse connective branches; Fig. 1J; Fig. S1C), suggesting that Stac-EVs might be released by FCs during tube fusion. Although direct visualization of EV exocytosis events was precluded by the small diameter of the lumen during tube fusion and by the small size of EVs, time-lapse movies allowed tracking of Stac-EVs to DT fusion points (Fig. S1A–C, Movie 2), suggesting that Stac-EVs are released during the tube fusion process. Supporting this notion, Stac-EVs were detected in the tracheal lumen when EGFP–Stac was expressed exclusively in FCs (Fig. S1D; 16% of embryos, n=31), and Stac-EVs were absent from non-fused tracheal metameres in short stop³ (shot³) mutants (n=23 embryos; Fig. 1D).

We did not detect Stac-EVs in the tracheal lumen of Arl3¹ embryos (n=50), which lack Stac-LROs (Fig. 1E,H; Caviglia et al., 2016), indicating that the FC-specific GTPase Arl3 is required for Stac-EV formation and corroborating the notion that Stac-EVs originate from Stac-LROs in FCs. Conversely, to ask whether the presence of Stac-LROs is sufficient to enable release of Stac-EVs, we generated ectopic Stac-LROs throughout the tracheal system by mis-expressing Arl3 in all tracheal cells (Fig. 1F; Caviglia et al., 2016). However, Arl3 mis-expression did not affect the number (Fig. 1H) or spatial distribution (Fig. 1J) of Stac-EVs in the DT lumen, suggesting that Stac-EVs are released exclusively by FCs even when Stac-LROs are ectopically induced in other tracheal cells.

By contrast, CD63–GFP-positive, mCherry–Stac-negative EVs were distributed more broadly throughout the lumen (Fig. 1J). CD63–GFP-positive EVs were present also in tube-fusion-defective stac^3B20 embryos, albeit at reduced numbers compared to controls (Fig. 1G,H; n=10 embryos), indicating that the formation of CD63–GFP-positive EVs does not depend on tube fusion or stac function. Moreover, Stac-EVs and CD63-positive EVs were found in the tracheae of secretion-defective γCOP¹⁰ mutants, suggesting that EVs are released independently of the secretory activity that drives luminal expansion (Fig. S1E–G; Grieder et al., 2008). Taken together, these findings suggest that CD63–GFP-positive, Stac-negative EVs are produced by all tracheal cells, whereas Stac-EVs are associated with FCs, and their formation depends on Arl3.

Based on these results, we hypothesized that Stac-EVs originate from MVBs in FCs. Supporting this idea, Stac-LROs were labeled by the MVB marker CD63–GFP (Fig. 2A), and mCherry–Stac accumulated inside and at the boundary of CD63–GFP-limited vesicular structures (Fig. 2B,C), suggesting that mCherry–Stac is present in intraluminal vesicles (ILVs) and at the outer membrane of MVBs. To test whether Stac can be incorporated into ILVs, we expressed EGFP–Stac in secondary cells of adult Drosophila testis accessory glands. These cells contain large MVBs and release EVs into the gland lumen (Fig. S2; Corrigan et al., 2014; Fan et al., 2020). Indeed, EGFP–Stac was present in ILVs inside MVBs in secondary cells, as well as in membranous EVs in the accessory gland lumen (Fig. S2D,E). Thus, Stac can be incorporated into ILVs and can be secreted in EVs into the accessory gland and tracheal lumen. Consistent with the hypothesis that Stac-MVBs release EVs during tracheal tube fusion, we observed Stac-MVBs between the apical membrane tips of approaching FC lumina during tube fusion (Fig. 2D–G).

Arl3 is necessary for the presence of luminal Stac-EVs (Fig. 1E,H) and for Stac-LRO formation in tracheal cells (Caviglia et al., 2016), and may act by modulating the activity of Rab GTPases that recruit Stac to LROs, analogous to the proposed role of Arl3 in mammalian cells (Ismail, 2011; Williams, 2011). To test whether Arl3 and Stac act in parallel or in the same pathway to promote Stac-LRO formation and tube fusion, we analyzed tracheal defects in amorphic Arl3¹ and stac^3B20 single mutant and in Arl3¹ stac^3B20 double mutant embryos (Fig. 3A–E). Arl3¹ and stac^3B20 embryos displayed tube fusion defects of similar expressivity, with on average two out of nine fused DT anastomoses (Fig. 3E). Fusion defects were not enhanced in Arl3¹ stac^3B20 double mutants (Fig. 3D,E), suggesting that Arl3 and Stac act in the same pathway to mediate Stac-MVB formation. Consistent with this notion, HA-tagged Arl3 partially colocalized with EGFP–Stac in tracheal cells (Fig. 3F).

To identify factors acting downstream of Arl3 to promote Stac-LRO formation, we generated a sensitized genetic background by mis-expressing Arl3 throughout the tracheal system in EGFP–Stac-expressing embryos. This led to ectopic Stac-LROs in most tracheal cells (44±24 Stac-LROs in DT metameres 4–6, mean±s.d.; n=10; Fig. S3). To identify Rab GTPases required for Stac-LRO formation, we reduced the gene dosage in this genetic background of 26 of the 33 annotated Drosophila Rab GTPases by introducing null mutations of the respective Rab loci (Chan et al., 2011; Kohrs et al., 2021) and analyzed the number of Stac-LROs (Fig. S3A). This approach revealed six Rab GTPases (Rab8, Rab10, Rab27, Rab30, Rab35 and Rab39) whose reduced gene dosage led to significantly reduced Stac-LRO numbers compared to those in control embryos (Fig. S3B). We focused on Rab27, Rab35 and Rab39, which are likely to regulate late-endosomal trafficking, and membrane docking of LROs and MVBs (Alzahofi et al., 2020; Biesemann et al., 2017; Caviglia et al., 2016; Neeft et al., 2005) or exosome release (Messenger et al., 2018), whereas Rab8, Rab10 and Rab30 are predicted to play general roles in endoplasmic reticulum morphology, Golgi sorting and retrograde transport (Bellec et al., 2018; English and Voeltz, 2013; Gillingham et al., 2014; Schuck et al., 2007).

Having identified Rab27, Rab35 and Rab39 mutations as suppressors of ectopic Stac-LRO formation, we tested whether these Rabs are also required for Stac-LRO formation in FCs of otherwise wild-type embryos. Embryos lacking maternal and zygotic Rab27 or zygotic Rab35 displayed significantly reduced numbers of Stac-LRO in FCs (47% and 53%, respectively, compared to control; Fig. 4A,C,D,F), suggesting that these GTPases participate in Stac-LRO formation. Surprisingly, however, Rab39⁻ embryos showed similar Stac-LRO numbers as controls (Fig. 4A,B,F), despite the suppression of ectopic Stac-LROs by Rab39⁻ in embryos mis-expressing Arl3 (Fig. S3), suggesting that Rab39 is not essential for forming Stac-LROs in wild-type FCs. Because loss of Rab27 or of Rab35 each caused a partial reduction of Stac-LROs, we generated Rab27⁻ Rab35⁻ double mutants to ask whether Rab27 and Rab35 cooperate to form Stac-LROs. Indeed, Rab27⁻ Rab35⁻ embryos showed significantly fewer Stac-LROs compared to Rab27⁻ and Rab35^– single mutants (30.5% and 27%, respectively; Fig. 4A,C–F), suggesting that Rab27 and Rab35 cooperate in Stac-LRO formation. Consistent with these findings, overexpressed YFP–Rab27 (Fig. 4N) and endogenously tagged YFP::Rab35 (Fig. 4O; Caviglia et al., 2016) partially overlapped with mCherry–Stac-positive LROs in FCs. Moreover, Rab27⁻ and Rab35⁻ embryos exhibited slightly reduced numbers of Stac-EVs. We found that 80% of Rab27⁻ and 76% of Rab35⁻ embryos lacked detectable EVs, whereas at least one EV was detectable in 40% of control embryos (n=50 embryos per genotype). Taken together, these results suggest that Rab27 and Rab35 cooperate to promote Stac-LRO formation and possibly the release of Stac-EVs.

We finally asked whether Rab27, Rab35 and Rab39 are required for tracheal tube fusion. While Rab27⁻ and Rab39⁻ homozygous (female) and hemizygous (male) flies were viable and fertile, Rab35⁻ flies were semi-lethal. Embryos lacking maternal and zygotic Rab27 or Rab39 and embryos lacking zygotic Rab35 did not show DT fusion defects (Fig. 4A–E,G–K). To assess more subtle tracheal defects, we analyzed dorsal branch (DB) fusion, which is more sensitive than DT fusion to genetic perturbations. Whereas nearly all ten DBs per embryo were fused in wild-type controls (mean of 9.9 fused DBs, n=25 embryos) and in most Rab mutant embryos (for example, mean of 9.6 fused DBs in Rab26⁻ embryos, n=14 embryos), Rab35⁻ and Rab39⁻ embryos showed DB fusion defects (mean of 9.1 and 8.8 fused DBs, n=18 and n=13, respectively; Fig. 4G,H,J,M). These defects were reproduced by expressing in tracheal cells dominant-negative versions of Rab35 (Rab35^DN; mean of 7.3 fused DBs, n=11; Fig. 4L,M) or Rab39 (Caviglia et al., 2016). By contrast, Rab27⁻ embryos did not show notable tracheal defects (mean of 9.8 fused DBs, n=16; Fig. 4I,M). However, double mutant embryos lacking zygotic Rab27 and Rab35 showed enhanced DB fusion defects (mean of 8.1 fused DBs, n=12) compared to Rab27⁻ and Rab35⁻ single mutants (Fig. 4I–K,M), indicating that Rab27 and Rab35 exert partially redundant functions.

We report the presence of membranous EVs in the lumen of developing tracheal tubes in Drosophila embryos. A subset of these EVs is associated with tracheal tube fusion sites and carries the Munc13-4 ortholog Stac. Stac associates with LROs that display features of MVBs, the formation of which, as well as of Stac-EVs, depends on the fusion-cell-specific GTPase Arl3. Moreover, we identified Rab27 and Rab35 as partially redundant regulators of Arl3-dependent Stac-MVB formation and tracheal tube fusion. We propose that tracheal anastomosis formation involves fusion of Stac-MVBs with the luminal membranes of FCs, resulting in release of Stac-EVs (Fig. S4). Our attempts to visualize tracheal EV exocytosis using the pH-sensitive fluorescent proteins pHluorin (Sankaranarayanan et al., 2000) and pHuji (Shen et al., 2014) fused to the lysosomal transmembrane protein Lamp1 were unsuccessful (data not shown). However, we were able to detect EVs that emerged at and remained associated with anastomosis sites upon completion of DT fusion. Moreover, the absence of Stac-EVs from fusion-defective tracheal metameres in shot mutant embryos supports the idea that Stac-EVs are released during tube fusion.

How is MVB formation regulated in FCs? Consistent with our finding that Rab27 and Rab35 act downstream of Arl3 to promote Stac-MVB formation, Rab27a, Rab27b and Rab35 (Ostrowski et al., 2010; Hsu et al., 2010) regulate MVB-dependent exosome secretion in mammalian cells. Moreover, the Stac homolog Munc13-4 is a key effector of Rab27a in LRO docking in hematopoietic cells and melanocytes (Alzahofi et al., 2020; Boswell et al., 2012), and regulates MVB maturation and exosome release in cancer cells (Messenger et al., 2018). Whereas Rab27 and Rab35 are required for Stac-MVB formation or for anchoring Stac at MVBs, Rab39 is dispensable for FC-specific Stac-LRO formation, suggesting that it acts in a successive step, such as transport or plasma membrane docking of LROs. While our findings define a core pathway of MVB maturation and EV release in tracheal cells, additional players, such as ESCRT complex components, are likely to be involved. MVBs also act as intermediates in intracellular lumen formation in tracheal terminal cells, which resemble FCs in forming seamless tubes (Mathew et al., 2020; Nikolova and Metzstein, 2015). Interestingly, Rab35 directs transport of apical components to promote terminal cell lumen growth (Schottenfeld-Roames and Ghabrial, 2012). Whether Rab35 also participates in MVB formation in tracheal terminal cells remains to be tested.

Despite substantial progress in characterizing EVs and their mode of action in cultured cells, studies of physiological functions of EVs in vivo have remained scarce. Exosomes are released into the seminal fluid in Drosophila testis accessory glands and modulate female reproductive behavior (Corrigan et al., 2014). Glia-derived exosomes stimulate growth of motor neurons and tracheae in Drosophila larvae through exosomal microRNA-dependent gene regulation in target cells (Tsai et al., 2019). Moreover, cardiomyocyte-derived EVs are taken up by macrophages and endothelial cells in zebrafish embryos (Scott et al., 2021), and may play a role in trophic support of endothelial cells during vasculogenesis (Verweij et al., 2018). It will be exciting to explore possible functions of tracheal EVs, including potential roles in long-range intercellular communication across the tracheal lumen. However, the luminal EVs that we discovered in late-stage embryos are unlikely to participate in such intercellular communication, as the presence at this stage of a luminal matrix (Dong et al., 2014; Tonning et al., 2006) appears to constrain EV mobility, and the cuticle covering the apical tracheal cell surface presumably prevents fusion of luminal EVs with the plasma membrane. Rather, Stac-EVs are likely to constitute remnants of MVB–plasma membrane fusion events during anastomosis formation. Stac-MVBs could serve as membrane reservoirs that facilitate lumen fusion. ILVs might back-fuse with the MVB limiting membrane, resulting in a rapid increase of membrane material available at the luminal membrane interfaces in FCs (Fig. S4). Such a mechanism has been described in dendritic cells, where MVBs carrying major histocompatibility complex class II (MHC II) in ILVs are reorganized after stimulation, resulting in tubular membrane extensions and transport of MHCII to the plasma membrane (Kleijmeer et al., 2001). Recruitment of ILVs to the MVB limiting membrane will increase the surface area and may lead to surface exposure of ILV membrane proteins that could promote membrane fusion. Exploring how MVB exocytosis is regulated in tracheal cells, and whether similar mechanisms contribute to vascular anastomosis formation in vertebrates, will be exciting subjects for future studies.

Arl3¹ (Kakihara et al., 2008) carries a deletion that removes 680 bp including the start codon of Arl3 and part of the overlapping 5′ UTR of the neighboring gene CG6678, suggesting that Arl3¹ is a null mutation (Kakihara et al., 2008). stac^3B20 (Caviglia et al., 2016) contains a T-to-A transversion in the 5′ splice site of the last intron, resulting in retention of the intron in stac mRNA. Translation of the mutant mRNA yields presumed truncated proteins lacking the carboxy-terminal 30 amino acids. Embryos carrying stac^3B20 in trans to a deletion [Df(3R)BSC493] of the stac locus show tracheal defects indistinguishable from stac^3B20 homozygotes, indicating that stac^3B20 is an amorphic mutation (Caviglia et al., 2016).

The following Drosophila strains used are described in FlyBase, unless indicated otherwise: Abd-B-Gal4, btl-Gal4, dysf-Gal4 (Caviglia et al., 2016), btl-Moesin-mRFP, γCOP¹⁰ (Grieder et al., 2008), shot³ (Lee and Kolodziej, 2002), UAS-PLCδ-PH-EGFP, UAS-EGFP-Stac, UAS-mCherry-Stac, UAS-Verm-mRFP, UAS-palm-mKate2 (Caviglia et al., 2016), UAS-CD63-GFP (Panáková et al., 2005; gift from the late Suzanne Eaton, Max-Planck Institute of Cell Biology and Genetics, Dresden, Germany), E-Cad::3xmTagRFP (Pinheiro et al., 2017; gift from Yohanns Bellaiche, Institut Curie, Paris, France), UAS-Arl3 (Kakihara et al., 2008; gift from Shigeo Hayashi, RIKEN Center, Kobe, Japan), UAS-Arl3-3xHA (Bischof et al., 2007), UAS-YFP-Rab27 (Zhang et al., 2007), YFP::Rab35 (Dunst et al., 2015; gift from Marko Brankatschk, Technical University Dresden, Germany). The collection of Rab null alleles has been described previously (Chan et al., 2011; Kohrs et al., 2021; gift from Robin Hiesinger, Freie Universität Berlin, Germany). Lethal Rab mutants were balanced using FM7 Dfd-GMR-nvYFP, CyO Dfd-GMR-nvYFP, or TM6B Dfd-GMR-nvYFP balancer chromosomes (Le et al., 2006). For analyzing the effects of autosomal Rab mutations on Stac-LRO number, btl-Gal4 UAS-EGFP-Stac females were crossed to males carrying the Rab⁻ mutation. For X-chromosomal Rab loci, females carrying the Rab⁻ mutation were crossed to btl-Gal4 UAS-EGFP-Stac males. Embryos (14–18 h) were collected at 22°C. Embryos carrying the Rab⁻ allele were identified by the absence of Dfd-YFP expression. The y¹ w¹¹¹⁸ strain was used as wild-type control.

UAS-OLLAS-pHluorin-Lamp1 and UAS-OLLAS-pHuji-Lamp1 constructs were generated by fusing the OLLAS-pHluorin (Sankaranarayanan et al., 2000) or OLLAS-pHuji (Shen et al., 2014) open reading frames to an N-terminal preprolactin signal peptide and C-terminally to the transmembrane domain and cytosolic domain of the human lysosomal transmembrane protein Lamp1 (Pulipparacharuvil et al., 2005), which localizes to Stac-LROs in FCs (Caviglia et al., 2016). The constructs were inserted into the pUASt-attB vector (using EcoRI and XbaI restriction sites; Bischof et al., 2007) and integrated into the attP40 and attP2 landing sites using PhiC31 integrase (Bischof et al., 2007). To generate flies carrying UAS-EGFP-Stac on the third chromosome, pUASt-attB-UAS-EGFP-StacA (Caviglia et al., 2016) was integrated into the attP2 landing site.

Embryos were collected 14–18 h after egg lay (h AEL) and 17h–18 h AEL at 22°C for analysis of DT and DB fusion defects, respectively. Embryos were fixed in 4% formaldehyde in phosphate-buffered saline (PBS) and heptane (1:1) for 20 min and devitellinized by shaking in methanol and heptane (1:1). The following primary antibodies were used: mouse anti-GFP (1:500; Sigma, G6539), chicken anti-GFP (1:500; Abcam, 13970), rat anti-HA (1:300; Roche, 3F10). Luminal chitin was detected using the chitin-binding domain from Bacillus circulans chitinase A1 conjugated with SNAP-Surface Alexa Fluor 488, 546 or 647 (1:500; NEB), produced as described previously (Caviglia and Luschnig, 2013).

For live imaging, dechorionated embryos were glued on a coverslip (0.17 mm, grade #1.5), embedded in Voltalef 10S oil and covered with a gas-permeable foil (Lumox, Sarstedt). Accessory glands were dissected from 3-day-old virgin males, stained with membrane dye MM4-64 (2 nM; Santa Cruz Biotechnology) and Hoechst 33342 (1 μg/ml; Sigma) to label plasma membrane and nuclei, respectively, and were mounted on slides in cold (4°C) PBS. Imaging was performed on a Leica SP8 confocal microscope with 40×/1.3 NA and 63×/1.4 NA objectives and HyD detectors, or on a Zeiss LSM710 confocal microscope with a 40×/1.1 NA objective. Images were processed using OMERO (5.4.10; https://www.openmicroscopy.org/omero/), Fiji (ImageJ; 1.53c; https://fiji.sc/) and Imaris (8.4.1, Bitplane), and were prepared in Affinity Designer (1.8.4; Serif). Cross-sections were performed using Imaris (8.4.1). Where indicated, images were deconvolved using the Leica HyVolution software in adaptive mode. Calculations were performed in R (3.5.1) using RStudio Interface (1.3.1093).

Fixed embryos were stained for luminal chitin. The chitin signal at DB (in stage 16 embryos) or DT (in stage 15 embryos) anastomoses was analyzed to determine whether the lumen was continuous or interrupted. The y¹ w¹¹¹⁸ strain was used as a wild-type control. In addition, Rab26⁻ embryos were analyzed as a control for the specificity of the DB tube fusion defects in Rab mutants.

Confocal z-stacks of tracheal metameres 4–6 were acquired in living embryos expressing EGFP–Stac in tracheal cells under the control of btl-Gal4. Stac-LROs were segmented in 3D by applying a smooth function, enhancing contrast (0.01% saturated pixels) and using a manually defined threshold in Fiji. The number and volume of the vesicles was determined using the 3D object counter plugin in Fiji (Bolte and Cordelières, 2006).

We thank Wilko Backer for expert technical help, Raphael Schleutker for help with image analysis and statistics, and Yohanns Bellaiche, Marko Brankatschk, the late Suzanne Eaton, Shigeo Hayashi and Robin Hiesinger for providing fly stocks and reagents. We thank Sara Caviglia, Mylène Lancino and Raphael Schleutker for comments on the manuscript.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259590