Glial response to hypoxia in mutants of NPAS1/3 homolog Trachealess through Wg signaling to modulate synaptic bouton organization

Synaptic structure and activity are sensitive to environmental alterations. Modulation of synaptic morphology and function is often induced by signals from glia. However, the process by which glia mediate synaptic responses to environmental perturbations such as hypoxia remains unknown. Here, we report that, in the mutant for Trachealess (Trh), the Drosophila homolog for NPAS1 and NPAS3, smaller synaptic boutons form clusters named satellite boutons appear at larval neuromuscular junctions (NMJs), which is induced by the reduction of internal oxygen levels due to defective tracheal branches. Thus, the satellite bouton phenotype in the trh mutant is suppressed by hyperoxia, and recapitulated in wild-type larvae raised under hypoxia. We further show that hypoxia-inducible factor (HIF)-1α/Similar (Sima) is critical in mediating hypoxia-induced satellite bouton formation. Sima upregulates the level of the Wnt/Wingless (Wg) signal in glia, leading to reorganized microtubule structures within presynaptic sites. Finally, hypoxia-induced satellite boutons maintain normal synaptic transmission at the NMJs, which is crucial for coordinated larval locomotion.

Introduction Animals need oxygen and food, not only to sustain life, but also for motility. In vertebrates, oxygen and nutrients are delivered through the vascular systems to organs and tissues throughout the body. To maintain proper nutrient and oxygen supply, and thus physiological functions, the vascular system is also highly coordinated with the nervous system during development. Indeed, the vascular and nervous systems resemble each other in terms of their anatomical structures and developmental processes [1]. In the brain, nerves and vessels, form close associations and are in physical contact through the third player astrocytes to form neurovascular units (NVU) [2]. Such organization is essential for controlling oxygen and glucose delivery through the blood vessels by neuronal activity, and this regulatory process is mediated through the coupled astrocytes [3]. However, some invertebrates lack the complex vascular systems [4]. In nematodes, oxygen is supplied simply by ambient diffusion to inner cells [5]. Insects such as Drosophila have evolved a prototype of the tracheal system to deliver oxygen and a primitive vascular system, the dorsal vessel, to facilitate nutrient delivery [6]. However, the physical association of nerves, trachea, and glial processes has also been demonstrated at the NMJs of adult Drosophila flight muscles [7].
Animals respond to changing oxygen levels by altering their oxygen delivery system. Insufficient oxygen levels (hypoxia) activate a broad range of genes to re-establish body homeostasis. One crucial regulator of these hypoxia-responsive genes is the sequence-specific DNAbinding transcription factor hypoxia inducible factor 1 (HIF-1) [8]. HIF-1 consists of α and β subunits that form heterodimers [9]. Whereas HIF-1β is expressed constitutively, HIF-1α protein levels are modulated by oxygen levels [10]. Under normal oxygen conditions (normoxia), oxygen-dependent prolyl hydroxylases (PHDs) catalyze hydroxylation of a conserved prolyl residue in the central oxygen-dependent degradation (ODD) domain of HIF-1α [11]. Hydroxylation of HIF-1α promotes interaction with Von Hippel Lindau (VHL), which is the substrate recognition subunit of the cullin2-based E3 ubiquitin ligase, leading to HIF-1α ubiquitination and proteasomal degradation. Under hypoxia, prolyl hydroxylation does not occur, HIF-1α proteins are stabilized and are translocated from the cytoplasm to the nucleus where they form heterodimers with HIF-1β to activate transcription of target genes. One major class of target genes encoding the Fibroblast Growth Factor (FGF) is involved in inducing angiogenesis in mammals. In Drosophila, the FGF member encoded by Branchless (Bnl) induces tracheal branching [12]. When oxygen levels are reduced, oxygen-starved cells express Bnl as a chemoattractant to guide the growth tracheal terminal branches toward them [13].
In addition to adaptations of the respiratory system, the nervous system also responds to hypoxia. Oxygen levels modulate the survival, proliferation, and differentiation of radial glial cells (RGCs) in the human cerebral cortex. Interestingly, physiological hypoxia (3% O 2 ) induces neurogenesis and differentiation of RGCs into glutamatergic neurons [14]. Hypoxia induces neurite outgrowth in PC12 cells through activation of A2A receptor [15]. Brief exposure to anoxia and hypoglycemia caused axonal remodeling in hippocampal neurons, including presynaptic protrusion of filopodia and formation of multi-innervated spines [16]. Under hypoxia or upon depletion of PHD2, upregulation of the actin cross-linker Filamin A (FLNA) induces generation of more immature spines [17]. Astrocytes have been shown to play a crucial role in ischemic tolerance via the activation of P2X7 receptors, which trigger upregulation of HIF-1α [18].
Neuronal PAS (NPAS) proteins containing a DNA-binding Per-Arnt-Sim domain function in vascular and nervous system development. In mice, NPAS1 is responsible for cortical interneuron generation [19], whereas NPAS3 is required for adult neurogenesis [20]. NPAS1 and NPAS3 also play key roles in lung development [21,22]. The homolog of NPAS1/3 in Drosophila, Trachealess (Trh), has been well studied for its involvement in formation of the respiratory tracheal system. Trh is a master regulator of tracheal cell fates, activating gene expression to induce tracheal development [23,24]. However, the role of Trh in the development of other tissues, particularly the nervous system, is unknown. In this study, we found altered synaptic bouton morphology at the NMJs of trh 1 /trh 2 mutants. By performing trh-RNAi knockdown and UAS-trh transgene rescue experiments, we show that trh is required in tracheal cells for normal bouton formation. Defective tracheal branching in the trh 1 /trh 2 mutant mimics the effect of hypoxic conditions during larval development, and supplying higher than normal oxygen levels restored normal bouton morphology. We further show that glial cells respond to hypoxia by elevating Wnt/Wg expression to mediate synaptic bouton reorganization through HIF1-α/Sima in Drosophila. Finally, we reveal that this morphological change may be linked to normal synaptic transmission and locomotion of larvae.

trh modulates synaptic bouton formation non-cell autonomously
To understand the possible role of Trh in synapse formation, we examined NMJ morphology in the trh mutant. Since both trh 1 and trh 2 loss-of-function alleles are homozygous lethal [23,25,26], we examined the trans-heterozygous trh 1 /trh 2 mutant that survived to adult stages and compared it to wild-type (w 1118 ) and heterozygous trh 1 /+ controls. Synaptic boutons of w 1118 and trh 1 /+ NMJs were evenly spaced along the axonal terminals, displaying the typical "beadson-a-string" pattern ( Fig 1A, upper and middle panels, enlarged images at right). Strikingly, the trh 1 /trh 2 mutant larvae exhibited aberrant NMJ morphology. Small synaptic boutons formed clusters or surrounded a normal large bouton ( Fig 1A, bottom panel); a phenotype described as "satellite boutons" [27]. This satellite bouton phenotype in the trh 1 /trh 2 mutant was detected at a high frequency; 20.7 ± 4.1% (n = 10) of total boutons were satellite ones, more than three-fold increases compared to 4.3 ± 1.7% (n = 10) for trh 1 /+ and 6.3 ± 1.7% (n = 10) for w 1118 (Fig 1B). Although the percentage of satellite boutons in the trh 1 /trh 2 mutant was greatly increased, the total bouton number was slightly higher than that observed in controls (74.1 ± 4.7, n = 10 in w 1118 ; 89.6 ± 5.2, n = 10 in trh 1 /+; and 95.5 ± 9.1, n = 11 in trh 1 /trh 2 , bottom panel in Fig 1B). Also, the muscle areas were not significantly different from each other (S1A Fig). Given the small size and clustering of satellite boutons in the trh 1 /trh 2 mutant, we examined whether these boutons express synaptic proteins normally. We found that the synaptic vesicle protein Synapsin (Syn in Fig 1A) was normally distributed relative to control, but the active zone protein Bruchpilot (Brp) was expressed at higher levels in the trh 1 /trh 2 mutant (S1B Fig). The postsynaptic glutamate receptor, as revealed by GluRIII (S1B Fig) and GluRIIA (S1C Fig) signals, as well as dPAK (S1C Fig) were also localized in satellite boutons, which were surrounded by the subsynaptic reticulum protein Dlg (S1D Fig). Thus, although the Brp signal intensity in the trh 1 /trh 2 mutant was stronger than in controls, the composition of synaptic proteins in satellite boutons was largely similar to that of normal-sized boutons.
To further confirm the necessity of tracheal trh for normal bouton formation, we performed a rescue experiment for the trh 1 /trh 2 phenotype. When UAS-trh expression was driven by tracheal btl-GAL4 in the trh 1 /trh 2 mutant, the satellite bouton phenotype was suppressed (1.8 ± 1.4%, n = 13, Fig 1E and 1F). Controls bearing only btl-GAL4 (14.3 ± 3.3%, n = 10) or UAS-trh (21.1 ± 3.3%, n = 10) still contained large amounts of satellite boutons. These results indicate that trh is required in the trachea for normal bouton formation.

Hypoxia induces satellite bouton formation
Apart from specifying the tracheal cell fate, Trh is also involved in the branching of tubular structures during post-embryonic stages [24]. Therefore, we examined the tracheal phenotypes in the trh 1 /trh 2 larvae (S2A Fig) and observed an increase in the number of terminal branches in the dorsal branch of the third segment (S2B Fig, 5.7 ± 0.15, n = 10 for trh 1 /+, and 7.5 ± 0.28, n = 11 for trh 1 /trh 2 ). Furthermore, we identified morphological defects such as tracheal breaks and tangles, suggesting structural defects in the trh 1 /trh 2 larvae (arrows in S2A Fig). Tracheal branching activity is enhanced under hypoxia [12]. Thus, the increased terminal branches in trh 1 /trh 2 could be a compensatory mechanism for defective trachea formation.
To understand whether trh 1 /trh 2 mutant cells are under hypoxia, we used the hypoxia biosensor GFP-ODD, in which the GFP is fused to the ODD domain of Sima, under the control of the ubiquitin-69E (ubi) promoter [29]. We first confirmed that GFP-ODD signal was low under normoxia (21% O 2 ) and enhanced under hypoxia (5% O 2 ) in wild-type late-stage embryos when tracheal tubules are already formed and functioning [29]. Indeed, enhanced GFP signal was ubiquitous under hypoxia in wild-type embryos with some pronounced focal GFP signals (Fig 2A,  Quantification of the GFP/RFP ratio revealed a significant difference between normoxia and hypoxia (0.18 ± 0.05, n = 6 at 21% O 2 and 0.84 ± 0.15, n = 6 at 5% O 2 , Fig 2B). We then examined whether oxygen supply is deficient in the trh 1 /trh 2 mutant by measuring the GFP/RFP ratios. We detected a higher GFP/RFP ratio (0.82 ± 0.15, n = 6) in the mutant compared to that in heterozygous trh 1 /+ (0.09 ± 0.02, n = 6) in the 21% O 2 condition, supporting that the trh 1 /trh 2 mutant senses reduced oxygen levels internally (Fig 2A and 2B).
We also examined whether the Sima protein levels are changed in hypoxia or in the trh mutant. We found ubiquitous increases in the Sima levels in the wild-type control under the 5% O 2 condition or in the trh 1 /trh 2 mutant (S3E and S3F Fig). The increases could be identified in glial processes along the peripheral nerves and in different subtypes of glia. Thus, glial Sima could play the role to mediate hypoxia in the trh 1 /trh 2 mutant and in the low O 2 condition to modulate synaptic bouton formation.

Wg signals mediate glial Sima activity to modulate bouton morphology
Next, we explored possible signals transduced from glia to neurons in response to hypoxia. The glia-secreted Wingless (Wg) signaling molecule regulates synaptic growth at Drosophila NMJs [34,35]. Therefore, we examined whether Wg can be induced under hypoxia in the trh 1 / trh 2 mutant. Wg signals were enriched around the synaptic boutons of wild-type NMJs ( Fig  4A). Whereas the pattern of Wg signals at trh 1 /+ NMJs was similar to that of w 1118 , we detected much higher levels of Wg signals at the trh 1 /trh 2 NMJ (Fig 4A). Quantification of Wg immunofluorescence intensities normalized to co-stained HRP in trh 1 /trh 2 (Wg/HRP: 0.49 ± 0.07, n = 9, Fig 4B) revealed~3-fold increases relative to w 1118 (0.18 ± 0.03, n = 8) and trh 1 /+ (0.15 ± 0.02, n = 9). We then examined whether glial Sima is required for the enhanced Wg expression in the trh 1 /trh 2 mutant (Fig 4C). The Wg level relative to the HRP level in the trh 1 /trh 2 mutant carrying repo-GAL4 (0.68 ± 0.16, n = 8 for repo-GAL4 trh 1 /trh 2 ) was also about 3 folds to the repo-GAL4 control (0.23 ± 0.04, n = 11, Fig 4D). When we reduced sima levels in the trh 1 /trh 2 mutant by repo-GAL4-driven UAS-sima-RNAi, Wg signals were suppressed to a level equivalent to that in the repo-GAL4 control (0.25 ± 0.05, n = 9 for repo-GAL4/sima-RNAi trh 1 /trh 2 ). Interestingly, sima-RNAi knockdown in glia of the repo-GAL4 control had no effect on the Wg level (0.25 ± 0.02, n = 10 for repo-GAL4/sima-RNAi), suggesting that Sima is induced in the trh 1 /trh 2 mutant to upregulate Wg expression but has no role in basal Wg expression in the wild-type. Taken together, we suggest that glial Sima is required for Wg upregulation at the NMJs of the trh 1 /trh 2 mutant.
As Wg signals are secreted from both glia and presynaptic neurons [34,36], we found that reduction of Wg signals from neurons also suppressed satellite bouton formation in the trh 1 / trh 2 mutant (Fig 5A and 5B, 10.1 ± 1.5%, n = 10 for elav-GAL4 trh 1 /trh 2 ; 6.0 ± 1.0%, n = 8 for elav-GAL4/wg-RNAi trh 1 /trh 2 ). Also, neuronal and glial but not muscle overexpression of Wg induced satellite bouton formation (S4A and S4B Fig). Glial and neuronal knockdown of Wg reduced total Wg levels at the trh 1 /trh 2 NMJ, although the reduction by neuronal knockdown was not significant (S4C and S4D Fig). Thus, neuronal Wg also contributes to satellite bouton formation. However, only glial overexpression of Sima induced higher levels of Wg, but not neuronal or muscle overexpression of Sima (S4E and S4F Fig). Thus, while neuronal expression of Wg contributes to the overall level at NMJs in the trh mutant, Sima-induced Wg expression is likely glial-specific.
Glial processes invade synaptic boutons to match the growth of NMJs [38], which intrigued us to assess whether glia in the trh 1 /trh 2 mutant exhibits morphological change. In a live fillet preparation for imaging NMJs, we found that glial processes labeled by GFP invaded the area of synaptic boutons in the trh 1 /trh 2 mutant, whereas glial processes were relatively restrained from the bouton areas in the control (Fig 5E). Quantification of the glial process overlaying the synaptic bouton area revealed significantly greater area of overlap in the trh 1 /trh 2 mutant relative to control (Fig 5F, 6.2 ± 1.0%, n = 10 for trh 1 /+; and 16.7 ± 3.7%, n = 10 for trh 1 /trh 2 ). This increased extent of glial processes in the synaptic area may facilitate signal transduction from glia to synaptic boutons for structural reorganization. Taken together, these results suggest that Wg plays a prominent role in the trh 1 /trh 2 mutant to transduce the hypoxia signal from glia to modify presynaptic bouton structure.
We further examined whether glial Sima and Wg have any role on modifying bouton morphology in the posterior A6 segment of the trh 1 /trh 2 larvae. With the satellite boutons at a basal level in the A6 segment of trh 1 /trh 2 (Fig 7B), we tested whether overexpression of Sima or Wg could induce satellite boutons in the trh 1 /trh 2 mutant. Overexpression of Sima by repo-GAL4 in trh 1 /trh 2 induced some satellite boutons (S5A and S5B Fig, 6.8 ± 2.3%, n = 10), which showed no significant difference to the trh 1 /trh 2 mutant carrying repo-GAL4 (2.8 ± 1.5%, n = 8). Overexpression of Wg by repo-GAL4 in trh 1 /trh 2 displayed a basal level of satellite boutons (2.6 ± 1.5%, n = 9). Also, glial wg-RNAi knockdown suppressed satellite bouton formation in the A3 segment of trh 1 /trh 2 (Fig 4F), but had no effect on the morphological phenotype in the A6 segment (S5C and S5D Fig). Thus, the analysis of these data suggests that Wg and Sima might have relatively specific roles to induce satellite bouton formation in the anterior A3 segment.

Discussion
Here, we demonstrate that Trh, a member of the NPAS protein family, non-cell autonomously regulates synaptic bouton formation at NMJs through a hypoxic response from glia. We observed small-sized and clustered satellite boutons at the NMJs of the trh mutant larvae or larvae reared at low oxygen levels. The abnormal bouton morphology at the trh NMJs could be suppressed by reducing the level of the hypoxia-inducible factor Sima in glia. We further show that Sima enhanced the Wg signal from glia to cause satellite bouton formation. Although normal synaptic transmission was detected at NMJs located in an anterior segment of larvae bearing satellite boutons, reduced synaptic transmission was found in a posterior segment lacking satellite boutons of the trh mutant, suggesting that glia-induced satellite bouton formation might be a homeostatic response in restoring normal synaptic transmission. Imbalanced synaptic activities at the anterior and posterior NMJs of the trh mutant might contribute to the uncoordinated stride cycles detected in the trh mutant, slowing larval crawl speed (S6 Fig). Thus, we provide a model for studying the glial responses that modulate synaptic bouton reorganization and activities during hypoxia.

Glia play a critical role in satellite bouton formation in the trh mutant
Our results suggest that Trh has a late developmental role in tracheal morphogenesis, in addition to its well-characterized role in early tracheal cell fate specification [23,24]. We observed defective tracheal structure in the trh mutant (S2A Fig), which may result in hypoxic conditions inside the larval body. The increase in terminal branch number (S2A and S2B Fig) may be a response to oxygen supply deficiency [12]. Moreover, the increases in Sima protein levels and ODD-GFP reporter expression indicate reduced internal oxygen levels (Fig 2A and 2B and  S3E and S3F Fig). Finally, the satellite bouton phenotype in the trh mutant was recapitulated by rearing larvae under hypoxia, and it was suppressed by rearing larvae under hyperoxia. Taken together, these observations suggest that cells in the trh mutant sense low oxygen levels caused by the defective tracheal system and respond by elevating Sima protein levels. It is not clear how profound this effect is for other types of larval cells. Based on our ODD-GFP and Sima immunostaining patterns (Fig 2A and 2B and S3E and S3F Fig), many types of cells are likely to be affected [40].
We suggest that glia is the major cell type mediating satellite bouton formation in the trh mutant under hypoxia. While Sima was increased ubiquitously, manipulating the levels of Sima or Ftg, the negative regulator of Sima, in glia modulates satellite bouton formation (Fig 3  and S3A and S3B Fig). Elevated Sima levels induce tracheal sprouting in tracheal cells, as well as protrusions in non-tracheal cells [12]. Interestingly, we also observed protrusion of glial processes into synaptic area in the trh mutant, indicative of a glial response (Fig 5E and 5F). Several types of cells in Drosophila have been shown to respond to hypoxia [32,41]. For instance, under hypoxia, elevated Sima levels induce the expression of Breathless (Btl, the FGF receptor) in tracheal cells that branch out seeking cells that express Branchless (Bnl)/FGF, with this latter process also being partially dependent on Sima [12,13]. In an alternative pathway, atypical soluble guanylyl cyclases can mediate graded and immediate hypoxia responses mainly in neurons [42,43]. Drosophila glia have not been reported to sense and respond to hypoxia, but mammalian astrocytes in the central nervous system have been shown to be involved in these processes. In a mouse model for middle cerebral artery occlusion, astrocyte activation was shown to play a crucial role in ischemic tolerance, which is mediated through P2X7 receptor-activated HIF-1α upregulation [18]. Under physiological hypoxia, reduced mitochondrial respiration leads to the release of intracellular calcium and exocytosis of ATPcontaining vesicles, thereby signaling the brainstem to modulate animal breathing [44]. Our results reveal a role for Drosophila larval glia in sensing hypoxia via the conventional HIF-1α/ Sima pathway. We also demonstrate that under hypoxia, glia modulate the formation of synaptic boutons (Fig 3E and 3F). These results clearly place the glia-modulated morphology of synaptic boutons in the context of hypoxia responses.

Sima elevates Wg signal expression in glia in satellite bouton formation
Our study further establishes that in response to hypoxia, Wg is a glial signal that modulates synaptic bouton formation. Two sources of Wg, presynaptic motor neurons and glia, are involved in synaptic growth and remodeling [34,36]. Our results suggest that Sima upregulates the level of Wg secreted from glia to modulate synapse formation in the trh mutant or in control larvae grown under hypoxia. In hypoxic macrophages, HIF-1α ediates the induction of Wnt11, which is a mammalian homolog of Wg [45]. It is likely wg is a direct target of Sima in Drosophila. We found the HIF1-α binding motif (CGTG) at the -269 nucleotide sequence in the wg promoter. Also, direct binding of Sima to this consensus site was reported in a systematic ChIP-seq experiment [46]. We further show that the level of Wg is controlled by glial Sima in the trh mutant (Fig 4C and 4D) and that Wg mediates Sima-induced satellite bouton formation at trh NMJs (Fig 3B). Also, glial overexpression of Sima upregulates Wg levels at NMJs (S4E and S4F Fig). Wg is also expressed from presynapses [35]. Thus, neuronal wg knockdown partially suppressed satellite bouton phenotypes in the trh mutant (Fig 5A and  5B), while neuronal wg overexpression in wild-type induced the phenotypes (S4A and S4B  Fig). These results are consistent with the idea that presynaptic Wg contributes to the overall pool of Wg at NMJs of the trh mutant. As a secreted morphogen, Wg functions in both presynaptic and post-synaptic sites [35][36][37]. At presynaptic terminals, the canonical Wg pathway induces microtubule loop formation to regulate synaptogenesis. We also detected an increase in microtubule loops in the trh mutant (Fig 5C and 5D), consistent with a role for Wg signaling in modulating synaptic reorganization. Postsynaptic Wg signaling leads to subsynaptic reticulum differentiation [35], which was not apparent in the trh mutant (S1D Fig), suggesting that Wg might be a component of the complex hypoxia response that induces synaptic bouton reorganization. Brief exposure to hypoxia induces immature spines and impaired synaptic function in hippocampal neurons [17]. The morphological change to satellite boutons at the A3 segment of the trh NMJs was not accompanied by altered synaptic transmission (Fig 8), which may be compensated during long-term hypoxia.
The satellite boutons, also named as bunch boutons, have been described in spastin mutants [47,48]. As an AAA ATPase, Spastin severs microtubules to facilitate transport to distal axon segments [49]. Accordingly, the spastin mutant also exhibits a lack of microtubules at terminal boutons [48]. In contrast, the trh mutant presented an increase of the more stabilized microtubule loops (Fig 5C and 5D). Microtubule loops have been linked to synaptic bouton stabilization, and an excess of microtubule loops has been associated with increased synaptic bouton formation [50,51]. The altered morphology of satellite boutons may be part of the structural changes necessary to maintain normal synaptic transmission under hypoxia. The trh and spastin mutants also exhibit differences in synaptic function, with loss of spastin function slightly enhancing spontaneous synaptic transmission release but reducing evoked synaptic transmission [48]. Thus, although the morphology of synaptic boutons at trh NMJs resembles that of spastin mutants, satellite boutons at trh NMJs retain synaptic functions, unlike the impaired synaptic transmission of spastin mutant boutons.

Differential morphological and physiological changes of anterior and posterior segments in the trh NMJs
The size of NMJs in muscles 6/7 decreases from the anterior to posterior segments, which could represent a coupling with muscle growth [52,53], thereby maintaining similar electrophysiological efficacy at anterior and posterior NMJs (Fig 8). Interestingly, our findings show that synaptic responses in the trh mutant differ, with satellite boutons only appearing in anterior segments (Fig 7A and 7B). Furthermore, synaptic transmission at trh NMJs remained normal in the anterior A3 segment but was impaired in the posterior A6 segment (Fig 8). These observations are consistent with the idea that satellite bouton formation is a part of a homeostatic response to restore synaptic activity. Why synapses are not reorganized in the posterior segments remains elusive. We failed to detect an upregulation of Wg in the A6 segment (S4C and S4D Fig), and glial Sima overexpression even in the A6 segment of the trh mutant failed to increase satellite boutons significantly (S5A and S5B Fig). Thus, the upregulation of Wg by Sima may be segment-dependent, which awaits further study. Motor neurons in the ventral nerve cord project much longer axons to muscles in posterior segments compared to anterior ones. It has been shown that axonal transport to posterior segments is more vulnerable to inefficient transport conditions. For example, mutation of long-chain Acyl-CoA synthetase impairs the balance between anterograde and retrograde transport, causing distally-biased axonal aggregations and affecting the growth and functioning of synapses [54]. It is possible that glia-derived Wg signals may not be efficiently transported to posterior segments during hypoxia. This polar difference in synaptic activity and bouton morphology may contribute to the uncoordinated movements of the trh mutant larvae. Alternatively, defective locomotion in posterior segments of the trh mutant is independent of the glial modulation of bouton morphological changes. Larval forward locomotion, propelled by peristaltic contraction, is controlled by different circuits targeting anterior and posterior segments. The GABAergic SEZ-LN1 neurons specifically control posterior A6 and A7 segmental muscle contraction by inhibiting A27h premotor neurons, which promotes longitudinal muscle contraction during larval forward crawling [55]. Specific alteration of the circuit in the posterior segments may lead to the locomotion defect in the trh mutant.

Hypoxia or hyperoxia rearing conditions
Larvae in a food vial were transferred at 1 day after egg laying (AEL) to a ProOx (model 110, BioSpherix, Lacona, NY) oxygen-controlled chamber. Oxygen or nitrogen was infused into the chamber to a desired concentration (5% or 50%), which was maintained until assay.

Image acquisition and processing
NMJs in muscle 6/7 of A3 segments (or A2-A6 in Fig 7 and A6 in S4 and S5 Figs) of wandering third-instar larvae were analyzed. Confocal images were acquired via LSM510 confocal microscopy (Carl Zeiss) using 40x water, 40x water immersion (for live tissue in Fig 5E), or 100x oil objectives. All presented images are projections of confocal z-stacks. Numbers of satellite boutons, total boutons, and microtubule loops were counted manually. The percentage of satellite boutons was calculated as the number of satellite boutons divided by the total boutons (satellite + normal ones) for each NMJ, and the average percentage is calculated from about 6-13 NMJs for each genotype. The immunofluorescence intensities of Wg and HRP were analyzed by ImageJ. HRPpositive regions were chosen to measure mean intensities of Wg and HRP. After subtracting the intensity with the background one, the ratio of Wg levels to HRP levels was presented as the normalized Wg intensity. The overlapping area of GFP and HRP projections was chosen by the "AND" operator in ImageJ, which was divided by HRP area for the percentage. Each dot in the bar graph represents the data from a single NMJ of a larva, and 6-13 NMJs from 2-5 independent experiments were pooled for quantification. Embryos were acquired by means of LSM510 confocal microscopy (Carl Zeiss) using a 20x objective, and were analyzed as previously described [29]. For Fig 2B, each dot in the bar graph represents data from a single embryo in which fluorescence was measured in at least 35 cells.

Electrophysiological recordings
Basal transmission properties were analyzed at NMJs of muscle 6/7 in specified segments of wandering third-instar larvae as previously described [61], with some modifications. The larval body wall was dissected in cold calcium-free HL3 solution and recorded in HL3 solution containing 0.4 mM CaCl 2 at room temperature.

Crawling behavior
Mid third instar larvae (feeding stage) were placed on black agar plates (2% agar with black food coloring in 25 × 20 cm 2 dishes) at room temperature for filming. Video recording by a Sony Xperia Z1 camera started after 1 min habituation and lasted for 5 min, and it was analyzed using Ctrax software [62]. The (x, y) positions were used to calculate the crawling distance between two successive frames, and crawling speed was derived by dividing total distance travelled by time. The change in angle of larvae between two frames was divided by time to represent rotational angles. The forward crawling assay was a modification of a previous study [39]. Larvae were transferred into a tunnel (~1 mm width) made in 2% black agar. Specimens were video-recorded for 3-10 minutes using a Leica S8 APO microscope. Kymographs were constructed using the MultipleKymograph plug-in for ImageJ (NIH). Only forward crawling was counted, and 7-10 steps for each of ten larvae were analyzed for each genotype.