The epithelial Na⁺ channelUNC-8 promotes an endocytic mechanism that recycles presynaptic components to new boutons in remodeling neurons

SUMMARY

Circuit refinement involves the formation of new presynaptic boutons as others are dismantled. Nascent presynaptic sites can incorporate material from recently eliminated synapses, but the recycling mechanisms remain elusive. In early-stage C. elegans larvae, the presynaptic boutons of GABAergic DD neurons are removed and new outputs established at alternative sites. Here, we show that developmentally regulated expression of the epithelial Na⁺ channel (ENaC) UNC-8 in remodeling DD neurons promotes a Ca²⁺ and actin-dependent mechanism, involving activity-dependent bulk endocytosis (ADBE), that recycles presynaptic material for reassembly at nascent DD synapses. ADBE normally functions in highly active neurons to accelerate local recycling of synaptic vesicles. In contrast, we find that an ADBE-like mechanism results in the distal recycling of synaptic material from old to new synapses. Thus, our findings suggest that a native mechanism (ADBE) can be repurposed to dismantle presynaptic terminals for reassembly at new, distant locations.

Graphical Abstract

In brief

Neurons in the brain use specialized connections (synapses) to communicate. As the brain develops and circuits are refined, synapses may be dismantled for reassembly at new locations. Cuentas-Condori et al. identify a conserved mechanism whereby synaptic components are recycled from old to new synapses.

INTRODUCTION

Neural circuits are actively refined during development, forming new synapses while removing others. In the mammalian visual circuit, for example, axons of retinal ganglion cells (RGCs) initially project to the thalamus to innervate multiple targets with widely dispersed presynaptic boutons. In an activity-dependent mechanism, distal boutons are eliminated, while proximal RGC presynaptic domains are enlarged. Notably, the denuded sections of RGC axons are retracted later. Thus, synaptic refinement in this circuit is temporally segregated from axonal pruning and therefore likely depends on a separate mechanism.¹

Presynaptic boutons in the adult nervous system are also subject to remodeling, and this inherent plasticity has been proposed to mediate recovery from injury.^2,3 Live imaging of mammalian neurons revealed that presynaptic varicosities are highly dynamic under basal conditions and responsive to activity.⁴ Thus, a better understanding of the mechanisms that govern presynaptic elimination and formation should advance the goal of manipulating presynaptic structural plasticity to restore circuits damaged by injury or neurodegenerative disease.

Because synaptic remodeling is observed across species⁵ and therefore likely to depend on conserved pathways, we are investigating the mechanism in the model organism C. elegans. In the newly hatched larva (L1), DD-class GABAergic motor neurons innervate ventral muscles. During the transition from the L1 to the L2 larval stage, ventral DD synapses are eliminated, and new DD inputs are established with dorsal muscles^5–8 (Figures 1A and S1A–S1C). The relocation of DD presynaptic boutons proceeds without axon retraction, pointing to an internal cell biological mechanism that drives synaptic reorganization but not axonal pruning.^6,9

Figure 1. UNC-8 localizes to remodeling DD presynaptic boutons.

(A) Re-location of presynaptic boutons (green) from ventral to dorsal DD neurites during early larval development (late L1 and L2 stages). Arrowheads denote the DD commissure.

(B) Split-GFP strategy to tag endogenous UNC-8 in DD neurons. The unc-8 locus was genetically engineered to produce a C-terminal UNC-8::GFP11_×7 fusion protein. An extrachromosomal array drives the complementary GFP1-10 peptide in DD neurons under the flp-13 promoter (Pflp-13).

(C) Ventral UNC-8::GFP puncta were evaluated (left) at developmental time points (arrows): early L1 (16 hours post lay [hpl]) and late L1 (24 hpl) and L2 stages (32 hpl) (right) in remodeling DD axons (see STAR Methods).

(D) Ventral UNC-8::GFP puncta at the L2 stage in the wild type (left) vs. unc-104/Kif1A mutant (right). Scale bar, 2 μm.

(E) UNC-8::GFP density (UNC-8 puncta/10 μm) at each time point shows progressive upregulation in the wild type (left; early L1 [1.28 ± 1.1, n = 26] and late L1 [2.93 ± 1.8, n = 34] and L2 [5.31 ± 1.7, n = 39] stages) but not in an unc-104/Kif1A mutant (right; late L1 [2.66 ± 1.8, n = 25] and L2 [3.31 ± 1.5, n = 20] stages). Data are mean ± SD. Kruskal-Wallis test. **p < 0.01, ****p < 0.0001; NS, not significant. All comparisons are relative to the 24 hpl time point in the wild type.

(F) Accumulation of ventral UNC-8::GFP in DD cell soma of the unc-104/Kif1A mutant. The key denotes relative UNC-8::GFP signal intensity. Scale bar, 2 μm (30 hpl, L2).

(G) Dual-color imaging during remodeling (late L1) detects UNC-8::GFP puncta (green) and neighboring mCherry::RAB-3 (magenta) at ventral presynaptic DD boutons. Scale bar, 2 μm. Inset scale bar, 1 μm.

(H) Cholinergic DA motor neuron presynaptic boutons (magenta) provide excitatory input to DD GABAergic motor neurons (green) in the dorsal nerve cord. Optogenetically activated Chrimson in DA neurons (magenta) evokes Ca⁺⁺ transients in postsynaptic DD neurons (I–N). GCaMP signals were collected from DD presynaptic boutons (ROI).

(I) Chrimson activation in DA neurons (magenta bars) evokes GCaMP transients in wild-type (WT) DD presynaptic boutons. Gray vertical bars depict ~1-s lag before imaging.

(J) Initial GCaMP signal (0.02 ± 0.3, n = 32) is elevated (0.25 ± 0.5, n = 32) after Chrimson activation. Non-parametric paired Wilcoxon test, **p = 0.002.

(K) Examples of upregulated GCaMP signals after Chrimson activation in the WT.

(L) Chrimson activation in DA neurons fails to evoke a GCaMP response in DD neurons of the unc-8 mutant.

(M and N) GCaMP signal (0.06 ± 0.4, n = 32) does not increase after DA activation (0.07 ± 0.5, n = 32). Non-parametric paired Wilcoxon test, p = 0.0803.

(K and N) A dashed line denotes DD cell soma. Scale bar, 4 μm. All animals were grown with ATR.

Importantly, DD presynaptic remodeling is transcriptionally regulated^10–14 and accelerated by synaptic function,^14,15 suggesting that the underlying mechanism depends on the intersection of a genetic program with circuit activity. This idea is consistent with the previous finding that the homeodomain transcription factor IRX-1/Iroquois drives expression of UNC-8, a member of the degenerin (DEG)/epithelial Na⁺ channel (ENaC) family, that promotes disassembly of ventral DD presynaptic domains.^10,11,15 A reconstituted UNC-8 channel preferentially gates Na⁺ in Xenopus oocytes,¹⁶ which suggests that the native UNC-8/DEG/ENaC could depolarize the neurons in which it is expressed. In turn, this effect is predicted to enhance Ca²⁺ import by presynaptic voltage-gated calcium channels (VGCCs), an idea supported by the observation that UNC-2/VGCC is required for UNC-8-dependent presynaptic disassembly in GABA neurons.¹⁵ This model parallels the finding that Drosophila Pickpocket/DEG/ENaC is localized to the presynaptic region in motor neurons to elevate intracellular Ca²⁺ for neurotransmitter release.^17,18 The radically different outcome of synaptic destruction that arises from UNC-8 function in C. elegans depends on the serine/threonine phosphatase Calcineurin (CaN). CaN is activated by intracellular Ca²⁺ and functions genetically upstream of UNC-8. Thus, UNC-8/DEG/ENaC, UNC-2/VGCC, and CaN may constitute a positive feedback loop to amplify Ca²⁺ influx to drive synaptic disassembly (Figure S2E).¹⁵ Because CaN is known to function as a key regulator of activity-dependent bulk endocytosis (ADBE),¹⁹ we investigated the possibility that an ADBE-like mechanism is involved in DD synapse elimination and reconstitution.

Synaptic vesicle (SV) membrane and proteins are recycled at presynaptic domains by parallel-acting pathways.²⁰ In the canonical mechanism, single SVs are retrieved from the presynaptic membrane as clathrin-coated vesicles. In highly active neurons, where the plasma membrane is rapidly expanded by fusion with multiple SVs, recycling is achieved by ADBE, in which large (80- to 100-nm) endosomes recover the SV membrane in a clathrin-independent mechanism.^19,20 ADBE depends on CaN, which induces assembly of the dynamin-syndapin complex.^19,21–23 The initial formation of bulk endosomes has been proposed to depend on branched actin polymerization triggered by syndapin recruitment of N-WASP (Neuronal Wiskott-Aldrich Syndrome protein) to the remodeling membrane. N-WASP is a well-known activator of the branched-actin nucleator Arp2/3.^24,25 Consistent with this finding, syndapin and Arp2/3 subunits are enriched in bulk endosomes derived from highly active neurons.²⁶ Other studies of neurosecretory cells determined that the maturation of bulk endosomes depends on an acto-myosin II ring that separates the invaginating bulk endosomes from the membrane.²⁷

Here we show that UNC-8/DEG/ENaC and TAX-6/CaN localize to the presynaptic regions of remodeling DD neurons, where UNC-8 promotes local Ca²⁺ transients that are strengthened by TAX-6/CaN function, results consistent with the proposal that UNC-8 and TAX-6/CaN act together to elevate presynaptic calcium.¹⁵ We show that canonical ADBE components, including CaN, dynamin, syndapin, and Arp2/3, are required for UNC-8-dependent removal of DD presynaptic components. Importantly, our results provide an explanation for the dramatically different outcomes of ADBE, which maintains synaptic function, versus our proposed bulk endocytic mechanism, which is involved in recycling components to new synapses. As noted above, in ADBE, bulk endosomes give rise to SVs that replenish the local pool. Our studies of remodeling DD neurons have revealed that ventrally located presynaptic components are also recycled, but in this case to distal synapses in the dorsal DD neurite. These findings are important because they suggest that a conserved endocytic mechanism could be widely utilized to couple synapse elimination with synapse formation during circuit refinement.

RESULTS

Kinesin-3-dependent transport of the DEG/ENaC protein UNC-8 to remodeling presynaptic boutons

To monitor UNC-8 during DD remodeling, we used a split-GFP strategy to label the endogenous UNC-8 protein²⁸ (Figure 1B). UNC-8::GFP puncta were sparsely detectable in DD neurons in early L1larvaebut robustly elevated as DD remodeling progressed in late L1 and early L2 larvae (Figures 1D and 1E). This finding is consistent with the idea that unc-8 is transcriptionally upregulated to elevate UNC-8 protein for DD remodeling.¹¹

Next, we evaluated the distribution of UNC-8::GFP puncta in a mutant that disrupts the kinesin-3 protein UNC-104/KIF1A, which normally transports vesicular cargo to the presynaptic domain.²⁹ Strikingly, UNC-8::GFP puncta are largely sequestered in DD cell soma in unc-104 mutants (Figures 1E, 1F, S1D, and S1E). These results demonstrate that trafficking of UNC-8 protein into remodeling DD neurites depends on UNC-104/KIF1A activity.

UNC-8::GFP puncta appeared in perisynaptic regions (i.e., adjacent to mCherry::RAB-3 clusters) in the remodeling DD axon (Figure 1G), as also observed for ppk DEG/ENaC proteins in Drosophila.¹⁷ Our results demonstrate that UNC-8 is developmentally upregulated and localizes to the vicinity of remodeling DD boutons in a kinesin-3-dependent manner. Interestingly, UNC-8::GFP remains in the ventral DD neurite (Figure S1F) after presynaptic boutons are relocated to the dorsal side, a result that could be important for maintaining the stability of dorsal DD synapses in the adult.

UNC-8 maintains Ca²⁺ levels in remodeling DD neurons

A reconstituted UNC-8 channel mediates Na⁺ influx.¹⁶ The resultant depolarization of local membrane potential in DD presynaptic boutons is predicted to enhance VGCC/UNC-2 channel activation and consequently elevate intracellular Ca²⁺ (Figure S2E).¹⁵ To test this idea in vivo, we tracked spontaneous Ca²⁺ transients at remodeling DD synapses (Figure S1G). In wild-type animals, we detected striking Ca²⁺ fluctuations in bouton-like structures (Figures S1H, S1I, and S1K; Video S1).³⁰ Live imaging revealed that, at basal states, GCaMP fluorescence is downregulated in unc-8 mutants (Figures S1J and S1K), suggesting that UNC-8 maintains Ca²⁺ levels in remodeling DD neurons (Video S1).

To determine whether UNC-8can also mediate evoked synaptic responses, we activated presynaptic cholinergic DA motor neurons with the light-sensitive opsin Chrimson³¹ (Figure 2H). This experiment detected consistent upregulation of Ca²⁺ at remodeling DD boutons with DA motor neuron activation in wild-type animals (Figures 2I–2K) in an all-trans-retinal (ATR)-dependent manner (Figure S1L). In unc-8 mutants, Ca²⁺ levels in DD neurons were not elevated with DA activation (Figures 2L, 2N, and S1M; Video S2). Notably, neither expression nor localization of endogenously tagged GFP::UNC-2 is detectably altered in remodeling DD neurites in unc-8 mutants vs. control (Figures S1N–S1P). Our results parallel previous findings in which evoked Ca²⁺ responses are reduced in presynaptic regions by the ENaC antagonistbenzamil, presumptively because of inhibition of a presynaptic DEG/ENaC at the Drosophila neuromuscular junction.¹⁸ Thus, our findings support the proposal that presynaptic UNC-8 mediates elevated Ca²⁺ levels in remodeling boutons and substantiates a role of synaptic activity, through Ca²⁺ signaling, in the disassembly of DD presynaptic domains (Figure S2E).¹⁵

Figure 2. Presynaptic TAX-6/CaN elevates Ca⁺⁺ transients in remodeling DD boutons.

(A) Split-GFP strategy for labeling endogenous TAX-6/CaN in DD neurons. Three copies of GFP11 were inserted at the tax-6 locus to produce TAX-6:: GFP11_×3. Pflp-13 drives GFP1-10 expression from an extrachromosomal array in DD neurons.

(B) Left: Before DD remodeling, in L1 larvae, TAX-6::GFP is enriched (arrowheads) in the ventral DD neurite but relocated to the dorsal cord (right) after remodeling, at the L4 stage. Scale bar, 2 μm.

(C) Quantification of TAX-6::GFP detects initial enrichment in the ventral cord (V = 0.79 ± 0.1 vs. D = 0.2 ± 0.1, n = 29) in L1 larvae and later enrichment on the dorsal side at the L4 stage (V = 0.27 ± 0.1, n = 21 vs. D = 0.73 ± 0.1, n = 21). Data are mean ± SD. One-way ANOVA with Tukey’s multiple comparisons test, ****p < 0.0001. (D and E) Top: relocation of TAX-6 (green) from ventral (L1 stage) to dorsal (L4 stage) DD neurites. (Bottom: co-localization (arrowheads) of endogenous TAX-6::GFP (green) and active-zone protein mRuby::CLA-1s (magenta) (D) on the ventral side before remodeling (L1 stage) and (E) on the dorsal side (L4 stage) after remodeling. Line scans show co-localization of TAX-6 (green) and CLA-1s (magenta) from insets. Scale bar, 2 μm.

(F) Chrimson was expressed in presynaptic DA neurons (magenta) to evoke Ca⁺⁺ transients in DD neurons (G–I).

(G–I) Chrimson activation in DA neurons (G, magenta bar) elevates GCaMP fluorescence in remodeling DD presynaptic boutons (ROI) (H) (before activation = −0.02 ± 0.2, n = 44; after activation = 0.14 ± 0.4, n = 44). Non-parametric paired Wilcoxon test, ****p < 0.0001. Gray vertical bars depict ~1-s lag before imaging. GCaMP fluorescence is elevated after Chrimson activation in the WT (0.32 ± 0.3, n = 32) in comparison with tax-6 mutants (0.18 ± 0.2, n = 44) (I). Data are mean ± SD. Mann-Whitney test, *p = 0.0109. All experiments were performed with ATR.

CaN upregulates Ca²⁺ in remodeling DD boutons

Because we have shown previously that the conserved phosphatase CaN functions with unc-8 in a common genetic pathway to promote presynaptic disassembly in remodeling GABAergic motor neurons,¹⁵ we used a split-GFP strategy²⁸ to determine whether TAX-6, the catalytic subunit of CaN in C. elegans, is presynaptically localized (Figure 2A). Before the remodeling period (L1 larval stage), endogenous TAX-6::GFP is enriched in the DD ventral neurite. After remodeling, at the L4 stage, TAX-6::GFP is brightest on the dorsal side (Figures 2B and 2C) with a diffuse signal and distinct TAX-6::GFP puncta (Figures 2D and 2E) that co-localized with mRuby::CLA-1, an active-zone protein³² (Figures 2D and 2E). Thus, our results suggest that TAX-6 is localized to presynaptic regions in remodeling DD neurons.

Incontrastto unc-8 mutants, in which Ca²⁺ transientsare largely abrogated (Figures 1J and 1K), intrinsic Ca²⁺ levels in tax-6 mutants are comparable with that of the wild type (Figures S2A and S2B). However, the intensity of the evoked Ca²⁺ response in tax-6 mutant animals is significantly reduced in comparison with the wild type (Figures 2F–2I, S2C, and S2D). These findings are consistentwiththeideathatTAX-6strengthenstheCa²⁺response, perhaps by tuning unc-8 activity (Figure S2E).¹⁵

CaN functions downstream of UNC-8 to promote presynaptic disassembly

We proposed that TAX-6/CaN functions in a positive feedback loop with ENaC/UNC-8 and VGCC/UNC-2 to elevate intracellular Ca²⁺ (Figure S2E).¹⁵ This model predicts that a loss-of-function mutation in tax-6/CaN should impair ENaC/UNC-8-dependent presynaptic disassembly. To test this idea, we exploited ventral D (VD) GABAergic neurons, which normally do not undergo synaptic remodeling. Forced expression of ENaC/UNC-8 in VD neurons, however, is sufficient to drive the removal of the SV-associated protein synaptobrevin/SNB-1::GFP from VD synapses (Figure 3B).^11,15 We used this experimental paradigm to show that RNAi knockdown of tax-6/CaN prevents UNC-8-dependent elimination of VD synapses (Figure 3C), a result consistent with the hypothesis that TAX-6/CaN functions with UNC-8 to dismantle the GABAergic presynaptic apparatus (Figure S2E). Although this RNAi experiment does not exclude the possibility that TAX-6/CaN is required in a different cell type, our results favor the parsimonious model of a cell-autonomous role in DD neurons for presynaptic remodeling.

Figure 3. Cell-autonomous dynamin activity is required for presynaptic disassembly in remodeling DD neurons.

(A) During ADBE, synaptic activity elevates intracellular Ca⁺⁺ to activate CaN/TAX-6. Dephosphorylation of dynamin by CaN promotes the formation of the dynamin-syndapin complex for membrane localization to drive branched-actin polymerization for bulk endosome formation and local SV recycling. (B) Left: UNC-8 overexpression (OE) in VD neurons induces presynaptic disassembly. Shown are control VD neurons (gray, VD11) with SNB-1::GFP puncta (arrowheads) vs. anterior region of UNC-8(OE) VD neurons (magenta, VD10) with fewer SNB-1::GFP puncta (dashed line). Right: paired analysis of neighboring VD neurons: control VDs (196 ± 139 a.u., n = 12) and UNC-8(OE) VD neurons (56.7 ± 67 a.u., n = 12). Wilcoxon matched-pairs signed-rank test, **p = 0.0024. (C) TAX-6 is required for UNC-8-dependent removal of presynaptic domains. Left: tax-6 RNAi-treated UNC-8(OE) VD neuron (magenta, VD5) vs. adjacent control VD neuron (gray, VD6), both with SNB-1::GFP puncta (arrowheads). Right: paired analysis of neighboring VD neurons treated with tax-6 RNAi: control VDs (126 ± 76 a.u., n = 17) and UNC-8(OE) VD neurons (86 ± 99 a.u., n = 17). Wilcoxon matched-pairs signed-rank test, p = 0.16. Scale bars, 10 μm. L4 stage.

(D) Transgenic strategy for cell-specific RNAi (csRNAi) knockdown of dyn-1/dynamin.

(E) Top left: control DD neuron (gray) shows perinuclear SNB-1::GFP (arrows). Bottom left: dyn-1(csRNAi)-expressing DD neurons (magenta) labeled with cytosolic mCherry, nuclear GFP (green arrowhead), and perinuclear SNB-1::GFP (arrows) retain ventral SNB-1::GFP puncta (white arrowheads). Right: dyn-1(csRNA)-treated DD neurons retain a greater fraction of ventral SNB-1::GFP fluorescence (0.31 ± 0.2, n = 18) than controls (0.15 ± 0.1, n = 18). Data are mean ± SD. Unpaired t test, ***p = 0.0006. Scale bar, 10 μm.

(F) Schematic of DYN-1 protein domains (see text). A magenta bar denotes the putative C-terminal dynamin phospho-box sequence³³ with phosphorylatable residues (S53, Y55, S59) in the WT and converted to alanine (A) in phospho-resistant dyn-1.

(G) Top: transgenic strategy for OE of dynamin (DYN-1) and phospho-resistant DYN-1 in DD neurons (STAR Methods). Center: a control DD neuron (gray) shows robust SNB-1::GFP puncta (white arrowheads). See inset for cell soma and line tracing. Bottom: DD neuron that overexpresses phospho-resistant DYN-1 (dyn-1AAA (OE)) is labeled with nucleus-localized TagRFP (magenta) (inset) and shows fewer SNB-1::GFP puncta than control DD neurons (white arrowheads). Scale bar, 10 μm; late L1 stage.

(H) Top: SNB-1::GFP in DD neurons was imaged during remodeling (arrow) to assess whether ventral (bottom left) OE of phospho-resistant DYN-1 (dyn-1AAA(OE)) accelerates SNB-1::GFP removal (0.61 ± 0.3, n = 21) vs. control DD neurons (0.92 ± 0.5, n = 62). Bottom right: OE of WT dynamin (dyn-1(OE)) (1.47 ± 0.7, n = 21) does not result in fewer SNB-1::GFP puncta vs. control DD neurons (1.21 ± 0.6, n = 34). Data are mean ± SD. Unpaired t test. **p = 0.0055, NS (not significant), p = 0.137. See inset for cell soma and line tracing.

Dephosphins, targets of CaN phosphatase activity, promote the removal of presynaptic components in remodeling DD neurons

In addition to amplifying Ca²⁺ import, CaN/TAX-6 could also regulate additional downstream effectors for presynaptic disassembly. One of the known roles of CaN at presynaptic termini is to trigger ADBE (Figure 3A). In this mechanism, elevated calcium activates CaN to function as a Ca²⁺-dependent phosphatase for dephosphorylation of a group of synaptic proteins known as dephosphins.^19,21 Importantly, the endocytic mechanism in ADBE does not require clathrin,¹⁹ which is necessary for SV endocytosis during mild stimulation.²⁰ The canonical role of ADBE is to recycle large amounts of membrane during periods of intense activity for the local replenishment of SVs for neurotransmission.¹⁹ We used either RNAi or a genetic mutant to determine that the conserved dephosphins amphiphysin/amph-1, epsin/epn-1, Eps15/ehs-1 and synaptojanin/unc-26 normally promote removal of the presynaptic domain in remodeling GABAergic motor neurons (Figures S3A–S3E). Notably, the CaN target that drives clathrin-mediated endocytosis, AP180/unc-11, and the clathrin light chain/clic-1 and the clathrin adaptor AP2/dpy-23 are dispensable for the removal of remodeling GABAergic synapses (Figure S3D). Thus, our findings support the hypothesis that conserved CaN targets (e.g., dephosphins) are required for a clathrin-independent endocytic mechanism that participates in the disassembly of the presynaptic apparatus in remodeling neurons.

Phospho-resistant dynamin functions in DD neurons to promote presynaptic disassembly

The GTPase dynamin, is a known dephosphin target of CaN and functions as a key effector of ADBE^22,34 and the canonical clathrin-dependent mechanism for recycling the SV membrane at active synapses.²⁰ We devised a cell-specific RNAi (csRNAi) strategy for selective knockdown of dyn-1 in DD neurons (Figures 3D and 3E). Removal of ventral SNB-1::GFP was impaired in dyn-1(csRNAi)-treated DD neurons, thus suggesting that dynamin normally functions in DD neurons to promote presynaptic disassembly (Figures 3D and 3E).

Because CaN-dependent activation of dynamin involves the dephosphorylation of residues in a proline-rich C-terminal “phospho-box” domain, we searched the phospho-proteome of C. elegans to identify three phosphorylated DYN-1 sites (S53, Y55, S59) in a proline-rich C-terminal domain (Figure 3F).³³ To determine whether the dephosphorylation state of DYN-1 modulates DD remodeling, we forced expression of a phospho-resistant mutant (S53A, Y55A, S59A; DYN-1AAA) in DD neurons, which, our model predicts, should accelerate the rate of presynaptic disassembly (Figure 3A). This experiment showed that ectopic expression of DYN-1AAA results in precocious removal of presynaptic SNB-1::GFP (Figures 3G and 3H). In contrast, overexpression of wild-type DYN-1 (DYN-1[OE]) does not enhance the elimination of SNB-1::GFP from DD synapses (Figure 3H). These results confirm that dynamin can act cell autonomously in DD neurons and that its dephosphorylation accelerates removal of DD synapses. Thus, our findings are consistent with the idea that CaN dephosphorylates dynamin and additional dephosphins to drive presynaptic disassembly in remodeling DD neurons. Because ADBE depends on CaN-dependent activation of dynamin, our results also substantiate the hypothesis that a bulk endocytic mechanism is active during DD synaptic remodeling. In contrast to the canonical function of ADBE of local recycling to sustain neurotransmission at nearby synapses (Figure 3A), our results suggest that a bulk endocytic mechanism functions in an overall mechanism that captures presynaptic components for reassembly at new locations.

Syndapin functions in a common pathway with UNC-8 to remove SV components in remodeling DD neurons

During ADBE, dynamin and the F-Bar (Fes/CIP4 homology-Bin/Amphiphysin/Rvs) protein syndapin function together to promote polymerization of branched-actin networks for endocytosis (Figure 4A).^22–24 RNAi knockdown of sdpn-1/syndapin impairs removal of SNB-1::GFP from ventral GABAergic synaptic terminals (Figure S3D). Similarly, the elimination of endogenous GFP::RAB-3 from ventral neurites of DD neurons is disrupted in sdpn-1 mutants (Figure 4B). Ectopic retention of ventral GFP::RAB-3 puncta in DD neurons of unc-8; sdpn-1 double mutants is not significantly different from that of either the unc-8 or sdpn-1 single mutants (Figure 4B). This finding argues that UNC-8 and SDPN-1 act in a common genetic pathway to remove SVs in remodeling DD neurons. In addition, RNAi knockdown of sdpn-1 (Figures S4G and S4H) antagonizes the UNC-8-dependent removal of SNB-1::GFP from ventral VD synapses, thus supporting the hypothesis that SDPN-1 acts downstream of UNC-8. Consistent with a presynaptic role of SDPN-1 for DD synaptic disassembly, SDPN-1::mCherry is closely associated with SNB-1::GFP during the remodeling window (Figure S4A). Although we cannot rule out a model in which SDPN-1 function is required in a different cell type, our combined results point to a cell-autonomous role in DD neurons.

Figure 4. Branched-actin polymerization functions downstream of UNC-8 to remodel DD boutons.

(A) F-BAR proteins associated with the plasma membrane, like syndapin, recruit the wave regulatory complex (WRC) via SH3 domains to activate Arp2/3 for branched-actin polymerization. The WRC components Sra1/GEX-2, Nap1/GEX-3, Abi2/ABI-1, WAVE1/WVE-1 and HSPC300/Y57G11C.1147 are conserved in C. elegans. The VCA domain (V, verprolin homology domain; C, central domain; A, acidic domain) of WAVE1 interacts with Arp2/3 to promote its activation.

(B) SDPN-1 and UNC-8 function in a common genetic pathway. Left: residual endogenous GFP::RAB-3 puncta (arrowheads) in ventral DD neurites in the WT, unc-8 and sdpn-1 single mutants, and unc-8; sdpn-1 double mutants in early L4 larvae. Right: ventral GFP::RAB-3 density is not significantly different among unc-8 (1.02 ± 0.5, n = 51), sdpn-1 (1.22 ± 0.9, n = 46) and unc-8; sdpn-1 double mutants (1.13 ± 0.9, n = 62) but is elevated in comparison with the WT (0.69 ± 0.6, n = 73). Data are mean ± SD. Kruskal-Wallis test. ** p < 0.01. * p = 0.019.

(C) Top: OE of the WRC-VCA domain (VCA(OE)) and cytosolic mCherry (magenta) in remodeling DD neurons labeled with SNB-1::GFP (green). Bottom left: ventral SNB-1::GFP puncta (arrowheads) in VCA(OE) and control DD neurons. Scale bar, 5 μm. Bottom right: proportion of ventral SNB-1::GFP fluorescence in control DDs (0.56 ± 0.1, n = 26) vs. VCA(OE) (0.40 ± 0.1, n = 42) during remodeling. Data are mean ± SD. Unpaired t test, ****p < 0.0001.

(D) Top: arx-5 sense and antisense transcripts co-expressed with either cytosolic mCherry or nucleus-localized GFP. Bottom left: arx-5(csRNAi) knockdown DDs with nuclear GFP (green arrowhead) and cytosolic mCherry (magenta) vs. unlabeled control DDs (gray). Bottom center: ventral SNB-1::GFP (white arrowheads) in control and arx-5(csRNAi) DD cells. Bottom right: proportion of ventral SNB-1::GFP fluorescence in control DDs (0.19 ± 0.1, n = 24) vs. arx-5(csRNAi) (0.43 ± 0.3, n = 24) DDs of L4 larvae. Data are mean ± SD. Unpaired t test, ****p < 0.0001. Scale bar, 10 μm.

(E) Ventral DD neurites imaged during remodeling (gray arrow) to monitor actin dynamics with LifeAct::mCherry (magenta) and SVs with GFP::RAB-3 (green).

(F) GFP::RAB-3 (green) associated with LifeAct::mCherry (magenta). Shown are GFP::RAB-3 clusters (arrowheads). Scale bar, 5 μm.

(G–K) Elevated actin dynamics during DD remodeling in controls but not in unc-8 mutants.

(G) During remodeling, LifeAct::mCherry associates with stable and transient GFP::RAB-3 puncta in control animals. Arrows and arrowheads point to locations of dynamic GFP::RAB-3 puncta. (G′) Line scans from kymographs show dynamic LifeAct::mCherry fluorescence during remodeling; n = 6 boutons from the kymograph in (G).

(G′′) Line scans of LifeAct::mCherry and GFP::RAB-3 from kymographs (arrowheads in G) showing reduced GFP::RAB-3 signal with transient elevation of LifeAct::mCherry (asterisk).

(H and I) LifeAct::mCherry fluorescence from GFP::RAB-3 regions (normalized to t₀ = 0) in (H) WT (n = 12 videos) and (I) unc-8 mutants (n = 11 videos).

(J) Fewer transient GFP::RAB-3 events in unc-8 mutants (1.2 ± 0.8, n = 11 videos) vs. WT (3.2 ± 2.4, n = 12 videos). Data are mean ± SD. t test, *p < 0.05.

(K) unc-8 mutants retain more GFP::RAB-3 puncta (2.37 ± 0.8, n = 19 snapshots) than the WT (1.74 ± 0.7, n = 20 snapshots) during the remodeling window (L2 stage). Data are mean ± SD. t test, *p < 0.05.

Branched actin polymerization promotes DD synaptic remodeling

Because the F-BAR protein SDPN-1/syndapin promotes branched actin polymerization,²⁴ we conducted a candidate screen of known actin-related effectors for roles in synaptic remodeling (Figures S5C–S5F). RNAi knockdown of components of the Arp2/3 complex and the wave regulatory complex (WRC) impairs the removal of SNB-1::GFP from ventral synapses in remodeling GABAergic neurons (Figures S4B–S4E).¹⁰ These findings are consistent with the hypothesis that SDPN-1/syndapin promotes synapse elimination by recruiting downstream effectors of actin polymerization, the WRC and Arp2/3 complex, to remodel GABAergic synapses (Figure 4A). Intriguingly, an additional Arp2/3 activator, N-WASP, is dispensable for this process (Figure S4E).

Additional evidence suggests that the WRC acts downstream of UNC-8; RNAi knockdown of the WRC component wve-1 impairs the removal of SNB-1::GFP from the ventral synapses of VD neurons that overexpress UNC-8 (Figures S4I and S4J). For endocytosis, F-BAR proteins interact with the WRC, which, in turn, deploys its VCA domain to activate the Arp2/3 complex^35,36 (Figure 4A). Remodeling DD neurons with forced VCA expression or VCA(OE) showed fewer ventral SNB-1::GFP puncta than controls (Figure 4C).³⁵ As an additional test of this model, we used csRNAi for DD-specific knockdown of ARX-5/p21, a key component of the Arp2/3 complex³⁷ (Figure 4D). We observed significant retention of SNB-1::GFP puncta at the presynaptic terminals of arx-5(csRNAi)-treated DD neurons, indicating the failure of SNB-1::GFP elimination in arx-5-depleted cells (Figure 4D). Finally, GFP-labeled ARX-5/p21 is closely associated with remodeling ventral DD presynaptic domains (Figure S4F). Thus, our results point to a cell-autonomous role of Arp2/3 and branched actin polymerization in the removal of presynaptic components of remodeling DD neurons.

Actin polymerization and transient RAB-3 particles are elevated during DD synaptic remodeling and depend on UNC-8

Live-cell imaging detected GFP::RAB-3 puncta surrounded by LifeAct::mCherry-labeled actin³⁸ in remodeling DD neurons (Figures 4E and 4F). We observed robust elevation of LifeAct:: mCherry fluorescence intensity and dynamics at presynaptic regions during the remodeling window (Figures 4G–4K and S5A–S5D; Video S3). Instances of high actin dynamics in axonal regions occurred with mobile GFP::RAB-3 particles that also appear to be short-lived (Figure 4G)(Video S3). Transient GFP::RAB-3 puncta are more common during the remodeling window than before DD remodeling ensues (Figure S5E) and are correlated in some instances with temporal elevation of LifeAct::mCherry (Figure 4G⁰⁰; Video S4). Notably, increased actin dynamics (Figure S5D) and transient GFP::RAB-3 puncta (Figure S5E) occur in concert with an overall reduction in the density of RAB-3::GFP puncta (Figure S5F). We detected a striking reduction in LifeAct::mCherry dynamics and mobile GFP::RAB-3 particles during the DD remodeling period in unc-8 mutants (Figures 4H–4J and S5G; Video S3). These defects are correlated with increased density of residual ventral GFP::RAB-3 puncta in unc-8 mutants (Figure 4K). Together, our findings support the hypothesis that UNC-8 normally promotes actin polymerization to accelerate removal of GFP::RAB-3 from remodeling DD synapses.

DEG/ENaC-dependent intracellular vesicles populate remodeling DD neurons

We have shown that UNC-8/DEG/ENaC elevates Ca²⁺ at the synapse (Figure 1), and that key elements of the ADBE mechanism, including branched actin polymerization, function downstream of UNC-8 (Figures 3 and 4).¹⁵ To test the hypothesis that UNC-8 promotes the formation of bulk endosomes, we used serial section transmission electron microscopy (TEM) to generate 3D reconstructions of the ventral neurites of remodeling DD neurons in the wild type and in an unc-8 mutant (Figure 5A; Video S5). In addition to SVs, dense core vesicles, and dense projections (e.g., presynaptic densities), we noted large, membrane-bound structures that resemble bulk endosomes (Figures 5A–5E and S6A–S6E).^19,39–41 These endosome-like structures were visible in the DD neurites of more than 30% of electron microscopy (EM) sections in the control but in only ~15% of EM sections in the unc-8 mutant (Figure 5E). Our determination that remodeling DD dendrites contain abundant, large, endosome-like structures and that this number is reduced in unc-8 mutants is consistent with the hypothesis that UNC-8 promotes the formation of bulk endosomes in remodeling DD neurons.

Figure 5. UNC-8 promotes the formation of endosome-like structures in remodeling DD neurons.

(A) L1 larvae were subjected to high-pressure freezing (HPF) during DD remodeling (27 hpl), and serial sections were imaged by transmission electron microscopy (TEM).

(B and C) Reconstructions of (B) WT and (C) unc-8 mutant DD2 neurons with enlargements shown below. Neuronal membrane, transparent white; SVs, blue; dense core vesicles, green; dense projections, gray; endocyte-like structures, magenta.

(D) Representative TEM sections containing presynaptic densities (stars) and endosome-like structures (arrowheads) from reconstructed WT and unc-8 mutant DD2 neurons.

(E) Percentage of TEM sections containing endosome-like structures in reconstructed DD2 neurons. Numbers denote fractions of sections with endocyte-like structures for the WT (133/396) and unc-8 (55/349) mutant. Fisher’s exact test, ****p < 0.00001.

UNC-8 promotes recycling of RAB-3 to nascent, dorsal synapses in remodeling DD neurons

Our finding of abundant endosome-like structures in remodeling DD axons and their dependence on UNC-8 suggested that SV components could be recycled by an ADBE-like mechanism for reassembly; in this case, at nascent presynaptic domains in dorsal DD neurites. For a direct test of this idea, we photoconverted endogenous Dendra2-tagged⁴² RAB-3 in ventral DD neurites prior to the remodeling period (Figures 6A–6C). Later, by the L3 stage, photoconverted Dendra2::RAB-3 had been efficiently removed from the ventral side, with ~65% recovered at dorsal DD neurites (Figures S7A–S7E). These findings confirm an earlier report that Dendra::RAB-3 overexpressed from a transgenic array is also recycled from ventral to dorsal synapses in remodeling DD neurons.⁴³

Figure 6. UNC-8 promotes RAB-3 recycling from ventral to dorsal DD boutons.

(A) Endogenous labeling of RAB-3 with photoconvertible protein Dendra2 in DD neurons. Pflp-13 drives flippase in DD neurons to fuse Dendra2 to the endogenous RAB-3 protein.⁴⁴

(B) UV irradiation of ventral DD neurites (box) produces photoconverted Dendra2::RAB-3 (magenta).

(C) Photoconversion before DD remodeling and imaging after remodeling at the L3 stage.

(D) Photoconverted Dendra2::RAB-3 in the dorsal nerve cord of WT and unc-8 animals in L3 larvae after DD remodeling. An asterisk denotes autofluorescence. Scale bar, 10 μm.

(E) unc-8 (32.3% ± 25.6%, n = 14) mutants recycle less photoconverted Dendra2::RAB-3 to the dorsal nerve cord than the WT (82.0% ± 40.8%, n = 12). Data are mean ± SD. Ordinary one-way ANOVA with Dunnett’s multiple comparison test. ***p = 0.0009.

(F) GFP::RAB-3 puncta in dorsal DD neurites in the WT and unc-8 mutants at the L4 stage.

(G) Quantification of dorsal GFP::RAB-3 density in the WT (3.14± 0.36, n = 21) and unc-8 mutants (2.91± 0.34, n = 24). Data are mean± SD. Unpaired t test, *p = 0.046.

(H) Photoconverted Dendra2::RAB-3 in the ventral nerve cords of the WT and unc-8 mutants at the L3 stage. Scale bar, 10 μm.

(I) WT (1.14% ± 2.1%, n = 12) and unc-8 mutants (1.32% ± 2.9%, n = 14) lack photoconverted signal on the ventral cord at the L3 stage. Data are mean ± SD. Mann-Whitney test, p = 0.46.

unc-8 mutants, however, show substantially reduced levels of recycled (i.e., photoconverted) Dendra2::RAB-3 at dorsal synapses (Figures 6D and 6E). In an independent experiment, we measured the accumulation of endogenous GFP::RAB-3 puncta at dorsal synapses and observed a significant but weaker effect in unc-8 mutants (Figures 6F and 6G). This disparity is a likely a consequence of our finding that dorsal DD synapses are normally comprised of ~44% nascent (green) Dendra2::RAB-3 (i.e., synthesized after photoconversion) and ~56% recycled (red) Dendra2::RAB-3 (STAR Methods; Table S2; Figure S7F). In this model, the unc-8 mutant impairs RAB-3 recycling but does not retard the incorporation of nascent RAB-3 at dorsal synapses.¹⁵ Interestingly, by the L3 stage, photoconverted Dendra2::RAB-3 is efficiently removed from ventral DD synapses in unc-8 mutants (Figures 6H and 6I). In contrast, at an earlier time point during the remodeling window (L2 stage), residual ventral GFP::RAB-3 particles are elevated in unc-8 mutants in comparison with the wild type (Figure 4K). These results suggest that RAB-3 removal is delayed but not completely blocked in unc-8 mutants, perhaps because of the activity of an additional parallel-acting pathway that we have reported previously.¹¹ Taken together, our results suggest that UNC-8 promotes an endocytic mechanism that recycles presynaptic components for reassembly at new DD synapses in the dorsal nerve cord.

RAB-11, a key recycling endosome component, functions downstream of UNC-8 to mediate synapse elimination and dorsal assembly

The conserved GTPase Rab11 is a key recycling endosome component and is also required for ADBE.^26,45 To test Rab11 for a role in remodeling, we used csRNAi to knock down rab-11 in DD neurons labeled with Dendra2::RAB-3 (Figure 7A). In control DD neurons, photoconverted Dendra2::RAB-3 was efficiently removed from the ventral side, and ~70% was recovered at nascent dorsal DD synapses (Figures 7B–7D). In contrast, rab-11(csRNAi)-treated DD neurons retained ventral photoconverted Dendra2::RAB-3 signal and showed lower levels (~40%) of recycled Dendra2::RAB-3 at dorsal synapses (Figures 7B–7D). For an additional test, we showed that csRNAi of RAB-11 in DD neurons impairs the removal and dorsal assembly of ventral SNB-1::GFP in remodeling DD neurons (Figures S7G and S7H). Together, these findings suggest that SV components, initially associated with ventral DD synapses, are recycled to nascent dorsal synapses in remodeling DD neurons in a mechanism that depends on the canonical recycling endosome component RAB-11.

Figure 7. RAB-11 promotes recycling of RAB-3 from old to new presynaptic boutons in remodeling DD neurons.

(A) csRNAi of rab-11 in DD neurons by co-expression of rab-11 sense transcripts with nucleus-localized TagRFP (magenta) and rab-11 antisense transcript with nucleus-localized GFP (green). Non-fluorescent DD neurons (gray) served as controls.

(B) Left: photoconverted Dendra2::RAB-3 (magenta) in control and rab-11(csRNAi) DD neurons in dorsal and ventral DD neurites. Arrowheads denote ventral retention of photoconverted Dendra2::RAB-3 in rab-11(csRNAi) neurons, and asterisks labels nucleus-localized TagRFP. Right: line scans of photoconverted Dendra2::RAB-3 dorsal (top) and ventral (bottom) DD neurites of control and rab-11(csRNAi) knockdown cells. Scale bar, 10 μm.

(C) rab-11(csRNAi) DD cells recycle less photoconverted signal to the dorsal cord (37.7% ± 24.8%, n = 10) than controls (74.0% ± 36.2%, n = 13). Horizontal lines show median and 25% to 75% distribution. Unpaired t test, **p = 0.0064.

(D) rab-11(csRNAi) DD cells retain more photoconverted signal on the ventral side (9.96% ± 10.5%, n = 10) than controls (1.09% ± 1.8%, n = 13). Horizontal lines show median and 25% to 75% distribution. Mann-Whitney test, *p = 0.0269.

(E) Left: control VD neurons (gray) with SNB-1::GFP puncta (arrowheads) and UNC-8(OE) VD cells (magenta) with fewer SNB-1::GFP puncta (dashed line). Right: paired analysis of ventral SNB-1::GFP fluorescence in neighboring cells, control (607 ± 688 a.u., n = 13) vs. UNC-8(OE) VD neurons (94.8 ± 129 a.u., n = 13). Wilcoxon matched-pairs signed-rank test, ***p = 0.0002. Scale bar, 10 μm.

(F) Left: SNB-1::GFP puncta (arrowheads) in control (gray, VD5) and UNC-8(OE) VD neurons (magenta, VD4) with rab-11 RNAi. Scale bar, 10 μm. Right: paired analysis of ventral SNB-1::GFP in neighboring cells, control VDs (439 ± 341 a.u., n = 23) and UNC-8(OE) VD neurons (360 ± 500 a.u., n = 23). Wilcoxon matched-pairs signed-rank test, p = 0.211.

Finally, we determined that RNAi knockdown of rab-11 was sufficient to prevent UNC-8(OE)-dependent elimination of ventral synapses, a result consistent with the hypothesis that RAB-11 normally functions downstream of UNC-8/DEG/ENaC (Figures 7E and 7F). Live-cell imaging during the remodeling window detected dynamic RAB-11::TagRFP particles that traffic along ventral and dorsal DD neurites and the commissures that connect them (Figures S7I and S7J; Video S6). Interestingly, we observed RAB-11::TagRFP particles closely apposed to stable GFP::RAB-3 puncta with striking examples of dynamic association (Figures S7J and S7K). These findings support the notion that RAB-11 functions downstream of UNC-8 in an ADBE-like mechanism to promote the relocation of synaptic proteins from ventral terminals to nascent, dorsal synapses.

DISCUSSION

We exploited the ready accessibility of C. elegans for live-cell imaging and genetic analysis to investigate developmentally regulated synaptic remodeling in the DD class of GABAergic neurons.^5,8 We show that presynaptic domains are dismantled and SV components recycled to new locations in a mechanism that involves genetic and activity-dependent pathways. Our results demonstrate that the DEG/ENaC UNC-8 is developmentally expressed to promote presynaptic disassembly and recycling in a process involving the Ca²⁺-dependent phosphatase CaN and its “dephosphin” targets, including the GTPase dynamin. Downstream effectors, the F-BAR protein syndapin, the WRC, and Arp2/3 trigger localized actin polymerization for DD synaptic remodeling. We determined that RAB-11, a key component of recycling endosomes, is required for the translocation of the SV components to new presynaptic domains in a pathway that functions downstream of UNC-8. Moreover, we used EM reconstruction to show that UNC-8 promotes the appearance of abundant, large intracellular membranous structures resembling endosomes in remodeling DD neurites.^19,39–41The mechanism we described is strikingly similar to that of ADBE, which functions during periods of high activity to recycle SV s for reuse at adjacent synapses.¹⁹ Our results suggest that the canonical ADBE mechanism can be repurposed during development for the radically different outcome of recycling the presynaptic apparatus for assembly at new locations.

The DEG/ENaC UNC-8 promotes a Ca²⁺-dependent mechanism of presynaptic remodeling

We have tested key predictions of our hypothesis that UNC-8 is expressed in DD neurons to promote synaptic remodeling and that the downstream mechanism depends on the elevation of intracellular Ca²⁺ by a positive feedback loop involving UNC-8/DEG/ENaC, UNC-2/VGCC, and TAX-6/CaN.^10,11,15 First, we showed that UNC-8 is expressed at the onset of DD remodeling and thus could function temporally to facilitate synapse elimination. Second, UNC-8 localizes to perisynaptic regions, and its trafficking depends on UNC-104/KIF1A, the anterograde kinesin-3 motor for presynaptic components.²⁹ Third, UNC-8 is required for elevated presynaptic Ca²⁺ in remodeling DD neurons.¹⁵ Fourth, the Ca²⁺-dependent phosphatase TAX-6/CaN is enriched at presynaptic boutons of GABAergic neurons. Fifth, TAX-6/CaN strengthens Ca²⁺ responses during the remodeling window. Together, these results support the hypothesis that UNC-8 elevates intracellular Ca²⁺ for activation of TAX-6/CaN and that this effect promotes presynaptic disassembly in remodeling DD neurons.¹⁵ The potential mechanism whereby TAX-6/CaN activates UNC-8 is unknown, but members of the DEG/ENaC family can be regulated by phosphorylation at conserved intracellular domains.⁴⁶

In addition to GABAergic DD neurons, UNC-8 is strongly expressed in cholinergic DA and DB motor neurons, which normally do not undergo synaptic remodeling.^11,16 In contrast, forced expression of UNC-8 in GABAergic VD neurons induces presynaptic disassembly. We suggest that DD and VD neurons may share cellular machinery for synaptic remodeling that is not available in DA and DB neurons. Alternatively, mechanisms that promote UNC-8 channel function⁴⁷ may be uniquely active in DD and VD neurons.

Although UNC-8 promotes synaptic disassembly and recycling, the overall remodeling mechanism is delayed but not prevented in unc-8 mutants. This finding is consistent with our earlier report that UNC-8 functions in parallel to at least one additional transcriptionally regulated pathway that depends on currently unknown downstream effectors.¹¹

A mechanism of bulk endocytosis promotes DD neuron synaptic remodeling

The remodeling pathway we identified resembles ADBE. During periods of high synaptic activity, ADBE mediates the recovery of the presynaptic membrane for recycling into the SV pool.¹⁹ For ADBE, elevated intracellular Ca²⁺ activates the phosphatase CaN, which, in turn, dephosphorylates a group of conserved proteins (dephosphins) that direct endocytosis.²¹ Dephosphorylation of the GTPase dynamin, for example, mediates the formation of a dynamin-syndapin complex to promote branched-actin polymerization, which drives the formation of bulk endosomes.¹⁹

Our results suggest that TAX-6/CaN functions downstream of UNC-8 to promote an ADBE-like mechanism that promotes recycling of the presynaptic apparatus to new synaptic locations. Strikingly, phospho-resistant dynamin is sufficient to accelerate DD remodeling, as predicted from a model where CaN-dependent dephosphorylation of dynamin facilitates endocytic removal of presynaptic domains. In addition, we showed that the F-BAR protein and known ADBE effector SDPN-1/syndapin localizes to remodeling synapses, functions in a common pathway with UNC-8, and mediates UNC-8-dependent elimination of ventral synapses in remodeling GABAergic neurons. Our finding that clathrin is not required for presynaptic disassembly is also suggestive of an ADBE-like mechanism because bulk endocytosis is clathrin independent.¹⁹ EM reconstruction confirmed that intracellular structures resembling bulk endosomes, an ultrastructural hallmark of ADBE,^39–41 are abundantly distributed in remodeling DD neurites and that the UNC-8 protein promotes their formation. Together, these findings support the hypothesis that an ADBE-like mechanism promotes DD synaptic remodeling. In this case, however, bulk endocytosis occurs in concert with the disassembly of the DD presynaptic domain as opposed to the canonical role of ADBE of sustaining presynaptic function.¹⁹ Finally, we have shown previously that UNC-8 acts in a common pathway with the apoptotic component CED-4.^15,48 In the future, it will be interesting to determine whether CED-4 is linked to the ADBE-like mechanism we report here.

Endosome-like structures have been observed previously in remodeling or unstable presynaptic terminals at the Drosophila neuromuscular junction (NMJ). For example, mutants of either spectrin or the dynactin complex show enlarged, clear vesicular structures adjacent to NMJs undergoing elimination.^49,50 Endosome-like structures are similarly abundant in the presynaptic regions of mutants that disable the BMP (bone morphogenic protein) receptor and result in smaller NMJs.^51,52 These observations suggest that a conserved bulk endocytosis-mediated mechanism could be utilized for synapse disassembly and remodeling in other species.⁵³

Actin polymerization promotes presynaptic refinement

We used live imaging to confirm that actin polymerization is elevated in presynaptic domains during DD remodeling and that transient RAB-3 clusters are detected in these regions. Furthermore, actin polymerization is reduced in unc-8 mutants and correlates with fewer transient RAB-3 events and delayed elimination of stable RAB-3 clusters. Notably, actin polymerization is required for bulk endocytosis in mammalian neurons and in neurosecretory cells.^25,27 The observed actin polymerization defect in unc-8 mutants is consistent with our finding that the F-BAR protein, SDPN-1/syndapin, which can recruit nucleators of branched-actin polymerization,²⁴ functions downstream of UNC-8. Our finding that actin polymerization is elevated during DD remodeling highlights the need for free barbed ends for Arp2/3-dependent nucleation, as also observed at the leading edge of migrating cells.⁵⁴ This requirement could explain an earlier report that the Ca²⁺-sensitive F-actin-severing protein gelsolin promotes DD synapse elimination.⁴⁸

Presynaptic proteins recycle from old to new boutons during circuit refinement

Our evidence indicates that the key recycling endosome component RAB-11 acts downstream of UNC-8 to promote DD presynaptic remodeling. We used a photoconvertible tag to confirm an earlier report of RAB-3 recycling from ventral to dorsal synapses⁴³ and to establish that rab-11 is required. The recently reported role of Rab11 in ADBE²⁶ strengthens the hypothesis that DD remodeling depends on an ADBE-like mechanism and that recycling endosomes are involved. Because our results suggest that only ~50% of the RAB-3 protein at dorsal DD synapses is derived from de novo synthesis, recycled RAB-3 is likely important for dorsal DD synaptic signaling. Interestingly, the unc-8 mutation impairs dorsal insertion of photoconverted RAB-3 but did not prevent its removal from ventral DD synapses later in development, likely because of a partially redundant remodeling pathway that acts in parallel to unc-8.¹¹

In vivo imaging revealed increased actin polymerization as well as mobile RAB-3 particles transiting between ventral DD presynaptic domains during remodeling. The appearance of transient SVs in inter-synaptic regions has been proposed to drive SV sharing across en passant boutons to maintain synaptic function.^55,56 Moreover, SV sharing between mammalian synapses is also actin dependent.⁵⁷ Notably, the extent of vesicle sharing declines with the distance between these en passant synapses.⁵⁵ Thus, we suggest that a pathway commonly used for local recycling may have been repurposed, in the case of remodeling DD neurons, for recycling of SV components to distal synapses in a mechanism that rewires a developing circuit.⁶

In summary, we have shown that a developmentally regulated pathway that dismantles presynaptic domains for reassembly at new locations depends on key elements of a homeostatic mechanism, ADBE, that normally acts to sustain local synaptic function. The widespread occurrence of synaptic refinement across species and the evolutionary conservation of the key components (DEG/ENaC, CaN, dynamin, syndapin, WRC, Arp2/3, and RAB11) that promote synaptic remodeling in C. elegans suggest that similar mechanisms may be utilized in the developing brain.

Limitations of the study

Our results do not exclude the possibility that an ADBE-like mechanism functions after synaptic disassembly so that ADBE merely captures dismantled presynaptic components for recycling to new dorsal synapses. When ADBE is disabled in this model, disassembled presynaptic markers are retained on the ventral side, which our current imaging methods would not distinguish from intact presynaptic boutons. Although genetic and Ca⁺² imaging data are consistent with the hypothesis that CaN/TAX-6 positively regulates UNC-8 activity, the mechanism of this effect is unknown. Finally, a fundamental understanding of the remodeling mechanism will depend on future studies to reveal underlying cell biological components that redirect recycling presynaptic components to nascent vs. existing synaptic locations.

STAR★METHODS

Detailed methods are provided in the online version of this paper and include the following:

RESOURCE AVAILABILITY

Lead contact

Further information and requests for reagents should be directed to and will be fulfilled by the lead contact, David M. Miller, III (david.miller@vanderbilt.edu).

Materials availability

Key C. elegans strains in this study have been deposited at the CGC.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Strains and genetics

Worms were maintained at 20°C–23°C using standard techniques.⁵⁸ Strains were maintained on NGM plates seeded with E. coli (OP-50) unless otherwise stated. The wild type (WT) is N2 and only hermaphrodite worms were used for this study.

Generation of unc-8:gfp11×7(syb1624) and tax-6:gfp11×3(sy2783b) animals

Sunybiotech used CRISPR/Cas9 to add 7 copies of GFP11 to the unc-8 locus and 3 copies of GFP11 to the tax-6 locus animals at the C terminus. To visualize the reconstituted GFP signal in DD neurons, complementary GFP1-10 was driven with the Pflp-13 promoter.²⁸

Generation of rab-3(syb2844) [Dendra2 FLP-ON]) II

Sunybiotech used CRISPR/Cas9 to add an FLP-ON cassette at the N terminus of the rab-3 coding sequence,⁴⁴ which contains the photoconvertible protein Dendra2. To visualize the endogenous RAB-3 tagged with Dendra2, we expressed flippase in DD neurons using the flp-13 promoter (Pflp-13). Animals were outcrossed three times before they were used for experiments.

METHOD DETAILS

Worm synchronization and staging

Staging was achieved by picking 30–40 L4 larvae onto a fresh plate in the afternoon and leaving at 23C overnight. The next morning, once the animals had become gravid adults, 20–30 worms were picked from original plate onto a fresh plate and allowed to lay eggs for 1 h at RT. Adults were then removed maintained at 23C to allow the eggs (embryos) to hatch and larvae to develop until selected timepoints for imaging which were defined in as “hours post lay” or “hpl.” Early L4 larvae were identified under DIC (Differential Interference Contrast) imaging in a compound microscope as animals with vulval shapes matching L4.0, L4.1 or L4.2 morphological categories.⁵⁹

Molecular Biology

Most plasmids were constructed using InFusion cloning. Plasmid pACC161 (Pflp-13::dyn-1AAA::SL2:3xNLSTagRFP) was built using Q5 site-directed mutagenesis kit. The pACC156 (Pflp-13::dyn-1:SL2:3xNLSTagRFP) plasmid was used as a template to change three residues in the phospho-box for a triple-alanine mutant.

GFP::RAB-3 density analysis

To analyze GFP:RAB-3 puncta density along the ventral cord, Maximum Intensity Projections were created from Z-stacks using NIS Elements. All images were subjected to background correction using the Rolling Ball algorithm. Analysis explorer was used to create a mask based on fluorescent thresholding for the GFP::RAB-3 signal. ROIs (Regions Of Interest) were defined for the interval spanning DD2 and DD3 ventral neurites. To calculate puncta density, each object was considered a RAB-3 punctum and the total number was normalized to a 10 μm neurite using the following equation: (# of objects detected in ROI/length of ROI)*10.

To quantify GFP::RAB-3 density in the dorsal nerve cord, ROIs were drawn within the anatomical DD1-DD3 region. Because curvature of the dorsal cord in the XY plane could artificially alter puncta density, all dorsal chord regions used in our analysis had to meet straightness criteria. To this end, the Angle measurement tool in NIS elements was used to measure the angle of intersection between two lines drawn tangent to the regions of the dorsal cord immediately anterior and posterior to the apex of the curve in question. Regions with an intersection measurement of less than 160° (where 180° represents a perfectly straight region and 90° represents a region curved at a right angle) were excluded from analysis. ROIs were then drawn along the longest possible region of the dorsal cord that adhered to these straightness criteria. If the regions anterior and posterior to a curve met straightness criteria, then two ROIs were drawn for that image. Off target fluorescent features (autofluorescence) and commissures were excluded from ROIs. ROI lengths were measured using the Distance Measurement: Polyline tool in NIS Elements and were recorded in μm. The Automated Measurement Results tab in NIS Elements was used to record the number of objects (puncta) in each ROI, and puncta density within the ROI was then calculated as described above. To designate a fluorescent signal as a punctum, a mask was generated in NIS Elements that labeled as puncta all fluorescent features in an image that met size, shape and intensity criteria for puncta as set by the experimenter. Once generated, the same masking criteria were used for all treatment and control images included in a given experiment.

Counting UNC-8:GFP11_×7 puncta

Worms were synchronized (see above) to track endogenous UNC-8 split-GFP puncta²⁸ before and during the remodeling window: early L1 (16hpl), late L1 (14hpl) and L2 stage (30hpl). Z-stacks were collected using a Nikon A1R laser scanning confocal microscope (40x, 1.4 N.A.). Worms were immobilized in a pool of 3mL of 100mM muscimol (TOCRIS biosciences #0289) + 7μL 0.05μm polybeads (2.5% solids w/v, Polysciences, Inc. #15913-10) on a 10% agarose pad. Maximum Intensity projections were created for each image using NIS Elements. The experimenter was blinded to developmental time to manually score each GFP punctum. UNC-8:GFP puncta density was calculated by normalizing the number of puncta to a 10 μm neurite.

To measure the UNC-8 split-GFP fluorescence in cell soma, maximum intensity projections were analyzed in FIJI. Cell soma were defined by LifeAct:mCherry fluorescence as the region of interest. Mean GFP fluorescence intensity measurements were recorded for wild-type and unc-104 mutant animals.

Presynaptic TAX-6 enrichment

Worms were synchronized (see above) to track endogenous TAX-6:GFP11_×3 NATF fluorescence²⁸ along the ventral and dorsal cords before at the L1 stage (16hpl), and after remodeling at the L4 stage. z stack images were acquired using a Nikon A1R laser scanning confocal microscope (40x, 1.4 N.A). Maximum intensity projections were created using FIJI and a 3-pixel wide line scan was drawn on the ventral and dorsal cord of each animal to determine average fluorescence. The relative dorsal (D) and ventral fluorescence (V) was normalized to the total fluorescence (D + V) for each DD neuron (DD1-DD4).

To detect endogenous TAX-6:GFP and mRuby:CLA-1s, Z-stacks were acquired using Nyquist Acquisition and small step size (0.2 μm) on a Nikon A1R confocal microscope. Z-stacks were subjected to 3D-deconvolution in NIS Elements and single plane images were used for line scans.

GCaMP6s imaging in remodeling DD axons

To detect GCaMP6s fluctuations in remodeling boutons (DD2-DD4), NC3569 animals were synchronized (26hpl) on an OP-50-1-seeded plate with either freshly added ATR or carrier (EtOH, no ATR). Image acquisition was performed on a Nikon TiE microscope equipped with a Yokogawa CSU-X1 spinning disk head, Andor DU-897 EMCCD camera, high-speed MCL piezo and 100X/1.49 Apo TIRF oil objective lens. Synchronized NC3569 worms at 26hpl were immobilized in a pool of 3μL of 100mM muscimol (TOCRIS biosciences #0289) + 7μL 0.05μm polybeads (2.5% solids w/v, Polysciences, Inc. #15913-10) on a 10% agarose pad.

To identify baseline GCaMP fluctuations, images 4–10fps (frames per second) were collected with 488 nm excitation in a single plane using a Perfect Focus System for 1 min. Ventral DD2-DD4 neurites were imaged for this experiment. To detect evoked GCaMP transients, triggered acquisition was used to excite GCaMP with 488 nm and activate Chrimson with 561nm lasers. Single plane movies were collected at 3.8 fps for 1 min using emission filters 525nm (+/− 25nm) and 646 (+/− 66nm), respectively. The sample was illuminated with a 561nm laser at 5 s intervals (e.g., every 20^th frame) to activate Chrimson expressed in cholinergic DA motor neurons of late L1 animals (26hpl) (Punc-4::ceChrimson:SL2:3xNLS::GFP) while maintaining constant illumination with a 488 nm laser to detect GCamP6s signal. This acquisition paradigm was applied for 10 cycles (1 min). In each case, imaging data were not collected during a 1-s period of optogenetic activation (200 ms) and channel switching (800 ms). Only neurites that were stable (not moving or sinking) for ≥2 cycles of acquisition were selected for further processing. For quantifying GCaMP6s fluorescence, ROIs were drawn on oblong, bouton-like structures near the base of the commissure and on a nearby region to capture background fluorescence. The same ROIs (bouton and background) were used to detect spontaneous and evoked GCaMP changes in videos from the same DD neuron. Mean fluorescence intensity of each ROI for each frame was exported into Excel for analysis. Background was subtracted from each frame and measurements were normalized to three time-points before Chrimson activation for the first acquisition cycle (F₀). At each time-point, F₀ was subtracted from the measured GCaMP fluorescence and then divided by F₀ to determine ΔF/F_0. Then, fluorescent values before and after each event of Chrimson activation were matched for paired comparisons. Because recent RNA-seq results indicate that the unc-49 transcript for the GABA receptor is expressed in DD/VD neurons,⁶⁰ it is possible that muscimol could affect their excitability although expression of UNC-49 protein in DD/VD neurons has not been previously reported.

Cell-specific RNAi

We used In-Fusion cloning to create two plasmids that express complementary strands of mRNA for each cell-specific RNAi (csRNAi) experiment. To confirm expression, each plasmid contains a transplicing sequence (SL2) followed by a fluorescent protein (cytosolic of nuclear-localized GFP and mCherry) downstream of the sense or antisense transcript. For example, for csRNAi knockdown of arx-5, dyn-1, rab-11 and rab-5 in DD neurons, we drove expression of one strand in DD neurons with the flp-13 promoter and the opposite strand in DD + VD neurons with the ttr-39 promoter (See Table S1). In this arrangement, expression of both the sense and anti-sense sequences is limited to DD neurons, thus resulting in a DD-specific knockdown of each targeted gene. csRNAi templates for each target gene were generated by PCR from genomic DNA with primers that excluded the ATG start codon as follows: 1210 bp of p21/arx-5, 1686 bp or dyn-1, 1174 bp of rab-11.1 and 1275bp of rab-5.

In this experimental design, csRNAi-treated DD neurons are marked by expression of both fluorescent markers (mCherry + GFP) whereas control DD neurons are unlabeled. To determine if remodeling of SNB-1:GFP (juIs137) was perturbed upon knockdown, ImageJ software was used to generate maximum intensity projections. All fluorescence intensity plots were created by drawing a linescan through the ventral and dorsal nerve cords of each worm and averaging the fluorescence intensity values across the anterior region of the DD neurons scored. Intensity of SNB-1:GFP was normalized to the total dorsal and ventral fluorescence.

Feeding RNAi

Clones from the RNAi feeding library (Source BioScience) were used in this study.⁶¹ RNAi plates were produced as previously described.¹⁰First, 15 mL culture tubes containing 2 mL LB broth with 10μL (10mg/mL) ampicillin were inoculated with individual bacterial colonies and grown in a 37°C shaker for 12–16 h. Control RNAi cultures were inoculated with bacterial colonies containing the cloning vector but no insert (“empty vector” control) for each feeding RNAi experiment. 250 μL of overnight culture was added to a mixture of 12.5 mL LB broth +62.5 μL ampicillin in a 50 mL conical tube. The new culture was then returned to the 37 C shaker until reaching log phase (OD > 0.8). The culture was then induced by adding 12.5 mL of LB broth, 62.5 μL ampicillin, and 100 μL 1M IPTG. After 3.5–4 h incubating in 37°C shaker, the culture was spun down using a tabletop centrifuge for 6 min at 3900 rpm to pellet the bacteria. The supernatant was discarded, and the pellet resuspended in 1mL M9 mixed with 8μL IPTG. The bacterial mixture was dispensed onto 4 × 60 mm NGM plates (250 μL each), and the plates were left to dry overnight in the horizontal laminar hood and then stored at 4°C for up to a week. To set-up each RNAi experiment, 3–5 L4 larvae (NC1852) were placed on each RNAi plate and maintained at 20°C. Four days later, F1 progeny was imaged as L4 animals. To count the number of puncta, see Dyn-1(OE) Analysis below.

Identifying remodeling genes

To screen candidate genes for a role in GABAergic synapse remodeling, we scored SNB-1:GFP puncta from the juIs1 [punc25::SNB-1:GFP; lin-15] transgene⁷ in unc-55; eri-1 mutants sensitive to feeding RNAi as previously described.¹⁰ We compared animals treated with each candidate gene RNA feeding strain vs. the blank RNAi (empty vector). Briefly, animals were anesthetized with 0.1% tricaine/tetramisole, mounted on a 2% agarose pad, and imaged with an inverted microscope (Zeiss Axiovert) using Micro Manager software and a 63× oil objective. Ventral puncta between VD3 to VD11 were counted following the mid-L4 stage for both RNAi control and RNAi-treated animals. Data were pooled at least from 2 separate experiments and the examiner was blinded to the treatment. One-Way ANOVA was performed with post hoc correction to compare each RNAi treatment against the control. Similarly, the scorer was blinded to genotype to test if Synaptojanin/unc-26 promotes synapse remodeling of SYD-2:GFP puncta from hpIs3[punc-25::SYD-2::GFP; lin-15] transgene⁶² in wild-type, unc-55 and unc-55; unc-26 double mutants.

Dyn-1(OE) analysis

Synchronized animals (28hpl) (28 h post laying) expressed either wild-type (WT) or phospho-resistant versions of dyn-1 in DD neurons under the Pflp-13 promoter in tandem with a nuclear-localized TagRFP protein to recognize cells expressing the array. Z-stacks were collected in a Nikon A1R laser scanning confocal spanning the depth (~3 μm) of the ventral nerve cord. Z-stacks were used to generate MaxIPs in NIS Elements and Analysis Explorer, to define a binary mask based on fluorescence and size to recognize cell soma that expressed nuclear TagRFP. DD1-DD3 neurons were categorized as Control (no red) or dyn-1(OE) (red nucleus) based on nuclear expression of TagRFP. An additional mask was defined in the GFP channel for ventral SNB-1:GFP (juIs137) puncta based on fluorescence and circularity. Each recognized punctum on the anterior process of all DD1-DD3 neurons was counted. Then, puncta counts for each DD dendrite was normalized to the length of the counted region to calculate SNB-1:GFP density as puncta/10 μm.

VCA-1(OE) analysis

Synchronized animals (26hpl) expressed the VCA domain of the Wave Regulatory Complex in tandem with cytosolic mCherry with the Pttr-39 promoter to mark GABAergic cells carrying the array. Z-stacks spanning the depth of both nerve cords were collected in a Nikon A1R laser scanning confocal microscope. Z-stacks were used to generate MaxIPs in FIJI and the segmented line tool (3-pixel wide) was used to draw ROIs on top of the ventral and dorsal nerve cords. mCherry-positive DD cells carry a VCA(OE) transgenic array. All fluorescence intensity plots were created by drawing a segmented line through the ventral and dorsal nerve cords of each worm and averaging the fluorescence intensity values across the anterior region of the DD neurons scored. Intensity of SNB-1:GFP (juIs137) was normalized to the total dorsal and ventral fluorescence.

UNC-8(OE) analysis in RNAi-treated animals

FIJI was used to quantify presynaptic markers in VD neurons for effects of UNC-8(OE) (over-expression). Z-stacks were collected for the full length of the ventral nerve cord (VD3 – VD11). mCherry-positive VD cells carry an UNC-8 cDNA transgenic array.¹⁵ Because UNC-8(OE) transgenic arrays are mosaic with expression limited to a random subset of VD neurons in each animal, data were collected from VD neurons (VD3-VD11) carrying the UNC-8 cDNA (mCherry-positive) vs. an adjacent control VD neuron that does not carry the array. Neighboring mCherry-positive [e.g., UNC-8(OE)] and mCherry-negative [control] VD neurons were compared to quantify differences in the fluorescence signal for SNB-1:GFP due to UNC-8 over-expression with RNAi knockdown of either tax-6, sdpn-1, wve-1 or rab-11. Intensity values were obtained from line scans anterior to the VD cell bodies of interest. Background fluorescence was obtained from a line scan of an adjacent region inside the animal and subtracted from the VD line scans.

Microscopy

Laser scanning confocal microscopy

Unless otherwise indicated for specific experiments, Larval or young adult animals were immobilized on 2–10% agarose pads with 15mM levamisole, 0.05% tricaine as previously described.⁶³ Z-stacks were acquired with a Nikon confocal A1R using Apo Fluor 40X/1.3 and 60X/1.4 N.A. oil objectives.

In-vivo actin dynamics at remodeling synapses

For measurements of actin dynamics at synaptic regions in DD neurons (Figures 6 and S6), endogenous GFP::RAB-3 was used to mark synaptic vesicle clusters and LifeAct:mCherry³⁸ to visualize actin dynamics. Live-imaging was performed in a Nikon TiE microscope equipped with a Yokogawa CSU-X1 spinning disk head, Andor DU-897 EMCCD camera or Photometrics Prime 95B sCMOS camera, high-speed piezo stage motor, a perfect focus system and a 100X/1.49 Apo TIRF oil objective lens. Synchronized animals at 16–18hpl (early L1) or 26–28hpl (late L1) were mounted on 10% agarose pads and immobilized in a pool of 3μL of 100 mM muscimol (TOCRIS biosciences #0289) + 7μL 0.05μm polybeads (2.5% solids w/v, Polysciences, Inc. #15913–10). Single-plane snapshots were collected at 1 fps using triggered acquisition. Movies were submitted to 2D-deconvolution on NIS-Elements using the Automatic algorithm and aligned with the NIS Elements alignment tool when necessary. To analyze GFP::RAB-3 dynamics, videos were limited to 60–150 s to reduce photobleaching and the GFP signal was binarized with manual thresholding to enhance local contrast. The resultant binarized video was then submitted for particle tracking to detect transient GFP::RAB-3 movement. The total number of transient events was normalized to events per minute per 10 μm neurite. Kymographs to visualize GFP::RAB-3 and actin dynamics (LifeAct:mCherry) were generated in NIS Elements. At sites of GFP::RAB-3 puncta, ROIs were defined in the 561nm channel to track LifeAct:mCherry transients (e.g., actin transients) in presynaptic regions. LifeAct:mCherry fluorescence throughout time were plotted and normalized with the first fluorescence value set to zero. GFP::RAB-3 density (puncta/10 μm) for each genotype was determined from the first frame of each video and submitted for 2D deconvolution with NIS elements. The line scan function was then used to define fluorescent GFP::RAB-3 peaks. Any peak at > 300AU fluorescent units was considered a GFP::RAB-3 punctum.

In-vivo dynamics of endogenous GFP::RAB-3 and TagRFP::RAB-11 during remodeling

We built a strain to follow endogenous GFP::RAB-3 and TagRFP::RAB-11 dynamics (Figures S7D—S7F) during the remodeling window 27–30 hpl. Animals were placed on 10% agarose pads and immobilized using a combination of 3μL of 100mM muscimol (TOCRIS biosciences #0289) and 7μL 0.05μm polybeads (2.5% solids w/v, Polysciences, Inc. #15913-10). Single plane movies were captured at 0.5fps (frames per second) on a Nikon TiE microscope with a Yokogawa CSU-X1 spinning disk head, Photometrics Prime 95B sCMOS camera, a high-speed MCL Piezo stage, a Perfect Focus system and a 100X/1.49 Apo TIRF oil objective lens. Movies were 2D-deconvolved using the automatic algorithm in NIS Elements and kymographs were created in FIJI with the Multi Kymograph tool. Kymographs were then exported and submitted to Kymobutler for identification of particle movement (https://www.wolframcloud.com/objects/deepmirror/Projects/KymoButler/KymoButlerForm).

Tracking endogenous Dendra2:RAB-3 during DD remodeling

This experiment consists of four steps: (1) worm synchronization, (2) photoconversion and collection of baseline data before remodeling, (3) worm recovery, and (4) analysis of photoconverted signal after remodeling. Larvae expressing endogenous Dendra2:RAB-3 in DD neurons, were synchronized by allowing young adults to lay eggs for 1 h (see above). Plates were maintained at 23 C L1 animals at 18 hpl were photoconverted on a Nikon TiE microscope with a Yokogawa CSU-X1 spinning disk head using a Mini Scanner equipped with a 405 nm, 100mW photo stimulation laser. Briefly, single worms were mounted on a 10% agarose pad with 2μL of anesthetic [from mixture of 3mL of 100mM muscimol (TOCRIS biosciences #0289) and 7μL 0.05μm polybeads (2.5% solids w/v, Polysciences, Inc. #15913-10)]. Wax-paraplast 50/50 v/v combination was applied to each corner of the coverslip to secure it. Using NIS elements, we defined a targeted ROI limited to ventral Dendra2:RAB-3 synaptic puncta visualized in the 488 nm channel. Dendra2:RAB-3 was photoconverted with 30μs dwell time with 0.5% power of the 405 nm laser. We collected a z stack using 488nm and 561nm lasers before and after Dendra2 photoconversion. In most cases, we photoconverted Dendra2:RAB-3 puncta in the anterior ventral neurites of either DD2, DD3 or DD4.

To recover worms, we added 90μL of M9 to the space between the coverslip and the agarose pad and cut the sealant from each corner using a razor blade. The coverslip and agarose pad were rinsed with the addition of M9 buffer and allowed to drip on top of an NGM 60mm plate. Typically, two rounds of 90μL M9 was used to rinse the coverslip and five rounds for the agarose pad. After a few minutes, when the NGM plates had dried, each plate was placed in a 23C incubator. Each treated animal was maintained on a separate NGM plate to facilitate matching imaging data obtained before and after remodeling. Twenty hours after photoconversion (38 hpl), we mounted each worm on a 10% agarose pad using 3uL of the anesthetic mixture (see above) and collected a z stack using 488nm and 561nm lasers to track the Dendra2:RAB-3 signal.

For analysis, we first used NIS Elements to 3D-deconvolve all Z-stacks and paired the images for each individual worm before (18hpl) and after remodeling (38 hpl). After deconvolution, we imported all images to FIJI and created maximum intensity projections. We drew a 3-pixel wide line scan on top of the Dendra2:RAB-3 signal and exported the traces into Prism from each fluorescent channel (488 and 561nm) and time-point. Traces from neighboring regions were used to define background fluorescence. We used Prism to determine the area under the curve for line scans of the ROI (Dendra2:RAB-3 puncta) as well as neighboring regions (background). We subtracted the background area and plotted the relative photoconverted signal (vs. 18hpl) in the dorsal and ventral side at 38 hpl.

Estimating nascent vs. recycled Dendra2:RAB-3 in dorsal DD synapses

Rv = ventral red fluorescence at 18 hpl

Rd = dorsal red fluorescence at 38 hpl

Gv = ventral green fluorescence at 18 hpl

Gd = dorsal green fluorescence at 38 hpl

Because photoconversion was incomplete, dorsal green fluorescence derives from both nascent (Gn) and recycled (Gy) Dendra2:GFP:

$$\mathrm{G}\mathrm{d}=\mathrm{G}\mathrm{n}+\mathrm{G}\mathrm{y}$$

We calculated the average fraction of red Dendra2:RAB-3 recycled to the dorsal side:

$$\mathrm{f}=\frac{\mathrm{R}\mathrm{d}}{\mathrm{R}\mathrm{v}}=0.65(\text{See}\text{S9E)}$$

Assuming: $f=\frac{Rd}{Rv}=\frac{Gy}{Gv}=0.65$

Then: $\mathrm{G}\mathrm{d}=\mathrm{G}\mathrm{n}+0.65\mathrm{G}\mathrm{v}$

And: $\mathrm{G}\mathrm{n}=\mathrm{G}\mathrm{d}-0.65\mathrm{G}\mathrm{v}$

Substituting paired values for Gd and Gv from measurements of the same DD neuron before (18 hpl) and after remodeling (38 hpl) (File S1) we calculated the average nascent fraction of green Dendra2:RAB-3 on the dorsal side:

$$\frac{\mathrm{G}\mathrm{n}}{\mathrm{G}\mathrm{d}}=0.61\pm 0.13\text{(See}\text{S9E)}$$

And the average fraction of recycled green Dendra2:RAB-3 on the dorsal side:

$$\frac{\mathrm{G}\mathrm{y}}{\mathrm{G}\mathrm{d}}=0.39\pm 0.13$$

Assuming ~50% photoconversion (see S9F), an equivalent amount of red (photoconverted) Dendra2:RAB-3 is recycled to the dorsal side.

Recalculating: Fraction of nascent (green) $\mathrm{D}\mathrm{e}\mathrm{n}\mathrm{d}\mathrm{r}\mathrm{a}2:\mathrm{R}\mathrm{A}\mathrm{B}-3=\frac{0.61}{1.39}=0.44$ Fraction of recycled (red + green) $\mathrm{D}\mathrm{e}\mathrm{n}\mathrm{d}\mathrm{r}\mathrm{a}2:\mathrm{R}\mathrm{A}\mathrm{B}-3=\frac{0.78}{1.39}=0.56$

Electron microscopy

20 young adults were placed on an OP-50 seeded NGM plate, allowed to lay eggs for 1 h and then removed. Timing for hours post laying (hpl) was calculated from the midpoint of the 1-h period. Cultures were maintained at 23 C. We monitored the endogenous GFP::RAB-3 marker with confocal microscopy to confirm that late-L1 stage larvae at 27hpl were remodeling (i.e., GFP-RAB-3 puncta in both ventral and dorsal DD neurites).

Strains were prepared using HPF (High Pressure Fixation) as previously described (Liu et al., 2021). Briefly, twenty to thirty staged L1 larvae at 27 hpl were placed in specimen chambers filled with Escherichia coli and frozen at −180°C, using liquid nitrogen under high pressure (Leica HPM 100). Samples then underwent freeze substitution (Reichert AFS, Leica, Oberkochen, Germany) using the following program: Samples were held at −90°C for 107 h with 0.1% tannic acid and 2% OsO₄ in anhydrous acetone, then incrementally warmed at a rate of 5 °C/h to −20°C, and kept at −20°C for 14 h, before increasing temperature by 10 °C/h to 20°C. After fixation, samples were infiltrated with 50% Epon/acetone for 4 h, 90% Epon/acetone for 18 h, and 100% Epon for 5 h. Finally, samples were embedded in Epon and incubated for 48 h at 65°C (Liu et al., 2021). Thin (70 nm) serial sections were acquired using an Ultracut 6 (Leica) and collected on formvar-covered, carbon-coated copper grids (EMS, FCF2010-Cu). Sections were post-stained with 2.5% aqueous uranyl acetate for 4 min, followed by Reynolds lead citrate for 2 min (Liu et al., 2021). Images were obtained using a JEOL JEM-1400F transmission electron microscope, operating at 80 kV. Micrographs were acquired using an AMT NanoSprint1200-S CMOS or BioSprint 12M-B CCD Camera and AMT software (Version 7.01). DD2 neurons were identified based on measurements taken from 27 hpl larvae with GFP::RAB-3 expressed in DD neurons, and determined using distance from the isthmus of the pharynx in conjunction with established GABA neuron synaptic morphology (White et al., 1986).^11,15

Serial sections (control = 396; unc-8 = 349) containing the anteriorly-directed ventral process of the DD2 neuron were obtained from one wild-type and one unc-8(tm5052) larval animal at 27 hpl. Although both strains were fixed at 27 hpl, the wild-type control showed a double exterior cuticle and lacked alae as expected for an animal undergoing the L1/L2 molt whereas the unc-8 mutant was likely an early L2 since it contained a single-layered cuticle and lacked the lateral alae characteristic of L1 larvae.⁶⁴ Serial sections were manually aligned using the TrakEM2 package of NIH FIJI/ImageJ software. The aligned stack was imported into 3Dmod/IMOD (University of Colorado; https://bio3d.colorado.edu/imod/) for segmentation and reconstruction of the sections into a 3D model of DD2. The following criteria were used for morphometric analyses: Synaptic vesicles were clear-core spherical structures with an average diameter of ~30 nm, dense-core vesicles had an average diameter of ~65 nm, and distinct endosomal-like structures were light or intermediate-core, spherical or irregular membrane organelles of varied size, as previously described.^19,39–41 Quantitative data were extracted from the 3D model using the imodinfo module of IMOD. Values were imported to Prism (GraphPad) for statistical analysis using Fisher’s exact test or t test. Videos of the model were created using the movie/montage module in IMOD to create TIFF frames and imported into FIJI/ImageJ to create an AVI movie file.

QUANTIFICATION AND STATISTICAL ANALYSIS

First, we used the Shapiro-Wilk test to determine if a sample is normally distributed. For comparisons between 2 normally distributed groups, Student’s T test was used and p < 0.05 was considered significant. ANOVA was used for comparisons of >2 samples and for normally distributed data followed by Dunnett’s multiple-comparison test. Standard Deviations between two samples were compared using an F-test and p < 0.05 was considered significant. If the samples were not normally distributed, we used a Mann-Whitney test to compare two groups and a Kruskal Wallis test to compare three or more groups. The specific test and N for each experiment are listed in figure legends. For Fisher’s Exact test in Figure 7E, sections with EVs: WT = 133; unc-8 = 55 and sections without EVs: WT = 263; unc-8 = 294. For Fisher’s Exact test in Figure S7D, sections with dense projections: WT = 21; unc-8 = 31 and sections without dense projections: WT = 375; unc-8 = 318.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
E. coli OP50-1	Caenorhabditis Genetics Center (CGC)	Wormbase ID: OP50-1
Chemicals, peptides, and recombinant proteins
All-trans retinal (ATR)	Sigma-Aldrich	R2500
Experimental models: Organisms/strains
See Table S1		N/A
Recombinant DNA
punc-4::Chrimson:SL2:3xNLSGFP	Cuentas-Condori et al., 2019	pACC92
pflp-13::LifeAct:mCherry	Cuentas-Condori et al., 2019	pACC12
pttr-39::VCA::SL2:mCherry	This work	pACC42
pflp-13::arx-5 sense:SL2:mCherry	This work	pACC50
pflp-13:: arx-5 antisense:SL2:3xNLS::GFP	This work	pACC51
pflp-13::sdpn-1b::mCherry	This work	pACC57
pflp-13::GCaMP6s:SL2:mCherry	Cuentas-Condori et al., 2019	pACC83
pflp-13::2xNLS::FLP_D5	This work	pACC98
pflp-13::dyn-1F::SL2:mCherry	This work	pACC106
pflp-13::dyn1R::SL2:3xNLS::GFP	This work	pACC107
pflp-13::mRuby:CLA-1s	This work	pACC123
pflp-13:: 2xNLS::FLP_D5:SL2:LifeAct:mCherry	This work	pACC141
pflp-13::RAB-11.1F::SL2:3xNLSGFP	This work	pACC150
pttr-39::RAB-11R::SL2:mCherry	This work	pACC151
pflp-13::dyn-1:SL2:2xNLS::TagRFP	This work	pACC156
pflp-13::TagRFP::RAB-11.1cDNA	This work	pACC157
pttr-39::RAB-11.1csRNAi_R::SL2:2xNLSTagRFP	This work	pACC158
pflp-13::dyn-1-AAA::SL2:2xNLSTagRFP	This work	pACC161
pmyo-2::RFP	This work	pSH4
pstr-1::GFP	This work	pSH21
pflp-13::GFP1-10	He et al.,28	pSH87
Software and algorithms
NIS Elements	Nikon	https://www.microscope.healthcare.nikon.com/products/software/nis-elements
ImageJ 2.0.0	NIH	https://imagej.nih.gov/ij/download.html

Highlights.

• GABAergic presynapses are dismantled during development for reassembly at new locations

• ENaC/UNC-8 promotes calcium transients that activate presynaptic disassembly

• ENaC/UNC-8 promotes activity-dependent endocytosis for recycling from old to new synapses