Transcriptional Control of Parallel-Acting Pathways That Remove Specific Presynaptic Proteins in Remodeling Neurons

Synapses are actively dismantled to mediate circuit refinement, but the developmental pathways that regulate synaptic disassembly are largely unknown. We have previously shown that the epithelial sodium channel ENaC/UNC-8 triggers an activity-dependent mechanism that drives the removal of presynaptic proteins liprin-α/SYD-2, Synaptobrevin/SNB-1, RAB-3, and Endophilin/UNC-57 in remodeling GABAergic neurons in Caenorhabditis elegans (Miller-Fleming et al., 2016). Here, we report that the conserved transcription factor Iroquois/IRX-1 regulates UNC-8 expression as well as an additional pathway, independent of UNC-8, that functions in parallel to dismantle functional presynaptic terminals. We show that the additional IRX-1-regulated pathway is selectively required for the removal of the presynaptic proteins, Munc13/UNC-13 and ELKS, which normally mediate synaptic vesicle (SV) fusion and neurotransmitter release. Our findings are notable because they highlight the key role of transcriptional regulation in synapse elimination during development and reveal parallel-acting pathways that coordinate synaptic disassembly by removing specific active zone proteins. SIGNIFICANCE STATEMENT Synaptic pruning is a conserved feature of developing neural circuits but the mechanisms that dismantle the presynaptic apparatus are largely unknown. We have determined that synaptic disassembly is orchestrated by parallel-acting mechanisms that target distinct components of the active zone. Thus, our finding suggests that synaptic disassembly is not accomplished by en masse destruction but depends on mechanisms that dismantle the structure in an organized process.


Introduction
The nervous system is actively remodeled during development as new synapses are constructed and others are removed to refine functional circuits.Although synaptic assembly has been extensively investigated, synapse elimination is a less understood phenomenon despite its widespread occurrence (Goda and Davis, 2003;Südhof, 2018).In some cases, synaptic remodeling is limited to a specific developmental stage in which activity drives circuit plasticity.These "critical periods" are indicative of the necessary role of genetic programs that define specific developmental windows for activity-induced remodeling.Thus, synaptic remodeling mechanisms are likely to depend on the combined effects of both transcriptionally-regulated and activitydependent pathways (Hensch, 2004;Hong and Chen, 2011;Kano and Watanabe, 2019).
In Caenorhabditis elegans, synapses in the GABAergic motor circuit are relocated by a stereotypical remodeling program during early larval development (Cuentas-Condori and Miller, 2020).Dorsal D (DD) motor neurons are generated in the embryo and initially synapse with ventral body muscles (Fig. 1A).During the first larval stage, presynaptic domains are removed from ventral DD processes and then reassembled in the dorsal nerve cord (Fig. 1B; White et al., 1978;Hallam and Jin, 1998).Postembryonic ventral D (VD) neurons are born during this early larval period (Sulston, 1976) and synapse exclusively with ventral muscles (Fig. 1B).In the resultant mature circuit, GABAergic output alternates between dorsal (DD) and ventral (VD) muscles for sinusoidal movement (White et al., 1976(White et al., , 1986)).
The COUP-TF transcription factor, UNC-55, is selectively expressed in VD neurons to prevent synaptic remodeling (Zhou and Walthall, 1998;Shan et al., 2005), in unc-55 mutants, VD neurons initially synapse with ventral muscles but then mimic the native DD remodeling program by relocating presynaptic domains to the dorsal nerve cord (Fig. 1C; Petersen et al., 2011;Thompson-Peer et al., 2012).The idea that UNC-55 normally blocks expression of genes that drive synaptic remodeling is supported by the finding that forced expression of UNC-55 in DD neurons is sufficient to prevent the native remodeling program (Shan et al., 2005).In earlier work, we exploited the synaptic remodeling phenotype of unc-55 mutants in cell-specific profiling experiments to identify UNC-55 targets.An RNAi screen detected a subset of unc-55-regulated genes that are required for synaptic remodeling.For example, the homeodomain transcription factor, Iroquois/IRX-1, is ectopically expressed in unc-55 mutant VD neurons which consequently undergo aberrant synaptic remodeling.RNAi knock-down of Iroquois/IRX-1, however, prevented the removal of GABAergic presynaptic domains from the ventral nerve cord in unc-55 mutants (Fig. 1D; Petersen et al., 2011).Similarly, the DEG/ENaC cation channel subunit gene, unc-8, is upregulated in unc-55 mutants and an unc-8 lossof-function allele also antagonized VD synaptic remodeling (Fig. 1E).Thus, these results argue that IRX-1 and UNC-8 are required for presynaptic elimination of remodeling GABAergic neurons.Additional experiments confirmed that both the Iroquois/IRX-1 and DEG/ENaC/UNC-8 normally promote the native DD remodeling program (Fig. 1F; Petersen et al., 2011;Miller-Fleming et al., 2016).
DEG/ENaC proteins function as cation channels and we have previously shown that UNC-8 gates sodium influx (Wang et al.,  (Petersen et al., 2011).E, Mutations that disable the DEG/ENaC cation channel, UNC-8, impair the removal of DD and VD GABAergic presynaptic domains in unc-55 mutants (Miller-Fleming et al., 2016).F, IRX-1 and UNC-8 are normally expressed in DD neurons to drive disassembly of the presynaptic apparatus.The COUP-TF transcription factor, UNC-55, blocks expression of IRX-1 and UNC-8 in VD neurons to prevent synapse elimination.
2013; Matthewman et al., 2016).The resultant membrane depolarization arising from UNC-8 channel activity is predicted to open local voltage-gated Ca 21 channels (VGCC) which we have shown function with UNC-8 to promote presynaptic disassembly.Based on these findings we have proposed that UNC-8 promotes presynaptic disassembly in a pathway that depends on intracellular calcium and neural activity (Miller-Fleming et al., 2016).Here, we show that DEG/ENaC/UNC-8 is transcriptionally-regulated by Iroquois/IRX-1 to remove a core group of presynaptic components including liprin-a/SYD-2, Synaptobrevin/ SNB-1, RAB-3, and Endophilin/UNC-57.Surprisingly, proteins involved in synaptic vesicle (SV) priming, UNC-13/Munc-13 and ELKS, are not disassembled by UNC-8 but are removed by a separate pathway regulated by IRX-1/Iroquois.Together, these findings show that remodeling of GABAergic synapses depends on the combined effects of neural activity (UNC-8) and developmentally-regulated transcription (IRX-1).Thus, our work shows that synaptic disassembly can be orchestrated by parallel-acting mechanisms that selectively target molecularly distinct components of the presynaptic apparatus for removal.

Materials and Methods
Strains and genetics C. elegans strains were cultured at either 20°C or 23°C as previously described on standard nematode growth medium seeded with OP50 (Brenner, 1974).The mutant alleles and strains used in this study are outlined in Tables 1, 2.

Confocal microscopy
Larval or young adult animals were immobilized on 2% agarose pads with 15 mM levamisole as previously described (Smith et al., 2010).Zstack images (Figs.3A,D, 7A-F, 8A-F) were collected on a Leica TCS SP5 confocal microscope using a 63Â oil objective (0.5 mm/step), spanning the focal depth of the ventral nerve cord GABA neurons and synapses.Leica application suite advanced fluorescence (LAS-AF) software was used to generate maximum intensity projections.Ventral nerve cord images were straightened using an ImageJ plug-in.Z-stack images (Figs. 4,5I,9,10) were acquired with Nikon confocal A1R using Apo Fluor 40Â/1.3 and 60Â/1.4N.A. oil objective (0.3 mm/step).
FIJI was used to quantify effects of UNC-8(OE) (overexpression) on presynaptic disassembly (Figs. 7,8).Z-stacks were collected for the full length of the ventral nerve cord.In Figure 7H-K, mCherry-positive VD cells carry an UNC-8 cDNA transgenic array (Miller-Fleming et al., 2016).Neighboring mCherry-positive [e.g., UNC-8(OE)] and mCherrynegative (control) VD neurons were compared with quantify differences in the fluorescence signal for SNB-1::GFP and UNC-13L::GFP arisng from UNC-8 overexpression.In Figure 8J,K, GFP-positive VD neurons carry an UNC-8 cDNA transgenic array.Neighboring GFP-positive [e.g., UNC-8(OE)] and GFP-negative (control) VD neurons were compared with quantify differences in the fluorescence signal.Because unc-8 cDNA 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 for Fig. 7H-K or GFP-positive for Fig. 8J,K) versus an adjacent control VD neuron that does not carry the array.For results shown in Figures 7H-K, 8J,K, 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.

Electron microscopy (EM)
Young adult hermaphrodites of each strain were prepared for high-pressure freeze (HPF) fixation as described (Rostaing et al., 2004;Miller-Fleming et al., 2016).A total of 10-15 animals were loaded into a specimen chamber filled with Escherichia coli.The specimens were frozen rapidly in a high-pressure freezer (Leica HPM100) at À180°C and high pressure.Freeze substitution was performed on frozen samples in a Reichert AFS machine (Leica) with 0.1% tannic acid and 2% OsO 4 in anhydrous acetone.The temperature was kept at À90°C for 107 h, increased at 5°C/h to À20°C, and kept at À20°C for 14 h.The temperature was then increased by 10°C/h to 20°C.Fixed specimens were embedded in Epon resin after infiltration in 50% Epon/acetone for 4 h, 90% Epon/acetone for 18 h, and 100% Epon for 5 h.Embedded samples were incubated for 48 h at 65°C.All specimens were prepared using the same fixation procedure and labeled with anonymous tags so that the examiner was blinded to genotype.Ultrathin (40 nm) serial sections were cut using an Ultracut 6 (Leica) and collected on formvar-covered, carboncoated copper grids (EMS, FCF2010-Cu).Grids were counterstained in 2% aqueous uranyl acetate for 4 min, followed by Reynolds lead citrate for 2 min.Images were obtained on a Jeol JEM-1220 transmission EM operating at 80 kV.Micrographs were collected using a Gatan digital camera at a magnification of 100,000.Images were quantified using NIH ImageJ software.Dorsal and ventral cords were distinguished by size and morphology.GABAergic synapses were identified by previously established criteria, including position in the cord as well as the morphology of the synapse.GABAergic synapses are larger than their cholinergic motor neuron counterparts, and the active zones in these synapses form a direct, perpendicular angle with muscle arms.In contrast, the presynaptic density in cholinergic synapses orient at an acute angle to the muscle, generally 30-45°and are often dyadic.Some images were collected at 30k to aid in identifying synaptic identity based on terminal position in the cord.Two colleagues with expertise in EM reconstruction of the C. elegans ventral nerve cord independently reviewed synapse images from each strain to verify identification.Each profile represents an image of a 40-nm section.A synapse was defined as a set of serial sections containing a presynaptic density with two flanking sections either side without presynaptic densities.SVs were identified as spherical, light gray structures with an average diameter of ;30 nm.SVs were considered docked if they were in direct contact with the membrane.Three to five animals were imaged for each genotype.Numbers of profiles for each genotype were (# analyzed/# imaged): wild type = 80/1330, unc-55; unc-8 = 37/745, unc-55;irx-1(csRNAi) = 54/613 for ventral GABAergic synapse evaluation.

Electrophysiology
The C. elegans dissection and electrophysiological methods were as previously described (Richmond and Jorgensen, 1999;Miller-Fleming et al., 2016).Animals were immobilized along the dorsal axis with Histoacryl Blue glue, and a lateral cuticle incision was made with a hand-held glass needle, exposing ventral medial body wall muscles.Muscle recordings were obtained in the whole-cell voltage-clamp mode using an EPC-10 patch-clamp amplifier and digitized at 1 kHz.The extracellular solution consisted of 150 mM NaCl, 5 mM KCl, 5 mM CaCl 2 , 4 mM MgCl 2 , 10 mM glucose, 5 mM sucrose, and 15 mM HEPES (pH 7.3, ;340 mOsm).The intracellular solution consisted of 120 mM KCl, 4 mM KOH, 4 mM MgCl 2 , 5 mM N-tris[hydroxymethyl] methyl-2-aminoethane-sulfonic acid, 0.25 mM CaCl 2 , 4 mM Na 2 ATP, 36 mM sucrose, and 5 mM EGTA (pH 7.2, ;315 mOsm).GABAergic miniIPSCs and hyperosmotic responses were acquired at a holding potential of À60 mV by pressure-ejecting extracellular saline containing an additional 500 mOsm of sucrose; 10 mM d-tubocurare (dTBC) was added to both the extracellular solution and the pressure ejection pipette to block cholinergic hyperosmotic currents.Data were acquired using Pulse software (HEKA) on a Dell computer.Subsequent analysis and graphing were performed using Pulsefit (HEKA), Mini analysis (Synaptosoft Inc.) and Igor Pro (Wavemetrics).

Molecular biology
Generation of the punc-25::UNC-13L::GFP transgenic line We used the In-Fusion cloning kit (Takara) to amplify the cDNA of the long isoform of UNC-13 (UNC-13L) from a plasmid provided by J. Kaplan (pTWM88).This fragment was ligated into a vector containing the punc-25 GABA promoter and a C-terminal GFP tag.The resulting plasmid, pTWM90, was injected into unc-13 (e51) null mutants at 25 ng/ ml with the co-injection marker pmyo-2::mCherry (2 ng/ml).This transgenic array was integrated by x-ray irradiation and outcrossed for three generations to generate stable transgenic lines for analysis.

Feeding RNA interference experiments
Bacteria producing either double-stranded irx-1 RNA or containing the RNAi empty vector were seeded on NGM plates and stored at 4°C for up to one week.Four L4 unc-55, unc-55; eri-1, or unc-55; unc-8 animals were grown on each single RNAi plate at 23°C until progeny reached the L4 stage.Progeny were picked to fresh RNAi plates and the ventral synapses were quantified.

Movement assays
Animals were first tapped on the tail to ensure that they were capable of forward locomotion, then tapped on the head to assess ability to execute backward locomotion.Animals were binned into the following categories: unc (uncoordinated: coil ventrally immediately on tapping), initiate backing (initiate backwards movement but stop), and wild type (sustain backward locomotion with at least two body bends).In Figure 6, the wild type and initiate backing categories were grouped into a single initiate backing category.

Experimental design and statistical analysis
For samples that are not normally distributed, the Mann-Whitney test was used to compare two groups and determine significance.For comparisons between three or more groups, we used the Kruskal-Wallis test with multiple comparison.For samples that are normally distributed, Student's t test was used to compare two groups and one-way ANOVA with Bonferroni correction for multiple comparisons among more than two groups.Figure legends specify the statistical test used in each case and the number of independent measurements (N) evaluated.

Results
The homeodomain transcription factor, iroquois/IRX-1, drives DEG/ENaC/UNC-8 expression in remodeling GABAergic neurons In previous work, we used gene expression profiling and an RNAi screen to identify protein-encoding genes that promote presynaptic disassembly in remodeling GABAergic neurons (Petersen et al., 2011).These studies determined that the homeobox transcription factor Iroquois/IRX-1 and the DEG/ENaC ion channel subunit UNC-8 promote removal of the presynaptic vesicular SNARE protein, Synaptobrevin/SNB-1 (Fig. 1D,E; Petersen et al., 2011;Miller-Fleming et al., 2016).Because both IRX-1/Iroquois and DEG/ENaC/UNC-8 are expressed in remodeling DD neurons, we investigated the hypothesis that Iroquois/ IRX-1 functions as a transcription factor to regulate DEG/ENaC/ UNC-8 expression.First, we used smFISH to confirm expression of unc-8 transcripts in remodeling (control) DD neurons (Fig. 2).We note that unc-8 is highly expressed in adjacent DA and DB cholinergic neurons as previously reported (Wang et al., 2013;Miller-Fleming et al., 2016).We then targeted irx-1 in DD neurons with csRNAi (see Materials and Methods) and detected significantly fewer unc-8 transcripts in comparison to untreated (control) DD neurons (Fig. 2A,B).These results are consistent with the idea that Iroquois/IRX-1 is required for DEG/ENaC/ UNC-8 expression in DD neurons.
GFP reporters for irx-1 and unc-8 are not expressed in wildtype VD neurons which normally do not remodel.However, forced expression of Iroquois/IRX-1 in VD motor neurons is sufficient to trigger VD remodeling and drive the elimination of VD presynaptic terminals (Petersen et al., 2011;Miller-Fleming et al., 2016).Thus, we next asked whether Iroquois/IRX-1 overexpression [irx-1(OE)] could also induce unc-8 expression in VD neurons.smFISH quantification confirmed that unc-8 transcripts are elevated in irx-1(OE) VD neurons in comparison to controls (Fig. 2C,D).Together, these results demonstrate that the transcription factor, Iroquois/IRX-1, is both necessary and sufficient for DEG/ENaC/UNC-8 expression in remodeling GABA neurons (Fig. 2E).
Iroquois/IRX-1 drives a separate parallel-acting remodeling pathway that does not require UNC-8 for synaptic removal Because irx-1 encodes a transcription factor, we reasoned that Iroquois/IRX-1 might also regulate other targets in addition to the unc-8 gene in the GABA neuron synaptic remodeling pathway.If Iroquois/IRX-1 regulates a downstream target that functions in tandem with UNC-8, then genetic ablation of irx-1 should enhance the retention of ventral presynaptic markers in unc-55; unc-8 double mutants.For this test, we used feeding RNAi for global knock-down of irx-1 because the irx-1 null allele is lethal (Petersen et al., 2011).The unc-8(tm5052) deletion allele used for these experiments is a likely null mutation (Miller-Fleming et al., 2016).We counted GFP puncta for the presynaptic proteins SNB-1::GFP, SYD-2::GFP and UNC-57::GFP in unc-55; unc-8 double mutants versus unc-55; unc-8 animals treated with irx-1-RNAi.This experiment revealed that RNAi knock-down of irx-1 increases the number of ventral SNB-1:: GFP, SYD-2::GFP, and UNC-57::GFP puncta (Fig. 4A-C) in unc-55; unc-8 double mutant animals.Together, these results suggest that Iroquois/IRX-1 drives an additional genetic pathway, independent of UNC-8, that also eliminates presynaptic terminals in remodeling GABAergic neurons (Fig. 4D).In the next series of experiments, we used a combination of ultrastructural analysis, electrophysiology and genetics to confirm that Iroquois/IRX-1 functions in tandem with unc-8 to dismantle the presynaptic apparatus.

Iroquois/IRX-1 removes fusion-competent SVs in remodeling GABAergic neurons
Because ventral GABAergic synapses with apparently normal ultrastructure are observed in unc-55; unc-8 and unc-55; irx-1 (csRNAi) (Fig. 5A-D), we next performed experiments to determine whether these synapses are functional.We first examined whether the docked SVs in unc-55; unc-8 animals are fusioncompetent by recording iPSCs from ventral muscles.For these experiments, we used dTBC to block cholinergic signaling.As previously reported, tonic release of ventral iPSCs was restored in unc-55; irx-1(csRNAi) animals, but iPSCs were not detected in unc-55; unc-8 mutants despite the presence of organized clusters of fluorescent presynaptic proteins (Fig. 3), electron dense active zones and docked vesicles in both strains (Fig. 5A-D; Petersen et al., 2011;Miller-Fleming et al., 2016).Previously, we determined that the postsynaptic GABA A receptor UNC-49 was also properly localized and functional in unc-55;unc-8 animals (Miller-Fleming et al., 2016), thus excluding the possibility that the absence of iPSCs in unc-55; unc-8 animals is because of a postsynaptic defect.The lack of tonic release in unc-55; unc-8 animals could be because of defective vesicle priming or downstream Ca 21 -sensing.To determine whether the morphologicallydocked SVs in unc-55; unc-8 animals are primed, we measured iPSCs in response to hypertonic sucrose perfusion which is sufficient to induce neurotransmitter release from the primed SV pool (Fig. 5E-H).Hyperosmotic treatment of unc-55; unc-8 animals failed to trigger ventral GABA release suggesting that these animals are defective in SV priming (Fig. 5G,H).As expected, hyperosmotic stimulation triggered robust iPSCs in wild-type and in unc-55; irx-1(csRNAi) animals, whereas ventral muscles in unc-55 mutants were unresponsive.To rule out the possibility of a general defect in SV fusion, we confirmed that preparations of unc-55 and unc-55; unc-8 mutants exhibited spontaneous cholinergic activity that could be abolished by dTBC (Fig. 5E,F).Together with our EM results (Fig. 5A-D ;Miller-Fleming et al., 2016), these data indicate that SVs at ventral GABAergic synapses in unc-55; unc-8 animals are capable of docking with the presynaptic membrane but are not fusion competent.We further conclude that SVs at ventral GABAergic synapses in unc-55; irx-1(csRNAi) animals can dock and are also fusion competent since endogenous miniature IPSCs were detected and hyperosmotic treatment evoked robust IPSCs in unc-55; irx-1(csRNAi) adults (Fig. 5F-H).Thus, our results are consistent with the idea that Iroquois/IRX-1 and UNC-8 eliminate a shared set of presynaptic proteins (e.g., RAB-3, Synaptobrevin/SNB-1) in the remodeling program but that Iroquois/IRX-1 selectively removes specific presynaptic components for vesicle priming and neurotransmitter release in a mechanism that does not require UNC-8.

A behavioral assay for functional GABAergic synapses in the motor circuit
As an additional test of our ultrastructural and electrophysiological results, we devised a behavioral assay to evaluate the functionality of GABAergic synapses in remodeling defective mutants (Fig. 6).Ventral synapses for both DD and VD neurons are dismantled in unc-55 mutants and reassembled in the dorsal nerve cord (Fig. 1D).The resultant imbalance of excess inhibitory GABAergic output to dorsal muscles versus excess excitatory cholinergic input to ventral muscles results in a striking behavioral phenotype in which unc-55 animals coil ventrally when tapped on the head instead of initiating coordinated backward locomotion (Walthall and Plunkett, 1995;Shan et al., 2005).We have shown that irx-1(csRNAi) restores fusion-competent GABAergic presynaptic densities to unc-55 mutants (Fig. 5G,H).If these restored synapses are functional, then the tapping assay  should detect improved backward locomotion.Indeed, unc-55; irx-1(csRNAi) animals (Fig. 6) show robust backward movement in comparison to unc-55 mutants (Petersen et al., 2011).This result is congruent with our previous finding that GABAergic release is restored to ventral cord synapses of unc-55 mutants by RNAi knock-down of irx-1 (Fig. 5F-H).In contrast, unc-55; unc-8 mutant animals show severely defective backward locomotion that is not significantly different from that of unc-55 mutants (Fig. 6).This finding is in agreement with our observation that hyperosmotic treatment fails to evoke GABA release (Fig. 5G,H) and reinforces the idea that ventral GABAergic synapses in unc-55; unc-8 mutants are not functional.Genetic ablation of unc-8 activity in unc-55; irx-1(csRNAi) does not further enhance backward locomotion (Fig. 6) as predicted by our conclusion that residual ventral cord GABAergic synapses in unc-55; unc-8 double mutants are dysfunctional (Fig. 5F,G) and by our finding that Iroquois/IRX-1 regulates expression of the unc-8 gene (Fig. 2).To summarize, the results of the behavioral assay suggest that although ultrastructurally normal ventral GABAergic synapses are visible by EM in both unc-55; unc-8 and in unc-55; irx-1(csRNAi) animals (Fig. 5A), GABAergic release is selectively reactivated by knock-down of irx-1, but not by genetic removal of unc-8 (Fig. 5F-H).This striking difference suggests that Iroquois/IRX-1 must drive the removal of key determinants of presynaptic neurotransmitter release that are not targeted by UNC-8.
Iroquois/IRX-1, but not DEG/ENaC/UNC-8, removes the SV priming protein UNC-13 in remodeling GABAergic neurons The cytosolic protein Munc13/UNC-13 functions as a conserved component of the presynaptic apparatus to mediate SV fusion (Brose et al., 1995;Augustin et al., 1999;Richmond et al., 1999;Kohn et al., 2000;Weimer et al., 2006;Südhof, 2012).Mammalian neurons express four UNC-13-related proteins whereas only two distinct UNC-13 proteins, a long (UNC-13L) and a short (UNC-13S) version, are expressed in C. elegans.Because SVs appear docked, but are incapable of fusion in unc-55; unc-8 mutants (Fig. 5A-D, F-H), we hypothesized that UNC-13 could be absent from ventral GABAergic synapses in these animals.To test this idea, we generated a strain expressing GFP-tagged UNC-13L protein in GABA neurons.We selected UNC-13L for this experiment because it co-localizes with UNC-10/RIM whereas the short isoform, UNC-13S, shows a diffuse distribution in GABAergic motor neurons (data not shown), as also reported for C. elegans cholinergic motor neurons (Hu et al., 2013).As previously observed for other presynaptic proteins (Fig. 3; Hallam and Jin, 1998;Petersen et al., 2011;Thompson-Peer et al., 2012;Miller-Fleming et al., 2016), UNC-13L::GFP is restricted to the ventral nerve cord before DD remodeling in early L1 larvae, but is detectable postremodeling in both the dorsal and ventral nerve cords in adults (Fig. 7A,B).This finding indicates that UNC-13L::GFP remodels to DD presynaptic domains in the dorsal nerve cord and is also a component of ventral VD synapses in the adult.We quantified the number of UNC-13L::GFP puncta in the ventral nerve cord and determined that UNC-13L::GFP is largely removed in unc-55 mutants (Fig. 7D,G) thus demonstrating that UNC-13L::GFP is disassembled from the ventral presynaptic domains of DD neuron and also VD neurons that undergo remodeling in unc-55 mutants.In contrast to other presynaptic markers (e.g., SNB-1:: GFP; Fig. 3A-C), UNC-13L::GFP is also eliminated from ventral GABAergic synapses in unc-55; unc-8 mutants (Fig. 7E,G).Thus, wild-type UNC-8 activity is not required for the removal of UNC-13L from remodeling GABAergic synapses.This finding suggests that although the ventral presynaptic active zone in unc-55; unc-8 mutant GABAergic neurons appears normal by EM (Fig. 5A-D ;Miller-Fleming et al., 2016), UNC-13L is not localized at these terminals thus likely accounting for their SV fusion defect (Fig. 5E-G).
Since UNC-8 expression in VD neurons was shown to drive elimination of SNB-1::GFP (Miller-Fleming et al., 2016), we devised an additional experiment to test the idea that removal of UNC-13L is UNC-8 independent.We confirmed that forced expression of UNC-8 in VD neurons is sufficient to remove SNB-1::GFP from ventral GABAergic synapses (Fig. 7H,I), but does not displace UNC-13L::GFP (Fig. 7J,K).Together, these results show that UNC-8 function is neither necessary nor sufficient for UNC-13L removal from remodeling GABAergic synapses.
In contrast to results obtained from unc-55; unc-8 mutants (Fig. 8D,G), a substantial number of ELKS-1 puncta are retained in the ventral nerve cord of unc-55 animals treated with irx-1 RNAi (Fig. 8F,G).This finding suggests that IRX-1/Iroquois promotes the removal of ELKS-1 in remodeling GABAergic neurons but that UNC-8 is not required.For a direct test of this model, we used a transgenic strategy to overexpress UNC-8 in VD neurons (Fig. 8H,I).Although UNC-8(OE) in VD neurons drives the removal of ventral SNB-1::GFP (Fig. 7H,I), we observed that UNC-8 (OE) is not sufficient to eliminate ELKS-1 (Fig. 8J,K).Thus, our results support the idea that Iroquois/IRX-1 removes ELKS-1 from ventral synapses of remodeling GABAergic motor neurons in a genetic pathway that is independent of UNC-8 (Fig. 8L,M).
Iroquois/IRX-1 and DEG/ENaC/UNC-8 drive the removal of RAB-3 from remodeling DD neuron GABAergic synapses We have exploited the ectopic remodeling phenotype of unc-55 mutant VD neurons to show that both IRX-1 and UNC-8 act to dismantle the presynaptic apparatus.In addition, our findings suggest that IRX-1 functions as a transcription factor to orchestrate the overall mechanism by activating expression of UNC-8 as well as another downstream pathway that functions in parallel to drive synapse removal (Fig. 8L).To determine whether a similar mechanism also drives synapse elimination in the native DD remodeling program, we used an endogenous GFP::RAB-3 marker that is specifically expressed in DD neurons (Fig. 9A).As expected, in the wild-type, endogenous RAB-3::GFP remodels from ventral to dorsal locations during early larval development; few ventral GFP::RAB-3 puncta are detectable by the L4 larval stage (Fig. 9B).In contrast, unc-8 mutants at the L4 stage show significant retention of GFP::RAB-3 signal (Fig. 9C), likely resulting from failed synapse elimination (Miller-Fleming et al., 2016).Similarly, csRNAi of irx-1 in DD neurons prevents the elimination of GFP::RAB-3 (Fig. 9D-F), indicating that IRX-1 is also necessary for the efficient removal of presynaptic GFP::RAB-3 in DD neurons (Petersen et al., 2011).Notably, significantly more GFP::RAB-3 puncta are retained in irx-1(csRNAi)-treated DD neurons than in unc-8 mutants (Fig. 9F), a result consistent with the idea that IRX-1 drives expression of both unc-8 and an additional pathway for synapse elimination.This model of parallelacting pathways predicts that irx-1(csRNAi) should enhance the synaptic removal defect of unc-8 mutants, which we also observe (Fig. 9E,F).Finally, genetic ablation of unc-8 does not enhance retention of residual GFP::RAB-3 by irx-1(csRNAi) (Fig. 9F), a result consistent with the idea the IRX-1 is required for unc-8 expression.
Iroquois/IRX-1, but not DEG/ENaC/UNC-8, removes ELKS-1 from ventral synapses in DD neurons Our results obtained from ectopic remodeling of VD neurons in unc-55 mutants showed that IRX-1, but not UNC-8, promotes removal of a specific subset of active zone proteins, ELKS-1 and 8).To determine whether ELKS-1 is similarly regulated in remodeling DD neurons, we used an endogenous GFP::ELKS-1 marker that is selectively expressed in DD neurons (Fig. 10A).In the wild-type, GFP::ELKS-1 is initially deposited at ventral DD synapses and then relocated to dorsal DD synapses as predicted for the DD remodeling program (Fig. 10B).GFP:: ELKS-1 is also removed from ventral DD synapses of unc-8 mutants (Fig. 10C,D), suggesting that UNC-8 is not required for ELKS-1 elimination.In contrast, we observed that irx-1 knockdown by irx-1(csRNAi) antagonizes the elimination of GFP:: ELKS-1 from ventral synapses of DD neurons (Fig. 10E,F).Thus, our results demonstrate that IRX-1 but not UNC-8 drives ELKS-1 removal in DD neurons.Because Munc13/UNC-13 is also selectively removed by IRX-1 in ectopically remodeling VD neurons (Fig. 7), we propose that Munc13/UNC-13 is similarly regulated in remodeling DD neurons (Fig. 10G).Our findings are notable because they show that distinct genetic pathways can dismantle the presynaptic apparatus in remodeling GABAergic neurons by targeting specific active zone components./ p = 0.723.Paired t test.All images from L4 animals, anterior to left.L, Working model: IRX-1 promotes an unc-8-independent pathway involving unknown downstream components (?) that removes UNC-13L from remodeling GABAergic synapses.M, IRX-1 knock-down blocks remodeling of GABAergic synapses which retain presynaptic structural components (green), RIM/UNC-10 (blue), and Munc13 (magenta) allowing docked vesicles to prime and fuse with the plasma membrane.Also depicted are the plasma-membrane SNAREs (gray) because these are required for functional synapses, which are removed by IRX-1 (Fig. 5G).

Activity-dependent active zone remodeling
The active zone region of the presynaptic terminal mediates SV fusion for neurotransmitter release (Südhof, 2012).This active zone function is defined by a core group of components including VGCCs, ELKS, Munc13/UNC-13, liprin-a/SYD-2, SYD-1, RIM/UNC-10, and RBP (Südhof, 2012).Notably, the composition and size of the SV release machinery can be modulated by synaptic activity.For example, additional copies of specific active zone proteins (i.e., ELKS, RBP, VGCCs, and Munc13) are incorporated into the presynaptic zones of Drosophila neuromuscular junctions (NMJs) in a homeostatic mechanism that elevates neurotransmitter release to compensate for reduced postsynaptic sensitivity (Böhme et al., 2019;Gratz et al., 2019).Elevated activity in Drosophila photoreceptors can also have the opposite effect of selectively removing a subset of these presynaptic proteins (liprin-a, RIM, and RBP) while leaving others intact (VGCCs and SYD-1) to adapt the synapse to different sensory inputs (Sugie et al., 2015).
Our findings point to a related effect in remodeling GABAergic neurons in C. elegans in which neuronal activity promotes the elimination of selected presynaptic components.In previous work, we determined that the DEG/ENaC channel UNC-8 functions in an activity-dependent pathway that dismantles the presynaptic active zone.Genetic results, for example, show that UNC-8 acts in a common pathway with the VGCC, UNC-2 (Miller-Fleming et al., 2016).Thus, we propose here that UNC-8 drives the removal of a core group of presynaptic proteins including; Synaptobrevin/SNB-1, liprin-a/SYD-2, Endophilin/UNC-57, and RAB-3, which depends on GABA neuron activity and cytoplasmic calcium.In contrast, synaptic elimination of Munc13 and ELKS does not require UNC-8 and is selectively regulated in a separate pathway driven by the transcription factor Iroquois/IRX-1.Additionally, our studies show that IRX-1 drives disassembly of the same core group of presynaptic proteins (Synaptobrevin/ SNB-1, liprin-a/SYD-2, Endophilin/UNC-57, RAB-3; Fig. 11).Although UNC-8 is a key downstream effector for this mechanism, our genetic evidence also indicates that Iroquois/IRX-1 must regulate at least one additional downstream gene to remove this core group of components from the presynaptic region (Figs. 4,9).Future studies are needed to define the additional downstream IRX-1 effectors that drive presynaptic disassembly.These IRX-1 targets could emerge from previously defined datasets of genes regulated by the UNC-55/COUP-TF transcription factor (Petersen et al., 2011;Yu et al., 2017) since UNC-55 controls IRX-1 expression in VD GABAergic neurons (Fig. 11;Petersen et al., 2011;He et al., 2015).
Presynaptic domains are remodeled within intact axons We have described an example of activity-dependent circuit refinement in C. elegans in which presynaptic termini are eliminated in a mechanism that does not perturb axonal morphology (White et al., 1978).Presynaptic domains are also selectively dismantled from intact axons in activity-dependent mechanisms that sculpt the developing mammalian visual circuit.Initially, retinal ganglion cells (RGCs) extend exuberant axonal projections to the lateral geniculate nucleus.Later, RGC inputs to each geniculate neuron are reduced.During this period, axonal pruning for at least one class of RGCs (BD-RGCs) is preceded by the internal reorganization of presynaptic boutons which are eliminated in distal axonal regions and simultaneously assembled in proximal locations (Hong et al., 2014).Axons denuded of presynaptic domains are then retracted in a later phase of refinement (Hong and Chen, 2011).Inputs to RGCs from rod bipolar cells (BCs) in the retina are also eliminated from stable axonaldendritic contacts during development (Morgan et al., 2011).Thus, the reorganization of presynaptic domains within intact RGC and BC axons is similar to the remodeling mechanism in C. elegans GABAergic neurons in which the presynaptic apparatus is dismantled without visible alterations in axonal morphology (White et al., 1978;Hallam and Jin, 1998).Presynaptic boutons are also actively assembled as well as removed within intact axonal processes in the adult brain (De Paola et al., 2006;Stettler et al., 2006).Notably, key components involved in presynaptic remodeling in C. elegans GABAergic neurons are highly conserved (Fig. 11).Together, these findings suggest that the molecular pathways that control presynaptic remodeling in C. elegans may also regulate circuit refinement and plasticity in mammals.
Remodeling and "silent" synapses Our work has revealed a synaptic remodeling mechanism that disables neurotransmitter release while leaving SVs and the presynaptic density intact.We showed, for example, that proteins with essential roles in SV priming, Munc13/UNC-13 and ELKS, can be selectively removed from the presynaptic apparatus in a genetic background that preserves the ultrastructural integrity of the active zone (Fig. 4A); thus, effectively "silencing" an otherwise normal appearing synapse (Fig. 5C).A potentially related phenomenon of synaptic silencing has been reported in the auditory circuit of the barn owl.Juvenile owls fitted with optical prisms learn to associate auditory cues with a new imposed visual location (Knudsen and Knudsen, 1989).Adaptation in this case involves innervation of a new midbrain region in the auditory localization circuit.Synaptic boutons are also maintained, however, in the nearby anatomic domain in which object association normally occurs (Mcbride et al., 2008) which could account for the restoration of normal responses to auditory cues in adult owls after the training prisms are removed (Mcbride and Debello, 2015).The retention of these inactive synaptic structures could correspond to more broadly observed "learning traces" that facilitate the ready reacquisition of quiescent behavioral responses (Knudsen, 2002).We thus suggest that the elucidation of mechanisms that disable synapses by removing specific functional components could reveal the molecular underpinning of presynaptic silencing mechanisms with key roles in learning and memory circuits.

Figure 1 .
Figure 1.A transcriptional program regulates GABAergic neuron synaptic remodeling.A, DD motor neurons innervate ventral muscles in early L1 stage larvae.GFP-tagged synaptobrevin (SNB-1::GFP; green puncta) marks GABAergic presynaptic domains.B, DD synapses are relocated to the dorsal nerve cord during early larval development as postembryonic VD class GABAergic motor neurons are generated to synapse with ventral muscles.These DD and VD connections are maintained in the adult motor circuit.C, In unc-55 mutants, both DD and VD presynaptic domains are relocated to the dorsal nerve cord.D, RNAi knock-down of the Iroquois family homeodomain transcription factor, IRX-1, antagonizes GABAergic neuron synaptic remodeling in unc-55 mutants(Petersen et al., 2011).E, Mutations that disable the DEG/ENaC cation channel, UNC-8, impair the removal of DD and VD GABAergic presynaptic domains in unc-55 mutants(Miller- Fleming et al., 2016).F, IRX-1 and UNC-8 are normally expressed in DD neurons to drive disassembly of the presynaptic apparatus.The COUP-TF transcription factor, UNC-55, blocks expression of IRX-1 and UNC-8 in VD neurons to prevent synapse elimination.

Figure 6 .
Figure6.A behavioral assay for functional GABAergic synapses in the motor circuit.Behavioral assays to detect backward locomotion.Young adult animals were tapped on the head and scored for wild type (gray) versus uncoordinated (teal) backward movement.unc-55 and unc-55; unc-8 animals coil ventrally with head tap indicating that a loss-of-function unc-8 mutation does not rescue backward locomotion in unc-55 mutants (NS, not significant; Fisher's exact test, n ! 100 animals per genotype).csRNAi knock-down of irx-1 restores backward locomotion to unc-55 animals and this effect is not enhanced in unc-55; unc-8; irx-1(csRNAi); * significantly different from wild type with p , 0.002; # significantly different from unc-55 with p , 0.002; 1 significantly different from unc-55; unc-8 with p , 0.002.

Figure 11 .
Figure11.Parallel-acting pathways dismantle the presynaptic apparatus in remodeling GABAergic neurons.A, top, Before remodeling, presynaptic proteins SNAREs, Rab3, RIM, endophilin, liprin-a, Munc13, and ELKS mediate SV fusion and neurotransmitter release on the ventral side of GABAergic neurons.B, After remodeling, presynaptic markers except RIM are removed from the ventral nerve cord.C, Synaptobrevin, Rab3, RIM, endophilin, and liprin-a are retained at ventral presynaptic regions of GABAergic neurons in unc-8 mutants.Note that unc-8 activity is not required for removal of ELKS and Munc13.D, With RNAi knock-down of the Iroquois/IRX-1 transcription factor, Synaptobrevin, Rab3, RIM, endophilin, liprin-a, Munc13, and ELKS persist in the ventral nerve cord to mediate GABA release from functional terminals.Plasma-membrane SNAREs are depicted because these synapses are functional.E, Transcriptional regulation of parallelacting pathways that drive presynaptic disassembly.In DD neurons (left), the transcription factor IRX-1/Iroquois activates expression of at least two downstream targets.(1) The DEG/ENaC channel subunit, UNC-8, which promotes removal of Synaptobrevin, endophilin, Rab3 and liprin-a but not the active zone proteins Munc13 or ELKS.(2) A second pathway (?), that promotes removal of Synaptobrevin, endophilin, Rab3 and liprin-a as well as Munc13 and ELKS.In VD neurons (right), the COUP-TF transcription factor UNC-55 blocks expression of IRX-1 thereby preventing the elimination of ventral presynaptic terminals.

Table 1 .
Mutant alleles and genotyping primers used in this study

Table 2 .
Strains used in this study