Transcriptional control of parallel-acting pathways that remove discrete 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 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 C. elegans (Miller-Fleming et al., 2016). Here, we report that the 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 fusion and neurotransmitter release. Our findings are notable because they highlight the key role of transcriptional regulation in synapse elimination and reveal parallel-acting pathways that orchestrate synaptic disassembly by removing specific active zone proteins.


INTRODUCTION
The nervous system is actively remodeled during development as new synapses are assembled and others are removed to refine functional circuits. 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 these developmental windows for activity-induced remodeling. Thus, synaptic remodeling mechanisms are likely to depend on the combined effects of both transcriptionally-regulated and activity-dependent pathways (Hensch 2004;Kano and Watanabe 2019).
In the nematode, C. elegans, synapses in the GABAergic motor circuit are relocated by a stereotypical remodeling program during early larval development. Dorsal D (DD) GABAergic motor neurons are generated in the embryo and initially synapse with ventral body muscles ( Figure 1A). During the first larval stage, presynaptic domains are removed from ventral DD processes and then reassembled in the dorsal nerve cord ( Figure 1B) (White, Albertson, and Anness 1978). Postembryonic Ventral D (VD) neurons are born during this early larval period and synapse exclusively with ventral muscles ( Figure 1B). In the resultant mature circuit, alternating GABAergic output to dorsal (DD) versus ventral (VD) muscles is required 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; 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 ( Figure 1C) (Shan et al. 2005;Zhou and Walthall 1998). The idea that UNC-55 blocks expression of genes that normally drive synaptic remodeling is supported by the finding that forced expression of UNC-55 in DD neurons antagonizes the native remodeling program (Shan et al. 2005). In earlier work, we exploited the ectopic 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, RNAi knockdown of the Iroquois homeodomain transcription factor, IRX-1, reduced removal of GABAergic presynaptic domains from the ventral nerve cord in unc-55 mutants ( Figure   1D) (Petersen et al. 2011). Similarly, a loss-of-function allele of the DEG/ENaC cation channel subunit gene, unc-8, also antagonized synaptic remodeling in unc-55 mutants ( Figure 1E).
Additional experiments confirmed that both the IRX-1 transcription factor and UNC-8 cation channel normally promote the native DD remodeling program ( Figure 1F) ).
DEG/ENaC proteins function as cation channels and we have previously shown that UNC-8 gates sodium influx (Matthewman et al. 2016;Wang et al. 2013). UNC-8 promotes presynaptic disassembly in a pathway that depends on intracellular calcium and neural activity ). Here we show that DEG/ENaC/UNC-8 is transcriptionally-regulated by Iroquois/IRX-1 to remove the presynaptic components liprin-a/SYD-2, Synaptobrevin/SNB-1, RAB-3 and Endophilin/UNC-57. Our findings indicate that these presynaptic proteins are also disassembled by a separate Iroquois/IRX-1-dependent pathway that functions in parallel to DEG/ENaC/UNC-8.
Thus, remodeling of GABAergic synapses depends on the combined effects of neural activity (UNC-8) and developmentally-regulated transcription (IRX-1). Finally, we show that the active zone proteins Munc13/UNC-13 and ELKS-1 are exclusively removed by Iroquois/IRX-1, but not UNC-8. Thus, our work suggests that synaptic disassembly in the GABAergic circuit is orchestrated by parallel-acting mechanisms that selectively target molecularly distinct components of the presynaptic apparatus during the remodeling period.

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). Because these studies determined that two of these proteins, the homeobox transcription factor IRX-1/Iroquois and the DEG/ENaC ion channel subunit UNC-8 are both involved in removing the presynaptic vesicle SNARE protein, synaptobrevin/SNB-1 Petersen et al. 2011), we decided to test the hypothesis that Iroquois/IRX-1 regulates DEG/ENaC/UNC-8 expression.
We have previously shown that GFP reporter lines for the irx-1 and unc-8 genes are highly expressed in DD motor neurons but not in VD motor neurons in a wild-type background Petersen et al. 2011) (Figure 1). Here, we used single-molecule Fluorescent In-Situ Hybridization (smFISH) to detect unc-8 transcripts in remodeling DD neurons and to determine if the irx-1 gene is necessary for unc-8 expression. DD neurons in which irx-1 is targeted by cell-specific RNAi (csRNAi) (See Methods) showed significantly fewer unc-8 transcripts in comparison to the wildtype (Figure 2A-B), suggesting that Iroquois/IRX-1 is required for DEG/ENaC/UNC-8 expression in DD neurons. Because forced expression of Iroquois/IRX-1 in VD motor neurons is sufficient to drive the elimination of VD presynaptic terminals (Petersen et al. 2011), we next asked if Iroquois/IRX-1 over-expression (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 wildtype ( Figure 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 ( Figure 2E). Iroquois/IRX-1 drives a DEG/ENaC/UNC-8-dependent mechanism of presynaptic disassembly as well as a separate parallel-acting pathway that does not require UNC-8 for synaptic removal.
We have previously shown that Iroquois/IRX-1 promotes the removal of Synaptobrevin/SNB-1::GFP from ventral presynaptic domains in remodeling GABAergic neurons ( Figure 1E) (Petersen et al. 2011). If Iroquois/IRX-1 activates UNC-8 expression as predicted by our smFISH results (Figure 2), then Iroquois/IRX-1 should also drive removal of the additional presynaptic markers liprin-a/SYD-2, endophilin/UNC-57 and RAB-3/GTPase which depends on UNC-8 function . To test for this possibility, we exploited unc-55 mutants in which the VD GABAergic presynaptic domains are eliminated due to ectopic activation of the native DD remodeling program (Zhou and Walthall 1998) ( Figure 1C). In this paradigm, removal of ventral GABAergic synapses in unc-55 mutants is prevented by mutations that disable the pro-remodeling program. For example, ventral SNB-1::GFP puncta are eliminated in unc-55 mutant animals but a significant fraction is retained in unc-55; unc-8 double mutants (Figures 1E and 3A). This result confirms our earlier finding that UNC-8 function is required for the efficient removal of presynaptic SNB-1::GFP in remodeling GABAergic neurons . Similarly, RNAi knockdown of irx-1 prevents the elimination of ventral SNB-1::GFP puncta in unc-55 worms (Figures 1D & 3A) (Petersen et al. 2011). Additional experiments showed that ablation of either unc-8 or irx-1 also blocks the removal of SYD-2::GFP and RAB-3::mCherry ( Figure 3B-C) in the ventral nerve cord of unc-55 mutants. Since Iroquois/IRX-1 induces unc-8 expression (Figure 2), these results are consistent with the hypothesis that Iroquois/IRX-1 drives an UNC-8-dependent mechanism to remove presynaptic terminals in GABAergic neurons.
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. This idea is consistent with our finding that wild-type levels of presynaptic components are not fully restored in unc-55; unc-8 double mutants thus pointing to an additional parallel acting mechanism for synaptic disassembly . We reasoned that if Iroquois/IRX-1 regulates a downstream target that functions in tandem with UNC-8, then genetic ablation of irx-1 in an unc-55; unc-8 double mutant should enhance the retention of ventral presynaptic markers in comparison to unc-55; unc-8 mutants. For this test, we used feeding RNAi for global knockdown of irx-1 because the irx-1 null allele is lethal (Petersen et al. 2011). We counted GFP puncta for the presynaptic proteins SNB-1::GFP, SYD-2::GFP and UNC-57::GFP in irx-1-  These experiments showed that RNAi knockdown of irx-1 increases the number of ventral SNB-1::GFP, SYD-2::GFP and UNC-57::GFP puncta ( Figure 3D-F) in unc-55; unc-8 double mutant animals. Together, these results demonstrate that Iroquois/IRX-1 drives an additional genetic pathway, independent of UNC-8, that eliminates presynaptic terminals in remodeling GABAergic neurons ( Figure 3G).

Iroquois/IRX-1, but not DEG/ENaC/UNC-8, removes fusion-competent synaptic vesicles in remodeling GABAergic neurons.
The retention of presynaptic markers (e.g., SNB-1::GFP, SYD-2::GFP) ( Figure 3D-F) in ventral GABA motor neuron processes of unc-55; unc-8 double mutants suggested that a reconstituted presynaptic apparatus in these animals might also restore synaptic release. We previously used electron microscopy (EM) to confirm the presence of ventral cord GABAergic synapses in unc-55; unc-8 adults ( Figure 4A) . In contrast, EM analysis did not detect GABAergic ventral cord synapses in unc-55 animals as expected since both DD and VD motor neuron synapses are relocated to the dorsal nerve cord in unc-55 mutants Walthall and Plunkett 1995). Since our assays with fluorescent presynaptic markers also showed that Iroquois/IRX-1 drives presynaptic disassembly (Figure 3), we used EM to ask if cell-specific RNAi (csRNAi) knockdown of irx-1 would prevent the removal of ventral GABAergic synapses in an unc-55 mutant. These experiments detected GABAergic presynaptic terminals in unc-55; irx-1(csRNAi) samples ( Figure 4A) (7 synapses/613 sections, n = 3 animals), thus, confirming that IRX-1 is necessary for the removal of synapses in remodeling GABA neurons (See Methods). Together, these EM data are consistent with our findings that ventral puncta for fluorescently-labeled SNB-1, SYD-2, RAB-3 and UNC-57 are absent in unc-55 mutants, but that knockdown of either unc-8 or irx-1 activity partially rescues these defects ( Figure 3A-C).
Our EM results determined that genetic ablation of unc-8 restores the ultrastructural morphology of some ventral GABAergic synapses to unc-55 animals ( Figure 4A), including the presence of presynaptic dense projections, synaptic vesicles and plasma-membrane docked vesicles . As a first test of the possibility that these synapses are functional, we used a behavioral readout that depends on ventral GABAergic synapses. Ventral synapses for both DD and VD neurons are dismantled in unc-55 mutants and reassembled in the dorsal nerve cord ( Figure 1D). The resultant imbalance due to 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 (Shan et al. 2005;Walthall and Plunkett 1995). We have shown that mutants that disable pro-remodeling genes (e.g., unc-8, irx-1) restore ventral GABAergic synapses to unc-55 mutants ( Figure 1C). If these restored synapses are functional, then the tapping assay should detect improved backward locomotion in double mutants with unc-55. In this assay, each animal is tapped gently on the head to induce backward movement. We determined, however, that unc-55; unc-8 double mutant animals display severely defective backward locomotion that is not significantly different from that of unc-55 mutants ( Figures 4B &   4S1). This finding suggests that that ventral GABAergic synapses in unc-55; unc-8 mutants are not functional. In contrast, unc-55; irx-1(csRNAi) animals ( Figure 4B) show robust backward movement in comparison to unc-55 mutants. This result suggests that GABAergic release is restored to ventral cord synapses by RNAi knockdown of irx-1 and thus that Iroquois/IRX-1 is required for the removal of functional GABAergic synapses in unc-55 mutants. ( Figure 4B,C) (Petersen et al. 2011). Genetic ablation of unc-8 activity in this genetic background (i.e.,) does not further enhance backward locomotion as predicted by our conclusion that residual ventral cord GABAerigic synapses in unc-55; unc-8 double mutants are not functional ( Figure 4B) and by our finding that Iroquois/IRX-1 regulates expression of the unc-8 gene (Figure 2). To summarize, the results of the movement assay suggest that although ventral GABAergic synapses are visible by  animals, GABAergic release is selectively reactivated by knockdown of irx-1, but not by genetic removal of unc-8.
To investigate the mechanism of this differential effect on synaptic function, we quantified the number and distribution of docked synaptic vesicles in unc-55; unc-8 and unc-55; irx-1(csRNAi) double mutants. EM analysis determined that  animals show similar numbers of docked synaptic vesicles in ventral cord GABAergic synapses ( Figure 4S2). This finding argues against a model in which synaptic dysfunction in unc-55; unc-8 mutants can be attributed to the failure of synaptic vesicles to contact the presynaptic membrane prior to neurotransmitter release.
We next examined whether the docked synaptic vesicles in unc-55; unc-8 animals are fusion-competent by recording iPSCs from ventral muscles using d-tubocurare (dTBC) to block cholinergic signaling. As previously reported, tonic release of ventral iPSCs was restored in unc-55; irx-1(csRNAi) animals, but this was not observed in unc-55; unc-8 mutants despite the presence of organized clusters of fluorescent presynaptic proteins (Figure 3 prior to DD remodeling in early L1 larvae, but is detectable post-remodeling in both the dorsal and ventral nerve cords in adults ( Figure 5S1B-C). 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 ( Figure 5A). In contrast to other presynaptic markers (e.g. SNB-1::GFP) ( Figure 3A-C), UNC-13L::GFP is also eliminated from ventral GABAergic synapses in unc-55; unc-8 mutants ( Figure 5A). Thus, wildtype 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-2016), UNC-13L is not localized at these terminals thus likely accounting for their synaptic vesicle fusion defect ( Figure 4C-D).
Since UNC-8 expression in VD neurons was shown to drive elimination of SNB-1::GFP , we devised an 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, but does not displace UNC-13L::GFP ( Figure 5S3). Together, these results show that UNC-8 function is neither necessary nor sufficient for UNC-13L removal from remodeling GABAergic synapses.
RNAi knock down of irx-1 in unc-55 mutants is sufficient to restore ventral GABAergic synaptic release ( Figure 4C-D). Thus, we next asked if UNC-13L::GFP is retained in the ventral nerve cord of unc-55; irx-1(csRNAi) animals. Indeed, we found that ventral UNC-13L::GFP puncta are detectable in both wild-type and in unc-55; irx-1(csRNAi) animals ( Figure 5A) thus indicating that Iroquois/IRX-1 function is required for the removal of UNC-13L from remodeling GABAergic synapses.
Based on previous work demonstrating that the RIM-binding protein ELKS-1 recruits the mammalian protein bMunc-13-2 to active zones and Drosophila ELKS homologue, Bruchpilot, recruits UNC-13L/Unc13A (Böhme et al. 2016;Kawabe et al. 2017), we next asked if ELKS-1 synaptic localization is differentially affected by the UNC-8 versus IRX-1-dependent synaptic remodeling pathways. We determined that expression of ELKS-1::tdTomato (Cherra and Jin 2016) in wild-type adult GABA neurons results in bright fluorescent puncta characteristic of DD synapses in the dorsal nerve cord and VD synapses in the ventral nerve cord ( Figure 5S2). Ventral ELKS-1::tdTomato-labeled puncta are largely absent in unc-55 mutants indicating that ELKS-1 is dismantled from the presynaptic domains of remodeling GABAergic motor neurons ( Figure 5B).
As an additional test of this idea, we expressed ELKS-1::GFP and SYD-2::mCherry in DD neurons for simultaneous dual color imaging. This experiment confirmed that more SYD-2::mCherry puncta are retained in the ventral nerve cords of adult unc-8 mutants than wildtype, thus confirming that unc-8 normally promotes the removal of SYD-2 from ventral DD synapses in the native remodeling program ( Figure 6). In contrast, few ventral DD ELKS-1::GFP are retained in either wild-type or unc-8 mutant adults, an observation consistent with our finding above ( Figure   5B) that UNC-8 function is not required for the removal of ELKS-1 from GABAergic synapses that remodel in unc-55 mutants. Overall, our results demonstrate that subsets of active zone proteins are targeted for removal by separate mechanisms that disassemble the presynaptic apparatus in remodeling GABAergic neurons (Figure 7).

Presynaptic disassembly in remodeling in GABAergic neurons
Synaptic plasticity is a key dynamic feature of the nervous system as neurons actively assemble new synapses while dismantling others (Chen et al. 2011;Nishiyama et al. 2007;De Paola et al. 2006). In contrast to synaptic assembly about which much is known, the molecular mechanisms that drive synaptic elimination are relatively unexplored (T. Südhof 2018; T. C. Südhof 2017). In this study, we investigated a developmentally-regulated mechanism of presynaptic disassembly in C. elegans (White, Albertson, and Anness 1978). Our findings revealed parallel-acting pathways that selectively remove different components of the presynaptic active zone in remodeling GABAergic synapses (Figure 7). We have shown that the conserved transcription factor Iroquois/IRX-1 drives expression of the DEG/ENaC channel subunit UNC-8 ( Figure 2) to remove the presynaptic proteins Synaptobrevin/SNB-, liprin-a/SYD-2, Endophilin/UNC-57 and RAB-3 ( Figure 3A-C). IRX-1 regulates an additional parallel-acting pathway that also removes these presynaptic components ( Figure 3D-G). In addition, Iroquois/IRX-1 promotes the disassembly of ELKS-1 and Munc13/UNC-13 in a separate pathway that does not require UNC-8 activity ( Figure 4, 5, 6 and 7). Together, our findings support the conclusion that synaptic disassembly can be transcriptionally-regulated and involve molecularly distinct mechanisms that differentially eliminate selected subsets of presynaptic proteins.

Activity-dependent active zone remodeling
The active zone region of the presynaptic terminal mediates synaptic vesicle (SV) fusion for neurotransmitter release (T. C. Südhof 2012). This active zone function is defined by a core group of components including Voltage-gated Ca ++ Channels (VGCCs), ELKS, Munc13/UNC-13, liprina/SYD-2, SYD-1, RIM/UNC-10, and Rim Binding Protein (RBP) (T. C. Südhof 2012). Notably, the composition and size of the SV release machinery can by 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) (Sugei et al. 2015). Our findings point to a parallel 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 activitydependent 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 .
Thus, we propose here that UNC-8-driven removal of Synaptobrevin/SNB-1, liprin-a/SYD-2, Endophilin/UNC-57 and RAB-3 also depends on GABA neuron activity and cytoplasmic calcium whereas synaptic elimination of Munc13 and ELKS, which does not require UNC-8, is selectively regulated in a separate pathway driven by the transcription factor Iroquois/IRX-1 (Figure 7).
Future studies are needed to define the downstream IRX-1 effectors that promote Munc13 and ELKS elimination. Because Iroquois/IRX-1 expression and synaptic remodeling in GABAergic neurons are blocked by the transcription factor UNC-55 (He et al. 2015;Petersen et al. 2011;Zhou and Walthall 1998), molecular regulators of Munc13 and ELKS elimination may be included in previously defined data sets of UNC-55-regulated genes (Petersen et al. 2011;Yu et al. 2017).

ACKOWLEDGEMENTS
We thank K. Shen for the ELKS-1::GFP and SYD-2::mCherry markers used in Figure 6 CryoCluster equipment, which has received support from the MRI program (NSF DMR-1229693).
Imaging experiments were performed in part in the Vanderbilt Cell Imaging Shared Resource (supported by NIH grants CA68485, DK20593, DK58404, DK59637 and EY08126). This work was supported by NIH grants R01NS081259, R01NS106951(DMM) and predoctoral fellowships from the NIH to TWFM (1F31NS084732), NSF to SP (DGE:1445197) and AHA to SP (19PRE34380582) and ACC (18PRE33960581).

DECLARATION OF INTERESTS
The authors declare no financial interests.

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 and 2.

Confocal Microscopy
Larval or young adult animals were immobilized on 2% agarose pads with 15mM levamisole as previously described. Z-stack images ( Figure 3A-C, 5, 5S1B-C) were collected on a Leica TCS SP5 confocal microscope using a 63X oil objective (0.5 μm/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 ( Figure 3D-F, 4S3, 5S1A, 5S2 and 5S3) were acquired with Nikon confocal A1R using 40X/1.3 and 60X/1.4 N.A. oil objective (0.5 μm/step).

Image Analysis
Synapse density counts (Figures 3A-C and 5) were collected by tracing segments of the ventral nerve cord using the segmented line tool in ImageJ. Distance in micrometers and gray value plot traces were used to count the number of peaks (synapses) that occur over the specified distance.
Synapses were defined as fluorescent peaks that reached a threshold of 25 arbitrary units of fluorescence intensity. NIS Elements 5.2 software was used to produce Figures 2, 3D-F, 4S3 and 5S1. Synaptic density for each marker (UNC-10::GFP, SYD-2::GFP, RAB-3::mCherry, UNC-57::GFP and SNB-1::GFP) was calculated using the General Analysis tool. First, images were preprocessed to subtract background using Rolling Ball Correction. Then, the intensity threshold was defined for each marker and binary objects were filtered by size and circularity. Each object along the nerve cord was considered a synaptic punctum. Density was defined as the number of puncta per 10 μm of dendrite.
FIJI was used to produce Figures 6 and 5S3. Line scans were drawn on the ventral cord anterior to the VD cell body of interest. Background fluorescence was obtained from a line scan from an adjacent region interior to the ventral cord. Average intensity was reported for both line scans and subtracted ( Figure 5S3). Puncta density was counted in the GFP and mCherry channel by a scorer blinded to the genotype ( Figure 6).

Single molecule mRNA Fluorescence In Situ Hybridization (smFISH)
smFISH was performed with custom unc-8 probes linked to Quasar® 670 (Biosearch Technologies). Synchronized larvae (from either late L1 or early L3 stage) were collected by washing plates with M9, fixed in 4% paraformaldehyde in 1X PBS for 45 min and permeabilized in 70% ethanol for 48 h. Hybridization followed the manufacturer's instructions (http://www.biosearchtech.com/stellarisprotocols) and was performed at 37°C for 16h in Stellaris RNA FISH hybridization buffer (Biosearch Technologies Cat# SMF-HB1-10) containing unc-8 probe at 1:100. For irx-1 cell specific RNAi (csRNAi) experiments, all DD motor neurons were marked with Punc-47::GFP (oxIs12) and specific DDs expressing the irx-1(csRNAi) constructs (pttr-39::irx-1 sense, pttr-39::irx-1 antisense) were co-labeled with Punc-25::mCherry to distinguish them from DD neurons that did not express the irx-1(csRNAi) transgenic array. For irx-promotor (pttr-39::IRX-1::GFP). In this setup, VDs and DDs were marked with Punc-47::mCherry (wpIs39) and individual DDs or VDs expressing irx-1(oe) were detected by expression of nuclearlocalized IRX-1::GFP (Petersen et al. 2011). In all cases, cell nuclei were stained with DAPI. Zstacks were collected in a Nikon spinning disk confocal microscope with optical filters for DAPI, Quasar® 670 and GFP using a 100X objective (NA=1.49) in 0.2 μm steps spanning the cell body and merged for quantification following 3D-deconvolution in Nikon elements. smFISH puncta were counted if they corresponded to circular fluorescent spots, exceeded the Quasar® 670 background signal and were located within either a GFP-labeled or mCherry-marked DD/VD cell body. At least 30 worms were scored for each group and the Mann-Whitney test used to determine significance (n >45 neurons). As a positive control, unc-8 smFISH staining was noted in adjacent DA and DB ventral-cord neurons for all samples to confirm successful hybridization.

Electron Microscopy
Young adult hermaphrodites of each strain were prepared for high-pressure freeze (HPF) fixation as described Rostaing et al. 2004). 10-15 animals were loaded into a specimen chamber filled with E. coli. The specimens were frozen rapidly in a high-pressure freezer (Leica HPM100) at -180°C and high pressure. Freeze substitution was performed on frozen samples in a Reichert AFS machine (Leica, Oberkochen, Germany) with 0.1% tannic acid and 2% OsO4 in anhydrous acetone. The temperature was kept at -90°C for 107 h, increased at 5°C/h to -20°C, and kept at -20°C for 14h. 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 4h, 90% Epon/acetone for 18h, and 100% Epon for 5 hours. Embedded samples were incubated for 48h at 65°C. All specimens were prepared in 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, carbon-coated 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 (Tokyo, Japan) transmission electron microscope operating at 80 kV. Micrographs were collected using a Gatan digital camera (Pleasanton, CA) at a magnification of 100k. Images were quantified using NIH ImageJ software. Dorsal and ventral cords were distinguished by size and morphology.

Electrophysiology
The C. elegans dissection and electrophysiological methods were as previously described . 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.

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/µl with the co-injection marker pmyo-2::mCherry (2 ng/ µl). 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 1 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 upon tapping), "initiate backing" (initiate backwards movement but stop), and "wild-type" (sustain backward locomotion with at least two body bends). In Figure 4S1, the "wild-type" and "initiate backing" categories were grouped into a single "initiate backing" category.