Differential regulation of polarized synaptic vesicle trafficking and synapse stability in neural circuit rewiring in Caenorhabditis elegans

Neural circuits are dynamic, with activity-dependent changes in synapse density and connectivity peaking during different phases of animal development. In C. elegans, young larvae form mature motor circuits through a dramatic switch in GABAergic neuron connectivity, by concomitant elimination of existing synapses and formation of new synapses that are maintained throughout adulthood. We have previously shown that an increase in microtubule dynamics during motor circuit rewiring facilitates new synapse formation. Here, we further investigate cellular control of circuit rewiring through the analysis of mutants obtained in a forward genetic screen. Using live imaging, we characterize novel mutations that alter cargo binding in the dynein motor complex and enhance anterograde synaptic vesicle movement during remodeling, providing in vivo evidence for the tug-of-war between kinesin and dynein in fast axonal transport. We also find that a casein kinase homolog, TTBK-3, inhibits stabilization of nascent synapses in their new locations, a previously unexplored facet of structural plasticity of synapses. Our study delineates temporally distinct signaling pathways that are required for effective neural circuit refinement.


Author summary
In this study, we identify pathways that regulate the formation and maintenance of synapses, the functional connections between neurons, in the nervous system of the nematode C. elegans. Our work characterizes the interaction between molecular motors kinesin and dynein, which carry cargo and move towards opposite ends of microtubules during synapse formation. We also address the role of a protein kinase gene TTBK-3 in maintaining synapse structure once synaptic components have reached the sites of new synapses. Our

Introduction
Neurons communicate through synapses, necessitating a system of checks and balances to achieve precise patterns of synaptic connectivity that execute neural circuit function. Large scale axonal growth and pruning mediate synapse formation with appropriate targets during development, shaping neuronal circuits during critical periods of plasticity [1]. Hyper-and hypo-connectivity in different brain regions is a widely observed phenomenon in children with autism spectrum disorders (ASDs) and related comorbid conditions [2]. Brain development defects during critical postnatal periods of plasticity are also thought to contribute to schizophrenia, which has a varying age of onset [3]. Structural synaptic plasticity is not purely a developmental phenomenon-synapse remodeling occurs in both normal and diseased adult brains in various contexts [4,5]. The mechanisms underlying synapse assembly and elimination have thus been the subject of intense study for several decades, although a majority of experimental models focused on synaptic plasticity that is coupled to neurite outgrowth and retraction [6,7]. With recent advances in in vivo imaging techniques, instances of synaptic rewiring that are independent of large scale neurite rearrangement have been identified in the mammalian central nervous system [4,8]. Elucidating the mechanisms underlying the cellular dynamics of such refinement, particularly in pre-synaptic terminals, is of general significance.
In the C. elegans locomotor circuit, a subset of type-D GABAergic motor neurons exhibit critical period synapse plasticity. Upon birth and in young larvae, the Dorsal D (DD) neurons initially form synapses with ventral body wall muscles. During an early developmental molt, these early synapses are disassembled, and new synapses are formed with dorsal body wall muscles, without overt changes in neuronal morphology ( Fig 1A) [9]. DD synapse remodeling is developmentally stereotyped, activity dependent, and uncoupled from neurite outgrowth, providing a tractable genetic framework to study the molecular mechanisms underlying structural synaptic plasticity. Numerous studies have provided insights into the conserved transcriptional programs that regulate the initiation of DD synapse remodeling and that maintain the temporal precision of synapse remodeling (reviewed in [10]).
Once circuit connectivity changes have been initiated in the DD neurons, the cellular execution of synapse assembly and disassembly takes place. Pre-synaptic terminals are eliminated from the DD ventral neurite, following which synaptic vesicles are transported to the DD dorsal neurite where they assemble to form new synapses that are stable for the lifetime of the animal. Previous work from our lab and others has found that dynamic microtubules (MTs) are required for synaptic vesicle transport to DD dorsal neurite during remodeling [11] and the patterning of new pre-synaptic terminals is achieved by the sequential action of anterograde and retrograde motors Kinesin-3/UNC-104 and dynein, respectively [12]. Synapse elimination from the DD ventral neurite is mediated in part by the cyclin Y homolog CYY-1 and the apoptotic cell death pathway [12,13].
In this study, we characterized multiple mutants isolated from a genetic screen for genes involved in DD synapse remodeling. We performed this screen on a mutant strain containing a gain-of-function (gf)mutation of alpha-tubulin tba-1, and a loss-of-function (0) mutation of the conserved MAPKKK dlk-1. This tba-1(gf)dlk-1(0) double mutant combination results in defective DD synapse remodeling due to a reduction in MT dynamics [11]. We identified mutations in the C. elegans α-and β-tubulin genes tba-1 and tbb-2 that reversed defects in MT architecture. We also show that novel mutations in the minus end directed motor dynein and its adaptor protein dynactin ameliorate defects in kinesin-mediated synaptic vesicle transport to the DD dorsal neurite during remodeling, highlighting the interdependence of the two motors even in cases of polarized cargo movement. We further find that a member of the casein kinase superfamily, TTBK-3, specifically acts after remodeling is complete to modulate nascent synapse stability on the dorsal neurite, a previously uncharacterized aspect of synaptic plasticity. Our observations indicate that the dynein motor complex and TTBK-3 act at distinct temporal windows to differentially regulate synapse rewiring.

Results
A forward genetic screen to identify novel regulators of DD synapse remodeling A missense mutation in C. elegans α-tubulin, tba-1(ju89) (henceforth tba-1(gf))results in a mild reduction in the synapse number of GABAergic motor neurons and a dampening of the amplitude of sinusoidal locomotion ( Fig 1B) [11,14]. In animals carrying both tba-1(gf)and a loss of function of the conserved MAPKKK DLK-1 (dlk-1(0)), DD remodeling is completely blocked such that the dorsal neurites contain almost no synapses, when visualized by the synaptic marker juIs1 (P unc-25 -SNB-1-GFP) (Fig 1B) [11]. Consistent with a lack of GABAergic innervation, tba-1(gf)dlk-1(0) double mutant animals are uncoordinated and coil dorsally when touched in the head (Fig 1C) [11]. We performed a suppressor screen on tba-1(gf)dlk-1 (0); juIs1 animals, first based on behavioral improvement, then by visual examination of synapses in DD neurons. We mapped the suppressor mutations to various genetic loci using whole genome sequencing and subsequent recombinant mapping (see methods). Below, we report the characterization of these mutations.
(ju964, ju966 and ju975) caused nonsense mutations at different amino acids of tba-1 (S1 Fig and S1 Table). This suggests that truncated versions of TBA-1 produced in these mutants were likely not functional. One missense mutation at the start codon ATG (ju980) also behaved similar to tba-1(0), with synapse formation and locomotion restored to normal in tba-1(ju980 ju89) dlk-1(0) animals ( Fig 1B, S1 Fig). The other four suppressors caused missense mutations at highly conserved regions of α-tubulin (S1 Fig and S1 Table). Two missense mutations (ju973 (L426F) and ju987 (A419T)) were in H12 of the C-terminal domain, and could possibly have reversed the MAP binding defects caused by tba-1(gf)(S1 Fig). Another missense mutation, ju965 (S138L), was close to the GTP binding pocket of α-tubulin [15] (S1 Fig). The last missense mutation, ju962 (S285F), was in the intermediate domain of α-tubulin, reported to be necessary for binding of the MT stabilizing drug, taxol ( Fig 1D and S1 Fig) [15]. Mutations in the intermediate domain or the GTP binding pocket possibly prevented the incorporation of tba-1 into MT polymers. While we cannot exclude the possibility that these mutations might also reduce TBA-1 protein levels, our results highlight the in vivo importance of specific residues within functional domains of α-tubulin.
tbb-2(ju1535) behaved as a neomorphic allele, as gk129, a null (0) mutation of tbb-2 did not cause overt defects in locomotion or synapse formation and also did not suppress the synapse remodeling defects of tba-1(gf)dlk-1(0) animals (S1 Fig, Fig 1D and 1E). tba-1(gf);tbb-2(0) animals were viable and displayed similar synapse formation defects to tba-1(gf)single mutant animals, whereas tba-1(0); tbb-2(0) animals were lethal due to their requirement in early embryonic development [14]. Taken together, these results suggest that while heterodimers of TBA-1 and TBB-2 were essential for embryonic development, in the absence of tbb-2, tba-1(gf) could form heterodimers with other β-tubulins that were then incorporated into MTs. tba-1 (gf)resulted in a change in MT architecture [11] and is a mutation altering the external surface of MTs, separate from domains responsible for GTP binding and heterodimer formation (S1 Fig) [14,15]. Thus, the suppression of tba-1(gf) by tbb-1(ju2535) may likely be through either modifying or reducing the incorporation of tba-1(gf);tbb-2(ju1535) heterodimers in MTs, in turn reducing the number of abnormal MTs.

Dynein-dynactin complex function is important for synapse remodeling
We mapped one of the suppressors, ju1279, to the C. elegans cytoplasmic dynein heavy chain, dhc-1. Cytoplasmic dynein is a large multi-subunit molecular motor that comprises of two catalytic heavy chains, as well as numerous light and intermediate chains. Dynein moves towards MT minus ends and is the primary motor involved in retrograde axonal transport, with mutations in dynein and its adaptor proteins implicated in several neurodegenerative disorders [21]. ju1279 converts a conserved proline residue to leucine (P262L) in the N-terminal region 1 of the tail domain of DHC-1 (Fig 2A and S2 Fig), which is responsible for dynein homodimerization and acts as a scaffold for subunit assembly [22]. dhc-1(ju1279) acted as a weak suppressor of tba-1(gf)dlk-1(0), since synapse remodeling was only partially restored in dhc-1(ju1279) tba-1(gf)dlk-1(0) triple mutant animals, with a significant reduction in dorsal neurite synapse number compared to tba-1(gf)animals (Fig 2B and 2C). We confirmed that dhc-1(ju1279) was causative by rescuing the suppression of tba-1(gf)dlk-1(0) using extra-chromosomal copies of a fosmid containing the full genomic region of DHC-1 (Fig 2B and 2C).
Another suppressor, ju993, changed valine 229 to isoleucine in the C. elegans p62 subunit of dynactin DNC-4 ( Fig 2D). Dynactin is a large multi-subunit protein complex that is essential for most cellular functions of cytoplasmic dynein, including MT binding and linking dynein to its cargo during fast axonal transport [26][27][28][29]. DNC-4, together with p25 and p27 subunits, interacts with actin-related proteins Arp1 and Arp11 to form the pointed end of the dynactin complex, which is positioned diametrically opposite dynein motor domains to primarily influence cargo binding [28][29][30]. In dnc-4(ju993) adult animals DD neurons formed synapses in the dorsal neurites, and the pattern and number of synapses were comparable to wild type. In tba-1(gf)dlk-1(0) animals, dnc-4(ju993) significantly increased the number of synapses in DD dorsal neurites (Fig 2E). Expression of wild type copies of DNC-4(+) rescued the suppression of tba-1(gf)dlk-1(0) by dnc-4(ju933) (Fig 2E). dnc-4(ju933) complemented the temperaturesensitive embryonic lethality of dnc-4(or633), which altered a conserved glutamic acid to lysine in the N-terminal region (S2 Fig) [31]. Additionally, dnc-4(or633) did not suppress the synapse remodeling defects of tba-1(gf)dlk-1(0) at the permissive temperature (20˚C), or when shifted to the restrictive temperature (25˚C) after embryonic development (Fig 2F), indicating that ju993 is a novel mutation of dnc-4. In conjunction with the effects on SV transport seen in dhc-1(ju1279) animals, and the similar levels of suppression seen in both dhc-1 and dnc-4 alleles (Fig 2C and 2E), we propose that altering dynein-dynactin complex interactions has a profound effect synapse remodeling.

Altered dynein activity promotes anterograde vesicle transport during remodeling
We next sought to understand the mechanism by which dhc-1(ju1279) suppressed the synapse remodeling defects of tba-1(gf)dlk-1(0). The remodeling defects in tba-1(gf)dlk-1(0) animals were primarily brought about by an increase in MT stability, resulting in reduced synaptic vesicle (SV) transport in the DD neurons during remodeling [11]. We first considered the possibility that dhc-1(ju1279) modifies MT dynamics in tba-1(gf)dlk-1(0), since an established role of cytoplasmic dynein is to stabilize dynamic MT plus ends by tethering them to the cell cortex [32]. However, dhc-1(ju1279) had no significant effect on the number of dynamic MTs or their direction of growth in both wild type and tba-1(gf)dlk-1(0) adults (S2 Fig), leading us to conclude that the suppression of synapse remodeling defects by dhc-1(ju1279) did not result from a change in MT dynamics.
Since the ju1279 allele affects the tail domain of DHC-1, which is structurally adjacent to the cargo binding domain of the dynein motor complex [30], we wondered whether SV transport was altered in the mutant animals. We assayed SV transport along the commissures of DD neurons during synapse remodeling in wild type and mutant animals using 4-dimensional (4-D) imaging (S1 Movie, Fig 3A). In wild type animals, most SVs moved in the anterograde direction, i.e., away from the cell body and towards their new location in the dorsal neurite ( Fig 3C). This proportion of anterogradely moving SVs was maintained in tba-1(gf)dlk-1(0) animals (Fig 3C), albeit with a strong reduction in the total number of mobile SVs (Fig 3B) [11]. Addition of dhc-1(ju1279) did not change mobile SV numbers; instead we observed a significant increase in the proportion of SVs moving anterogradely in both dhc-1(ju1279) single and dhc-1(ju1279) tba-1(gf)dlk-1(0) triple mutant animals (Fig 3B and 3C). We also imaged SV transport in dhc-1(js319) animals during DD remodeling, and did not find any increase in anterogradely moving SVs. Together, these data indicate that an increase in anterogradely moving SVs in dhc-1(ju1279) tba-1(gf)dlk-1(0) triple mutant animals ameliorates a reduction in the total number of SVs reaching the dorsal neurite to promote synapse formation during remodeling.
We previously reported two suppressors (ju972 and ju977) to be novel alleles of the anterograde motor, Kinesin-1/UNC-116 [11]. unc-116(ju972) strongly suppressed defective remodeling in tba-1(gf)dlk-1(0) by increasing total SV transport during remodeling, without altering the proportion of anterogradely moving SVs (Fig 3B-3D) [11]. Kinesins and dynein are classically thought of being in a "tug-of-war" during bi-directional cargo transport, highlighting the interdependence of the two motors for axonal transport in either direction (Fig 3D) [33][34][35]. We then wanted to see whether modifying dynein function using dhc-1(ju1279) would have any effect on SV transport in unc-116(ju972) animals. In both wild type and tba-1(gf)dlk-1(0) backgrounds, dhc-1(ju1279); unc-116(ju972) double mutant animals displayed a significant increase in both total number of mobile SVs and the proportion of anterogradely moving SVs, resulting in a strong anterograde bias in SV transport (Fig 3B and 3C). These observations led us to hypothesize that dhc-1(ju1279) likely weakens the interaction between the dynein complex and SVs, shifting the balance of bidirectional cargo transport in the anterograde direction ( Fig 3D).
TTBK-3 belongs to a large expansion of the Casein-kinase (CK1) superfamily in C. elegans, sharing 32% identity in the kinase domain to human tau-tubulin kinases (TTBK1 and TTBK2) [36]. Tau-tubulin kinases were first identified by their ability to phosphorylate the MT-associated protein tau and tubulin; mammalian TTBK1 is highly enriched in the nervous system, while TTBK2 is more broadly expressed [37,38]. ju978 generated a STOP codon in the C-terminal end of TTBK-3, which could result in a truncated protein with an intact kinase domain that lacked a coiled-coil domain further downstream (Fig 4A and 4D). Since ttbk-3(ju978) suppressed tba-1(gf)dlk-1(0) to a similar extent as ttbk-3(tm4006), we asked whether the catalytic activity of ttbk-3 was required for synapse remodeling. Overexpressing kinase-dead versions of TTBK-3 (K115A or D209A) [39] failed to rescue tba-1(gf)dlk-1(0); ttbk-3(ju978) (Fig 4D). We also obtained similar results using a mutant TTBK-3 lacking the C-terminal coiled-coil domain (Fig 4D). These results suggest that both the catalytic activity of the kinase domain and the coiled-coil domain are likely required for TTBK-3 function in synapse remodeling.
As reported by a recent study, expression of ttbk-3 was weak and extremely variable [40] and we were not able to reliably examine its neuronal expression pattern using extrachromosomal arrays of GFP driven by the endogenous ttbk-3 promoter. Expression of GFP tagged TTBK-3 in D motor neurons (P unc-25 -TTBK-3-GFP) showed both diffuse and punctate GFP accumulation in the cell body and neurites during the L3 and L4 developmental stages, with a more diffuse distribution of GFP observed in adult animals (S3 Fig). GFP expression was almost undetectable outside the cell body during the period of DD synapse remodeling in L1 and L2 animals. To test if TTBK-3 acted cell autonomously in the DD neurons to regulate remodeling, we overexpressed TTBK-3 under a DD neuron specific promoter (P flp-13 ), which rescued the suppression of defective remodeling by ttbk-3(ju978) to a similar degree as fulllength ttbk-3. As a control, expression of TTBK-3 from a muscle specific promoter (myo-3) failed to do so (Fig 4B and 4C), supporting the conclusion that TTBK-3 is required in DD neurons.

TTBK-3 modulates nascent synapse stability after DD remodeling
Next, we focused on how ttbk-3 could regulate synapse remodeling. Tau is a MAP that plays important roles in axonal transport, MT dynamics and neurite outgrowth during development, and is misregulated in neurodegenerative disease [41]. We tested a null mutation of the C. elegans homolog of tau, ptl-1, and found that ptl-1(0) did not have any effect on synapse formation or remodeling in wild type animals, and also failed to suppress the synapse remodeling defects of tba-1(gf)dlk-1(0) animals (S3 Fig). We also tested a null allele of ttbk-7 (R90.1), the closest homolog of mammalian TTBK1/2 (65% sequence identity), which also failed to retrograde directions, with more SVs moving towards the dorsal neurite (anterograde). b) unc-116(ju972) (black stars) modifies the MT binding domain of kinesin to enhance both anterograde and retrograde SV transport. c) Almost all SVs move in the anterograde direction in dhc-1(ju1279) (green star) and dnc-4(ju993) (red star) animals, possibly due to a disruption in the interaction between the dynein-dynactin complex and SVs.

Fig 4. Kinase activity of ttbk-3 is required for suppressing
We next imaged MT dynamics and SV transport and found that neither changed in tba-1 (gf)dlk-1(0); ttbk-3(tm4006) animals when compared to tba-1(gf)dlk-1(0) animals (S3 Fig). Taken together with the expression pattern in DD neurons, this data suggested that ttbk-3 was not involved in the early stages of synapse formation during remodeling. We then assayed the temporal requirement of ttbk-3 in synapse remodeling using GFP tagged TTBK-3 expressed under a heat-shock inducible promoter (P hsp-16.2 -TTBK-3-GFP). Following heat-shock in young adult animals TTBK-3-GFP formed punctate aggregates in neurons, the pharynx, the intestinal lumen and posterior intestinal cells (S3 Fig). To assay ttbk-3 requirement, we induced TTBK-3 expression at various larval stages, and observed behavioral and synapse remodeling phenotypes in the induced animals at adulthood (Fig 5A). Heat-shock treated wild type animals (with or without TTBK-3-GFP) did not coil dorsally and had normal dorsal neurite synapse formation (Fig 5B and 5C). On the other hand, heat shock treatment of L4 stage tba-1(gf)dlk-1(0); ttbk-3(ju978) animals expressing TTBK-3-GFP significantly reduced the suppression of both behavioral and synapse remodeling defects by ttbk-3(ju978), when compared to non-transgenic animals undergoing the same heat shock treatment (Fig 5D and 5E). We observed no difference in the extent of suppression between transgenic and non-transgenic animals that were heat-shocked at any other developmental stage, suggesting that ttbk-3 specifically played a role in synapse remodeling at the L4 stage, and was not required for the induction of synapse remodeling at the L1-L2 stage. We previously reported the presence of transient dorsal neurite synaptic puncta in tba-1(gf)dlk-1(0) animals from the L2-L4 stage, which were then eliminated as the animal achieved adulthood [11]. Indeed, the intensity of SNB-1-GFP puncta observed along the dorsal neurite at the L4 stage was higher in ttbk-3 (tm4006) animals, both in the wild type and tba-1(gf)dlk-1(0) backgrounds (S4 Fig). This suggested the existence of a mechanism to regulate the stabilization and maintenance of new synaptic sites formed on the dorsal neurite during remodeling. Since adding back wild type TTBK-3 at the L4 stage removed dorsal neurite synapses in tba-1(gf)dlk-1(0); ttbk-3(ju978) animals, we concluded that TTBK-3 antagonizes the stability of nascent synapses that are formed during synapse remodeling.

Discussion
Modifications to synapse architecture occur throughout the lifetime of an animal, either dependent or independent of large scale neurite rewiring [4][5][6][7][8][9]. In this study, we dissect the mechanisms underlying a C. elegans model of structural synaptic plasticity, where synapses are eliminated and re-assembled along different neurites of the same neuron during a developmentally defined time scale [9][10][11][12].
https://doi.org/10.1371/journal.pgen.1006844.g004 defects of tba-1(gf)dlk-1(0) animals following EMS mutagenesis. Since we performed a nonclonal screen for homozygous viable adult animals, we were not able to isolate suppressors that might cause either lethality or sterility. Of the viable suppressors that we characterized, more than half were missense or nonsense alleles of tba-1 that either reverted tba-1(gf)function back to wild type or resulted in null mutation of tba-1. Multiple mutations in human α-tubulin have been implicated in neuronal disorders like lissencephaly and ALS, and their effects on MTs have only been studied using in vitro overexpression models [42][43][44]. All the missense tba-1 alleles that we identified in this screen alter invariant residues adjacent to disease-linked mutations in the C-terminal H12, the GTP binding domain and the loop between H8 and S7 (annotated in S1 Fig) [42][43][44], highlighting the importance of these three regions for α-tubulin functionality in vivo.
The non-tubulin suppressor mutations displayed incomplete suppression of the synapse remodeling defects of tba-1(gf)dlk-1(0). On their own, none of these alleles had any effect on synapse formation or animal behavior. This might result from multiple factors. First, as was the case with motor proteins and their adaptors (unc-116, dhc-1 and dnc-4 alleles), the screen identified novel mutations that subtly altered protein function in such a way that embryonic development was completed and the animals were viable. Next, loss of function alleles in ttbk-3(ju978) behaved superficially wild type, likely because of the presence of other redundant kinases, which might also explain the incomplete suppression on tba-1(gf)dlk-1(0). Finally, any suppressors with strong behavioral deficits were likely missed since the design of our screen aimed to identify animals with wild type behavior first, and then assayed for restoration of DD synapse remodeling.
A bottleneck to successful DD synapse remodeling is the motor dependent transport of SVs from the ventral to the dorsal neurite. Consistently, a majority of the suppressors altered either anterograde or retrograde motor function to compensate for a reduction in SV transport in tba-1(gf)dlk-1(0). We previously identified kinesin-1 mutations that increased motor motility, and here, we characterized dynein-dynactin mutations that could alter cargo binding, with both strategies increasing anterograde SV transport along the DD neuron commissure during remodeling. However, while there was an overall increase in bidirectional SV transport in unc-116(ju972) animals, dhc-1(ju1279) selectively increased anterograde SV transport. This likely reflects the distinct motor domains targeted by ju972 and ju1279 on kinesin and dynein, respectively, with MT-motor interactions affecting bidirectional transport and variations in motor-cargo binding biasing transport in the direction with more engaged motors. Our results thus highlight both the interdependence and the competition between anterograde and retrograde motors during bidirectional cargo transport in vivo. The combination of unc-116(ju972) and dhc-1(ju1279) had an additive effect on SV transport, with an increase in both the total and anterogradely moving vesicle pools. Interestingly, dhc-1(ju1279); unc-116(ju972) did not enhance the number of dorsal neurite synapses in tba-1(gf)dlk-1(0) compared to unc-116 (ju972) alone (S4 Fig), indicating that the synapse formation defects caused by tba-1(gf)alone [14] could not be suppressed simply by an increase in the number of available SVs.
Studies from various experimental models and cultured neurons indicate that nascent synapse stability is dependent on the transport of sufficient numbers of SVs to the site of new synapse formation [45][46][47]. In tba-1(gf)dlk-1(0) animals, the lack of sufficient SVs at sites of new synapse formation on the dorsal neurite resulted in their destabilization and subsequent elimination at the L4 stage [11]. Here, we found that TTBK-3, a C. elegans member of the casein kinase superfamily, promotes synapse destabilization along the dorsal neurite, and loss of ttbk-3 was sufficient to maintain newly formed synapses in tba-1(gf)dlk-1(0). Synapse destabilization by TTBK-3 is likely phosphorylation dependent, and future studies will be required to identify potential substrates. The coiled-coil domain was also required for TTBK-3 function, leading us to speculate a possible regulatory role for the domain on kinase function. We previously showed that the initiation of dorsal neurite synapse formation was mediated by dlk-1 at the L2 stage; however ttbk-3 does not appear to play a role at this stage [11]. ttbk-3 is also not involved in ventral neurite synapse elimination during DD neuron remodeling, which is complete by L3 stage [11][12][13]. These results demonstrate that developmental synapse remodeling is marked by distinct phases of synapse formation, elimination and maintenance, each under tight spatio-temporal control to achieve precise neuronal connectivity.

C. elegans culture
Strains were maintained at 20˚C on NGM plates unless noted otherwise [48]. Information on alleles and genotypes of strains is summarized in S2 Table. Plasmid and transgene generation Plasmids were generated using Gateway technology (Invitrogen). DNA for dnc-4 and ttbk-3 constructs was amplified from purified genomic or cDNA by PCR using Phusion HF DNA polymerase (Finnzyme) (S4 Table), and subcloned into PCR8 entry vectors. Transgenic animals were generated by microinjection, following standard procedures [49], using plasmids of interest at various concentrations (listed in S3 Table) and Pgcy-8-GFP (80-90 ng/μl) or Pmyo-2-mCherry (2.5 ng/μl) as co-injection markers. A minimum of 2-3 transgenes were generated for each construct described in this study. For rescue experiments using dhc-1, dnc-4 and ttbk-3 constructs, the data from 3 transgenes was pooled in statistical analyses. A list of the plasmids used in this study and the transgenic arrays generated from them is listed in S3 Table. Primer information for the DNA clones generated in this study is listed in S4 Table. Fluorescent imaging of synapses L4 animals were cultured at 20˚C overnight, and day 1 adults were imaged using a Zeiss LSM 710 confocal microscope. Animals were anaesthetized in 0.6 mM levamisole on 2% agar pads for image acquisition. Z-stacks were generated from slices of 0.6 μm thickness. Images were processed using Zen lite software. Synaptic puncta number was quantified manually using a Zeiss Axioplan 2 microscope equipped with Chroma HQ filters.

SNB-1::GFP trafficking
L2 stage animals (14-18 hrs post hatching when maintained at 20˚C) were collected for analysis, and anesthetized using 30 mM muscimol on 10% agarose pads. 4-D imaging was performed using a Yokogawa CSU-W1 spinning disc confocal head placed on a Leica DMi8 confocal microscope equipped with a piezo Z stage for fast Z-acquisition controlled using MetaMorph (Molecular Devices). The entire DD commissure was visualized in 5-6 slices and images were collected for a total of 20 frames. The resulting movies were analyzed using Metamorph to generate kymographs for analysis of number and direction of movement of synaptic vesicles.

EBP-2::GFP image acquisition and analysis
Animals were anaesthetized in 30mM muscimol on 10% agarose pads for image acquisition. Live imaging for monitoring EBP-2 dynamics was done using a Yokogawa CSU-W1 spinning disc confocal head placed on a Leica DMi8 confocal microscope controlled by MetaMorph. 200 single plane images were taken serially at an exposure time of 113ms with an interval of 230ms between each frame, and analyzed using Metamorph software (Molecular Devices) to generate kymographs for analysis.
Analysis of mutants from tba-1(gf) dlk-1(0) suppressor screen tba-1(gf)dlk-1(0) animals were mutagenized using Ethyl Methane Sulphonate (EMS) following standard procedures [48]. F2 animals with improved locomotion were selected as putative suppressors in a non-clonal screen. Several suppressors were determined to be intragenic loss of function mutations in tba-1(gf)by Sanger sequencing the genic region of tba-1 for any additional mutations. Two of the extragenic suppressors, ju972 and ju977, were determined to be extragenic and mapped to the gene unc-116 following whole genome sequence analysis by MAQGene [50]. The results of whole genome sequencing of the remaining suppressors were analyzed using a Galaxy (https://usegalaxy.org/) workflow, and the causative mutations were identified by linkage analysis of the suppression to the SNPs identified in the whole sequence analysis.

Heat shock induced expression of ttbk-3
Transgenic animals expressing P hsp-16.2 TTBK-3-GFP; P myo-2 mCherry in the wild type and tba-1(gf)dlk-1(0); ttbk-3(0) backgrounds were selected by positive pharyngeal mCherry expression. L1, L2, L3, L4 and young adult animals were heat shocked at 34˚C for 2 hours in an incubator. Heat shocked animals were maintained at 20˚C after heat shock until they reached day 1 adulthood, when they were imaged using a Zeiss LSM 710 confocal microscope.

Statistical analysis
Statistical analysis was performed using GraphPad Prism 5. Normal distribution of samples was tested using D'Agostino & Pearson omnibus normality test. Significance was determined using One way ANOVA followed by Tukey's multiple comparison tests and two way ANOVA followed by Bonferroni posttests for multiple samples.
Supporting information S1