A forward genetic screen identifies Dolk as a regulator of startle magnitude through the potassium channel subunit Kv1.1

The acoustic startle response is an evolutionarily conserved avoidance behavior. Disruptions in startle behavior, particularly startle magnitude, are a hallmark of several human neurological disorders. While the neural circuitry underlying startle behavior has been studied extensively, the repertoire of genes and genetic pathways that regulate this locomotor behavior has not been explored using an unbiased genetic approach. To identify such genes, we took advantage of the stereotypic startle behavior in zebrafish larvae and performed a forward genetic screen coupled with whole genome analysis. We uncovered mutations in eight genes critical for startle behavior, including two genes encoding proteins associated with human neurological disorders, Dolichol kinase (Dolk), a broadly expressed regulator of the glycoprotein biosynthesis pathway, and the potassium Shaker-like channel subunit Kv1.1. We demonstrate that Kv1.1 and Dolk play critical roles in the spinal cord to regulate movement magnitude during the startle response and spontaneous swim movements. Moreover, we show that Kv1.1 protein is mislocalized in dolk mutants, suggesting they act in a common genetic pathway. Combined, our results identify a diverse set of eight genes, all associated with human disorders, that regulate zebrafish startle behavior and reveal a previously unappreciated role for Dolk and Kv1.1 in regulating movement magnitude via a common genetic pathway.


Introduction
Defects in initiating or executing movements are associated with a range of disorders. While some neurological disorders are primarily defined by motor impairments, several disorders defined primarily by cognitive deficits include motor features. For example, the eyeblink response, in which a patient reacts to a startling stimulus, is disrupted in a variety of neurodevelopmental and psychiatric disorders, including obsessive compulsive disorder, schizophrenia, posttraumatic stress disorder, and autism spectrum disorder [1][2][3][4]. The eyeblink response is one component of the startle response, which in humans is a whole-body defensive maneuver to shield the upper body from impact and in aquatic vertebrates, including zebrafish, is critical to evade avoid predators [5,6]. A combination of electrophysiological, lesion, and imaging studies have uncovered the core neural circuity underlying startle behavior in human and various vertebrate animal models [7][8][9][10]. Yet despite its critical role in animal survival and its link to several neurological disorders, the repertoire of genes and genetic pathways that regulate startle behavior has not been explored using an unbiased genetic approach.
Over the past several decades, the larval zebrafish has emerged as a powerful vertebrate model organism for unbiased genetic screens to identify genes critical for basic locomotion [11][12][13] and more recently for more complex behaviors, including visual behaviors and sleep [14,15]. However, an unbiased genetic screen to identify the genes critical for the execution of the startle response has been absent. By five days post fertilization (dpf), in response to an acoustic stimulus, zebrafish larvae undergo a characteristic short latency C-start (SLC), consisting of a sharp C-shaped turn and swimming away from the stimulus [16] (Fig 1B-1F and S1 Video). The behavioral circuit for the acoustic startle response is well characterized (reviewed in [9,17], Fig 1A) and is functionally similar to the human startle circuit [5]. Central to the zebrafish acoustic startle circuit are the Mauthner cells, a bilateral pair of reticulospinal neurons in the hindbrain. The Mauthner cells are necessary and sufficient for this short latency escape behavior [16,18,19]. Hair cell activation following an acoustic stimulus leads to activation of the eighth cranial nerve, which, along with the spiral fiber neurons, provides excitatory input to the Mauthner cell. The Mauthner cell directly activates contralateral primary motor neurons and excitatory interneurons to drive unilateral body contraction and turning away from the acoustic stimulus [20,21]. Inhibitory input prevents Mauthner cell firing at subthreshold stimuli or when the other Mauthner cell has already fired [22]. Because the circuit is well defined and the behavior is robust, this system is ideal for investigating how genes regulate behavior.

Fig 1. The zebrafish larval startle response is amenable to circuit and genetic analysis. (A)
The acoustic startle response is driven by an action potential from the Mauthner neuron (green), which activates motor neurons in the spinal cord to drive a contralateral body bend. Excitatory and inhibitory neurons in the hindbrain and in the spinal cord impinge upon the Mauthner cells to ensure motor neurons fire on only one side. (B-F) A representative acoustic startle response in a 5 dpf (days post fertilization) wild type larva. An acoustic stimulus is delivered at 0 ms (B), which elicits a rapid turn (C) followed by a counter We previously performed a forward genetic screen to identify functional regulators of the acoustic startle response. This approach identified a distinct set of genes with previously unrecognized roles critical for startle habituation [23,24], startle sensitivity [25], and sensorimotor decision making [26]. Here, we describe kinematic mutants identified from this screen that display defects in executing swim movements of the startle response. These mutants fall into two general categories. One group of mutants display a "weak" acoustic startle response characterized by shallow bends and minimal displacement. Using high-throughput sequencing, we have identified causative mutations in five weak startle mutants. In all five lines, the affected genes control muscle or neuromuscular junction (NMJ) function and are associated with locomotor disorders in humans.
The second group of mutants perform high amplitude bends, resulting in an "exaggerated" response. For one of the exaggerated startle mutant lines, we identified a causative mutation in PR domain containing 12b (prdm12b), a transcription factor that controls development of inhibitory neurons in the spinal cord and has previously been shown to be important for regulation of movement in fish [27]. The remaining two exaggerated mutant lines harbor mutations in dolichol kinase (dolk), which encodes a glycosylation pathway enzyme, and in potassium voltage-gated channel, shaker-related subfamily, member 1a (kcna1a), which encodes the potassium channel subunit Kv1.1. We demonstrate that dolk and kcna1a likely act in a common pathway to control startle movement magnitude as Dolk is required for Kv1.1 protein localization. Additionally, we demonstrate that Kv1.1 and Dolk act in the spinal cord to control the magnitude of body bends. Thus, through our forward genetic screen we identified a number of genes that are essential to regulate body movement. Furthermore, we demonstrate how a broadly expressed protein, Dolk, selectively regulates behavior through the Kv1.1 protein.

A forward genetic screen for regulators of the larval startle response
We previously performed a forward genetic screen for regulators of the acoustic startle response [23][24][25][26]. In brief, using a high-speed camera and automatic tracking, individual F3 larvae were exposed to a series of startling stimuli (for details on mutagenesis and the breeding scheme to obtain F3 larvae see [23]). This portion of the screen focused on kinematic mutants, characterized by significant changes in any of the stereotypic parameters characteristic of the startle response (Table 1). Putative mutant lines were retested in the subsequent generation to confirm genetic inheritance. This screen identified a group of eight mutants defective in kinematic parameters (including response latency, turn angle, distance, and displacement). These mutants fall into two categories: five mutant lines display shallow bends of decreased turning angle compared to their siblings (one representative mutant line shown in in Fig 1G-1K and 1Q-1T, S2 Video); we refer to these as "weak" startle mutants. Three mutant lines perform numerous high amplitude bends in response to an acoustic stimulus. These turns often result bend (D) and swimming away (projection of 90 ms response in E, color coded by time). Automated tracking of the curvature of the larva throughout the behavior (F) reveals the response latency (blue bar = latency in F,K,P). (G-K) ryr1b mutants display a weak startle response with reduced bend and counter bend angles (H,I) and reduced displacement (J), largely due to minimal swimming after the counter bend (K). (L-P) prdm12b mutants display an exaggerated startle response with increased bend and counter bend angles (M,N). The duration of each bend is longer than in wild type as well (P). (Q-T) Quantification of response latency (Q; manual measurement); max C1 head angle (R; automated measurement); distance traveled along escape trajectory (S; automated measurement); and displacement from initial head position to final head position (T; automated measurement). Each point is the average response over ten trials for an individual larva. n�10 larvae, � p = 0.002, �� p<0.0001 (one-way ANOVA with Tukey correction for multiple comparisons). https://doi.org/10.1371/journal.pgen.1008943.g001

PLOS GENETICS
in larvae swimming in a figure eight pattern (one representative mutant line shown in Fig 1L-1T and S3 Video); we refer to these as "exaggerated" startle mutants. Combined, these eight mutant lines offer an opportunity to reveal genetic regulators of movement kinematics.
To identify the molecular mechanisms underlying these behavioral phenotypes, we first set out to identify the causative mutations. In all kinematic mutant lines, approximately 25% of progeny from carrier incrosses display the mutant behavioral phenotype, consistent with the causative mutations being recessive, monoallelic, and causing highly penetrant phenotypes. In addition, when behaviorally mutant larvae were raised for each of seven lines, no animals survived past two weeks, indicating these lines harbor homozygous lethal mutations. For the eighth line (kcna1a p181 ), we found rare escapers (~5%) that lived >two months, but these fish did not grow past a juvenile stage (<1 cm in length). This lethality can likely be attributed to defects with swimming and hence deficits in prey capture and feeding. For each mutant line, after behavioral testing, pools of behaviorally mutant larvae and behaviorally wild type siblings were collected for high-throughput DNA sequencing. We performed whole genome sequencing (WGS) and homozygosity analysis on mutant and sibling pools for four of the lines, as described in more detail in [23]. Using known single nucleotide polymorphisms (SNPs) present in our wild type background, we identified regions of homozygosity in the mutant pools that were heterozygous in the sibling pool. Within these regions of homozygosity, potentially detrimental exonic SNPs not observed in wild type fish were identified. Nonsense mutations, particularly in genes known to function in neurons or muscle, were prioritized. Individual larvae displaying mutant or wild type behavior were then sequenced for potentially causative SNPs. If a potential SNP was observed as homozygous in 100% of mutant larvae (>20 individuals) and heterozygous or homozygous wild type in all siblings (>20 individuals), we considered the SNP to likely be causative. For the remaining four mutant lines, we performed whole

PLOS GENETICS
exome sequencing (WES) [28]. To identify linkage (SNPs with allelic frequencies~100% in mutants and~33% in siblings, based on ratio of heterozygous to homozygous wild type larvae) and potentially causative mutations, we used the online tool SNPTrack [29]. Confirmation of potentially causative mutations was performed as described above for candidates from WGS. Five lines (neb, cacna1ab, ryr1b, rapsn, and slc5a7a) contain nonsense mutations (or a splice mutation resulting in nonsense mutation) in genes known to regulate neuron or muscle function (see Table 1), strongly indicating we have identified the correct mutation (see also discussion of identified genes below). Two of the mutant lines (prdm12b and kcna1a) have missense mutations, and we or others have generated nonsense alleles that display the same phenotype. For the remaining mutant line (dolk) containing a nonsense mutation, we generated an independent second mutant allele to confirm that mutations in dolk are causative for the exaggerated locomotor phenotype (see below). Based on these data, we are confident we have identified mutant alleles of eight genes critical for regulating proper kinematic behavior during the acoustic startle response. We note that all genes have a human disease associated counterpart and that six of the eight genes are associated with human movement disorders, further underscoring conservation of disease associated genes in zebrafish [30] (Table 1).
Based on their molecular identities, the affected genes can be subdivided by their likely site of action. Below, we report on several genes that act in skeletal muscle, at the neuromuscular junction, or in inhibitory spinal neurons, and we then focus on two genes likely to act in the same pathway to regulate movement magnitude.

Genes controlling muscle function modulate acoustic startle kinematics
Of the five weak startle mutant lines, two have mutations in genes that are required for muscle function. p413 mutants contain a nonsense mutation in the nebulin (neb) gene (Table 1), which encodes a protein necessary for sarcomere assembly and subsequent function [31]. Previous work in zebrafish demonstrated that a loss-of-function neb allele displays defects in sarcomere assembly, leading to reduced swim movement [32]. The second mutant line we identified, p414, contains a splice mutation in the ryanodine receptor 1b (ryr1b) gene (Table 1 and Fig 1G-1K and 1Q-1T). RyR1 is required for calcium release at the sarcoplasmic reticulum, which drives muscle contraction. A previously characterized zebrafish allele of ryr1b, called relatively relaxed, displays a decreased touch response and reduced Ca 2+ transients in fast muscle [33]. Interestingly, RyR1b is expressed primarily in fast muscle while RyR1a is expressed primarily in slow muscle [33]. Consistent with this finding, ryr1b p414 mutants display a drastic startle response defect, which is dependent on fast muscle. However, spontaneous movement, which is dependent on slow muscle, is not significantly different from siblings (Table 1). This result emphasizes the importance of examining different behaviors, especially ones that utilize different circuitry and muscle groups [20,34,35], when characterizing locomotion in mutant animals.

Genes acting at the neuromuscular junction regulate acoustic startle kinematics
Three genes identified in our screen are required at the neuromuscular junction (NMJ). This includes a new mutant allele, p416, of the gene receptor-associated protein of the synapse (rapsn), originally identified as twitch once mutants [11], and subsequently shown to be caused by mutation in the rapsn locus [36]. rapsn mutants display reduced spontaneous swimming and touch responses [11] caused by decreased acetylcholine receptor clustering at NMJs [36]. In addition, we identified a nonsense mutation (p415) in the calcium channel, voltage-dependent, P/Q type, alpha 1A subunit, b (cacna1ab) gene, which encodes Ca v 2.1. A previously identified mutant allele of cacna1ab, fakir, displays a weak touch response [11,37]. This weak touch response in fakir larvae can be attributed to defects in transmission at the NMJ [38]. This likely accounts for the reduction in startle response and spontaneous movement we observe in p415 (Table 1). Since rapsn and cacna1ab were both previously identified in a screen for escape response following a tail touch [11], which is driven by the Mauthner cells [39], we predicted they could be identified in the Mauthner-dependent acoustic startle screen described here.
The fifth weak startle mutant, p417, demonstrates a unique behavior (S1 Fig). Unlike sibling larvae that initiate the startle response at the head (S1C-S1G Fig), in p417 mutants, the startle response often initiates with a turn in the tail instead (S1H- S1L Fig and S4 Video). Subsequent body bends are uncoordinated, rather than a smooth progression from head to tail. The C1 angle, distance, and displacement are also drastically reduced in these mutants, and mutants undergo very little spontaneous movement (Table 1). We identified the causative p417 mutation as a nonsense mutation in slc5a7a, which encodes the high-affinity choline transporter (CHT) (S1A and S1B Fig). Uptake of choline by the high-affinity choline transporter is the rate limiting step for acetylcholine synthesis [40], so Slc5a7 is essential for cholinergic transmission.
There are a number of cholinergic neuronal populations in the acoustic startle circuit. Most notably, spinal motor neurons release acetylcholine at the NMJ to activate skeletal muscle in the trunk and tail [41]. We hypothesized the slc5a7a mutant phenotype of reduced turn angle and minimal displacement is caused by reduced muscle contractions due to reduced acetylcholine release at the NMJ. To test this, we generated a transgenic line in which slc5a7a is expressed under the control of a motor neuron specific promoter, mnx1/hb9 [42] (S1M Fig). Compared to slc5a7a mutants lacking the transgene (S1N-S1P Fig), mutant larvae expressing the transgene (slc5a7a -/-; Tg(hb9:slc5a7a), display a significantly increased turn angle and distance traveled following an acoustic stimulus. Thus, motor neuron specific expression of slc5a7a partially rescues the mutant phenotype, providing strong evidence that Slc5a7a plays a critical role in regulating coordinated swim movements, presumably by increasing acetylcholine level release at NMJs.

prdm12b regulates development of V1 interneurons and modulates acoustic startle kinematics
In addition to the five weak startle mutants described above, we identified three exaggerated startle mutants. In the p419 mutant line in which the C1 turning angle of the startle response is dramatically increased (165% of wild type siblings; Table 1 and Fig 1L-1P and S3 Video), we identified a missense mutation in the coding sequence of PR domain containing 12b (prdm12b). Prdm proteins are transcription factors, and several family members have been shown to play roles in nervous system development [43]. Zebrafish prdm12b was previously identified in a screen for genes that express in the nervous system [44]. Morpholinos targeting prdm12b [27] and a prdm12b CRISPR mutation [45] result in larvae with aberrant, exaggerated touch behaviors, similar to the exaggerated acoustic startle response we observe in the p419 mutants. prdm12b is required for development of engrailed 1b positive, V1 interneurons in the spinal cord [27]. V1 interneurons are glycinergic and have been shown to regulate the speed of locomotor movements in mammals [46]. In zebrafish larvae, ablation of V1 interneurons results in an exaggerated touch response [47], reminiscent of the exaggerated startle response we observe in prdm12b mutants. Thus, Prdm12b is required during development for the specification of inhibitory spinal neurons utilized for the acoustic startle response to achieve coordinated and controlled movement.

The glycosylation pathway protein Dolk regulates acoustic startle magnitude
In a second exaggerated movement mutant, p420, we identified a nonsense mutation in the dolichol kinase (dolk) gene (Fig 2A and 2B). Similar to prdm12b mutants, the dolk mutant C1 turn angle is dramatically increased (153% of wild type siblings; Table 1 and Fig 2C-2M). In addition to exaggerated C1 turn angles in response to an acoustic stimuli, dolk mutants swim in "figure eights," resulting in reduced displacement (Fig 2K and 2O). Dolk catalyzes the CTPdependent phosphorylation of dolichol to dolichol monophosphate. Dolichol monophosphate , with the predicted amino acid change from the screen mutation indicated. The deletion in the CRISPR allele causes a frame shift in the sequence that results in a premature stop at amino acid 50. In contrast to siblings (C-G), dolk mutants (H-L) display an exaggerated startle response, with an increased bend and counter bend angle (I,J), resulting in larvae swimming in a "figure eight" (K). Blue bar = latency in G,L. (M-O) Kinematic parameters of the acoustic startle response in dolk siblings and mutants (homozygotes of the screen identified mutation, p420/p420, and transheterozygotes from the screen mutation and CRISPR, p420/p421). Each point represents average of ten trials for an individual fish. n�6 larvae, �� p<0.001, � p = 0.008 (one-way ANOVA with Tukey correction for multiple comparisons). https://doi.org/10.1371/journal.pgen.1008943.g002

PLOS GENETICS
is an essential glycosyl carrier for C-and O-mannosylation and N-glycosylation of proteins [48]. To confirm that the dolk nonsense mutation identified in our screen (p420) is causative, we generated a second mutant dolk allele using CRISPR/Cas9. In this dolk p421 CRISPR allele, a 1 bp deletion results in a frameshift and premature stop codon at position 50. Both mutations occur early in the protein sequence (Fig 2B), and we predict they likely eliminate all protein function. Transheterozygous (p420/p421) dolk mutants display the same exaggerated phenotype as p420 homozygotes (Fig 2M-2O), confirming that mutations in dolk cause the behavioral deficits observed in the original p420 mutants.

The potassium channel subunit Kv1.1 regulates acoustic startle magnitude
Given Dolk's critical role in protein glycosylation, we predict that in dolk mutants, many proteins exhibit reduced glycosylation. In 2 dpf zebrafish, over 160 unique glycosylated proteins have previously been cataloged via mass spectrometry [49], so predicting the number and identities of Dolk's behavior-relevant glycosylation targets is challenging. To identify the relevant glycosylated downstream protein(s), we instead turned to our remaining exaggerated mutant line. In this line, we identified a missense mutation in kcna1a, which encodes the voltage potassium Shaker-like channel subunit Kv1.1 (Fig 3A and 3B). We have previously characterized this allele for its role in regulating startle habituation [24]. Cell culture studies have shown that Kv1.1 is highly glycosylated, and glycosylation is required for Kv1.1 channel function [50-52]. We confirmed the kinematic defect in the screen p181 allele (Fig 3C-3O) is due to the missense mutation in kcna1a using a previously generated kcna1a CRISPR allele (p410), which harbors an early stop codon and is a presumed null [24]. p181/p410 transheterozygotes display the same exaggerated movements as mutants homozygous for p181 (Fig 3M-3O), and p181 homozygous larvae are behaviorally not significantly different from p410 homozygous larvae (Fig 3M-3O). Combined, this strongly suggests that the screen missense p181 allele also represents a null allele for the kinematic behavior. Thus, kcna1a and dolk are both required for executing controlled movements, possibly via Dolk-dependent glycosylation of Kv1.1.

Dolk is required for axonal localization of Kv1.1
Glycosylation is well known to regulate protein activity, localization, and/or stability [53]. To address whether dolk is critical for Kv1.1 localization or stability, we examined Kv1.1 protein localization in dolk mutants using a previously validated antibody [24]. In mammals, Kv1.1 is broadly expressed throughout the brain and primarily localizes to axons and axon terminals [54]. In 5 dpf wild type larvae, we observe Kv1.1 protein localization to axon tracts throughout the brain (Fig 4A and 4B) and the spinal cord ( Fig 4C). Kv1.1 strongly accumulates at the Mauthner axon cap (Fig 4B), where spiral fiber and inhibitory neurons synapse onto the Mauthner axon initial segment (AIS). In contrast, in dolk mutants, Kv1.1 protein localization is strongly reduced or absent in axon tracts and instead is prominent in somata throughout the brain and spinal cord (Fig 4D-4G). This is particularly apparent in the large Mauthner soma (Fig 4E and 4F). Little or no Kv1.1 is detectable at the axon cap (Fig 4E and 4F) in dolk mutants. Since we do still observe Kv1.1 expression in dolk mutants, Dolk may be largely dispensable for Kv1.1 protein stability yet indispensable for correct Kv1.1 cell surface localization. This might reflect an indirect role of Dolk on Kv1.1 localization via additional Dolk glycosylation targets. Alternatively, this might reflect a more direct role of Dolk-dependent glycosylation on Kv1.1 localization (see Discussion). As previous studies have shown that Kv1.1 membrane localization is required for its channel function, the most parsimonious explanation for the behavioral phenotype in dolk mutants is the lack of proper Kv1.1 localization.

Kv1.1 acts in the spinal cord to regulate swim movement magnitude
We next asked where in the startle circuit Kv1.1 acts to regulate the magnitude of the startle response. Expression of Kv1.1 throughout the hindbrain and spinal cord is consistent with the idea that Kv1.1 may act at various sites in the startle circuit. kcna1a mRNA is observed in Mauthner neurons as early as 2 dpf, and developmental expression of Kv1.1 correlates with the sibling and mutants (homozygotes of the screen mutation, p181/p181, transheterozygotes of the screen mutation and CRISPR, p181/p410, and homozygotes of the CRISPR mutation, p410/p410). Each point represents average of ten trials for an individual fish. n�17 larvae, �� p<0.001 (one-way ANOVA with Tukey correction for multiple comparisons). https://doi.org/10.1371/journal.pgen.1008943.g003

PLOS GENETICS
Mauthner cell acquiring single spike properties [55,56]. Furthermore, inhibition of Kv1 channels by local bath application of dendrotoxin-I onto the Mauthner cell lowers the threshold required to induce multiple spikes in the Mauthner cell [55,56]. Based on these previous findings, it is possible that kcna1a functions in Mauthner neurons to regulate the magnitude of the startle response. We explored this possibility using whole-cell recordings to determine the electrophysiological properties of Mauthner cells in sibling and kcna1a mutant larvae. Averaged measurements of membrane resting potential (V resting ), input resistance (R in ), rheobase (amount of current necessary to trigger an action potential), and action potential threshold (V threshold ) are not significantly different between siblings and kcna1a mutant larvae (S2A Fig). Next, we asked whether loss of kcna1a function changes the weight of key synaptic inputs to the Mauthner cells. Auditory afferents known as "large myelinated club endings" (club endings) that terminate on the lateral dendrite of the Mauthner cell are critical inputs for the acoustic startle response [57,58]. To test whether this synaptic input was altered, we examined  Finally, we failed to detect any significant difference in the halfwidth action potential between siblings and kcna1a mutants. Combined, these results reveal that neither changes of the excitability of Mauthner cells nor changes in the weight of critical auditory synaptic inputs can explain the exaggerated startle behavior observed in kcna1a mutants.
While Mauthner cell function and club endings appear unaffected in kcna1a mutants, it is possible that other inputs onto the Mauthner cells or synapses of the Mauthner cells are disrupted in kcna1a mutants. To determine whether kcna1a acts in a Mauthner-dependent pathway, we ablated Mauthner cells in kcna1a mutant larvae. Following Mauthner cell ablation, wild type larvae presented with an acoustic stimulus perform a longer latency, Mauthner-independent turning maneuver [16]. Thus, analyzing behavior in kcna1a mutants in which Mauthner neurons are ablated can clarify whether the exaggerated phenotype is dependent on Mauthner cell function and/or the function of neurons 'upstream' of Mauthner cells. We chemogenetically ablated Mauthners in kcna1a siblings and mutants using the nitroreductase (NTR)/metronidazole (Mtz) system [59]. Specifically, we utilized 62a:Gal4 [60] to drive Mauthner expression of UAS:NTR-RFP. With this genotype, expression was mosaic, so that some 62a:Gal4, UAS:NTR-RFP larvae had expression in both Mauthner neurons, some in only one Mauthner neuron, and some in neither Mauthner neuron. This mosaicism allowed us to compare identical genotypes for ablation in larvae with initial Mauthner expression of NTR-RFP and without expression of NTR-RFP. Prior to Mtz treatment, larvae were scored for presence or absence of Mauthner cell NTR-RFP expression (Fig 5A and 5B). Following Mtz treatment, in larvae initially lacking Mauthner NTR-RFP expression, Mauthner neurons were intact (Fig 5C). Conversely, following Mtz treatment in larvae with initial Mauthner NTR-RFP expression, Mauthner neurons were undetectable (Fig 5D). We then presented fish with intact (controls) or ablated Mauthner cells with an acoustic stimulus. As expected, compared to controls, sibling and kcna1a mutant larvae in which Mauthner cells were ablated display a longer latency response characteristic of Mauthner-independent turning maneuvers ( Fig 5E) [16]. Importantly, kcna1a mutant larvae in which Mauthner cells were ablated still performed exaggerated responses and figure eight turns (Fig 5F-5H). These responses were, aside from latency, indistinguishable from responses in mutants with intact Mauthner cells (S5 Video). Thus, the exaggerated behavioral response in kcna1a mutants is independent of activity in neurons 'upstream' of the Mauthner cell or in the Mauthner cell itself, further supporting the notion that Dolk and Kv1.1 primarily act outside the Mauthner-dependent hindbrain startle circuit to regulate startle magnitude.
We next tested whether dolk and kcna1a act in the spinal cord to regulate startle magnitude. We first determined whether dolk and kcna1a mutants exhibit any abnormal swim movements in the absence of external stimuli. kcna1a mutants execute normal swimming behaviors, including slow scoots and turns (Fig 6A and 6B) and when compared with their siblings, initiate a similar number of spontaneous movements (Fig 6G). However, after kcna1a mutant larvae begin a scoot maneuver, they sometimes undergo a series of rapid, high amplitude turns, similar to the behavior observed in response to acoustic stimuli (Fig 6C and S6 Video). This exaggerated "thrashing" movement is not observed in kcna1a siblings (n = 17 larvae, >800 responses), while kcna1a mutant larvae exhibit this behavior 22% of the time during spontaneous swim movements (Fig 6H). dolk mutant larvae also display exaggerated spontaneous swim movements (Fig 6H; 76% of spontaneous movements). Additionally, compared to siblings, dolk mutants initiate fewer spontaneous movements (Fig 6G), though each swim bout tends to be much longer (sibling average distance per swim bout = 10.9 a.u., mutant average distance per swim bout = 65.0 a.u.). Thus, kcna1a and dolk regulate the magnitude of both acoustic startle responses and spontaneous swim movements.
To address whether Dolk and Kv1.1 act primarily in spinal circuits to drive movement, we removed brain input to the body by severing the spinal cord. Zebrafish larvae can be "spinalized" by severing the head from the trunk and leaving only the spine intact. These spinalized larvae are unable to perform acoustic startle responses but are still capable of initiating spontaneous movement [61], though the rate of movement is reduced. We therefore transected the spinal cords of sibling and dolk/kcna1a mutant larvae ( Fig 6D) and observed their spontaneous movement. Compared to intact kcna1a mutant larvae, spinalized kcna1a mutant larvae continue to exhibit exaggerated swim bouts not observed in spinalized siblings (Fig 6E-6H and S7 Video). Spinalized dolk mutant larvae also continue to exhibit exaggerated swim bouts ( Fig  6H). Combined, these results demonstrate that Dolk and Kv1.1 are required in spinal locomotor circuits to control swim movement magnitude.
Finally, we asked in which motor circuits Dolk and Kv1.1 act to control swim movements. Kv1 in Drosophila [62] and Kv1.1 in mice [63,64] play important roles in preventing overactivity of motor neurons. In zebrafish larvae, we observe expression of Kv1.1 in motor neuron axons (Fig 7A). These motor neurons innervate both fast and slow muscle. In wild type animals, slow motor circuits turn off as fast motor circuits are recruited during fast movement [47]. We predicted that if loss of kcna1a mutants causes hyperactivity in motor circuits, perhaps slow motor circuits remain engaged during the fast startle response, leading to

PLOS GENETICS
Dolk regulates startle magnitude through Kv1.1 overactivation of muscle contraction and the exaggerated turns observed in these mutant larvae. To test this idea, we generated double mutants in dolk, where Kv1.1 is drastically mislocalized in the spinal cord (Fig 4G), and ryr1b, which specifically disrupts fast muscle [33]. If slow motor circuits are hyperactive in dolk mutants, we would expect to see a rescue of the ryr1b behavioral phenotype in the double mutants. Instead, we observe dolk -/-;ryr1b -/mutant larvae display a very similar startle phenotype to ryr1b mutants. This indicates that Dolk, and likely also Kv1.1, primarily acts within fast motor circuits to regulate movement magnitude.
Fast locomotor circuits are comprised of fast motor neurons and both excitatory and inhibitory interneurons [47]. While it is possible dolk and kcna1a are solely required in motor neurons to drive correct movement magnitude, we observe Kv1.1 expressed broadly in the spinal cord, including in commissural axons and ipsilateral axon tracts (Fig 4C). Within fast motor , we predict loss of kcna1a induces hyperactivity in specific neurons, making it unlikely that loss of kcna1a in V1 neurons leads to the exaggerated response phenotype. However, it is possible that loss of kcna1a affects neurons upstream of V1 interneurons, leading to their inhibition. prdm12b and kcna1a mutants have similar but distinct acoustic startle phenotypes (Fig 7E-7G); while the initial turn angle is increased for both single mutants (Fig 7E), the "figure eight" movement, leading to reduced displacement ( Fig 7G) is more pronounced in kcna1a mutants. We predicted that if kcna1a primarily acts upstream of V1 interneurons, kcna1a/prdm12b double mutants would display the prdm12b phenotype. Alternatively, if kcna1a acts primarily downstream or in parallel to V1 interneurons, double mutants would likely display the kcna1a phenotype. Double mutants display a strong exaggerated "figure eight" startle response, similar to kcna1a mutants (Fig 7E-7G). This suggests kcna1a may act in parallel or downstream to V1 interneurons, either solely in motor neurons, and/or in other interneurons.
One class of excitatory interneurons that act in combination with V1 interneurons to regulate fast locomotor circuits are the glutamatergic V2a Chx10+ interneurons. When V2a interneurons are silenced, the C1 turn angle and tail beat frequency following the acoustic stimulus are reduced [68]. V2a interneurons can be visualized in the spinal cord in chx10:Gal4, UAS: GFP (chx10::GFP) transgenic larvae (Fig 7H-7J"). Within the spinal cord where axons of V2a neurons project, we observe Kv1.1 protein localization (Fig 7H-7I"). To confirm Kv1.1 is expressed in V2a interneurons, we performed Kv1.1 staining in dolk -/-;chx10::GFP larvae in which Kv1.1 is retained in the nucleus of cells throughout the spinal cord. We observed colocalization of Kv1.1 in a subset of GFP+ soma (Fig 7J). These results suggest that Kv1.1 may regulate movement magnitude through V2a interneurons.

Discussion
Movement initiation, magnitude, and duration all depend on neural circuits acting in a coordinated manner to drive muscle contraction. Here, we describe mutations in eight genes identified in a forward genetic screen in larval zebrafish that enable and ensure coordinated movements. The driving force for carrying out this screen was to discover, in an unbiased way, genes that are required to regulate the acoustic startle response. With this unbiased approach, we predicted we would identify both genes already implicated in locomotion and genes without a prior known role in locomotion. This is in fact what we found. All of the eight genes we identified in our screen are associated with human disorders, and six are associated with motor disorders (Table 1). For example, NEB mutations in humans are associated with nemaline myopathy, a congenital muscle disease characterized by muscle weakness [31]. We identified neb mutants based on their weak response to an acoustic stimulus, and a previously characterized zebrafish neb mutant was shown to have defects in sarcomere assembly [32],

PLOS GENETICS
Dolk regulates startle magnitude through Kv1.1 similar to human patients with nemaline myopathy. Human mutations in SLC5A7 are associated with congenital myasthenic syndrome [69], which is characterized by muscle weakness and fatigue. slc5a7a mutants from our screen display a weak acoustic startle response similar to neb mutants. While Neb protein acts within the muscle, Slc5a7 acts within motor neurons at the NMJ. Therefore, while the mutants have a similar phenotype, suggesting a similar site of action, the proteins act in different cell types. This underscores the importance of understanding the normal biological function of a protein when trying to understand the underlying cause of a genetic disorder.
Even when the function of a protein is known, the protein may act in many different neurons or cell types to regulate movement. For example, Kv1.1 is broadly expressed in the zebrafish nervous system ([55] and our work) and in the mammalian nervous system [54]. In humans, heterozygous disease alleles of the Kv1.1-encoding gene KCNA1 have been shown to be causative for episodic ataxia (EA) Type 1 [70,71]. EA patients display frequent attacks of incoordination that are usually preceded by stress or startle. This phenotype is attributed to dysfunction in the cerebellum and peripheral nerves [72]. Work in mice heterozygous for a human KCNA1 disease allele has demonstrated a cerebellar defect of action potential broadening in basket cells [73], and studies in patients with EA have demonstrated increased nerve excitability when the median nerve of the wrist was stimulated [67]. Thus, Kv1.1 acts at multiple points in the mammalian nervous system to regulate nerve activity.
In this study, we uncovered a brain-independent role for Kv1.1 in regulating swim movements in larval zebrafish. Spinalizing kcna1a mutant larvae fails to restore wild type motor function, suggesting that Kv1.1 plays a critical role in spinal motor circuits. Is Kv1.1 acting in interneurons, motor neurons, or both to drive movement? In zebrafish larvae, Kv1.1 is expressed throughout the spinal cord (Fig 4), including on axons of spinal motor neurons ( Fig  7A). Kv1 plays an important role in regulating motor neuron activity across species. In Drosophila, there is a single Kv1 channel, Shaker, which was the first potassium channel ever cloned [74][75][76]. Another potential site of action for Kv1.1 in the spinal cord is in excitatory interneurons that activate motor neurons. One excitatory class is the V2a glutamatergic Chx10+ interneurons. In larval zebrafish, V2a interneurons are required for producing a normal acoustic startle response. When V2a interneurons are silenced, the C1 turn angle and tail beat frequency following the acoustic stimulus are reduced [68]. Recently, V2a interneurons in larval zebrafish have been categorized as Type I or Type II, similar to characterizations in mice, based on morphological and electrophysiological properties. From this work, type II V2a interneurons are predicted to regulate amplitude control [77]. We find that in the zebrafish spinal cord, Kv1.1 is expressed in a subset of V2a interneurons (Fig 7H-7J). It is therefore feasible that loss of Kv1.1 in type II V2a interneurons could result in hyperexcitability or broadening of the action potential, leading to increased recruitment or activation of motor neurons. This could explain the movement amplitude defects observed in kcna1a mutant larvae. Classes of locomotor interneurons are mostly conserved between fish and mammals [78], and single nucleus RNA-seq performed in mouse spinal cord neurons has revealed that Kcna1 is enriched in V2a interneurons [79]. Moreover, in mice, V2a interneurons are critical to coordinate left-right alternating movement, particularly at high speeds [80]. Kv1.1 may play a critical role in zebrafish in regulating locomotion through limiting excitability of V2a interneurons.
While expression in different neuronal cell types provides specificity to Kv1.1's function, post-translational modifications critically regulate Kv1.1 activity. We have previously shown that the palmitoyltransferase Hip14 palmitoylates Kv1.1, and both Hip14 and Kv1.1 are essential for zebrafish larvae to habituate to repeated acoustic stimuli [24]. Here, we demonstrate a role for a glycosylation pathway protein Dolk in regulating Kv1.1 localization. In cell culture, glycosylation of Kv1.1 is required for proper channel function [50-52]. Interestingly, in these studies, glycosylation appeared dispensable for channel localization to the cell membrane.
Here, we find that in dolk mutants, where Kv1.1 is predicted to not be glycosylated, cell surface localization of Kv1.1 is drastically disrupted. This effect may be a direct consequence due to the lack of Kv1.1 glycosylation or may be an indirect effect caused by reduced glycosylation of trafficking proteins. For example, Leucine-rich glioma-inactivated 1 (LGI1) is a glycoprotein that promotes axonal localization of Kv1.1 in hippocampal cells [81]. Whether the effect is direct or indirect, loss of Dolk clearly drastically affects Kv1.1 cell surface localization, providing compelling in vivo evidence that Kv1.1 localization is dependent, directly or indirectly, on Dolk-dependent glycosylation.
How does loss of a ubiquitous protein cause nervous system specific defects? Protein glycosylation is required in all cell types, and Dolk is a critical part of the protein glycosylation pathway. dolk mRNA has been detected in all human tissues assayed, with the highest expression observed in the brain [82]. Patients with dolk disease alleles display a range of phenotypes. In some patients, homozygous loss of function mutations in dolk cause metabolic and heart defects that are often lethal early in life [82,83]. Patients with similar homozygous, loss of function dolk alleles have also been identified that at the time of diagnosis in childhood only present with neurological conditions, including seizures [84]. Recently, a dolk variant has been identified as a potential risk gene in a patient with autism spectrum disorder [85]. The neurological symptoms observed in patients with dolk mutations underscore the importance of protein glycosylation in the nervous system [86]. In fact, defects in the nervous system are a predominant feature of congenital disorders of glycosylation [87]. Zebrafish dolk mutants do not display obvious morphological defects by 5 dpf except for failure to inflate their swim bladders, which is common in mutants with swim defects [11]. Homozygous dolk mutants die by 14 dpf, although we have not investigated the precise cause for this lethality. Lethality could be due to a failure to feed as zebrafish larvae must perform coordinated swim movements for prey hunting and feeding. Lethality could also be due to a broader requirement of dolk in vital organ systems.
In contrast, the behavioral phenotype in dolk mutants is highly specific and fairly unique. While Kv1.1 is likely not the only disrupted protein in dolk mutants, the similarity in behavioral phenotypes of kcna1a and dolk mutants suggests Kv1.1 is a critical protein affected by loss of dolk. It would be interesting to investigate whether similar dysregulation of Kv1.1 occurs in patients with neurological phenotypes arising from dolk mutations.
Together, this work demonstrates the utility of zebrafish for understanding genetic mechanisms driving regulation of locomotion. While the neural circuits driving movement are not identical among species, many of the genes required for circuit development and function are conserved. This is especially apparent from the number of genes identified from our screen that are associated with locomotor disorders in humans. In addition, while startle behaviors differ between species, the behavioral phenotypes associated with gene mutations are in some instances remarkably similar. For example, mice with kcna1 mutations are hypersensitive to acoustic stimuli, display increased response latency during the acoustic startle response, and demonstrate increased amplitude of the response [88], similar to the acoustic startle phenotypes we observe in kcna1a mutant zebrafish larvae ( [23] and this study). Further understanding of how loss of critical genes affects locomotor behaviors will further shape our models of how molecular pathways act in behavioral circuits to drive movement.

Ethics statement
All animal protocols were approved by the University of Pennsylvania Institutional Animal Care and Use Committee (IACUC).

Behavior testing
Behavioral experiments were performed using 5 dpf larvae (except for Mauthner ablation experiments performed at 6 dpf) and analyzed using FLOTE software as described previously [16]. Larvae were tested in groups in a 6x6 laser cut grid. The acoustic stimuli used for acoustic startle experiments were 25.9 dB based on larval response rates and previous measurements of the speaker [25]. 10 stimuli were given with a 20 sec interstimulus interval. Before FLOTE tracking and analysis, ImageJ was used to remove background from videos to prevent erroneous measurements from the outline of the wells. Latency measurements in Table 1 were determined manually, as many mutants do not have swim bladders and lay on their sides and FLOTE did not always reliably calculate the movement latency. Angle measurements by FLOTE that were incorrect due to larvae laying on their sides and turning in the wrong plane were discarded. For measurements of spontaneous movement, only larvae that moved in a three-minute period were counted. Videos of acoustic startle response were acquired at 1000 fps, and videos of spontaneous movement were acquired at 125 fps. Images in figures have background removed, except for the spinalized larvae, and some images were mirrored to show an initial rightward turn for ease of comparison between genotypes.

Mutagenesis, WGS/WES, and transgenesis
ENU mutagenesis was performed using TLF and WIK as described previously [23]. Cloning of p181 is described in [24].
Alleles p415, p417, and p420 were cloned using WGS as described previously [25]. Alleles p413, p414, p416, and p419 were cloned using WES as described previously [28]. For SNP analysis, we presumed our sibling pools (wild type behavior) consisted of 66% heterozygous (+/-) larvae and 33% (+/+) larvae, which results in a 1:2 ratio of mutant:wild type chromosomes in the analysis. To identify causative mutations and maintain mutant lines, mutations were genotyped using the following primers/reactions: To confirm the splice site mutation in ryr1b p414 , a forward primer was designed in exon 4 and a reverse primer was designed in exon 5. PCR performed on cDNA from ryr1b homozygous mutant larvae resulted in a larger product (245 bp), caused by retention of an intron, than the product from wild type larvae (153 bp).
To generate dolk p421 , a CRISPR-Cas9 Alt-R gRNA (IDT) was designed using CHOPCHOP [91]. gRNA was combined with Cas9 protein (PacBio) and injected into one-cell stage embryos. F1 progeny from injected embryos were screened for mutations by sequencing the region flanking the gRNA site and using the ICE Analysis tool (v2) (Synthego) to identify Cas9 induced indels. Mosaic injected carriers were outcrossed to establish stable lines.
Unless noted otherwise, homozygous mutants were identified by abnormal kinematic behavior and siblings (heterozygous carriers and wild type) were designated as larvae performing normal startle responses. All phenotypes described in the paper display 100% penetrance, based on sequencing of at least twenty larvae displaying the mutant phenotype and at least twenty larvae displaying the wild type phenotype in presumed siblings. We do not observe phenotypes in heterozygous carrier larvae in any of the mutants described. Unless otherwise noted, all mutant alleles discussed in the text are from the forward genetic screen.

Electrophysiology
To perform electrophysiological recordings from kcna1a siblings and mutants, larvae (5-8 dpf from an incross of kcna1a p181/+ ;Tol056) were anesthetized using a 0.03% Tricaine solution pH adjusted to 7.4 with NaHCO 3 . Larvae were then paralyzed using d-tubocurarine (10 μM, Sigma) in external solution in mM: 134 NaCl, 2.9 KCl, 2.1 CaCl 2 , 1.2 MgCl 2 , 10 HEPES, 10 Glucose, pH adjusted to 7.8 with NaOH [49]. Then, larvae were transferred dorsal-side down to a Sylgard-coated small culture dish (FluoroDish, WPI) and immobilized with tungsten pins. The brain was exposed ventrally as described in [17]. Following surgery, the petri dish containing the larva was placed on the stage of an Axio Examiner upright microscope (Carl Zeiss AG) adapted for electrophysiology recordings. Larvae were superfused with external solution throughout the recording session. Mauthner cells were identified by far-red DIC optics and GFP fluorescence. The patch pipette (3-4 MO) was filled with internal solution in mM: 105 K-Methanesulfonate, 10 HEPES, 10 EGTA, 2 MgCl 2 , 2 CaCl 2 , 4 Na 2 ATP, 0.4 Tris-GTP, 10 K 2 -Phosphocreatine, 25 mannitol, pH adjusted to 7.2 with KOH. Whole-cell recordings under current-clamp configuration were performed using a Multiclamp 700B patch-clamp amplifier connected to a Digidata 1440A (Molecular Devices) digitizer. The liquid junction potential was estimated in -16 mV using Clampex 10.6 (Molecular Devices). The rheobase, defined as the minimum amount of positive current needed to elicit an action potential, was determined by delivering a 20 ms current pulse. Voltage threshold was measured as the membrane potential value at which the depolarizing-current step elicits an action potential. The input resistance was estimated using the voltage deflection caused by a hyperpolarizing-current step of -1 nA and 20 ms duration, followed by derivation of resistance with Ohm's law. To activate the auditory afferents to the Mauthner cell, a "theta" septated glass pipette filled with external solution was used as bipolar electrode. The bipolar electrode was placed at the posterior macula of the ear, where the dendritic processes of auditory afferents contact the hair cells [58].

Mauthner cell ablation
Larvae (obtained from a cross of kcna1a p181 /+;Tol056/+;UAS:NTR-RFP/+ and kcna1a p181 /+; 62a:Gal4/+) were sorted at 60 hpf for GFP+ (Tol056) and NTR-RFP+ on a fluorescent dissecting scope. At 4 dpf, larvae were behavior tested and sorted into mutants and siblings. Because 62a>NTR-RFP is mosaic, larvae expressing NTR-RFP in the brain were then mounted in agarose and screened on a compound epifluorescent scope for specific expression of NTR-RFP+ in the Mauthner cells. Only fish where both or neither Mauthner cells displayed expression were used. At 104 hpf, larvae were treated with 10mM Metronidazole/Mtz (Sigma-Aldrich) in 0.2% DMSO in E3. After 24 hours, Mtz was washed out and larvae were mounted and screened again to confirm ablation. At 144 hpf, larvae were behavior tested. Representative images of live fish in Fig 5 were obtained from a Zeiss LSM 880 confocal microscope.

Larvae spinalization
5 dpf larvae were anesthetized with Tricaine (Sigma-Aldrich) in Hank's balanced salt solution supplemented with 1x Glutamax and 1x sodium pyruvate. A sliver from a double edge razor was used to cut through the spinal cord and musculature through into the yolk, posterior to the brain, roughly at the halfway point of the swim bladder. Larvae were mounted in 2% low melt agarose in supplemented Hank's and agarose covering the tail was removed. Larvae were allowed to recover 30 min-2 hours in Hank's solution without Tricaine, until spontaneous movement was observed.

Statistics
Statistical analyses were performed using GraphPad Prism (tests used are listed in figure legends). All plots display mean and standard deviation.
Supporting information S1 Data. Raw data numbers for all graphs.