We sought to examine the relationship between the intracellular scaffold ZO1 and neuronal Connexins (Cxs) by utilizing the stereotyped synapses of the zebrafish Mauthner cell. This circuit drives a fast escape response to threatening stimuli using both electrical and chemical connections (Eaton et al., 1977; Jacoby and Kimmel, 1982; Liu and Fetcho, 1999; Wolman et al., 2015). Each animal has two Mauthner cells that receive multimodal sensory input that relay information to the spinal cord to coordinate circuits to elicit fast turns. We focus on two populations of stereotyped electrical synapses made by Mauthner cells: (1) ‘club ending’ (CE) synapses (Bartelmez and Hoerr, 1933; Pereda et al., 2004; Yao et al., 2014), which are mixed electrical/glutamatergic chemical synaptic contacts formed between auditory afferents of the eighth cranial nerve and the Mauthner cell's lateral dendrite (Figure 1A,B) and (2) en passant electrical synapses formed between the Mauthner axon and Commissural Local (CoLo) interneurons found in each spinal-cord segment (Figure 1A,M/CoLo synapses) (Satou et al., 2009). Neuronal GJs at both CEs and M/CoLo synapses are made of heterotypic channels formed by Cx35.5, encoded by the gene gap junction delta 2a (gjd2a), and Cx34.1, encoded by gjd1a, both homologous to mammalian Cx36 (gjd2). We previously found that Cx35.5 and Cx34.1 are localized asymmetrically to the pre- and postsynaptic sides at CE and M/CoLo synaptic contacts (Figure 1B; Miller et al., 2017). Throughout we use the terms pre- and postsynaptic to reference the neuronal, cell-biological compartment in which a Connexin is localized. At CE synapses, the auditory afferent axons are presynaptic to the postsynaptic Mauthner lateral dendrite; while at M/CoLo synapses, the Mauthner axon is presynaptic to the postsynaptic CoLo.

We first determined the localization of ZO1 and the Connexin proteins at Mauthner electrical synapses using immunofluorescence and confocal imaging. We stained 5 day post fertilization (dpf) larvae, a time at which the Mauthner circuit elicits a mature startle response, with antibodies against the human ZO1 protein and those that distinguish the zebrafish Cx35.5 and Cx34.1 (Figure 1C–L; Miller et al., 2017). We observed extensive colocalized signal for these three proteins at CE and M/Colo electrical synapses, with each protein apparent in the stereotyped shape and position of the neuronal gap junctions (GJs) at these contacts. We identified CEs unambiguously as large (1.5–2 µm) oval areas localized in the distal portion of the lateral dendrite of the Mauthner cell (Figure 1C; Yao et al., 2014). M/Colo synapses were identified by their regularly spaced sites of contact in the spinal cord (Figure 1K). Next, we examined the role of ZO1 at electrical synapses using CRISPR/Cas9-induced mutations to knock out gene function. Mammalian ZO1 is encoded by the gene tight junction protein 1 (tjp1), while zebrafish have two homologous genes, tjp1a and tjp1b. Using a CRISPR-based screen, we found that mutations in tjp1b/ZO1b, but not tjp1a/ZO1a, caused a failure of Connexin localization at M/CoLo synapses (Figure 1; Figure 1—figure supplement 1; Marsh et al., 2017; Shah et al., 2015). We examined the effect of these mutations on CEs and found that tjp1b/ZO1b^-/- mutants lack most of the detectable fluorescent staining for both Cx35.5 and Cx34.1, as well as ZO1, at the stereotyped synaptic contact sites (Figure 1D,L). Quantitation of Cx35.5, Cx34.1, and ZO1 fluorescence at CEs and M/CoLo contacts confirmed that staining for all three antibodies was greatly diminished in tjp1b/ZO1b^-/- mutants (Figure 1M,N). By contrast, homozygous tjp1a/ZO1a ^-/- mutants had extensive Connexin and ZO1 staining at these contacts, while tjp1a^-/-; tjp1b^-/- double mutants were indistinguishable from tjp1b^-/- single mutants (Figure 1—figure supplement 1A–G). We conclude that ZO1b protein, encoded by the tjp1b gene, is localized to electrical synapses and required for the robust localization of both Cx35.5 and Cx34.1 at synaptic contacts.

While tjp1b/ZO1b^-/- mutants showed significantly diminished levels of ZO1 and Connexin staining at presumptive synaptic locations, we wondered whether neurons were still attempting to assemble GJs. Indeed, we detected both Cx35.5 and Cx34.1 by western blot from brain homogenates of tjp1b/ZO1b^-/- mutant animals (Figure 1—figure supplement 1H). We therefore examined CEs using higher contrast and magnification to assess GJ structure as detectable by immunolabeling and individually stained for ZO1 or Connexin to avoid confounding the image analysis due to bleed through of signal amongst stained proteins. We found that in tjp1b/ZO1b^-/- mutants, each of the three antibodies revealed structures located at the stereotyped position of CE contacts and had morphologies reminiscent of wild-type animals, albeit with much dimmer fluorescence intensity (Figure 1E–J; note that image contrast was increased in mutants (H-J)). While we observed the stereotypical oval-shaped CE structures in mutants, the staining for each protein was weak and irregular in its distribution, suggesting the residual staining in mutants might represent incomplete, abortive synaptic structures (see electrophysiology below). Consistent with a reduced presence of GJ proteins, we also observed a reduced number of CEs detected by immunolabeling in tjp1b/ZO1b^-/- mutants (Figure 1—figure supplement 1I). We found similar results at M/CoLo synapses, although their smaller size precluded an analogous detailed analysis (Figure 1K,L,N). In contrast to tjp1b/ZO1b^-/-, the staining of tjp1a/ZO1a^-/- mutants was indistinguishable from wildtype (Figure 1—figure supplement 1E–G,I). These observations suggest that neurons of the Mauthner cell network in tjp1b/ZO1b^-/- mutants persist in attempting to create electrical synapses, despite their inability to robustly localize neuronal Connexins at synaptic contacts. We conclude that ZO1 is localized to electrical synapses where it plays a critical role in neuronal GJ formation.

Given that Connexin localization was dependent on ZO1b, we sought to determine if the converse was true – did ZO1 localization require Connexins? Using previously generated mutations in gjd2a/Cx35.5 and gjd1a/Cx34.1 (Miller et al., 2017), we examined the localization of Connexin and ZO1 proteins in mutants by immunolabeling (Figure 2A–L). First, we found that gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutant animals revealed a complete loss of detectable staining for the mutated protein. In addition, there was a failure of the non-mutated Connexin protein to robustly localize to the electrical synapse although low levels of staining were present. In line with these observations, by using brain homogenates and western blots, we found a complete loss of the Connexin affected by each mutation, but no effect on the non-mutated Connexin protein (Figure 1—figure supplement 1H). By contrast, ZO1 staining in the Connexin mutants was robust at the synaptic contact sites with the stereotyped appearance, distribution, and position clearly evident (Figure 2A–L). By comparing the relative ZO1 fluorescence between wild-type and Connexin mutant animals, we found that ZO1 was present at synaptic contacts at approximately half the normal level (Figure 2M,N). We examined gjd2a^-/-; gjd1a^-/- double mutants and found that ZO1 still robustly localized to CE and M/CoLo contact sites (Figure 2M,N; Figure 2—figure supplement 1A–F). These results reveal two critical organizational principles about electrical synapses in the Mauthner cell: (1) ZO1b localizes to putative electrical synaptic sites largely independent of Connexin proteins and (2) each Connexin requires the other for robust localization to the synapse. Based on these data, we conclude that ZO1 can localize to neuronal GJs independent of the presence of channel-forming proteins, yet ZO1 is absolutely essential for proper Connexin localization.

To further examine the electrical synapse structure of Connexin mutants, we assessed CEs at higher contrast (Figure 2D–I). We found that: (1) immunofluorescence for the mutated Connexin is completely lost at the synapse, (2) the non-mutated Connexin is detectable but with weak and irregular labeling, suggestive of incomplete, abortive structures, and (3) ZO1 labeling resembles wildtype with a distribution and morphology that appears normal across the expanse of the putative synaptic contact. Consistent with these findings, the number of CEs detected by ZO1 labeling in Connexin mutants was indistinguishable from that observed in wildtype, while those detected by antibodies for the mutated Connexin were significantly reduced (Figure 1—figure supplement 1I). In addition, the zebrafish genome contains two additional homologous Connexin genes, gjd2b/Cx35.1 and gjd1b/Cx34.7. We found that animals that were homozygous mutant for these two genes had normal Connexin and ZO1 labeling at CEs (Figure 2—figure supplement 1G–L). Similarly, we previously found there was no effect on M/CoLo synapses in the gjd2b/Cx35.1 and gjd1b/Cx34.7 mutants (Miller et al., 2017). Together, these results support a hierarchical relationship in the formation of neuronal GJs, in which ZO1 localizes to electrical synaptic contact sites where it is essential to robustly localize neuronal Connexins.

We next sought to examine the functional consequences of ZO1 and Connexin mutants, so we explored the properties of synaptic transmission at CEs during whole-cell recordings of the Mauthner cell. In wildtype zebrafish, extracellular stimulation of CE afferents near the posterior macula where they contact hair cells (Figure 3A) evoked a mixed synaptic response in the Mauthner cell composed of an early and large GJ-mediated electrical component followed by a delayed and smaller glutamatergic chemical response (Figure 3B; Yao et al., 2014). We first aimed to establish the functional consequences of removing the Connexins on this mixed synaptic response. Consistent with the presence of Cx35.5 and Cx34.1 at CEs, synaptic responses in gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutant zebrafish lacked a detectable electrical component, while exhibiting a response with the same delay as the chemical component of the mixed synaptic potential of wildtypes (Figure 3C–E). By contrast, electrical transmission was unaffected in gjd2b/Cx35.1^-/- and gjd1b/Cx34.7^-/- mutant zebrafish (Figure 3E; Figure 3—figure supplement 1A-B), as expected given our immunolabeling results (Figure 2—figure supplement 1G-L). The apparent chemical response in gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutants was blocked by extracellular application of a combination of the AMPA and NMDA glutamate receptor (GluR) antagonists cyanquixaline (CNQX) and D-2-Amino-5-phosphonovaleric acid (DAP5)(gray traces in Figure 3C,D; Figure 3—figure supplement 1C). No change in membrane potential was observed after application of the blockers. To confirm that the chemical synaptic potential observed in gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutants arises from the stimulation of CEs lacking electrical transmission, we examined if blocking GJs with meclofenamic acid (MA) could reproduce the observed synaptic phenotype. We found that application of MA to wildtype recapitulated the gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutant phenotype, as the characteristic mixed synaptic response was replaced by a larger chemical synaptic response that was sensitive to GluR antagonists (Figure 3F; Figure 3—figure supplement 1C). We also observed a small hyperpolarization of the Mauthner cell (from −80.2 ± 0.7 mV in control to −84 ± 1 mV in MA; p=0.04, n = 5), likely resulting from the action of MA on other membrane channels. We conclude that together Cx35.5 and Cx34.1 generate functional neuronal GJs at CEs.

We next examined the properties of synaptic transmission in ZO1 mutants. Strikingly, and consistent with the requirement for Connexin localization at contact sites (Figure 1), tjp1b/ZO1b^-/- mutants exhibited a failure in electrical transmission. The synaptic phenotype was indistinguishable from gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutants, and similarly consisted of a single, delayed response that was blocked by GluRs antagonists (Figure 3G; Figure 3—figure supplement 1D). This functional deficit was specific for ZO1b, as tjp1a/ZO1a^-/- mutant fish exhibited normal mixed synaptic responses (Figure 3H,I). These findings indicate the specificity of the mutants to electrical synapses as glutamatergic transmission at CEs remained intact. Accordingly, neurotransmitter release properties (Figure 3—figure supplement 1E) and the localization of GluR2/3 receptors at these terminals were not affected in mutant fish (Figure 3—figure supplement 2). We conclude that the presence of the ZO1b scaffold protein is critical for electrical transmission at CEs, even though its absence does not prevent the formation of glutamatergic synapses coexisting within the same contact.

To investigate the extent of the deficit on electrical transmission within the brainstem network, we investigated whether Connexin and ZO1 mutations affected other synaptic contacts onto the Mauthner cell. The Mauthner cell receives mixed synaptic inputs from a variety of descending and ascending sensory modalities, including visual information from the optic tectum and somatic information from the spinal cord (Dunn et al., 2016; Faber and Pereda, 2011; Kimmel et al., 1981; Korn and Faber, 2005). For this purpose, we recorded spontaneous synaptic responses that represent the activity of most, if not all, synaptic inputs received by this cell. Electrically and chemically mediated spontaneous responses are easily differentiated due to their dramatically different duration (Figure 4A). Thus, for automated detection purposes, we defined fast spontaneous events (<1.1 ms), which represent the electrical coupling of presynaptic spikes, as electrical responses and slow spontaneous events (>1.1 ms) as chemical responses. Fast events were nearly absent in tjp1b/ZO1b^-/-, gjd2a/Cx35.5^-/-, and gjd1a/Cx34.1^-/- mutants (Figure 4B,C), and were dramatically reduced in wild-type fish by application of MA (Figure 4C), confirming that they represent electrical transmission. Slow events that remained after losing the electrical component were dramatically reduced by application of GluRs antagonists (Figure 4D), confirming that they represent chemical responses. We conclude that the deficit in electrical transmission observed in mutant zebrafish is likely to be widespread within hindbrain and spinal cord circuits.

Given the primary role of the Mauthner cell in triggering escape responses, we also investigated possible changes in cellular excitability in mutant fish. Interestingly, the frequency of slow, chemical responses was increased in mutants and MA-treated animals (Figure 4D), even though we observed no changes on presynaptic neurotransmitter release properties (Figure 3—figure supplement 1E). This suggests that the lack of electrical transmission could have enhanced the detection of smaller amplitude chemical responses. We posit this could arise from electrical coupling influencing neuronal excitability, as the loss of neuronal GJs in mutants would increase the input resistance of the Mauthner cell (Alcamí and Pereda, 2019). Thus, during whole cell recordings, we determined the input resistance (R_in), rheobase, resting potential (V_rest), and firing threshold (V_threshold) of the Mauthner cells from wildtype and mutant zebrafish (Table 1). The input resistance, the main determinant of neuronal excitability, was increased in tjp1b/ZO1b^-/-, gjd2a/Cx35.5^-/-, and gjd1a/Cx34.1^-/- mutants. Accordingly, the rheobase, a parameter negatively correlated with neuronal excitability, was decreased in these mutant animals (Table 1). We found that the change in R_in correlated with the lack of electrical transmission and was not observed in fish that retained electrical transmission (Figure 4E,F; note however that R_in in tjp1a/ZO1a^-/- was found to slightly increase). Differences in magnitude of the effects observed on R_in could be due to distinct compensatory mechanisms in each case or to the mutations affecting channels contributing to leak conductance in the Mauthner cell. Thus, the lack of electrical transmission in tjp1b/ZO1b^-/- fish rendered synaptic transmission exclusively mediated by a relatively delayed glutamatergic response and more excitable Mauthner cells. Both deficits likely influence the behavioral responses generated by the Mauthner cell and its associated network.

Since tjp1b/ZO1b^-/- mutants had functional deficits in electrical transmission, but not chemical, we assessed the consequences on the behavioral output of the Mauthner cell network. Animals were placed into individual chambers of a multi-well testing stage and presented with acoustic-vibrational stimuli to elicit Mauthner-dependent startle responses (Wolman et al., 2015). Movements were captured with a high-speed camera (1000 frames per second) and analyzed with FLOTE software to automatically track and measure the kinematics (body movements) of responses (Burgess and Granato, 2007). In this paradigm, wild-type fish exhibit two types of escape responses: (1) Mauthner-cell-dependent short-latency C-bends (SLCs, hereafter referred to as ‘startles’) and (2) Mauthner-cell-independent long-latency C-bends (LLCs). These two behavioral responses were automatically distinguished using well-established kinematic parameters (Burgess and Granato, 2007). Larvae generated from crossing tjp1b/ZO1b^+/- heterozygous animals were tested and analyzed blind to genotype with subsequent post-hoc identification. We found that tjp1b/ZO1b^-/- mutants startled to strong acoustic stimuli (25.9 dB) as frequently as their wildtype siblings (Figure 5A). While these turns were classified as startles, we found that the tjp1b/ZO1b^-/- mutants initiated their responses ~ 2 ms slower than their wildtype siblings (Figure 5B). A delayed behavioral response in consistent with our electrophysiological findings indicating that the mutant escape network operates with longer synaptic delays due to the lack of electrical transmission (Figure 3). In mutant animals, we found that the escapes were often performed with the normal startle kinematic parameters, particularly the maximum angle of the turn and the maximum angular velocity of the response, albeit occurring later than in wild-type siblings (Figure 5C–H). However, we found a subset of mutant responses (~15%) that showed abnormally shallow and slow turns (Figure 5C,D red arrows). These abnormal responses suggested deficits in performing the stereotyped C-bend elicited by the Mauthner-cell network, and so we reanalyzed the video data from these responses. In these startles, we observed that the mutants displayed abnormal postures where the body would bend slightly to one side, creating ‘kinked’ or ‘S-shaped’ postures (Figure 5I–L). We note that the phenotypes observed in the tjp1b/ZO1b^-/- mutants are strikingly similar to those we previously observed in gjd2a/Cx35.5^-/- and gjd1a/Cx34.1^-/- mutants (Miller et al., 2017). Such kinked body shapes are reminiscent of startles following CoLo neuron ablation (Satou et al., 2009). Based on these data, we conclude that electrical synapses are essential for generating the speed and coordination of the Mauthner-induced startle response.

The electrophysiological analysis of tjp1b/ZO1b^-/- mutants also revealed that Mauthner cells showed increased excitability (Figure 4), suggesting animals may be hypersensitive with a lower threshold of response for environmental stimuli. To examine this possibility, we presented larvae with 60 pseudo-randomized stimuli, 10 at six different intensities with a 20 s inter-stimulus interval to eliminate habituation (Wolman et al., 2015). We then assessed the frequency with which animals responded to the stimuli with turns (Figure 5M–O). As stimulus intensity increased, both wildtype siblings and tjp1b/ZO1b^-/- mutants increased the likelihood of performing a startle response. However, across the mid-range of stimulus intensities, tjp1b/ZO1b^-/- mutants were more likely to respond with a startle than their wild-type siblings (Figure 5M,O). The increased tendency of mutants to perform Mauthner-dependent startles came at the expense of Mauthner-independent LLC responses, which were nearly absent in mutants (Figure 5N). We conclude that electrical synapses alter the sensitivity of Mauthner cells to environmental stimuli and contribute to the innate startle threshold, which alters the probability of eliciting a startle or LLC response. These results lend additional evidence for the critical role of ZO1b for creating functional electrical synapses, ultimately contributing to appropriately balanced neural network function and behavior.

Mutant zebrafish revealed a hierarchical relationship between ZO1b and neuronal Connexins (Figures 1 and 2), so we next investigated the mechanisms underlying this relationship. ZO1 is a membrane-associated guanylate kinase (MAGUK) scaffold protein and contains PSD95/Dlg/ZO1 (PDZ) protein-protein interaction domains that bind to PDZ binding motifs (PBMs) (Zhu et al., 2016). Previous work demonstrated that the C-terminal four amino acids of mouse Cx36 and perch Cx35 compose PBMs that are essential for interacting with ZO1 (Flores et al., 2008; Li et al., 2004). Given that the PBM sequence is conserved in zebrafish Cx35.5 and Cx34.1 proteins (Figure 6A), we tested whether these Connexins could mediate binding to zebrafish ZO1b. We cloned full-length sequences of tjp1b/ZO1b, gjd2a/Cx35.5, and gjd1a/Cx34.1 and used heterologous expression to test for interactions between the scaffold and Connexins. HEK293T cells were co-transfected with mVenus-ZO1b and full-length Cx35.5 or Cx34.1. Using western blot analysis with antibodies specific to each Connexin, we found that both Cx34.1 and Cx35.5 were individually detected in mVenus-ZO1b immune complexes (Figure 6B, lanes 1,3) compared to control immunoprecipitates (Figure 6—figure supplement 1A, lanes 1,2,5,6). We further found that removing the presumptive PBMs, by deleting the C-terminal four amino acids from both Connexins, resulted in a loss of co-purifying Connexins from mVenus-ZO1b immunoprecipitates (Figure 6B, lanes 2,4; Figure 6—figure supplement 1A, lanes 4,8). Control blots demonstrated the ability of the Connexin antibodies to equally recognize both full-length and ∆PBM versions of the proteins (Figure 6B; Figure 6—figure supplement 1A, bottom input panels). We conclude that zebrafish Cx35.5 and Cx34.1 can interact with ZO1b in a PBM-dependent manner.

The ZO1b scaffold has three PDZ domains that could mediate the interaction with neuronal Connexins (Figure 6A). Previous studies testing the three mammalian ZO1 PDZ domains demonstrated that Cx35/36 PBMs exclusively bound to ZO1-PDZ1, while PDZ2 and PDZ3 did not interact (Flores et al., 2008; Li et al., 2004). Given the high degree of conservation of the zebrafish ZO1 PDZ1 domain, including the amino acids of the putative PBM binding pocket (Figure 6A), we tested whether zebrafish ZO1b-PDZ1 could directly interact with zebrafish neuronal Connexins. To examine this question, we isolated the minimal domains of each zebrafish protein, produced them in bacteria, and performed in vitro binding studies. We found that purified ZO1b-PDZ1 could be pulled down with a GST-Cx34.1 or a GST-Cx35.5 C-terminal intracellular-tail (Figure 6C, lanes 2,4), but not with control GST protein (Figure 6C, lane1). Further, this interaction was significantly decreased when the predicted PBM in the Connexin tails were removed (Figure 6C, lanes 3,5). We next used an overlay assay to compare the ability of ZO1b-PDZ1 and ZO1b-PDZ2 to bind to immobilized GST-Connexin tails. Similar to the binding assays, significant amounts of ZO1b-PDZ1 bound to GST-Cx34.1 and GST-Cx35.5 tails in a PBM-dependent manner (Figure 6—figure supplement 1B, left panels). By contrast, little ZO1b-PDZ2 bound to the GST-Cx34.1 or GST-Cx35.5 tails, particularly when compared to the non-neuronal Cx43 C-terminal tail (Figure 6—figure supplement 1B, right panels), which is known to bind PDZ2 (Flores et al., 2008; Li et al., 2004). We conclude that ZO1b utilizes the PDZ1 domain to directly interact with the neuronal Connexin PBMs.

Since ZO1b and the neuronal Connexins colocalized at electrical synapses (Figure 1), we next investigated whether these proteins interacted in vivo. We utilized adult zebrafish brains that maintain widespread electrical synapses, including those in Mauthner cell (Kimmel et al., 1981), and provide an abundant source to derive ZO1b immunoprecipitates. Homogenates derived from wild-type fish brains were immunoprecipitated with anti-ZO1 and control antibodies (mIgG). Immunoprecipitates demonstrated that Cx34.1 copurified with ZO1, whereas Cx35.5 did not copurify with the anti-ZO1 antibody (Figure 6D, lanes 1,2). To confirm that copurification of Cx34.1 was dependent upon ZO1b, we replicated the experiment using homogenates from tjp1b/ZO1b^-/- mutants and found that Cx34.1 was lost in these immmunocomplexes (Figure 6D, lane4). We conclude that ZO1 preferentially interacts with Cx34.1 in vivo. We observed multiple ZO1 bands upon Western analysis of the immunoprecipitates, several of which were lost in the tjp1b/ZO1b^-/- mutants (Figure 6D, lanes 2,4). Since the ZO1 antibody was made against human protein, we reasoned it might be detecting the ZO1b protein and the highly similar ZO1a protein, so we sought evidence to confirm that ZO1b was the primary scaffold for Cx34.1 in vivo. Upon examining ZO1 immunocomplexes from wildtype, tjp1b/ZO1b^-/-, and tjp1a/ZO1a^-/- brains, we found that only ZO1b deficiency resulted in concomitant loss of Cx34.1 (Figure 6—figure supplement 1C). Taken together, we conclude that ZO1b preferentially interacts with Cx34.1 in vivo, despite the fact that the scaffold can interact with either neuronal Connexin.

Next, we determined the functional relevance for a preferential ZO1b/Cx34.1 interaction at zebrafish electrical synapses. We previously observed that Cx35.5 localizes presynaptically in axons, while Cx34.1 localizes postsynaptically in dendrites (Miller et al., 2017). Since ZO1b preferentially interacts with Cx34.1 in vivo (Figure 6), we speculated that ZO1b might share a similar dendritically compartmentalized localization. To directly examine this, we visualized the localization of ZO1b protein after inserting a V5 epitope at the N-terminus of the endogenous tjp1b locus (Figure 7—figure supplement 1A). We found that in transgenicV5-tjp1b larvae, V5 antibody staining colocalized with both Cx34.1 and Cx35.5 at CEs and M/CoLo synaptic contacts (Figure 7—figure supplement 1B–E). Moreover, we found no effect on the stereotyped patterns of Connexin staining at Mauthner electrical synaptic contact sites in homozygous V5-tjp1b/V5-tjp1b animals (Figure 7—figure supplement 1D,E), suggesting that V5-ZO1b is functional. While V5-tjp1b larvae permitted ZO1b visualization at electrical synapses, we could not discriminate whether it was localized asymmetrically at synaptic contacts due to the small size of these structures and the resolution limits of light microscopy. To overcome this, we utilized an alternate approach to address ZO1b compartmentalization, exploiting the fact that the Mauthner cell is both the postsynaptic partner at CEs and the presynaptic partner at the M/CoLo synapses (Figure 7A). We generated chimeric embryos by blastula transplantation (Kemp et al., 2009), extracted GFP and V5-ZO1b expressing transgenic cells from donor embryos, and transferred them to wild-type hosts (Figure 7B), allowing us to address V5-ZO1b localization within the Mauthner cell (Figure 7C–E). In animals containing only a donor-derived, V5-ZO1b-expressing Mauthner cell, we found that V5 staining was present at the CEs with Connexin staining, demonstrating that ZO1b was within the dendrite of the Mauthner cell and localized postsynaptically at electrical synapses (Figure 7D). Conversely, when we examined the M/CoLo synapses of these same embryos, we found that V5 staining was not present at these synaptic contacts despite the Mauthner cell expressing V5-ZO1b protein (Figure 7E). We note that in the non-chimeric, V5-ZO1b transgenic animals, V5 staining was observed at the M/CoLo synapses (Figure 7—figure supplement 1C,E), suggesting that ZO1b at these synapses derives from the postsynaptic CoLo. Our transplant experiments produce chimeric larvae in which only Mauthner, only CoLo, or both cells are derived from the transgenic donor embryos (Figure 7—figure supplement 1F–H). In larvae in which only CoLo expresses V5-ZO1b, we found that V5 staining is present at M/CoLo synapses (Figure 7—figure supplement 1G), confirming ZO1b’s postsynaptic localization. We conclude that ZO1b is robustly compartmentalized within the somato-dendritic compartment of the neuron and asymmetrically localized on the postsynaptic side of the electrical synapse.

We then addressed whether ZO1b functions postsynaptically to facilitate Connexin localization at the electrical synapse. We utilized chimeric animals and again took advantage of Mauthner cell morphology. However, in these experiments, we transplanted GFP-expressing cells from tjp1b/ZO1b^-/- mutant donors into wildtype hosts, producing animals in which ZO1b was specifically removed from the Mauthner cell (Figure 7F–I). At CEs, the removal of ZO1b exclusively from the Mauthner cell resulted in the loss of staining for ZO1, Cx34.1, and Cx35.5 (Figure 7H; Figure 7—figure supplement 1I). The data indicates that ZO1b is required postsynaptically for the cell autonomous localization of Cx34.1 and non-autonomously for presynaptic Cx35.5 localization. By contrast, when we examined the same chimeric animals with ZO1b removed from the Mauthner cell but focused on the M/CoLo synapses, in which Mauthner is presynaptic, there was no effect on either Connexin or ZO1 staining (Figure 7I; Figure 7—figure supplement 1J), indicating that ZO1b is dispensable presynaptically. As further support for this compartmentalized scaffold function, we reasoned that resupplying ZO1b to the Mauthner cell in an otherwise tjp1b/ZO1b^-/- mutant animal would be sufficient to rescue Connexin localization at CEs, but not at M/CoLo synapses. To test this prediction, we transplanted GFP-expressing cells from wildtype donor embryos into tjp1b/ZO1b^-/- mutant hosts and identified animals with donor-derived Mauthner cells. In such chimeras, where the tjp1b/ZO1b gene is functional only in the Mauthner cell, CEs had normal staining for both postsynaptic ZO1 and Cx34.1 and also for presynaptic Cx35.5. By contrast, there was no rescue of staining for these proteins at M/CoLo synaptic contacts (Figure 7J,K; Figure 7—figure supplement 1J). We conclude that ZO1b is both necessary and sufficient postsynaptically for building the structure of the neuronal GJ channels. Taken together, we find that ZO1b is exclusively localized to the postsynaptic compartment where it functions both cell-autonomously and non-autonomously to localize Connexins and build functional neuronal gap junctions.

Despite the continuous nature of electrical transmission, the thousands of channels that make up GJ plaques found at electrical synapses are maintained by the active turnover of Connexin proteins (Flores et al., 2012), similar to neurotransmitter receptors at chemical synapses (Carroll and Zukin, 2002; Chen et al., 2000; Ehlers, 2000; Lüscher et al., 1999). Our findings show that ZO1b is required for the robust localization of Connexins, suggesting it functions to stabilize Connexin hemichannels at the synaptic site. Additionally, ZO1 can localize to electrical synapses in the absence of the Connexins, although its apparent concentration was diminished. This suggests that building robust neuronal GJ structures involves a reciprocal interaction between ZO1 and the Connexins. Strikingly, our results reveal that ZO1b is compartmentalized to the dendrite and functions asymmetrically at postsynaptic sites of the Mauthner cell circuit. This localization is consistent with ZO1b’s preferential in vivo interaction with Cx34.1, as this Connexin is also localized and required postsynaptically at Mauthner cell electrical synapses (Miller et al., 2017). Despite the apparent autonomous ZO1b/Cx34.1 postsynaptic interaction at the GJ hemiplaque, tjp1b/ZO1b-/- mutants revealed that presynaptic Cx35.5 localization is also affected non-autonomously in the neighboring cell. This trans-synaptic interaction likely occurs via the Connexins themselves, as mutations to the postsynaptic Cx34.1 prevented the robust localization of the presynaptic Cx35.5. Our results are complementary to recent analysis of the mouse rod/cone network, where removing Cx36 from one neuron of a coupled pair results in the failure of Connexin localization in the adjacent neuron (Jin et al., 2020). Taken together, our results reveal that ZO1b acts as a postsynaptic molecular scaffold that localizes Cx34.1 to the GJ hemiplaque, which in turn ensures Cx35.5 stabilization at presynaptic hemiplaques (Figure 7C). Whether presynaptic Cx35.5 lacks an in vivo scaffold, or utilizes another unidentified presynaptic scaffolding protein, remains unresolved. If the molecular function and organization of ZO1 revealed here applies to all electrical synapses, including those formed by homotypic channels between various homologous cellular processes (dendro-dendritic, somato-somatic, or axo-axonic), remains to be determined in future studies. Never-the-less, we posit that the molecular organization of the electrical synapse can be asymmetrically compartmentalized, thereby enabling preferential biochemical interactions at each side of the junction.

Our results support the prediction that ZO1 likely plays distinct functional roles at GJs in different tissue types. ZO1 was first described at epithelial tight junctions and later shown to interact with various Connexins, notably with Cx43, a widespread Connexin expressed in many non-neuronal cell types (Giepmans, 2004). Evidence from cell expression systems suggested that ZO1 played critical functions on the periphery of Cx43-contaning GJ plaques forming part of the ‘perinexus’ to facilitate newly inserted hemichannels at each side of the junction (Rhett and Gourdie, 2012). Further, the ZO1/Cx43 interaction was critical for GJ communication before the channel ‘ages’ and is subsequently removed during channel turnover, a process governed by Cx43 phosphorylation (Laird, 1996; Laird, 2006; Márquez-Rosado et al., 2012; Solan and Lampe, 2016; Thévenin et al., 2017). However, preventing the ZO1/Cx43 interaction does not prevent GJ formation (Hunter and Gourdie, 2008; Hunter et al., 2003; Hunter et al., 2005). By contrast, our results demonstrated that ZO1b’s presence was asymmetric and required for robust Connexin localization and synaptic function. Beyond our observations here, ZO1 is likely to have homeostatic functions in the modulation of Connexin usage at electrical synapses. For example, analysis of the interactions between ZO1 and Cx36 (and its fish homologs) revealed the interactions occurs via a different PDZ domain than the interaction with Cx43 and the interaction with Cx36 has lower affinity and faster kinetics than that of Cx43 (Flores et al., 2008; Li et al., 2004). This suggests ZO1 has a more dynamic interaction with neuronal Connexins, which may serve the plastic, activity-dependent regulation of electrical transmission observed in fish and mammalian electrical synapses (Haas et al., 2011; Landisman and Connors, 2005; Mathy et al., 2014; Pereda and Faber, 1996; Pereda et al., 1998; Turecek et al., 2014; Yang et al., 1990). Thus, our findings suggest that ZO1’s function at electrical synapses differs from its role at Cx43-containing GJs, perhaps serving a specialized function in synaptic communication.

Our results highlight that neural circuit function requires functional electrical and chemical synapses to create an appropriate behavioral response. Our data indicates that the lack of electrical transmission in ZO1b and Connexin mutants did not prevent the formation of co-existing glutamatergic synapses at CEs on the Mauthner cell. The remaining chemical transmission supported a behavioral response organized by the Mauthner-cell network, although importantly, the response showed deficits in performance and altered sensitivity. Given that the startle behavior mediates predator avoidance, these defects would likely be detrimental to survival (Hecker et al., 2020). The lack of effect on glutamatergic transmission contrasts a wealth of data supporting a strong, interdependent relationship between the formation of electrical and chemical synapses during development in both invertebrate and vertebrate nervous systems (Jabeen and Thirumalai, 2018). For example, in the leech, knockdown of Innexins at a developmental stage where synaptic contacts are solely electrically coupled prevents the formation of later-forming chemical transmission (Todd et al., 2010). Similarly, in the developing mouse neocortex, dominant negative constructs of Cx26 prevent the initial electrical coupling and subsequent chemical synapse formation amongst sister excitatory neurons within ontogenetic columns (Yu et al., 2012). Our findings indicate that this deficit does not occur at zebrafish CEs when Cx35.5, Cx34.1, and ZO1b are removed. Whether this is due to the differences in the GJ proteins used in the Mauthner cell or instead due to mechanisms that specify CE formation, remains unknown. One intriguing possibility is that other proteins may act as a common synaptic control mechanism that is independent of GJ-forming proteins. Indeed, mutations in the scaffold Neurobeachin caused parallel defects in the formation of both electrical and chemical synapses of the Mauthner circuit (Miller et al., 2015), yet the mechanism by which this coordination occurs remains to be elucidated. Alternatively, rather than functional (conductive) GJ channels, other structural components of the electrical synapse may be sufficient to trigger chemical synapse formation via protein-protein interactions. We posit such interactions would apply to mammalian electrical synapses, which can co-exist with neighboring glutamatergic synapses at distances comparable to those found in fish mixed synapses (Nagy et al., 2018) and may mediate similar functional interactions.

Identifying the functional relevance of an intracellular scaffolding protein as a critical part of the electrical synapse draws parallels with our understanding of chemical transmission, where neurotransmitter receptors are clustered and modified by a rich network of postsynaptic proteins that dynamically shape synaptic structure and function. This chemical synapse protein network is known as the ‘postsynaptic density’ (PSD), a term resulting from its structural identification by EM (Cohen, 2013; Palay, 1956) and is composed of hundreds of unique proteins (Grant, 2019). Mirroring those findings, EM images of electrical synapses in mammals (Llinas et al., 1974) and fish (Brightman and Reese, 1969) revealed the presence of clearly identifiable electrodense structures, first described as ‘semi-dense material’ by Sotelo and Korn, 1978. These electrodense structures are localized intracellularly and form an undercoating band at neuronal GJs. The identified structures are presumably formed by a proteinaceous ‘organelle’ that resembles that found at PSDs (Feng et al., 2019). As evidence grows for an array of proteins localized to electrical synapses, we propose that ZO1 is member of such an organelle.

Our data provide enticing hints that the molecular framework of the electrical synapse extends beyond the ZO/Connexin interaction. In particular, the fact that ZO1 can localize to sites of synaptic contact independent of the Connexins, and that there remains immunofluorescent staining at Mauthner cell synapses in tjp1b/ZO1b^-/- mutants, both imply the existence of additional proteins that contribute to this synaptic structure. We presume the remaining ZO1 staining in tjp1b/ZO1b^-/- mutants comes from either the paralogous ZO1a protein (tjp1a) or from the related ZO2 (tjp2a, tjp2b) and ZO3 (tjp3) proteins. However, our previous analysis of ZO2/ZO3 mutants in zebrafish did not reveal overt defects in Connexin localization (Marsh et al., 2017), yet both proteins are localized to mammalian electrical synapses (Li et al., 2009). Whether these related scaffolds have functional roles at electrical synapses that were undetected in our initial screen remains to be determined. Beyond the ZO-family, other molecules can directly interact with Cx36 and/or localize at mammalian electrical synapses, such as cell adhesion molecules, cytoskeletal interacting proteins, and molecules that regulate Connexin post-translational modifications (Lynn et al., 2012; Martin et al., 2020; O'Brien and Bloomfield, 2018). Yet, the molecular roles of these proteins at the electrical synapse remain poorly defined. Therefore, as opposed to simple aggregates of intercellular channels, we propose that electrical synapses are complex synaptic structures at which communicating pre- and postsynaptic Connexin hemichannels are governed by a yet to be determined mechanism that builds an asymmetric molecular scaffold. Based on its analogy to the known functions of glutamatergic PSDs, we propose to name this organizational organelle the ‘electrical synapse density’, or ‘ESD’ (Lynn et al., 2012; Miller and Pereda, 2017; Figure 7C).

Chemical synapses are organized to match their specific functional requirements by combining presynaptic release properties with unique combinations of postsynaptic receptors. Electrical synapses also provide a variety of specific synaptic functions, yet we know little about the source of their diversity. Work in C. elegans exposed the large variety of Innexin expression in neurons contributing to neural circuits, with dozens of potentially unique cellular combinations that are altered during development and following environmental stress (Bhattacharya et al., 2019). These observations have greatly increased the appreciation of the incidence and potential functional diversity of electrical transmission in invertebrates. Our results indicate that the complexity of electrical synaptic transmission must also include their molecular scaffolds. Scaffold diversity may be more relevant for the function of vertebrate electrical synapses, which in contrast to C. elegans, are formed by a smaller number of GJ-forming proteins, with most electrical communication being reliant on Cx36-related proteins investigated here. By promoting and regulating channel trafficking and regulatory molecules, ZO1, and other scaffolding proteins, could support a variety of functions by governing channel function and the local regulatory environment. Thus, future investigations will further expose the molecular composition and functional roles of ZO1 and the ESD. Unraveling the functional complexity of the ESD will lead to a deeper understanding of the diversity of the molecular organization underlying electrical transmission and its contributions to brain function.

Fish were maintained in the University of Oregon’s and the Albert Einstein College of Medicine fish facilities with approval from Institutional Animal Care and Use Committees of each institution. Zebrafish, Danio rerio, were bred and maintained at 28°C on a 14 hr on and 10 hr off light cycle. Animals were housed in groups, generally of 25 animals per tank. Development time points were assigned via standard developmental staging (Kimmel et al., 1995). All fish used for this project were maintained in the ABC background developed at the University of Oregon. Most fish had the enhancer trap transgene zf206Et (M/CoLo:GFP) in the background (Satou et al., 2009), unless otherwise noted. Mutant lines were genotyped for all experiments. All immunohistochemistry, electrophysiological, and behavioral experiments were performed at 5 dpf. At this stage of development, zebrafish sex is not yet determined (Wilson et al., 2014). Protein extractions were performed from both male and female adult brains and combined.

HEK293T/17 verified cells were purchased from ATCC (CRL-11268; STR profile, amelogenin: X). Cells were expanded and maintained in Dulbecco's Modified Eagle's Medium (DMEM, ATCC) plus 10% fetal bovine serum (FBS, Gibco) at 37°C in a humidified incubator in the presence of 5% CO₂. Low passage aliquots were cryopreserved and stored according to manufacturer’s instructions. Cells from each thawed cryovial were monitored for mycoplasma contamination using the Universal Mycoplasma Detection Kit (ATCC, 30–1012K).

Anesthetized, 5–6 days post fertilization (dpf) larvae were fixed for 3 hr in 2% trichloroacetic acid in PBS. Fixed tissue was washed in PBS + 0.5% Triton X-100, followed by standard blocking and antibody incubations. Primary antibody mixes included combinations of the following: rabbit anti-Cx35.5 (Fred Hutch Antibody Technology Facility, clone 12H5, 1:800), mouse IgG1 anti-Cx35.5 (Fred Hutch Antibody Technology Facility, clone 4B12, 1:250), rabbit anti-Cx34.1 (Fred Hutch Antibody Technology Facility, clone 3A4, 1:250), mouse IgG2A anti-Cx34.1 (Fred Hutch Antibody Technology Facility, clone 5C10A, 1:350), mouse IgG1 anti-ZO1 (Invitrogen, 33–9100, 1:350), mouse IgG2a anti-V5 peptide (Invitrogen, R960-25, 1:50), and chicken anti-GFP (abcam, ab13970, 1:350- 1:500). All secondary antibodies were raised in goat (Invitrogen, conjugated with Alexa-405,–488, −555, or −633 fluorophores, 1:500). Tissue was then cleared stepwise in a 25%, 50%, 75% glycerol series, dissected, and mounted in ProLong Gold antifade reagent (ThermoFisher, P36930). Images were acquired on a Leica SP8 Confocal using a 405-diode laser and a white light laser set to 499, 553/554/557 (consistent within experiments), and 631 nm, depending on the fluorescent dye imaged. Each laser line’s data was collected sequentially using custom detection filters based on the dye. Quantitative images of the Club Endings (CEs) were collected using a 63x, 1.40 numerical aperture (NA), oil immersion lens, and images of M/Colo synapses were collected using a 40x, 1.20 NA, water immersion lens. For each set of images, the optimal optical section thickness was used as calculated by the Leica software based on the pinhole, emission wavelengths, and NA of the lens. Within each experiment where fluorescence intensity was to be quantified, all animals (including 3–5 wildtype controls) were stained together with the same antibody mix, processed at the same time, and all confocal settings (laser power, scan speed, gain, offset, objective, and zoom) were identical. Multiple animals per genotype were analyzed to account for biological variation. To account for technical variation, fluorescence intensity values for each region of each animal were an average across multiple synapses.

For high-contrast imaging of the CEs, fixed samples were washed three times with PBS, incubated at 4°C overnight, and hindbrains were dissected out. Dissected hindbrains were washed, blocked, and stained as above with the following primary antibodies: rabbit anti-Cx35.5 (12H5, 1:500), mouse anti-Cx35/36 (EMD Millipore, MAB3045, 1:250), mouse IgG2A anti-Cx34.1 (5C10A, 1:200), mouse IgG1 anti-ZO1 (33–9100, 1:200), rabbit anti-GFP (Invitrogen, G10362, 1:200), and chicken anti-GFP (Invitrogen, A10262, 1:200). Secondary antibodies were raised in goat (Invitrogen, conjugated with Alexa-405,–488, −546,–555, −633, or −647 fluorophores, 1:200). Samples were then transferred onto a slide in the dark and mounted with Fluoromount-G (Southern Biotech, 0100–01), covered using the standard ‘bridge’ procedure, and sealed with nail polish. Samples were imaged on LSM 710 and LSM 880 Zeiss microscopes using appropriate laser wavelengths and detection filters. Image stacks of roughly 20 µm were collected using a 63x, 1.40 numerical aperture (NA), oil immersion lens. For each Mauthner, laser strengths and gains were adjusted to achieve maximum visualization of CE staining.

Electrophysiological responses were obtained during whole-cell recordings of Mauthner cells (M-cells) in wt, Cx and ZO-1 mutant zebrafish larvae (5–7 dpf). For this purpose, fish were first anesthetized with a 0.03% solution of MS222 (pH adjusted to 7.4 with NaHCO₃) and later transferred to external solution containing d-tubocurarine (10 μM, Sigma). The external solution (in mM): 134 NaCl, 2.9 KCl, 2.1 CaCl₂, 1.2 MgCl₂, 10 HEPES, 10 Glucose, pH adjusted to 7.8 with NaOH (Yao et al., 2014). Zebrafish larvae were put on their backs onto a Sylgard-coated small culture dish (FluoroDish, WPI) and kept in place using fine tungsten pins. The hindbrain was then exposed ventrally following the dissection approach previously described (Koyama et al., 2011). Following this procedure, the larvae were placed on an Axio Examiner upright microscope (Carl Zeiss AG) equipped with a recording set-up and superfused with external solution during the entire recording session. The M-cells were identified by GFP expression and/or far-red DIC optics. Patch pipettes (3–4 MΩ) were filled with internal solution (in mM): 105 K-Methanesulfonate, 10 HEPES, 10 EGTA, 2 MgCl₂, 2 CaCl₂, 4 Na₂ATP, 0.4 Tris-GTP, 10 K₂-Phosphocreatine, 25 mannitol, pH adjusted to 7.2 with KOH. Whole-cell recordings under the current-clamp configuration were performed with a Multiclamp 700B amplifier and a Digidata 1440A (Molecular Devices) digitizer. The liquid-liquid junction potential was estimated in −16 mV using Clampex 10.6 (Molecular Devices) and was subtracted from the measured values. The electrode’s resistance was compensated using the bridge balance feature of the amplifier. To activate the auditory afferents terminating as CEs on the M-cell, a septated (theta) glass pipette was filled with external solution and positioned near the posterior macula of the ear, where the dendritic processes of auditory afferents contact the hair cells (Yao et al., 2014). The maximal amplitude of the electrical and chemical components was estimated by applying shocks of increasing intensity until the amplitude of the electrical component did not further increase and before additional responses with longer latency were evoked. To estimate the Paired-Pulse Ratio (PPR) of the chemical component one, a stimulating-pulse was applied to record 10–20 traces in basal conditions. Then two-stimulating pulses (2 ms apart) were applied to record (10–20 traces) facilitation of the chemical component. Traces that clearly showed a chemical component were averaged for basal and facilitated conditions. The trace in basal conditions was subtracted from the facilitated trace. The PPR was then calculated using the amplitude of the chemical component of the facilitated trace divided by the amplitude of the chemical component in basal conditions. Spontaneous electrical and chemical synaptic responses were assessed during offline analysis of continuous (10 s long) recordings using Clampfit (Axon instruments) to automatically identify events based on their duration (<1.1 ms for electrical spontaneous events and >1.1 ms for chemical spontaneous events). Potential erroneous identification of spontaneous events by the software was monitored manually by verifying the duration of the events. The assignment of electrical vs. chemical nature of spontaneous synaptic events by their duration was confirmed pharmacologically. For synaptic transmission blockade, CNQX was first dissolved in DMSO to have a stock solution of 10 mM. The pharmacological agents used to block synaptic transmission were added to the external solution: Meclofenamic Acid (200 μM, Sigma), CNQX and DAP5 (20 μM, Tocris Biosciences). The M-cell input resistance was estimated by applying a hyperpolarizing-current step of −1 nA and 20 ms in duration and measuring the voltage deflection caused, followed by derivation of resistance with Ohm’s law. The rheobase, defined as the minimum depolarizing current of infinite duration necessary to evoke an action potential, was determined by delivering a 20 ms current pulse of increasing intensity. Voltage threshold for action potential generation was determined by applying a depolarizing-current step.

Startle behavior of 5dpf larvae was analyzed as described previously (Marsden et al., 2018). Briefly, larvae were adapted to the testing temperature and lighting conditions for 30 min and then transferred to individual wells of a custom, laser-cut acrylic multi-well testing arena, illuminated from below with an infrared (IR) LED array and from above with a white light LED bulb to simulate daylight conditions. A total of 60 acoustic stimuli, 10 at each of 6 intensities, were delivered pseudorandomly using an acoustic-vibrational shaker (Bruel and Kjaer) with an inter-stimulus interval of 20 s to eliminate habituation to repeated stimulation (Wolman et al., 2015). The intensity of each stimulus was calibrated using a PCB Piezotronics accelerometer (model #355B04) and signal conditioner (model #482A21), and voltage outputs were converted to dB using the formula dB = 20 log (V/0.775). Behavioral responses were captured at 1000 frames per second with an IR-sensitive Photron mini-UX50 high-speed camera. After testing, larvae were fixed in methanol for subsequent genotyping, thus all testing and analysis was performed blind to genotype. Behavioral responses were tracked and analyzed using FLOTE software (Burgess and Granato, 2007), with short and long latency C-bend responses (SLCs and LLCs, respectively) automatically defined based on the kinematic parameters of the response. Startle sensitivity index was calculated by measuring the area under the curve of stimulus intensity versus SLC frequency for each larva.

Cell lines were obtained from ATCC, identity ensured by using exclusively low-passage cells, and were confirmed to be mycoplama free. Full-length Cx34.1 and full-length Cx35.5 were cloned into the pCMV expression vector. Full-length ZO1b was cloned into the pCMV expression vector with an NH₂-terminal mVenus tag and a COOH-terminal 8xHIS tag. Low passage HEK293T/17 cells were seeded 24 hr prior to transfection (1 × 10⁶ cells/well of a six-well dish), and the indicated plasmids were co-transfected using Lipofectamine 3000 (Invitrogen) following the manufacturer’s instructions. Cells were collected 36–48 hr post-transfection and lysed in 0.25 ml solubilization buffer (50 mM Tris [pH7.4], 100 mM NaCl, 5 mM EDTA, 1.5 mM MgCl2, 1 mM DTT and 1% Triton X-100) plus a protease inhibitor cocktail (Pierce). Lysates were centrifuged at 20,000 x g for 30 min at 4°C, and equal amounts of extract were immunoprecipitated with 0.5 ug rabbit anti-GFP (Abcam, Ab290) overnight with rocking at 4°C. Immunocomplexes were captured with 25 µl prewashed Protein A/G agarose beads for 1 hr with rocking at 4°C. Beads were washed three times with lysis buffer, and bound proteins were boiled for 3 min in the presence of LDS-PAGE loading dye containing 200 mM DTT. Samples were resolved by SDS-PAGE using a 4–15% gradient gel and analyzed by Western blot using the following primary antibodies: rabbit anti-GFP (Abcam Ab290), rabbit anti-Cx34.1 3A4-conjugated-680LT, and mouse anti-Cx35.5 4B12. Compatible near-infrared secondary antibodies were used for visualization with the Odyssey system (LI-COR).

Brains from wildtype, tjp1a/ZO1a^-/-, or tjp1b/ZO1b^-/- euthanized adult fish (4–15 months old) were removed, snap frozen in liquid nitrogen and stored at −80C until use. Brains were homogenized in 1 ml of HSE buffer (20 mM Hepes [pH7.5], 150 mM NaCl, 5 mM EDTA, 5 mM EGTA, and 1 mM DTT) plus a protease inhibitor cocktail using a glass homogenizer. Detergent was added to the homogenate (final 2% octyl ß-D-glucopyranoside, Anatrace) and solubilized overnight with rocking at 4°C. Solubilized homogenate was cleared by centrifugation at 20,000 x g for 30 min at 4°C, then pre-cleared for 1 hr at 4°C with Protein A/G beads before immunoprecipitation. The protein concentration for each homogenate was measured by Bradford assay. Pre-cleared homogenates (2 mg/IP) were immunoprecipitated with 0.5 µg mouse anti-ZO1 (Life Technologies, 33–9100), control mouse IgG (Jackson ImmunoResearch), mouse anti-Cx34.1 5C10, or mouse anti-Cx35.5 4B12 antibody overnight with rocking at 4°C. Immunocomplexes were captured with 25 ul prewashed Protein A/G agarose beads for 1 hr with rocking at 4°C. Beads were washed three times with HSE buffer, and bound proteins were boiled for 3 min in the presence of LDS-PAGE loading dye containing 200 mM DTT. Immune complexes were examined by western analysis using the following primary antibodies: mouse anti-ZO1, rabbit anti-Cx34.1 3A4-conjugated-680LT, rabbit anti-Cx35.5 12H5-conjugated-680LT, or mouse anti-Cx35.5 4B12. Compatible near-infrared secondary antibodies were used for visualization.

The Cx34.1-tail (aa256-299), Cx34.1-tail ∆PBM (aa256-295), Cx35.5-tail (aa267-309), and Cx35.5-tail ∆PBM (aa267-305) were cloned into the pGEX expression vector allowing for an NH₂-terminal GST tag. ZO1b-PDZ1 (aa105-207) and ZO1b-PDZ2 (aa298-387) were cloned into a modified pET expression vector (pBH) to allow for an NH₂-terminal 6xHis tag followed by a TEV cleavage site (vectors kindly provided by Ken Prehoda). Plasmids were transformed in E. coli BL21(DE3) cells and plated on selective LB plates. Single colonies were picked to inoculate 2 ml starter cultures and grown overnight. Overnight cultures were inoculated into 250 ml selective LB and grown for ~3 hr at 37°C with shaking until OD600 reached 0.8–1 followed by 4 hr induction with 0.4 mM IPTG. Cell pellets were collected by centrifugation at 6000 RPM for 5 min at 4°C and frozen at −20°C until test samples confirmed expression. Pellets were resuspended in sonication buffer (50 mM NaPO4 [pH7.4], 300 mM NaCl, and 1 mM PMSF). After adding a dash of lysozyme, the mixture was incubated on ice for 30 min. Resuspended bacteria were sonicated on ice at 50% amplitude, 1 s/1 s pulse on/off, four times for 20 s. Debris was cleared by centrifugation at 16,000 x g for 30 min at 4°C. For GST fusions, supernatant was added to 200 ul pre-washed glutathione agarose resin and incubated overnight with rocking at 4°C. Beads were washed three times with sonication buffer and stored at 4°C. Purity and amount loaded onto resin was determined by SDS-PAGE followed by Coomassie stain. For 6xHIS fusions, supernatant was brought to a final concentration of 20 mM imidazole and incubated with pre-washed His60 resin overnight with rocking at 4°C. Resin was washed with sonication buffer containing 20 mM imidazole. Protein was eluted from the resin with sonication buffer containing 250 mM imidazole. The protein was concentrated and exchanged into imidazole-free buffer using an Amicon centrifugal filter unit (10K MWCO) and stored at 4°C on ice. Protein concentration was estimated by A205 (https://spin.niddk.nih.gov/clore/) (Anthis and Clore, 2013), and purity was determined by SDS-PAGE followed by Coomassie stain.

Equal amounts of GST fusions (10 µl bed of resin) were aliquoted and the storage buffer was removed. To each sample 15 µl of 6xHIS-ZO1b-PDZ1 (7 mg/ml) was added, gently mixed and incubated at room temperature for 15 min. Resin was washed three times with cold wash buffer (50 mM NaPO₄ [pH7.4], 300 mM NaCl). After the last wash, all buffer was removed and resin was resuspended in 10 µl LDS-PAGE dye with 200 mM DTT. Samples were boiled for 3 min and resolved by SDS-PAGE using a 4–20% gradient gel. Samples were analyzed by Western blot using rabbit anti-TEV cleavage site primary antibody (ThermoFisher, PA1-119) and visualized with a compatible near-infrared secondary antibody. A portion of the GST fusion resin was analyzed by Coomassie stain to demonstrate equal loading.

Equal amounts of GST fusion were resuspended in LDS-PAGE dye plus 200 mM DTT, boiled for 3 min, resolved by SDS-PAGE on a 4–15% gradient gel, transferred to nitrocellulose, and blocked with 5% milk in TBS (20 mM Tris [pH7.4], 150 mM NaCl). Blocked membranes were incubated with TBS-T buffer (TBS + 0.1% Tween-20) alone (mock overlay), or TBS-T containing 1 µM 6xHIS-ZO1b-PDZ1 or 1 uM 6xHIS-ZO1b-PDZ2 overnight with shaking at 4°C. Membranes were processed for western analysis using rabbit anti-TEV cleavage site primary antibody and a compatible near-infrared secondary antibody to detect bound PDZ protein. Equal loading of GST fusions was determined by Stain-Free imaging technology.

A single guide RNA (sgRNA) targeting the 5’ region of the endogenous tjp1b coding sequence (sequence in Key Resources Table) was designed using the CRISPRscan algorithm (Moreno-Mateos et al., 2015) and synthesized as previously described (Shah et al., 2015). The sgRNA was generated using the T7 megascript kit (ThermoFisher, AMB13345). The V5-tjp1b single stranded donor oligo (ssODN) was designed to repair into the endogenous tjp1b locus and was synthesized complimentary to the coding strand. The ssODN contained 40 bp homology arms which flanked an XbaI restriction site, V5 sequence, and a 5x glycine linker, respectively (sequence in Key Resources Table). Upon correct repair, the inserted sequence was designed to disrupt the endogenous sgRNA recognition site to prevent further double stranded breaks after repair. Injection mixes were prepared in a pH 7.5 buffer solution of 300 mM KCl and 4 mM HEPES and contained a final concentration of 200 pg/nL ssODN, 200 pg/nL gRNA, and 1600 pg/nL Cas9 protein (IDT, 1081058). Injection mixes were incubated at 37 °C for 5 min immediately prior to injection to promote formation of the Cas9 and sgRNA complex. Finally, 1 nL of solution was injected into embryos at the one-cell stage. Injected embryos were raised to adulthood and outcrossed to wild-type animals. Animals carrying insertions were identified and verified using PCR and Sanger sequencing.

Cell transplantation was performed at the high stage approximately 3.3 hr into zebrafish development using standard techniques (Kemp et al., 2009). Embryos were chemically de-chorionated with protease (Sigma Aldrich, 9036-06-0) prior to transplantation. Cells were transplanted using a 50 mm wide glass capillary needle attached to an oil hydraulic. For ‘V5-tjp1b+/- into wildtype’ transplants (Figures 7C, D and Figure 7—figure supplement 1B-H) cells from animals heterozygous for V5-tjp1b in the M/CoLo:GFP background were transplanted into non-transgenic wildtype hosts. For ‘tjp1b-/- into wildtype’ transplants (Figure 7G and H), genotyped animals homozygous for the tjp1b^Δ16bp mutation in the M/CoLo:GFP background were crossed and progeny were transplanted into non-transgenic wildtype hosts. For 'wildtype into tjp1b-/-' transplants (Figure 7I and J), transgenic M/CoLo:GFP wildtype animals were crossed to use as donors, and non-transgenic, homozygous tjp1b^Δ16bp animals were crossed to produce hosts. Approximately 20 cells were deposited ∼10–15 cell diameters away from the margin, with a single donor embryo supplying cells to 3–5 hosts. At 5 dpf, larvae were fixed in TCA and processed for immunohistochemistry.

For fluorescence intensity quantitation, confocal images were processed and analyzed (in part or in full) using FiJi (Schindelin et al., 2012) software. To quantify staining at M/Colo synapses, a standard region of interest (ROI) surrounding each M/CoLo site of contact was drawn, and the mean fluorescence intensity was measured. For the quantification of staining at the club endings, confocal z-stacks of the Mauthner soma and lateral dendrite were cropped to 36.08 µm x 36.08 µm centered around the lateral dendritic bifurcation. Using the SciPy (Virtanen et al., 2020) and scikit-image (van der Walt et al., 2014) computing packages, the cropped stack was then cleared outside of the Mauthner cell, a 3³ median filter was applied to reduce noise, and a standard threshold was set within each experiment to remove background staining. The image was then transformed into a max intensity projection and the integrated density of each stain within the Mauthner cell was extracted. Where counts were used to quantify antibody labeling of CEs in high-contrast images, CEs were identified as large fluorescently labeled oval areas of approximately 1.5–2 microns on the distal portion and fork of the Mauthner cell’s lateral dendrite, which were thoroughly examined with the confocal microscope. For Figure 7—figure supplement 1I, J, the presence or absence of electrical synapses on Mauthner was quantified as counts of stereotyped electrical synapse structures dually labeled for Cx34.1 and Cx35.5.

Standard deviations and errors were computed using Prism (GraphPad) or Excel (Microsoft) software. Figure images were created using FiJi, Photoshop (Adobe), and Illustrator (Adobe). Statistical analyses were performed using Prism (GraphPad) and either an unpaired t-test with Welch’s correction or a one-way analysis of variance with Bonferrroni’s multiple comparison test was performed. For all experiments, values were normalized to wildtype control animals, and n represented the number of fish used. Fish were excluded from analysis if Mauthner morphology/GFP staining was abnormal.

Electrophysiology traces of spontaneous synaptic events were analyzed using Clampfit 10.7 software. Electrophysiology traces were transferred to Canvas for illustration and to OriginPro software for quantification and statistical analyses. For the different parameters measured using electrophysiological recordings, means and standard error of the mean (SEM) were illustrated using Prism (GraphPad) and computed using OriginPro software. For statistical analyses comparing electrophysiological recordings in wildtype and mutant zebrafish, Mann-Whitney and Kruskal-Wallis non-parametric tests were performed using OriginPro software.