Convergent NMDA receptor—Pannexin1 signaling pathways regulate the interaction of CaMKII with Connexin-36

Ca2+/calmodulin-dependent protein kinase II (CaMKII) binding and phosphorylation of mammalian connexin-36 (Cx36) potentiate electrical coupling. To explain the molecular mechanism of how Cx36 modifies plasticity at gap junctions, we investigated the roles of ionotropic N-methyl-D-aspartate receptors and pannexin1 (Panx1) channels in regulating Cx36 binding to CaMKII. Pharmacological interference and site-directed mutagenesis of protein interaction sites shows that NMDA receptor activation opens Cx36 channels, causing the Cx36- CaMKII binding complex to adopt a compact conformation. Ectopic Panx1 expression in a Panx1 knock-down cell line is required to restore CaMKII mediated opening of Cx36. Furthermore, blocking of Src-family kinase activation of Panx1 is sufficient to prevent the opening of Cx36 channels. Our research demonstrates that the efficacy of Cx36 channels requires convergent calcium-dependent signaling processes in which activation of ionotropic N-methyl-D-aspartate receptor, Src-family kinase, and Pannexin1 open Cx36. Our results add to the best of our knowledge a new twist to mounting evidence for molecular communication between these core components of electrical and chemical synapses. Siu et al use FRET imaging in neuronal cell lines to explore the interplay between the gap junction channel protein Connexin-36 (Cx36) and NMDA receptor and Pannexin 1-mediated calcium signaling. They demonstrate that an increase in intracellular calcium promotes binding of CaMKII to Cx36, leading to increased gap junction opening, thus providing a pathway by which components of chemical and electrical synapses communicate.

E lectrical synapses in neurons coexist alongside chemical synapses in mixed synapses throughout the vertebrate nervous system 1,2 . In the mammalian brain, most electrical synapses are composed of connexin-36 (Cx36) gap junction channels (GJC) 3 . Cx36 GJCs show functional plasticity similar to chemical synapses 4,5 , in which the interaction of Cx36 with Ca 2+ -activated calmodulin (CaM) and calmodulin protein kinase II (CaMKII) is considered analogous to the interaction of ionotropic N-methyl-D-aspartate (NMDA) receptors with CaM/CaMKII 6 .
CaMKII binding and phosphorylation of Cx36 have been identified as drivers of adaptive electrical coupling in neurons of teleosts and rodents. The protein motifs found in the cytoplasmic and carboxyterminal domains of Cx36 share a remarkable similarity to segments of the regulatory subunit of CaMKII 6,7 . Similar to the NR2B subunit of the NMDA receptor, both Cx36 binding sites exhibit phosphorylation-dependent interaction and autonomous activation of CaMKII, suggesting that the functional efficacy of both modes of interneuronal communication share common molecular features. Here, we explored the idea that Cx36 and NMDA receptor activities could be synchronized by sharing molecular features. Specifically, we hypothesized that ionotropic NMDA receptors mediate both synaptic potentiation and Cx36 activity by the timing of neuronal activity when glutamate release and a post-synaptic depolarization coincide temporally with calcium influx. The ATP-release channel Pannexin1 (Panx1) was tested as a potential mediator between NMDA receptors and Cx36. Panx1 is localized in glutamatergic synapses 8,9 in the proximity of NMDA receptors and has been linked to the propagation of calcium waves 10 . Panx1, together with metabotropic and ionotropic NMDA receptors and Src family kinases (SFKs) have been implicated in forming signaling complexes [11][12][13] .
Here, we describe the identification of a signaling pathway in which activation of ionotropic NMDA receptors modulate the interaction between Cx36 and CaMKII. Individual steps in the signaling process include activation of the ATP-release channel Panx1 [14][15][16] , SFKs 17 , and elevated intracellular calcium ([Ca 2+ ] I) . Förster resonance energy transfer (FRET) was used to identify steps at which signaling converges onto the Cx36-CaMKII interaction complex. This technology allowed to determine changes to the compact conformation of the three-dimensional gap junction plaque (GJP) superstructure at an nm-resolution when the interaction between the dodecameric Cx36 GJC and the dodecameric CaMKII holoenzyme was tuned to activation or deactivation. Functional dye transfer assays and ethidium bromide fluorescence recovery after photobleaching 18 were used to correlate functional changes with structural rearrangements in the protein complexes.
Results emphasize that the ionotropic NMDA receptormediated rise in [Ca 2+ ] I was an essential signal requirement to increase a compact conformation of the Cx36-CaMKII complex and to open Cx36 channels. Mutations in protein domains targeting the interaction between Cx36 and either CaM or CaMKII, and pharmacological interventions disrupting the NMDA receptor or CaMKII activation, abolished [Ca 2+ ] I signaling the opening of Cx36 channels located in GJPs. Inhibition of SFK, main actors in pathways leading to the generation of Ca 2+ signals, and modulators of NMDA receptor 17 , the CaMKII holoenzyme 19 , and Panx1 20 , blocked Cx36 channel opening in this study. The genetic ablation of Panx1 was effective in closing Cx36 as was the blockage of NMDA receptors, SFK, or CaMKII, an effect that was rescued by overexpression of Panx1. We conclude that ionotropic NMDA receptors and Panx1 provide a signaling mechanism endowing mixed chemical and electrical synapses with the flexibility to control synaptic outputs.
Cx36, similar to NMDA receptors, can bind the calciumactivated CaM/CaMKII complex, and autophosphorylation at amino acid T286 is required for orchestrating the binding of CaMKIIα to Cx36 6 . Pharmacological intervention with the CaMKIIα inhibitor KN93 decreased FRET eff (non-stimulated: 5.1 ± 0.4, n = 21; KN93: 2.8 ± 0.5, n = 20, p = 0.001) (Fig. 2b), likely through changes in the CaMKIIα structure, relaxing the binding complex, or by blocking the CaM/CaMKII complex 22 . Interestingly, FRET eff of the autophosphorylation negative CaMKIIα mutant T286A was much higher than controls. This suggested that the mutation locked the interaction between Cx36 and CaMKII in a compact conformation and that this step did not require autophosphorylation. When [Ca 2+ ] I increased, FRET eff decreased, most likely indicating that Cx36 detached from the complex (non-stimulated: 8.9 ± 1.0, n = 20; ionomycin: 5.5 ± 0.7, n = 22, p = 0.01) (Fig. 2b). The CaMKIIα T286D mutant and the wild-type protein showed a similar but slightly dampened response to [Ca 2+ ] I increase (non-stimulated: 4.1 ± 0.5, n = 22; ionomycin: 6.8 ± 0.7, n = 22, p = 0.004), which was consistent with reports indicating that nearly full activation of this mutant can be achieved by minimal further stimulation by Ca 2+ /CaM or when the binding protein mutes the mutant 23 (Fig. 2b). We concluded that Ca 2+ -induced structural changes caused by the binding of Cx36 to CaMKII require CaMKIIα autophosphorylation at T286.
Calcium is required for the interaction of Cx36 with CaMKIIα at the GJP. The fluorescent calcium reporter Oregon Green BAPTA-AM (OGB) was used next to visualize changes to [Ca 2+ ] I levels after treatment. Sample traces of treated Neuro2a cells, reporting fluorescence changes over 120 s, confirmed ionomycininduced Ca 2+ influx (boxed to show the most substantial changes). DMSO or BAPTA treatments caused negligible differences compared to the control (Fig. 3a). Together, the results demonstrated a causal relationship between [Ca 2+ ] I and the strength of the Cx36-CaMKIIα interaction at GJPs.
Next, an ethidium bromide uptake and recovery after photobleaching assay evaluated the impact of activation and inhibition of NMDA receptors on the gap junction function. Neuro2a cells expressing Cx36-EGFP and DsRed-CaMKII showed significant ethidium bromide redistribution after stimulation of NMDA receptors (non-stimulated: 10.3 ± 3.6, n = 21; NMDA: 37.8 ± 7.0, n = 22, p = 0.003). The pharmacological blockage of NMDARs with MK801 suppressed the ethidium bromide transfer through GJPs and inhibited the enhanced ethidium bromide transfer caused by NMDA (MK801: 13.1 ± 4.8, n = 20, p = 0.94; MK801 and NMDA: 17.9 ± 4.9, n = 20, p = 0.36 (Fig. 4f). The quantification of the GJP frequency and size showed a significant reduction in the number of gap junctions when NMDARs were blocked for 30 min with MK-801 or 10 min with AP5, suggesting that [Ca 2+ ] I changes can modulate gap junction communication rapidly (Supplementary Fig. S5e, f). The data establish NMDA receptors as a gateway for Ca 2+ entry that enhances the binding of Cx36 to the CaM/CaMKII holoenzyme.
The adoption of a compact structure of the Cx36-CaMKII interaction complex is modulated by an NMDA receptor-Src family kinase-Panx1 signaling pathway. Initial experiments were directed towards Ca 2+ entry through ionotropic NMDA receptors and how this affects CaMKII-Cx36 interaction. In a separate investigation line, we turned to whether posttranslational modification of Panx1 by SFKs affects CaMKII-Cx36 interaction and increases coupling after stimulation of ionotropic NMDA receptors. This change in focus was motivated by research demonstrating a signaling pathway between metabotropic NMDA receptors and Panx1, and the regulation of Panx1 channel function by SFK-mediated phosphorylation at the two tyrosine residues Y308 and Y198 11 . In ethidium bromide uptake and recovery after photobleaching experiments, activation of Cx36 GJC by stimulation of the NMDA receptor increased the redistribution of ethidium bromide between Neuro2a cell pairs after photobleaching (non-stimulated: 5.8 ± 3.0, n = 20; NMDA: 32.7 ± 7.4, n = 23, p = 0.0014) (Fig. 7a). The increase in dye in the photobleached regions of interest below GJPs was suppressed by co-application of the Src kinase inhibitor 4-amino-5-(4-chlorophenyl)-7-(t-butyl)pyrazolo   (Fig. 7b, c). Double transfected Neuro2a cells treated with PP2 and PP3 showed regular cell morphology (Fig. 7d).

Discussion
The activity-dependent strengthening or weakening of chemical synaptic transmission is regarded as a fundamental mechanism by which neural circuits are formed or broken. A principal molecular mechanism by which excitatory synapses are strengthened is through the interaction of CaMKII and NMDA receptor subunits. Evidence for similar use-dependent facilitation of electrical synaptic strength was first revealed at auditory mixed synapses on the goldfish Mauthner cells 30,31 . Like chemical synaptic potentiation, this facilitation was shown to depend on CaMKII binding to the channel protein, in this case, one or more teleost homologs of Cx36 32 . We previously demonstrated that the mammalian neuronal gap junction protein Cx36 also exhibits use-dependent facilitation, the so-called "run-up" 7,33 . The Cx36 protein interacts with and is phosphorylated by CaMKII in vitro, similar to the CaMKII interaction with the NR2B subunit of ionotropic NMDA receptors 6 . We further showed that phosphorylation by CaMKII strengthens junctional currents of Cx36 channels, thereby conferring functional plasticity on electrical synapses formed by this protein 7 .
Here, using neuroblastoma cells as an efficiently manipulatable neural cell line, we identify components of a signaling pathway that modifies the CaMKII interaction with Cx36 at the GJP. The signaling pathway consists of ionotropic NMDA receptors, SFK, and Panx1 and is distinct from protein interactions that we and others have previously reported involving Cx36. These include binding to Ca 2+ loaded CaM 34 , which mainly occurs at the ER/Golgi complex 18 , the interaction with tubulin facilitating microtubuledependent transport to the GJP 24 , or the removal of Cx36 from the GJP by binding to caveolin-1 35 . Our present interpretation is that these known interactions primarily contribute to on-demand transport to the GJP and endocytosis of Cx36.
The findings reported here were accomplished with FRET measurements of fluorescent protein-tagged construct pairs as a sensitive method to detect local pathways of protein interactions at the cell membrane. This method revealed increased compact conformation of protein-protein complexes, interpreted as tighter interactions between the pairs, when [Ca 2+ ] I was elevated either with Ca 2+ ionophore or by activation of ionotropic NMDA receptors. Interestingly, both GJP size and frequency have been reduced after the loss of CaM/CaMKII binding or when NMDARs were blocked. These observations suggest multi-faceted roles of CaMKII including the transport/insertion/removal of Cx36 at GJPs, resembling a potential similarity to feedback mechanisms found at NMDARs. CaMKII phosphorylation of NMDARs plays a central role in controlling the number and activity of the receptors and determining the strength of excitatory synaptic transmission 36 . The results made in our minimalistic neuroblastoma cell model provide mechanistic insight into how a critical function of electrical synapses is modulated by crosstalk with molecular components that are generally hallmarks of chemical synapses. Neuronal gap junctions are often found in close proximity to glutamatergic synapses, where they can be considered as functional mixed synapses 2 . Using established methods of manipulating NR1/NR2B subunit activation (by relieving the magnesium block 37,38 , inhibition or activation of extracellular receptor sites, or pharmacological modulation 39,40 ), we found that ionotropic NMDA receptors constitute an important entry point for [Ca 2+ ] E . The increase in [Ca 2+ ] I was identified as an essential driver of the transition of the Cx36-CaMKII complex to a more compact state.
Since the 90s, electrophysiological studies in lower vertebrates demonstrated NMDA receptor-mediated strengthening of electrical coupling between neurons in mixed synapses 31,41 and implicated CaMKII 42 . Later, regulation and plasticity of electrical transmission by NMDA receptor-initiated sequence of events were found in the mammalian brain in synaptic loci and at extra-synaptic loci proximal to gap junctions 43 . Our previous report of "run-up plasticity" has demonstrated that Cx36 binds Ca 2+ -activated CaMKII in vitro, using binding domains with similarities to the binding sites found in the NR2B subunit of the NMDA receptor. We concluded that this was evidence for a shared molecular basis of use-dependent plasticity 6,7 . The results identify short-range structural rearrangements which increase the adoption of a compact state of the Cx36-CaMKII holoenzyme complex as a fundamental requirement to open Cx36 channels. We predict that recent advances in cryoEM technologies provide an opportunity to solve the differences of compact and open states of the Cx36-CaMKII complex at an atomic resolution.
An additional way in which activation of the signaling complex may be modulated is through intracellular Mg 2+ , which was shown by Palacios-Prado et al. 44 to inhibit Cx36-dependent gap junctional intercellular communication in cell lines exogenously expressing Cx36, and in brain slice recordings from neurons in the mesencephalic nucleus known to express endogenous Cx36. The authors argued that Mg 2+ concentration changes sufficient to alter coupling might occur physiologically by ATP fluctuations, which reduces Mg 2+ and thus increase coupling. It was suggested that the effect of Mg 2+ on coupling strength was due to direct interaction with the channel. However, Mg 2+ also competes with Ca 2+ for binding to calmodulin 45 , and the effects reported in that study are consistent with an interference with the step-wise assembly of the Cx36-CaMKII signaling complex.
The involvement of Panx1 in regulating Cx36 was unexpected. To our knowledge, this is the first observation of functional interaction between members of the connexin and pannexin gene families. Previously, several stimuli have been reported to activate Panx1 channels [46][47][48] , including ionotropic NMDARs activated during anoxia or following energy deprivation 49,50 . The Thompson group also reported that Panx1 activation required SFKs, which corroborated that the link between NMDARs and Panx1 were SFKs 12 . The discovery that the activation of ionotropic NMDA receptors, Panx1, and SFKs increase intracellular calcium levels, which drive Cx36 bound to CaMKII to an open state, demonstrates molecular and perhaps mechanistic interaction between chemical and electrical communication in nerve cells. They also suggest a possible involvement of Cx36 in the postsynaptic metabotropic NMDARs, Panx1, and SFKs signaling complex reported by the Thompson group.
In primary neurons, a signaling pathway involving ionotropic NMDA receptors and Panx1 is expected to be restricted to the postsynaptic cell membrane based on known protein localization data. Ionotropic NMDA receptors are located close to Cx36 at mixed synapses in excitatory neurons 1,2 . Panx1 proteins were found in rodent hippocampal and cortical principal neurons accumulating at high concentration in proximity to postsynaptic densities 9 . In contrast, CaMKII and SFKs can be found in both pre-and postsynaptic compartments. The restricted localization of both NMDA receptors and Panx1 in postsynaptic compartments suggests that the signaling complex in primary central nervous system neurons is not a simple mirror image across the GJP. This finding adds to growing evidence for functional asymmetry caused by differences in post-translational modifications of individual Cx36 proteins 51 , or in the complement of Cx36-associated proteins found at the electrical synapse density 36,52,53 . Taken together, the outcomes of this study add to the best of our knowledge a new twist to how electrical and chemical synapses communicate. They suggest that interactions between the two modes of neuronal communication exist, are direct, interdependent, and might facilitate synchrony and adaptive plasticity at mixed synapses. TALEN Panx1 KD cells. Transcription Activator-Like Effector Technology (TALEN) cloning protocols were carried out described by Reyon and colleagues 54 . The ZiFit Targeter software (http://zifit.partner.org/ZiFiT/) was used to design constructs targeting a single NcoI restriction site located close to the start of the coding region in exon 1 of the mouse Panx1 gene ( Supplementary Fig. S7a) (Ensembl: ENSMUSG00000031934). Constructs were assembled in the expression vectors JDS74 and JDS78 using the protocol introduced by Reyon and colleagues 54 . Neuro2a cells were transfected with both TALEN constructs at an equimolar ratio, followed by treatment with the nucleoside antibiotic blasticidin (10 μg/ml) for 2 weeks. Stable cell lines were established by using a limited dilution protocol in which single-cell suspensions of blasticidin-resistant cells were repeatedly (3×) diluted and expanded from individual cell colonies. Upon reaching 80-100% confluency, genomic DNA was extracted using the GeneJET Genomic DNA purification kit (Thermo Fisher Inc., Mississauga, ON, Canada). Polymerase chain reaction (PCR) amplification of the genomic DNA followed by restriction digest of a 226 bp amplicon with NcoI to confirm the loss of the NcoI site at the start of the Panx1 open reading frame. PCR products were cloned into pJET1.2 plasmid (Thermo Fisher Inc., Mississauga, ON, Canada) and sequenced for confirmation of indel mutations. A cell line with a 17 bp deletion at the start codon of the Panx1 protein was selected for further analysis (Supplementary Fig. S7).

Methods
FRET analysis. Protein interactions at GJPs were investigated using the acceptor bleach protocol 18,21 . Neuro2a cells were transfected with DsRed-monomer and EGFP/ECFP-tagged expression vectors. Baseline readings were established before initiating the acceptor bleach protocol (five initial readings). In this protocol, DsRed-tagged proteins were photobleached using the 555 nm laser line at 100% intensity, while the intensity change of CFP or GFP tagged proteins was recorded using the 405 or 455 nm laser lines respectively. The experimental protocol was terminated upon reaching 10% of the acceptor channel baseline intensity. FRET efficiency was then calculated by using the FRET efficiency equation 1 where D post is the average intensity after bleaching, and D pre is the average intensity before the bleach after subtracting the values of the background (noise).
Ethidium bromide recovery after photobleaching assay. The assay has been reported previously 18,25 . Here, this assay was used to correlate functional changes with structural rearrangements of the GJP in the presence of stimulants or blockers. Transfected Neuro2a cells cultured for 48 h in 3.5 cm MatTek cell culture dishes were incubated with 10 μM of ethidium bromide in supplemented growth medium for 10 min at 37°C and 5% CO 2 prior to imaging . Transfected cells in MatTek dishes were placed in a live-cell imaging chamber (regulated at 37°C and 5% CO 2 ) and imaged with Zeiss 700 confocal microscope. Cell pairs expressing Cx36-EGFP were selected, and a time-lapse baseline image was recorded (defined as 100%).
One cell of each cell pair was bleached using the 555 nm laser line with 100% laser power intensity (Supplementary Fig. S2a). The number of iterations (~40 iterations at 100% intensity) in the bleaching protocol was terminated upon achieving a 40-50% reduction of baseline fluorescence. objective was used to measure the Ca 2+ signal changes over 300 cycles at 1-s intervals. The 488 nm laser line at 20% laser intensity was used to measure the cells. Experimental drug treatments were delivered after the first 10 cycles. Images were processed and analyzed in ImageJ with the Heatmap Histogram and Time Series Analyzer V3 plugin after background subtraction. Images were pseudo-colored using ImageJ's Thermal LUT.
Arclight S245 fluorescent voltage probe imaging. Membrane potential changes in double transfected Neuro2a cells expressing Cx36-DsRed and Arclight S245 55-57 were imaged 48 h post-transfection. Cells were imaged using a Zeiss Observer Z1 Spinning Disk Confocal Microscope equipped with a live-cell imaging chamber set to 37°C and 5% CO 2 . Evolve TIRF camera with a Plan-Apochromat 63×/1.4 Oil DIC M27 objective was used to measure the fluorescence change of Arclight S245 over 300 cycles at 2-s intervals. The 488 and 561 nm laser lines were at 15% and 10% laser intensity during the experiments. The first 30 cycles were used to establish the baseline for each experimental run, while drug treatments were delivered through tubing with a syringe during the 31st-40th cycle. The Definite Focus module of the microscope was used to maintain the plane of view throughout the experiment. Images were processed and analyzed in ImageJ with the Time Series Analyzer V3 plugin after background subtraction. The fluorescence scaled to baseline was calculated using the Eq. 3 Fluorescence ðscaledÞ ¼ F time F average ðbaselineÞ and the fluorescence change (ΔF/F) was calculated using Eq. 4  10 Panx, scrambled control SC Panx1 (100 µM, 10 min), all at 37°C and 5% CO 2 . The final solvent concentration of DMSO was 0.002-0.02%. All reagents were obtained from Sigma-Aldrich.
Statistics and reproducibility. Statistical analysis and data presentation were performed using the IBM-SPSS and R-software packages. Data were presented as (mean ± SEM). All data were analyzed with the Shapiro-Wilk normality test and Levene's homogeneity of variance test. Data with normal and equal invariance were subjected to independent t-tests, and non-normal distributed data were subjected to Mann-Whitney U (2-tailed) tests. Outliers were determined and shown on figures but have been excluded from statistical analyses. The sample sizes (n) reported corresponding to cells or cell pairs as indicated. The minimum number of independent experimental replicates was n ≥ 3.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information file), Source data for Figs. 1-7, and Supplementary Figs. S1-S73 can be found in Supplementary Data 1. Raw image files will be made available by the corresponding authors upon reasonable request.