Expression vectors contained full-length rat Cx36 [NM_019281, amino acids (aa) 1–321], CaM (NM_031969, aa 1–149), and CaMKIIa (NM_012920.1, aa 1–478) in variants of pEGFP-N1 expression vectors (Clontech Laboratories Inc., Mountain View, CA, USA) (Zoidl et al., 2002; Alev et al., 2008). Full-length in-frame cloning of CaM, the calcium insensitive mutant CaM_E1234Q (Maune et al., 1992; VanScyoc et al., 2002), as well as genetically encoded organelle markers (pER-DsRed2, pGolgi/DsRed2, pMito-DsRed2) (Fliegel et al., 1989; Rizzuto et al., 1989; Watzele and Berger, 1990) was completed in two steps. The open reading frames were synthesized as gBlocks [Integrated DNA Technologies Inc. (IDT), Coralville, IA, USA] and cloned into the TA cloning vector pJet1.2 (Thermo Fisher Inc., Mississauga, ON, Canada). Then the coding regions were isolated by restriction digest and cloned in-frame into pECFP-N1 or the pDsRed2-monomer expression vector. Mutant Cx36 plasmids (G276A, W277A, R278A) were generated using the Q5^TMSite-Directed Mutagenesis Kit according to the manufacturer’s protocol [New England Biolabs Inc. (NEB), Boston, MA, USA]. Oligonucleotides were designed using the NEBaseChanger tool (NEB) and synthesized by IDT. Table 1 summarizes the oligonucleotides utilized in this study (mutations depicted in bold and underlined). All plasmid constructs used on this study were sequence verified (Eurofins MWG Operon LLC, Huntsville, AL, USA).

Neuroblastoma 2a (Neuro2a) cells (Olmsted et al., 1970) were cultivated in DMEM with 2 mM glutamine, 1% non-essential amino acids (NEAA), 1% penicillin and streptomycin (PS), and 10% fetal bovine serum (FBS) at 37°C in a humidified atmosphere with 5% CO₂. Cells were seeded in 24-well plates or glass-bottom dishes (MatTek Corporation, Ashland, MA, USA) and transfected with 200 ng (single transfection) or 400 ng (double transfection) endotoxin-free plasmid DNA, using the Effectene transfection protocol (Qiagen Inc., Valencia, CA, USA). For western blot, whole cell protein lysates were prepared 48 h after transfection. The 20 μg protein was separated by 10% SDS-PAGE, transferred to 0.2 μm Midi format nitrocellulose membrane and processed using the iBind^TM Western System (Bio-Rad Inc., Mississauga, ON, Canada). Primary antibodies were diluted 1:1,000 [mouse anti-GFP, Roche; rabbit anti-GFP (FL), Santa Cruz Biotechnologies, Dallas, TX, USA] and 1:20,000 (mouse anti-β-actin; Sigma-Aldrich Chemie GmbH, Munich, Germany). The secondary antibodies (LI-COR Biosciences, St. Lincoln, NE, USA) were diluted 1:20,000 (donkey anti-rabbit IRDye680LT) or 1:20,000 (goat anti-mouse IRDye800CW). Signals were detected using the Odyssey^® CLx Infrared Imaging System (LI-COR Biosciences).

Transfected cells were fixed with 4% paraformaldehyde for 20 min at RT, washed with PBS, and mounted with ProLong^® Antifade Mountant (Thermo Fisher Inc., Mississauga, ON, Canada) for imaging. Samples were visualized using a Zeiss LSM 700 confocal microscope with a Plan-Apochromat 63x/1.4 Oil DIC M27 objective and the ZEN 2010 program to control all hardware parameters. Images were collected by line averaging (4×) at high resolution (2048 × 2048 pixels) using single planes or z-stacks. The GJ frequency was quantified as described previously (Qu et al., 2009), by counting the number of EGFP-expressing cell pairs against the number of cell pairs having GJPs in images from randomly selected, non-overlapping visual fields collected from experimental replicates (n > 3). The GJ area was quantified using ImageJ. Here, the ImageJ free hand tool was used to draw the contours of GJs between cell pairs, followed by quantification of the GJ area using the measure tool. Mander’s overlap coefficients were calculated using the co-localization analysis if the ZEN 2010 program (Manders et al., 1993). Images were exported and further processed using ImageJ or the Imaris Scientific 3D/4D Image Processing & Analysis Software (Bitplane USA, Concord, MA, USA). Images were combined using Adobe Photoshop CS6 for presentation.

The acceptor bleach protocol was established by testing different combinations of fluorescently tagged Cx36 and CaM proteins using a Zeiss LSM 700 confocal microscope (data not shown). In the experiments shown, Neuro2a cells were transiently transfected with DsRed-monomer and CFP tagged expression vectors and processed as described above. Baseline readings were measured prior to the acceptor bleach protocol. Here, DsRed-tagged proteins were bleached using the 555 nm laser line (set to 100% intensity) and the intensity change of CFP tagged proteins was recorded using the 405 nm laser line. The experiment was completed when the bleaching of the acceptor channel reached 10% of the initial intensity. FRET efficiency was then calculated by using the FRET efficiency formula: FRET_eff = (D_post -D_pre)/D_post, where D_post is the average intensity after the bleach, and D_pre is the average intensity before the bleach (Okamoto et al., 2004) after subtracting the values of the background (noise). The FRET distance (FRET_d) between the interacting proteins was calculated using the formula: R = R_o ((1/E) - 1)^1/6, where R_o is the reference distance between the two fluorescent tags (DsRed and ECFP: 5.09) (Erickson et al., 2003), and E is the FRET efficiency. The mean values of the intensity change and distance were calculated and formulated into graphical representations. Error bars denote the standard error of the mean (SEM) of each protein–protein interaction.

This assay is a variation of the ethidium bromide uptake assay previously reported (Prochnow et al., 2009; Kurtenbach et al., 2013). Transfected Neuro2a cells grown for 48 h in 3.5 cm MatTek cell culture dishes were incubated with 10 μM of ethidium bromide in complete growth medium for 10 min at 37°C and 5% CO_2. MatTek chambers were placed in a live-cell imaging chamber and imaged at 37°C using a Zeiss 700 confocal microscope. Cell pairs expressing Cx36–EGFP were selected and a time-lapse baseline image was recorded (defined as 100%). Next, one cell was bleached using the 555 nm laser line with the intensity set to 100% laser power. The number of iterations in the bleaching protocol was terminated when 40–50% reduction in fluorescence was achieved. In this process, a volume of 36.2 μm × 36.2 μm × 1.6 μm was bleached using a non-confocal imaging mode. The recovery of ethidium bromide fluorescence in the center plane of this volume after bleaching was measured over time in confocal imaging mode. In all experiments, three regions of interest (ROI: R0, R1, R2) were selected for analysis. R0 was placed outside the cell pair to determine background the fluorescence. ROIs were placed inside the bleached cell close (R1) and distant (R2) to the GJP. The fluorescence recovery in percent was calculated using the formula:

Prior to imaging transfected cells were treated with the Ca²⁺ chelator BAPTA-AM (Thermo Fisher) (Ueda and Shah, 1992), ionomycin (Sigma-Aldrich) (Liu and Hermann, 1978), carbenoxolone (CBX; Sigma-Aldrich) (Li et al., 2001) or by applying the CaM antagonist W-7 (Sigma-Aldrich) (Weiss et al., 1982). The final concentrations in complete growth medium were 24 μM of BAPTA-AM, 10 μM of W-7, 50 μM of CBX, for 10 min or 2 μM ionomycin for 5 min at 37°C and 5% CO₂. The final solvent concentration of ethanol in ionomycin treated cells was 0.002%.

A pET15 based (EMD Millipore, Etobicoke, ON, Canada) plasmid for expression of hexahistidine-tagged human calmodulin (His₆-CaM) was a gift from Prof. Mitsu Ikura (University of Toronto). From a 1 L fermentation in Luria broth, milligram quantities of >95% pure His₆-CaM were obtained by a two-step Ni-NTA affinity chromatography (Qiagen Inc., Toronto, ON, Canada) and gel filtration chromatography (Sephadex-75 HiLoad 16/60 column, GE Life Sciences) procedure. Final buffer conditions were 5 mM Tris-HCl, 100 mM NaCl, 0.05% NaN₃, pH 7.5. The CaM–Cx36 complex pursued for structural determination was expressed as a 169 aa fusion protein consisting of His₆-tagged human CaM (5–150), a short linker (GASTAAGS), and residues 276–292 of rat Cx36. A gene fragment encoding this fusion protein was synthesized by DNA2.0 (Menlo Park, CA, USA) and supplied in a T5-based expression vector (pJ441-kan). ¹⁵N, ¹³C labeled His₆-CaM–Cx36 was obtained from a 1 L fermentation in M9 media containing 1 g (99%, ¹⁵N) ammonium chloride, 3 g (99%, ¹³C-glucose) and 0.5 g (98%, ¹³C–¹⁵N) algal extract (Cambridge Isotopes, Tewksbury, MA, USA). Final buffer conditions for NMR spectroscopy were 1.5 mM protein in 5 mM Tris-HCl, 100 mM NaCl, 5 mM CaCl₂, 0.05% NaN₃, pH 7.5 with 10% (v/v) ²H₂O.

Spectra were acquired at 298 K with a 700 MHz Bruker Avance spectrometer equipped with a TXI cold probe. Backbone and side chain chemical shift assignments were obtained using a suite of standard heteronuclear, triple resonance experiments (HNCACB, CBCACONH, HNCO, HNCACO, HNCA, CCH-TOCSY, HCCH-TOCSY) with non-uniform Poisson gap sparse sampling at 10–15%. Distance restraint data were obtained from ¹⁵N-edited and ¹³C-edited 3D-NOESY spectra (120 ms mixing time). All datasets were processed with NMRPipe (Delaglio et al., 1995) and analyzed with CcpNmr Analysis (Skinner et al., 2015).

CcpNmr Analysis was used to integrate NOEs and calculate distance restraints based on an r^-6 function spanning 1.8–5.5 Å with a reference distance set at 3.2 Å. These restrains were used as input to CYANA (Buchner and Güntert, 2015) along with Ca²⁺ coordination distance restraints, hydrogen bonding distance restraints and torsion angle restraints predicted by the DANGLE method embedded in CCPN Analysis. Allowed backbone torsion angles and side chain rotamers were enforced by the ramaaco and rotameraco methods embedded in the CYANA suite. The final representative ensemble of 20 structures based on lowest CYANA target function was selected from an initial calculation of 400 structures. As this ensemble demonstrated good geometry and bonding, no additional refinement was pursued.

The structural ensemble was deposited in the Protein Data Bank under the accession number 2N6A.

Isothermal titration calorimetry (ITC) measurements were performed on MicroCal VP-ITC calorimeter (MicroCal Inc., Northampton, MA, USA). The recombinant H₆-CaM (affinity purified as described above) was titrated with Cx36 derived peptides reflecting the wild type (GSGWRKIKLAVRGAQAKRKSVYEIR; CanPeptide Inc., Montréal, QC, Canada) or the W227A mutant (GSGARKIKLAVRGAQAKRKSVY; BioBasic, Markham, ON, Canada). An optimal titration was achieved with 25 μM CaM in the reaction cell and 250 μM Cx36 peptide in the syringe, each in a buffer containing 50 mM NaCl, 5 mM BisTris pH 7.0, 5 mM CaCl₂. After an initial injection of 2 μL, the bulk of the titration consisted of 34 successive 8 μL injections with an equilibration delay of 300 s. Heats of dilution were determined by titrating the same peptide solutions into buffer alone. Titration profiles corrected for the heat of dilution were fitted into one-binding-site model using MicroCal Origin v5.0. All values reported were calculated based on three individual experiments.

Statistical analysis and data presentation was performed using the IBM-SPSS and R-software packages. Results shown derive from experimental replicates with n ≥ 3. All data were analyzed for data distribution and subjected to Mann–Whitney U tests for independent samples or a paired t-test, when appropriate.

Previous work (Burr et al., 2005) and the sequence alignment of Cx36 orthologs representing major vertebrate phyla ranging from primitive fish to humans suggested the existence of a highly conserved potential CaM binding motif in the carboxy-terminus of Cx36 resembling a variation of the 1–8–14 subclass. This motif overlapped with a binding site for the calcium-dependent calmodulin kinase II (Figure 1A) (Alev et al., 2008).

The role of W277 in position 1 of the putative CaM-binding site was investigated after sequential alanine-scanning mutagenesis of positions 276 to 278 of the rat Cx36 protein. Western blot analysis confirmed exogenous expression of Cx36 wild-type and Cx36 mutants (G276A, W277A, and R278A) in single transfected Neuro2a cells 48 h post transfection (Figure 1B). A similar result was obtained with CaM and the Ca²⁺ insensitive mutant of CaM (CaM_E1234Q; Maune et al., 1992) generated for this study (Figure 1B). The localization of Cx36, Cx36 mutants, and CaM was investigated in double-transfected cells 48 h post transfection. The formation of GJPs was observed in cells co-expressing CaM with wild-type Cx36 and mutant proteins (Figure 1C). A quantification of the GJP size revealed no significant differences in the GJP sizes by the mutations introduced (Figure 1D) (values in μm²; WT: 1.72 ± 0.20, G276A: 1.22 ± 0.09, p = 0.05, W277A: 1.37 ± 0.11, p = 0.19, R278A: 1.13 ± 0.11, p = 0.01). The number of cells with GJP formation in transfected cells was significantly reduced for mutant W277A, but not when WT was compared to G276A and R278A (Figure 1E) (values in %; WT: 13.68 ± 2.52, G276A: 11.24 ± 2.03, p = 0.5, W277A: 3.63 ± 0.72, p = 1.9 × 10^-5, R278A: 8.69 ± 1.55, p = 0.12).

Two types of experiments were performed to clarify the distribution of Cx36 and W277A in Neuro2A cells. When single transfected Neuro2A cells expressing Cx36–EGFP or Cx36–DsRed monomer were paired and analyzed 48 h after transfection, 3D imaging of cell pairs showed mixed GJPs (see inset in Figure 2A; Supplementary Figure S1), and limited presentation of Cx36 in mixed vesicles. Cell pairs expressing the W277A mutation showed mixed GJPs (see Supplementary Figure S2), and an extensive display of vesicular structures exchanged between cell pairs, suggesting that the mutation altered GJP stability (Figure 2A).

Next, the subcellular localization of Cx36 was visualized using Neuro2a cells double transfected with EGFP tagged Cx36 and genetically encoded cell organelle markers for endoplasmic reticulum (ER), Golgi, and mitochondria (Figure 2B, Supplementary Figure S3). Wild-type Cx36 was predominantly localized in vesicles and at GJPs. Outside of GJPs, both 2D images (data not shown) and 3D reconstructions of images from double transfected cells showed Cx36 in vesicles of variable size embedded in, or attached to ER and Golgi (Supplementary Figure S4). Co-localization analysis revealed a significant overlap of Cx36 with ER (0.695 ± 0.0413, n = 30, p = 9.837 × 10^-6) and Golgi organelles (0.651 ± 0.0355, n = 30, p = 4.209 × 10^-5). The Mander’s overlap coefficient for Cx36 and mitochondria was significantly lower (0.356 ± 0.0413, n = 30) demonstrating that exogenously expressed, tagged wild-type Cx36 is not redistributed into this organelle type (Figure 2C; Supplementary Figure S5). The distribution of Cx36 in cell pairs, together with the localization of tagged Cx36 in ER, Golgi and GJPs is consistent with the subcellular distribution of other connexin proteins. When Cx36–W277A was expressed, the Mander’s overlap coefficient was significantly higher than wild-type Cx36 in all organelles (ER: 0.814 ± 0.012, p = 0.009; Golgi: 0.811 ± 0.017, p = 1.6 × 10^-5; mitochondria: 0.812 ± 0.010, p = 2.22 × 10^-9). The distinct subcellular distribution of the W277A mutant suggests a mechanism that interferes with the transport, formation and/or stabilization of GJPs in transfected Neuro2a cells.

The interaction of Cx36–DsRed with CaM–ECFP in living cells was investigated using Fluorescence Resonance Energy Transfer (FRET) (Figure 3). The Cx36–ECFP and Cx36–DsRed assembly into mixed GJPs (Figure 2A) served as a positive FRET control. The FRET efficiency value (FRET_eff expressed in %) was 11.50 ± 0.93. The interaction of Cx36 with the Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) has been shown previously (Alev et al., 2008; Del Corsso et al., 2012). Here, Cx36–ECFP and CaMKII–EYFP served as the second positive control demonstrating that intracellular calcium elevation by treatment with 2 mM extracellular calcium [Ca²⁺]_E and 2 μM ionomycin increased the FRET_eff significantly at GJPs (values in %: non-stimulated: 5.1 ± 0.4; stimulated: 8.1 ± 0.5, p = 0.0002), but not at vesicular structures (non-stimulated: 6.7 ± 0.9; stimulated: 8.1 ± 0.9, p = 0.33) (Figure 3A).

The FRET_eff for Cx36–DsRed and CaM–ECFP was comparable at GJPs and in cytoplasmic vesicles under physiological conditions without calcium stimulation (GJP: 3.23 ± 0.26; vesicles: 3.65 ± 0.54, p = 0.54) (Figure 3B). When the fluorescent protein tags (FRET pair I: Cx36–DsRed, CaM–ECFP; FRET pair II: Cx36–ECFP, CaM–DsRed) were switched (GJP: 3.35 ± 0.49, vesicles: 3.09 ± 0.40, p = 0.82) FRET_eff was not significantly different, suggesting that the protein tags had equal impact.

No significant changes to FRET_eff were observed at GJPs when intracellular calcium levels were raised (non-stimulated: 3.3 ± 0.5, stimulated: 2 mM [Ca²⁺]_E with 2 μM ionomycin; 2.8 ± 0.6, p = 0.33). Intracellular calcium reduction (24 μM BAPTA-AM; 2.8 ± 0.9, p = 0.42), or blocking of CaM with 10 μM (W-7; 2.7 ± 0.6, p = 0.44) had no impact on FRET_eff. Co-transfection with the calcium insensitive CaM_E1234Q mutant (non-stimulated: 1.1 ± 0.3; stimulated: 1.3 ± 0.4, p = 0.85) reduced FRET_eff (Figure 3C). At vesicles, raising intracellular calcium ([Ca²⁺]_I) levels significantly increased FRET_eff (non-stimulated: 3.10 ± 0.4, stimulated: 6.38 ± 0.71, p = 0.002), and led to a significant reduction after BAPTA-AM (0.30 ± 0.30, p = 0.9 × 10^-4), or W-7 application (0.88 ± 0.56, p = 3.1 × 10^-5) (Figure 3D). Similar to our measurements at the GJP, CaM_E1234Q at the vesicles showed a significantly reduced response upon [Ca²⁺]_I increase (non-stimulated: 2.18 ± 0.54, stimulated: 0.45 ± 0.27, n = 21, p = 0.005) (Figure 3D).

Next, FRET-Distance (FRET_dis) for Cx36–CFP and CaM–DsRed was compared at GJPs and the vesicles. Typically, if two interacting proteins are within 10 nm distance, FRET can be achieved (threshold indicated by the dashed line). No significant difference at the GJPs (non-stimulated: 9.11 ± 0.24, stimulated: 9.61 ± 0.33, p = 0.33), but a significant reduction in FRET_dis was observed at cytoplasmic vesicles when [Ca²⁺]_I was increased, (non-stimulated: 9.27 ± 0.22, stimulated: 8.09 ± 0.19, p = 0.002). When CaM_E1234Q was used to test the interaction with Cx36 WT-CFP, neither at the GJP (non-stimulated: 10.93 ± 0.63, stimulated: 10.13 ± 0.53, p = 0.21) nor the vesicles (non-stimulated: 9.96 ± 0.58, stimulated: 10.81 ± 0.46, p = 0.23) was significantly different (Figure 3E). All values were below 10 nm except for CaM_E1234Q, which indicated FRET was not achieved between wild-type Cx36 and CaM_E1234A. We concluded that changes in [Ca²⁺]_I are relayed to Cx36 before reaching GJPs, and that the interaction between Cx36 and CaM is critically dependent on the capacity of CaM to bind calcium.

The role of the amino acid tryptophan in position 277 (W277) of Cx36 was investigated in calcium-dependent CaM binding together with the flanking amino acids 276 and 278. FRET_eff was determined for tagged Cx36 wild-type and Cx36 mutants in vesicles at physiological conditions, and after increasing [Ca²⁺]_I by stimulation with 2 mM [Ca²⁺]_E and 2 μM ionomycin. The FRET pair Cx36–ECFP and Cx36–DsRed served as the positive control (Figure 4). FRET_eff significantly increased in Neuro2a cells co-expressing CaM–DsRed and wild-type Cx36–ECFP (all values in %; non-stimulated: 3.65 ± 0.54, stimulated: 8.82 ± 1.19, p = 0.0003), the Cx36 mutants G276A (non-stimulated: 6.61 ± 0.63, stimulated: 9.74 ± 1.02, p = 0.02), or R278A (non-stimulated: 2.84 ± 0.29, n = 15; stimulated: 6.54 ± 1.02, n = 17, p = 0.003) (Figure 4A). In contrast, FRET_eff for W277A and CaM–DsRed were unaffected by the intracellular Ca²⁺ change (non-stimulated: 3.18 ± 0.39, n = 18; stimulated: 2.90 ± 0.69, p = 0.45).

This result was consistent with changes to the FRET distances (FRET_dis) between tagged Cx36 and CaM after stimulation with 2 mM [Ca²⁺]_E and 2 μM ionomycin. Wild-type Cx36 (all values in nm; non-stimulated: 9.15 ± 0.26, stimulated: 7.69 ± 0.22, p = 0.0003), mutant G276A (non-stimulated: 8.16 ± 0.23, stimulated: 7.48 ± 0.15, p = 0.02), or mutant R278A (non-stimulated: 9.33 ± 0.19, stimulated: 8.26 ± 0.24, p = 0.003) showed a significant decrease in FRET_dis, indicating that binding between both protein pairs increased (Figure 4B). In contrast, the FRET_dis for the Cx36 mutant W277A and CaM increased when intracellular Ca²⁺ was raised (non-stimulated: 9.22 ± 0.20, stimulated: 10.61 nm ± 0.91, p = 0.45). In summary, the FRET analysis suggested that tryptophan 277 is critical for CaM binding and that binding is sensitive to intracellular Ca²⁺ levels.

The W277A mutation decreased the GJP frequency, but not the GJP size (Figures 1D,E), and caused a reduced response to [Ca²⁺]_I elevation (Figures 4A,B). To investigate impaired gap junction communication prompted by the W277A mutation, the non-invasive uptake and redistribution of ethidium bromide across GJPs under normal cell growth conditions was quantified using a method based on gap-FRAP (Abbaci et al., 2007). Pairs of Neuro2a cells expressing Cx36–ECFP or W277A–ECFP, and forming GJPs were selected for imaging (Figures 5A–D, left column). Sample traces for each experimental condition show the fluorescence in regions of interest (R1, red trace; R2, black trace) over the entire time course (Figures 5A–D, right column). Regions of interest were placed close to the GJP (R1) and at a more distant location (R2) in the same cell to capture recovery of ethidium bromide after photobleaching. Quantifications shown below (Figure 5E) demonstrated that treatment with 2 mM [Ca²⁺]_E and 2 μM ionomycin caused a significant ethidium bromide fluorescence recovery in Neuro2a cells expressing the Cx36–ECFP protein (values in %: non-stimulated: 5.82 ± 4.22, stimulated: 47.54 ± 13.56, p = 0.0045). Blocking of CaM with W-7 led to a significant reduction of fluorescence recovery when intracellular calcium levels were increased suggesting that CaM binding is critical for this response (non-stimulated: 16.60 ± 8.01, stimulated: 3.77 ± 3.41, p = 0.1427). Treatment with the gap junction blocker carbenoxolone (CBX) modulated fluorescence recovery (non-stimulated: -8.66 ± 4.65, stimulated: 8.91 ± 1.89, p = 0.0032). When Neuro2a cells expressed Cx36–W277A, no significant recovery was seen after the same treatment (non-stimulated: 8.12 ± 4.21, stimulated: -0.55 ± 8.35, p = 0.425). It was concluded that W277A expression impairs dye redistribution between cell pairs in a way similar to a gap junction blocker. Control experiments using non-transfected Neuro2a cells with and without ionomycin treatment showed limited recovery (Neuro2a without ionomycin treatment: -15.72 ± 17.52, Neuro2a with ionomycin treatment: 3.40 ± 1.23, p = 0.001). The recovery observed after CBX or ionomycin treatment likely reflects alterations to the biophysical properties of the cell membranes caused by the compounds applied.

Following a strategy used successfully in earlier structure determination (Ogura et al., 2012), a single-chain protein hybrid consisting of human CaM, a short glycine-rich linker, and the CaM binding region of rat Cx36 (Figure 6A) was expressed and uniformly ¹³C, ¹⁵N labeled for NMR spectroscopy. Amide ¹H, ¹⁵N HSQC spectra of this CaM–Cx36 hybrid protein (Supplementary Figure S6) and a sample of ¹⁵N-labeled CaM in a 1:1 complex with a wild-type 25 amino acid long Cx36 carboxy-terminal tail peptide were similar suggesting that the structure of the CaM–Cx36 hybrid was an accurate representation of a binary Cx36–CaM interaction.

The Cx36–CaM hybrid protein structure was determined using 1100 distance restraints from ¹⁵N- and ¹³C-edited 3D-NOESY spectra, supplemented with a ¹²C-filtered, ¹³C-edited 2D-NOESY of a ¹³C-CaM/¹²C-Cx36 peptide complex (Table 2). The ensemble of 20 lowest energies structures is shown in Supplementary Figure S7. From the representative structure of the Cx36–CaM complex, CaM binds Cx36 in a well-known compact form (Figure 6B) with both Ca²⁺ saturated lobes enveloping a helical ligand with amphipathic features (Figure 6C). Based on the spacing of hydrophobic amino acids with W277 at anchor position 1, Cx36 resembles a 1–8–14 class ligand except that position 14 is occupied by K280 where a hydrophobic acid would typically be found. Accordingly, the Cx36–CaM protein structure compares favorably to other crystal structures of compact CaM complexes bearing 1–8–(14) ligands from smooth muscle myosin light chain kinase (Meador et al., 1992) (smMLCK; PDB:1CDL, 1.63 Å) and NMDA receptor NR1 C1 region (Ataman et al., 2007) (NMDAR NR1, PDB:2HQW, 1.81 Å) (Figure 6D). A contact map of Cx36–CaM from position 8 onward is shown in Figure 6E. The guanidino group of arginine at position 9 (R285 in Cx36) is balanced by ionic contacts from E84 in CaM. Likewise, the hydrophobic amino acid position 11 (A287 in Cx36) is supported by F19, L32, and V91 in CaM. At the penultimate contact position 14, the helix of Cx36 terminates earlier than smMLCK and NMDAR NR1 ligands. Despite this difference, three methionines in CaM (M36, M51, and M72) still contribute important hydrophobic contacts to Cx36. A lysine at position 14 in Cx36 is protruding from the CaM binding cleft toward a series of acidic amino acids in CaM (E47, D50, and E54). Taken together, a continuous network of contacts is observed between CaM and Cx36 thereby establishing Cx36 as bona fide ligand of CaM.

Finally, ITC was performed to determine the affinity of CaM for a 25 amino acid long peptide derived from the Cx36 cytoplasmic tail (Figure 7). As predicted from the NMR structure, the Cx36 peptide bound Ca²⁺-CaM stoichiometrically with a K_d of 0.53 ± 0.05 μM (n = 3). Highlighting the role of the amino acid W277 at the position 1 of the 1–8–(14) motif, no binding was observed for the peptide bearing a W277A substitution (n = 3). Consistent with the structure and biological assays described throughout this study, peptide binding was only observed with Ca²⁺ saturated CaM.

The novel details of CaM binding to rat Cx36 build on the original findings using mouse Cx36 and two perch Cx35 variants (Burr et al., 2005). The Cx36 binding motif is highly conserved across species from humans to lower vertebrates, like lamprey and zebrafish, and overlapping with a CaMKII binding site (Alev et al., 2008). In contrast, the CaM binding sites of Cx35/Cx36 are distinct from other connexins including Cx32, Cx43, Cx44, Cx46, and Cx50 in terms of ligand class (1–8 versus 1–5–10), location of the ligand binding sites, and resemble to date the only connexins activated by CaM (Zhou et al., 2007, 2009; Dodd et al., 2008; Myllykoski et al., 2009; Chen et al., 2011; Xu et al., 2012; Mruk et al., 2014).

The similarity of the Cx36 binding motifs across different species is reflected in comparable dissociation kinetics and thermodynamic characteristics found in mouse, the two perch Cx35 variants (Burr et al., 2005) and the rat Cx36 protein sequence investigated in this study. In more general terms, connexins bind CaM with dissociation kinetics and thermodynamic characteristics covering more than a 100-fold range (Dodd et al., 2008; Myllykoski et al., 2009; Zhou et al., 2009; Chen et al., 2011). Together, the specific localization of binding motifs, distinct subclasses and different dissociation kinetics suggest functional adaptations specific for each connexins protein.

The low affinity of Cx36 to CaM, at ∼500 nM, may have arisen due to the substitution of K290 at position 14 of typical 1–8–14 ligands like smMLCK, NMDAR-NR1, or MARCKS, which CaM encounters in the compact state. Furthermore, the less extensive array of hydrophobic contacts arising from the shorter side chains throughout Cx36, namely A283 and A287, cannot be compensated by larger hydrophobic amino at position 14 (K290 in Cx36). Regardless of the precise structural basis, a lower affinity may help Cx36 quickly sample CaM, CaMKII and possibly other proteins, depending upon the numerous signaling contexts it participates in throughout the life cycle of this protein.

Generally, CaM binding to connexins has been investigated in the context of gap junction communication, mainly focusing on implications for channel functions in the cell membrane. On the basis of the subcellular distribution of Cx36 and mutants thereof, we have expanded this focus, showing that binding already occurs before Cx36 reaches the cell membrane forming gap junctions, or after removal from the GJP. Interestingly, the W277A mutation caused retention in vesicles and showed a reduced gap junction coupling, reflecting in part similar observations made for Cx43 (Zhou et al., 2007). Further, gap junctions made by W277A are less frequent and more prone to removal as annular plaques found distributed in cell pairs, suggesting that decorating Cx36 with CaM is critical and most likely serving multiple purposes during the life cycle of this connexin.

We have previously proposed a model predicting the sequential, cooperative processing of interaction between CaM, CaMKII, and Cx36 at the GJP (Alev et al., 2008). The main results of this study are more consistent with a sequential and competitive process, sequestering calcium-dependent actions between vesicles and GJPs. To emphasize the new findings, we propose a revised model for plasticity at electrical synapses. As highlighted in Figure 8, both CaM and CaMKII have the capacity to bind Cx36 at vesicles, with CaMKII binding at physiological calcium levels, and CaM binding only when intracellular calcium is elevated. At GJPs, CaMKII is calcium-responsive, with CaM showing no further strengthening of binding in response to calcium increase. This sequence of actions differs significantly from the activation of NMDA receptors (Bayer et al., 2001), and lead us to conclude that electrical synapses share a portfolio of interacting proteins, but mechanisms of achieving plasticity do not appear to fully recapitulate those found at chemical synapses. Instead, CaMKII binding at Cx36 at GJP seems to be the critical interaction conferring calcium-dependent plasticity at electrical synapses, while calcium-dependent sensitization of CaM appears restricted to spatially separate compartments. This could endow electrical synapses with a tunable molecular machinery using calcium as the main regulator for activation through CaMKII, while interaction of Cx36 with CaM open new avenues for linking this channel to intracellular calcium signaling pathways.

What could be the function(s) of CaM binding to Cx36? The work by Burr et al. (2005) and our data suggest that CaM binding to Cx36 will occur when [Ca²⁺]_in levels are elevated. Such conditions typically are found during synaptic activity in neurons (Gamble and Koch, 1987). Since elevation of [Ca²⁺]_in increases Cx36–CaMKII, but not Cx36–CaM interaction at GJP, we conclude that calcium signals converge through CaM onto Cx36 before the protein reaches the GJP, ruling out a major role in plasticity.

It is reasonable to speculate that Cx36–CaM binding at vesicles may be sensitive to competition with other proteins with higher CaM affinity, suggesting that CaM binding could be orchestrated and/or sequestered by calcium affinities of CaM relative to CaMKII, or other, yet unidentified calcium binding proteins, or those activated or sensitized by calcium. Prior to its arrival at GJPs, Cx36–CaM could promote vesicular transport to the cell membrane and/or post-translational modifications affecting the oligomerization of connexins in the Golgi complex (Musil and Goodenough, 1993), as well as the assembly or disassembly of Cx36 into GJPs, potentially in a way similar to glutamate receptor trafficking to dendritic spines (Korkotian and Segal, 2007; Loo et al., 2014). Alternatively, calcium-dependent binding of CaM to Cx36 could link functional Cx36 channels in the ER directly to intracellular calcium signaling as suggested for Panx1, Panx3 or the transient receptor potential vanilloid 1 channel (TRPV1) (Vanden Abeele et al., 2006; Gallego-Sandin et al., 2009; Ishikawa et al., 2011), or by limiting calcium leakage from the ER as exemplified by CaM binding to Sec61 alpha (Erdmann et al., 2011). Addressing these consequences in future experiments will close this important knowledge gap.