Localized calcium signaling and the control of coupling at Cx36 gap junctions.

A variety of electrical synapses are capable of activity-dependent plasticity, including both activity-dependent potentiation and activity-dependent depression. In several types of neurons, activity-dependent electrical synapse plasticity depends on changes in the local Ca2+ environment. To enable study of local Ca2+ signaling that regulates plasticity, we developed a GCaMP Ca2+ biosensor fused to the electrical synapse protein Connexin 36. Cx36-GCaMP transfected into mammalian cell cultures formed gap junctions at cell-cell boundaries and supported Neurobiotin tracer coupling that was regulated by protein kinase A signaling in the same way as Cx36. Cx36-GCaMP gap junctions robustly reported local Ca2+ increases in response to addition of a Ca2+ ionophore with increases in fluorescence that recovered during washout. Recovery was strongly dependent on Na+-Ca2+ exchange activity. In cells transfected with NMDA receptor subunits, Cx36-GCaMP revealed transient and concentration-dependent increases in local Ca2+ upon brief application of glutamate. In HeLa cells, glutamate application increased Cx36-GCaMP tracer coupling through a mechanism that depended in part on Ca2+, calmodulin-dependent protein kinase II activity. This potentiation of coupling did not require exogenous expression of glutamate receptors, but could be accomplished by endogenously expressed glutamate receptors with pharmacological characteristics reminiscent of NMDA and Kainate receptors. Analysis of RNA Sequencing data from HeLa cells confirmed expression of NMDA receptor subunits NR1, NR2C and NR3B. In summary, Cx36-GCaMP is an effective tool to measure changes in the Ca2+ microenvironment around Cx36 gap junctions. Furthermore, HeLa cells can serve as a model system to study glutamate receptor-driven potentiation of electrical synapses.Significance We have developed a Connexin 36-GCaMP3 fusion construct that effectively reports the Ca2+ microenvironment in the vicinity of Cx36 gap junctions. This tool will be valuable to investigate the dynamic changes in Ca2+ that are responsible for some forms of electrical synapse plasticity. Furthermore, we have discovered that a widely used model system for in vitro studies, the HeLa cell, endogenously expresses glutamate receptors that effectively drive intracellular Ca2+, calmodulin-dependent protein kinase II signaling. This signaling can be exploited in many types of studies.


Introduction
Electrical synapses provide a means to organize neurons into networks from which high-order activity emerges. Plasticity is a fundamental property of electrical synapses, altering the strength of electrical communication between coupled neurons and potentially playing a large role in regulation of network states (Coulon and Landisman, 2017). Electrical synapse plasticity is also a critical element of microcircuit functions. For example, individual auditory afferent terminals that form mixed chemical and electrical synapses onto Mauthner neurons display a high degree of variability in both electrical and chemical synaptic strength (Smith and Pereda, 2003), influencing the effect of any one auditory neuron on Mauthner cell responses. Both components can be modified by activity, resulting in either potentiation or depression of individual elements, with interdependence of chemical and electrical synapse activity (Pereda et al., 2003b;Smith and Pereda, 2003).
In the fish Mauthner neurons, high-frequency stimulation of the afferent nerve results in potentiation of electrical synapses (Yang et al., 1990;Pereda and Faber, 1996). Activity-dependent potentiation of electrical synapses has also been observed in mammalian AII amacrine cells (Kothmann et al., 2012) and inferior olive neurons (Turecek et al., 2014). Common among these neural networks is the reliance on activation of NMDA receptors, influx of extracellular Ca 21 and activation of Ca 21 , calmodulin-dependent protein kinase II (CaMKII) activity (Pereda et al., 1998(Pereda et al., , 2003aKothmann et al., 2012;Turecek et al., 2014) to potentiate coupling. In retinal AII amacrine cells, the potentiation has been demonstrated to result from CaMKII-dependent phosphorylation of Cx36 (Kothmann et al., 2012). Ca 21 -dependent depression of electrical synapses has also been reported, depending on NMDA receptors in inferior olive neurons (Mathy et al., 2014) and voltage-dependent Ca 21 channels in thalamic reticular neurons (Sevetson et al., 2017), suggesting that the control of coupling by Ca 21 entry can be subtle.
Because of the central role of Ca 21 signaling in activitydependent modulation of electrical synapse strength, we wished to investigate the relationship between the Ca 21 microenvironment around Cx36 gap junctions and their functional plasticity. To accomplish this, we developed a Cx36 construct fused to the genetically-encoded Ca 21 biosensor GCaMP3 (Tian et al., 2009). This construct is an effective tool to examine Ca 21 signaling in the context of gap junction functional plasticity.
Unless otherwise indicated, restriction enzymes, DNA polymerases and enzymes used for cloning were obtained from New England Biolabs. To produce C-terminal EGFP-tagged Cx36, the mouse Cx36 cDNA was amplified by PCR using Pfu Ultra DNA polymerase (Agilent) with primers T7 and JOB 284 (all primer sequences are listed in Table 1), the latter of which eliminates the stop codon of Cx36 and adds a KpnI site in frame with the EGFP-N1 vector. The PCR product was cloned into EcoRI and KpnI sites of the EGFP-N1 vector by conventional cloning. A Cterminal GCaMP3-tagged Cx36 was produced in a similar fashion using primers T7 and JOB 285, the latter of which eliminates the stop codon of Cx36 and adds a BclI site. The PCR product was digested with SmaI and BclI and cloned into AfeI and BglII sites of the GCaMP3 vector. Resulting clones were fully sequenced. Cx36-GCaMP is available through Addgene (plasmid #123604).
A series of clones was created using the PB513B-1 dual expression vector of the Piggybac Transposon System (System Biosciences LLC). The vector was modified to accept a second user-derived sequence by deleting the GFP open reading frame driven by the EF1a promoter by digestion with Bsu36I and BgmBI and inserting a linker containing PacI and SbfI restriction sites followed by an IRES sequence derived from the vector pIRES DsRedT3-KR24 (Moshiri et al., 2008). The modification was completed by three-way assembly of PCR products amplified with Phusion polymerase using primers KBM 1 and KBM 2 for the IRES and primers KBM 3 and KBM 4 to replace a deleted portion of the EF1a promoter. Cold Fusion cloning mix (System Biosciences) was used for the assembly.
The full open reading frame of Cx36-GCaMP, including the SV40 poly A signal, was amplified by PCR using primers KBM 7 and KBM 8 and inserted into the SwaI site following the CMV promoter of the modified PB513 vector using Gibson Assembly. The full coding region of NMDA receptor NR1 was amplified by PCR using primers JOB 405 and JOB 406 and cloned into the PacI and SbfI sites following the EF1a promoter of the same construct to generate a dual-expression construct expressing both NR1 and Cx36-GCaMP. The NR2 subunits were each cloned into the SwaI site of the modified PB513 vector using Gibson Assembly as described above to generate separate single-expression clones. Resulting clones were fully sequenced.

Cell culture and transfection
Human Embryonic Kidney 293T/17 cells (HEK293, catalog #CRL-11268; ATCC) used for calcium imaging and immunolabeling experiments were grown in DMEM with sodium pyruvate 1 10% fetal bovine serum (FBS) with penicillin/streptomycin/fungizone. Cells for all experiments were used between passages 4 and 10 relative to the original cell line obtained from ATCC. Cell culture reagents were obtained from Gibco/ThermoFisher. Cells were plated onto 35-mm culture dishes containing coverslips coated with poly-L-lysine and Laminin (Sigma) and grown to 70-80% confluency. Cells were transiently transfected with Cx36-GCaMP/NR1 and various combinations of NR2 clones in Piggybac vectors, up to a maximum of 4 mg of DNA per dish, using GenePorter H (Genlantis) following the manufacturer's protocol. For all transfections including clones in the modified PB513 vector, Piggybac transposase (PB200, System Biosciences) was included in the transfection mix. Following transfection, cells were grown in MEM (without essential amino acids) 1 sodium pyruvate, 10% FBS, penicillin/streptomycin/fungizone and 100 mM APV (DL-2-amino-5-phosphonopentanoic acid; Sigma) to prevent toxicity from activation of transfected NMDA receptors. Cells were cultured for 24 h following transfection, rinsed twice with MEM without APV, and used for experiments.
HeLa cells (catalog #CCL2; ATCC) used for tracer coupling experiments were grown in MEM with essential amino acids 1 sodium pyruvate, 10% FBS and penicillin/ streptomycin/fungizone. Cells for all experiments were used between passages 7 and 20 relative to the original cell line obtained from ATCC. Cells were plated onto tissue culture coated coverslips (ThermoFisher) and grown to 70-80% confluency. Cells were transiently transfected with clones in pcDNA vectors or Piggybac vectors as were HEK293 cells and cultured for 24 h. Before use in experiments, cells were rinsed twice with MEM.
Cell perfusion and imaging Both HeLa and HEK293 cells were tested in perfusion and imaging experiments. HEK293 cells were superior for this application since most of the transfected Cx36-GCaMP was present at gap junctions and these were well organized at vertical cell-cell interfaces (Fig. 1A). HeLa cells displayed a more variable proportion of Cx36-GCaMP at cell-cell versus intracellular membrane compartments, making selection of regions of interest for quantification more difficult. To perform the imaging experiments HEK293 cells on 12-mm coverslips transfected with clones for Cx36-GCaMP, NMDA NR1, and NMDA NR2 subunits (NR2A, NR2B, or NR2C) were placed in perfusion chambers (Model RC-25, Warner Instruments) and perfused with cell maintaining solution (CMS) containing 150 mM NaCl, 6.2 mM KCl, 1. 2 mM NaH 2 PO 4 , 1.2 mM MgSO 4 , 2. 5 mM CaCl 2 , 10 mM glucose, 10 mM HEPES (pH 7.4), and 1 mM glycine (CMS 1 glycine) at 37°C using a VC 3 8-channel gravity perfusion system (ALA Scientific Instruments). Cells were imaged using a Zeiss Axiovert 200 microscope with 40Â/0.5 NA objective and an ORCA 100 digital camera (Hamamatsu Photonics) using 470BP40 excitation and 535BP40 emission filters. Live images were captured using HCImage software (Hamamatsu) set to record with 4 Â 4 binning and 0.4-to 0.8-s time intervals. Two to five individual gap junctions on each coverslip were marked as regions of interest for recording. Approximately 20-s baseline measurement was obtained under perfusion with CMS 1 glycine before switching into solutions Research Article: New Research containing glutamate (30 mM, 100 mM, or 1 mM) for 40 s, followed by return to CMS 1 glycine. The same perfusion timing was used for applications of 5 mM ionomycin (Fisher Chemical) in CMS 1 glycine. Ionomycin experiments also included 5 min pre-incubation with 200 nM Na 1 -Ca 21 exchanger antagonist SEA 0400 (2-[4-[(2,5-difluorophenyl) methoxy]phenoxy]À5-ethoxybenzenamine; R&D Systems) or 100 nM SR-ER Ca 21 -ATPase antagonist thapsigargin (R&D Systems) in CMS 1 glycine, followed by a perfusion experiment with drug in both perfusion solution and ionomycin solution.
Raw fluorescence intensity data from regions of interest were exported from HCImage as tabular data and analyzed in Excel (Microsoft) and Prism (GraphPad). Declining fluorescence intensity of the baseline due to photobleaching was fit with a first-order exponential function or a linear function for each region of interest. This baseline function was used as F 0 to calculate DF/F 0 without further correction for background fluorescence, since background regions of interest often showed some changes in fluorescence intensity due to scattered signals from nearby fluorescent Cx36-GCaMP structures. For each response, we calculated the peak value of DF/F 0 and the integrated area under the response curve, consisting of the sum of DF/F 0 values for each time point contained within the response normalized to 1-s time intervals.
In each experiment, a minimum of five loaded regions along the scraped edge of the cells were identified for complete single cell loading of Neurobiotin with identifiable transfer of tracer away from the loaded cell as an indicator of a coupled region for measurement. Images were collected using HCImage software and analyzed with SimplePCI (Hamamatsu) and MATLAB (MathWorks) software. Intensity of Neurobiotin/Cy3-streptavidin signal was measured in 2-mm circles, with the brightest regions of individual cells selected. Cell-to-cell distance was measured from center to center of adjacent cells. Tracer diffusion was estimated by fitting data from regions of cells extending out from a loaded cell along the scraped edge of the coverslip (Fig. 1B) using a linear compartmental diffusion model (Zimmerman and Rose, 1985) as implemented for neural networks (Mills and Massey, 1998;O'Brien et al., 2004). In this model we assume the cells are arranged in a linear compartment chain that is connected by Cx36 gap junctions and may be characterized by a rate-limiting diffusion coefficient k. Independent measurements of k were made for each of the five to eight loaded regions examined within a single experiment. Three to six experiments were performed and all values of k used for comparisons.

Gene expression analysis
In order to better understand the glutamate receptor and connexin gene expression in HeLa cells, we examined RNASeq dataset GSM759888 (https://www.ncbi.nlm.nih. gov/geo/query/acc.cgi?acc=GSM759888) deposited in GEO (Cabili et al., 2011). The bam file of this dataset was mapped against Hg19 human reference genome using Samtools v1.9 (Genome Research Ltd.; http://www.htslib.org; Li et al., 2009) on macOS v10.13/10.14 to generate a bam indexed (bai) file. The files were analyzed using Integrative Genomics Viewer (IGV) v2.4.14 (Broad Institute and Regents of the University of California; https://software.broadinstitute.org/software/igv/ home; Robinson et al., 2011). Each gene of interest was manually searched and an image of read count coverage was exported. IGVtools "count" function was then used to generate the raw read count and normalized read count.

Statistical analyses
Tracer coupling experiments were performed with control and drug-treated conditions on the same experimental days and batches of cells. Three experiments were performed, except in one case in which a drug treatment did not work, and three additional experiments with control and certain drug treatments were performed. All measurements of k were used from the experiments. A criterion of 63 SDs from the mean was used to exclude outliers among the individual measurements for data analysis. Treatments were compared with one-way or two-way ANOVA with appropriate multiple comparison tests. Tukey's multiple comparison tests were used when conditions were compared with more than one other condition; Dunnett's multiple comparisons were used to compare a specific condition in one construct to the same condition in another construct. A summary of all statistical tests performed for this study is presented in Table 2.

Development of a gap junction calcium sensor
In order to investigate calcium signaling in the microdomain surrounding Cx36 gap junctions we developed a fusion of mouse Cx36 with GCaMP3 at its cytoplasmic C terminus. Similar to fluorescent protein fusions to the C termini of connexins, this construct, called Cx36-GCaMP, formed gap junctions at cell-cell boundaries when transfected into HEK293 cells (Fig. 1A) or HeLa cells (data not shown). To determine whether Cx36-GCaMP formed functional gap junctions we performed tracer coupling experiments. Because of high background tracer coupling in HEK293 cells due to endogenous connexins, these experiments were performed in transiently-transfected HeLa cells. HeLa cells transfected with control (EGFP) or connexin constructs (Cx36-EGFP or Cx36-GCaMP) were loaded with Neurobiotin along the edge of a scrape (Fig.  1Bi) and tracer diffused to connected cells via gap junctions. Fits of Streptavidin-Cy3 fluorescence versus distance from the loaded cell in cell-to-cell spacings with a linear compartmental diffusion model (Fig. 1Bii) allowed calculation of the diffusion coefficient k for Neurobiotin tracer transfer. Figure 1C shows diffusion coefficients measured in cells transfected with Cx36-GCaMP compared with cells transfected with EGFP-N1 (no gap junction control) and Cx36-EGFP. HeLa cells without an added connexin (EGFP control) support some tracer coupling due to the presence of an endogenous connexin.  Ca 21 remained elevated while ionomycin perfusion continued ( Fig. 2A), but recovered rapidly on washout in normal CMS. To better understand the etiology of the recovery we investigated the influence of the SR/ER calcium ATPase and the Na 1 /Ca 21 exchanger on calcium dynamics. Perfusion of the SR/ER calcium ATPase inhibitor thapsigargin (100 nM; Fig. 2B, gray bar) caused an immediate, small increase in intracellular Ca 21 , followed by a more variable delayed rise in intracellular Ca 21 (n = 5 gap junctions in one experiment) . Application of 5 mM ionomycin after 5-min pretreatment with thapsigargin ( Fig.  2C; n = 15 gap junctions in three experiments) resulted in a delayed and reduced relative rise in intracellular Ca 21 with delayed recovery (compare response to the mean ionomycin response in control conditions shown by the black line; ionomycin treatment indicated by the black bar). Note that the intracellular Ca 21 used as the baseline F 0 is elevated, so the peak Ca 21 may not be substantially lower than in control conditions. Treatment with the Na 1 / Ca 21 exchange inhibitor SEA 0400 (200 nM) also delayed the rise in intracellular Ca 21 and prevented recovery (  significantly influence the kinetics of Ca 21 rise and recovery.

NMDA receptor activation stimulates Ca 26 signals at Cx36-GCaMP gap junctions
To investigate the coupling of NMDA receptor activation to calcium signals in the vicinity of Cx36 we transfected HEK293 cells with Cx36-GCaMP, NMDA receptor NR1 subunit and one of the NMDA receptor NR2 subunits NR2A, NR2B or NR2C. Figure 3 shows immunostaining of HEK293 cells transfected with Cx36-GCaMP and NR1 (Fig. 3A), or these two plus NR2A (Fig. 3B), NR2B (Fig.  3C), or NR2C (Fig. 3D). Expression level did vary among cells, and expression of very high levels of protein in a cell frequently caused delocalization of Cx36-GCaMP away from gap junctions. Only well-formed gap junctions were used for imaging Ca 21 responses, despite quite robust responses observed in cells diffusely expressing Cx36-GCaMP.
Triple-transfected cells were superfused with CMS and switched for 40 s into a solution containing 30 mM, 100 mM, or 1 mM glutamate. CMS in all conditions contained 1 mM glycine to act as a co-agonist for NMDA receptor activation. Figure 4A-C shows example raw single Cx36-GCaMP gap junction responses to 100 mM glutamate stimulation. NMDA receptors containing NR2A, NR2B or NR2C all drove transient increases in GCaMP fluorescence, indicative of Ca 21 increases in the microenvironment surrounding the gap junction. Baseline-subtracted average responses from 4 to 11 gap junctions are shown in Figure 4D-F. All three types of NMDA receptors drove robust transient increases in local Ca 21 within the first few seconds of glutamate application followed by gradual declines. There were no consistent differences in the kinetics of Ca 21 signals recorded at Cx36-GCaMP gap junctions in response to glutamate stimulation of NMDA receptors containing NR2A, NR2B or NR2C. Figure 5A,B shows concentration-response relationships of the response peak of 8-25 gap junctions in two to eight experiments in cells expressing NR1 and NR2A or NR2B to 30 mM, 100 mM, and 1 mM glutamate. Both NMDA receptor types drove concentration-dependent increases in peak response that were largely similar to each other, with both appearing to saturate between 100 mM and 1 mM glutamate. Because changes in signaling to the gap junction are likely to depend on the total Ca 21 encountered during NMDA receptor responses, we also compared integrated areas under the response curve (Fig. 5C,D) during the course of the responses. These also showed concentration-dependent increases that saturated between 100 mM and 1 mM glutamate.
We further characterized the glutamate receptors endogenously expressed in HeLa cells through pharmacological experiments using the effect of 5-min incubation with 100 mM glutamate on tracer coupling of Cx36-GCaMPtransfected cells as an assay. Figure 7 shows that several selective and poorly selective glutamate receptor inhibitors prevented the increase in coupling caused by incubation with 100 mM glutamate (one-way ANOVA with Tukey's  . Effects of glutamate application on tracer coupling in HeLa cells expressing EGFP or Cx36-GCaMP with or without added NMDA receptor subunits. The diffusion coefficient (k) for Neurobiotin tracer diffusion is shown for 5-min preincubation plus 10-min tracer diffusion time in control media (Con) or control media plus 100 mM glutamate (100 Glu). All data are shown from three experiments; bars show mean values; error bars show 95% confidence limits of the mean; ppp , 0.01, ppppp , 0.0001 versus same transfection composition in control media. multiple comparison tests). Each of the following significantly reduced coupling below the 100 mM glutamate condition: 100 nM of the selective kainate receptor antagonist ACET (p , 0.0001, Glu n = 42 measurements in six experiments, Glu1ACET n = 23 measurements in three experiments, mean effect size À0.00104, 95% CI À0.00079 to À0.00128), 10 mM of the poorly selective AMPA/kainate receptor antagonist CNQX (p , 0.0001, Glu n = 42 measurements in six experiments, Glu1CNQX n = 19 measurements in three experiments, mean effect size À0.00076, 95% CI À0.00050 to À0.00102), 40 mM of the poorly selective AMPA/kainate receptor antagonist GYKI 53 655 (p , 0.0001, Glu n = 42 measurements in six experiments, Glu1GYKI n = 15 measurements in three experiments, mean effect size À0.00080, 95% CI À0.00051 to À0.00108), or 10 mM of the selective NMDA receptor antagonist CPP (p , 0.0001, Glu n = 42 measurements in six experiments, Glu1CPP n = 19 measurements in three experiments, mean effect size À0.00104, 95% CI À0.00078 to À0.00130). Of these treatments, only CNQX did not fully block the increase caused by 100 mM glutamate, yielding a very slight increase in coupling (p = 0.038, Con n = 39 measurements in six experiments, Glu1CNQX n = 19 measurements in three experiments, mean effect size 0.00027, 95% CI 0.00001-0.00054). This indicates that glutamate receptors in HeLa cells that influence Cx36 coupling have pharmacological characteristics of kainate and NMDA receptors, and may as well have some characteristics of AMPA receptors.
To further investigate the glutamate receptor expression in HeLa cells, we analyzed HeLa RNA Sequence dataset GSM759888 deposited in GEO (Cabili et al., 2011), searching for all of the annotated human ionotropic and metabotropic glutamate receptor genes. Sequence reads were found aligning to a kainate receptor Grik1a antisense transcript, and to transcripts of NMDA receptor subunit genes Grin1 (NR1), Grin2c (NR2C), and Grin3b (NR3B ;  Table 3). No other glutamate receptor gene transcripts were detected. This further confirms that NMDA receptor genes are expressed in HeLa cells, but does not support the expression of kainate-type receptors, as suggested by the pharmacological properties of the glutamatedriven coupling changes.
We also examined the expression of connexin genes to gain insight into the connexin background that contributes to coupling in HeLa cells (Table 3). Substantial numbers of reads mapped to Gja1 (Cx43) and Gjc1 (Cx45) genes, and a smaller number of reads mapped to Gja9 (Cx59)-MYCB readthrough transcript, Gjb3 (Cx31), Gjc2 (Cx47), and Gjd3 (Cx31.9). Thus, there is likely a mixed background of connexins contributing to the coupling detected in control (EGFP-transfected) cells, and as well in those transfected with connexin constructs.
Because glutamate receptor signaling that increases coupling in Cx36 gap junctions depends, at least in part, on CaMKII activity (Pereda et al., 1998;Kothmann et al., 2012), we examined whether CaMKII was involved in the glutamate-driven increase in Cx36-GCaMP coupling in HeLa cells. Because of the presence of endogenous glutamate receptors that influenced coupling, we did not add additional NMDA receptor subunits. Figure 8 shows tracer coupling of HeLa cells transfected with Cx36-GCaMP or with EGFP. Treatment with 10 mM CaMKII inhibitor KN-93 significantly reduced the increase in coupling caused by 100 mM glutamate incubation (two-way ANOVA with Tukey's multiple comparison tests: p , 0.0001, Glu n = 20 measurements in three experiments, Glu1KN-93 n = 22 measurements in three experiments, mean effect size À0.00129, 95% CI À0.00094 to À0.00163), although coupling was still slightly enhanced compared with control (p = 0.021, Con n = 17 measurements in three experiments, Glu1KN-93 n = 22 measurements in three experiments, mean effect size 0.00040, 95% CI 0.00041-0.00076). In contrast, while coupling was significantly increased by inhibition of PKA with 10 mM Rp-8-cpt-cAMPS (p , 0.0001, Con n = 17 measurements in three experiments, Rp n = 22 measurements in three experiments, mean effect size 0.00103, 95% CI 0.00067-0.00139), 10 mM KN-93 did not block this increase (p = 0.998, Rp n = 22 measurements in three experiments, Rp1KN-93 n = 19 measurements in three experiments). Thus, the enhancement of coupling caused by activation of endogenous glutamate receptors depended in part on CaMKII, while the coupling revealed by PKA inhibition did not.

Discussion
Plasticity is a fundamental property of electrical synapses (Curti and O'Brien, 2016) and a great deal of this plasticity results from activity-dependent processes Haas et al., 2016). Calcium signaling has been identified as a critical element of electrical synapse activity-dependent plasticity, controlling protein kinase and phosphatase activities that modify synapse strength (Haas et al., 2016;Coulon and Landisman, 2017;Sevetson et al., 2017;O'Brien, 2019). We have designed a gap junction-localized calcium sensor to study the microenvironment that is most relevant for regulation of electrical synapse plasticity in hopes to further the understanding of dynamic regulation of electrical synapse strength.
Cx36-GCaMP responds robustly to influx of extracellular Ca 21 and reports dynamic changes in its concentration in the immediate vicinity of the gap junction. The Glutamate metabotropic receptor 1 0 0 Grm2 Glutamate metabotropic receptor 2 0 0 Grm3 Glutamate metabotropic receptor 3 0 0 Grm4 Glutamate metabotropic receptor 4 0 0 Grm5 Glutamate metabotropic receptor 5 0 0 Grm5-As1 GRM5 antisense RNA 1 0 0 Grm6 Glutamate metabotropic receptor 6 0 0 Grm7 glutamate metabotropic receptor 7 0 0 Grm7-As1 GRM7 antisense RNA 1 0 0 Grm7-As2 GRM7 antisense RNA 2 0 0 Grm7-As3 GRM7 antisense RNA 3 0 0 Grm8 Glutamate metabotropic receptor 8 0 0 Grik1 Glutamate ionotropic receptor kainate type subunit 1 0 0 Grik1-As1 GRIK1 antisense RNA 1 0 0 Grik2 Glutamate ionotropic receptor kainate type subunit 2 0 0 Grik3 Glutamate ionotropic receptor kainate type subunit 3 0 0 Grik4 Glutamate ionotropic receptor kainate type subunit 4 0 0 Grik5 Glutamate ionotropic receptor kainate type subunit 5 0 0 Gria1 Glutamate ionotropic receptor AMPA type subunit 1 0 0 Gria2 Glutamate ionotropic receptor AMPA type subunit 2 0 0 Gria3 Glutamate ionotropic receptor AMPA type subunit 3 0 0 Gria4 Glutamate ionotropic receptor AMPA type subunit 4 0 0 Grid1 Glutamate ionotropic receptor delta type subunit 1 0 0 Grid1-As1 GRID1 antisense RNA 1 0 0 Grid2 Glutamate ionotropic receptor delta type subunit 2 0 0 Grin2a Glutamate Gap junction protein beta 6 (Cx30) 0 0 GJB7 Gap junction protein beta 7 (Cx25) 0 0 GJC3 Gap junction protein gamma 3 (Cx29) 0 0 GJD2 Gap junction protein delta 2 (Cx36) 0 0 GJD4 Gap junction protein delta 4 (Cx40.1) 0 0 dynamic nature of the Ca 21 microenvironment is particularly apparent in the response to administration of ionomycin to HEK cells expressing Cx36-GCaMP (Fig. 2). While Ca 21 rises to a peak shortly after onset of ionomycin administration, its level plateaus and is rapidly reduced on washout by cellular processes, which include both extrusion via Na 1 -Ca 21 exchangers and sequestration in intracellular compartments such as endoplasmic reticulum and mitochondria. It is likely that the plasma membrane Ca 21 -ATPase, which we did not examine, also plays a significant role in extrusion of Ca 21 from the cell. Intracellular stores in the ER play a complex role. Inhibiting the SERCA pump with thapsigargin raised intracellular Ca 21 only a small amount, but potently damped the rate of rise of Ca 21 on ionomycin addition. This suggests that the rise in control conditions includes elements of Ca 21 -induced Ca 21 release from the ER stores, or that the elevated cytoplasmic Ca 21 has activated the plasma membrane Ca 21 -ATPase or Na 1 -Ca 21 exchanger. It is likely that all of these processes are involved in the dynamics of Ca 21 in the vicinity of the Cx36 gap junctions. Pertinent to understanding activity-dependent plasticity are the responses of Cx36-GCaMP to glutamate administration in the presence of NMDA receptors. All three types of NMDA receptors tested, those containing NR2A, NR2B, and NR2C, drove qualitatively similar Ca 21 responses at Cx36-GCaMP gap junctions. Responses peaked in the first few seconds of glutamate administration and often declined in the continued presence of glutamate. The decline no doubt reflects both the desensitization kinetics of the NMDA receptors and the Ca 21 removal processes described above. Despite these processes, Ca 21 remained elevated above the baseline level throughout the duration of the glutamate application for glutamate concentrations of 100 mM or higher. This suggests that signaling driven by elevated Ca 21 may be activated continuously when glutamate is present, at least at higher concentrations. Kothmann et al. (2012) demonstrated that activation of non-synaptic NMDA receptors associated with Cx36 drove an increase in CaMKII activity, resulting in phosphorylation of Cx36 and increased coupling. In an analogous fashion, activation of synaptic NMDA receptors drives potentiation of the electrical synapses in the Mauthner cell mixed synapses through CaMKII activity (Pereda et al., 1998). Phosphorylation of the connexins involved has not been evaluated in that system, but the presence of Cx35 (the fish orthologue of mammalian Cx36) and the highly homologous Cx34.7 at these synapses (Pereda et al., 2003a;Rash et al., 2013) strongly suggests that connexin phosphorylation is the mechanism responsible for activity-dependent potentiation. Such activity-dependent potentiation could be mimicked in the HeLa cell expression system. Tracer coupling supported by Cx36-GCaMP was enhanced by glutamate application in HeLa cells expressing NR2A-containing NMDA receptors. In a surprising twist, tracer coupling was also enhanced by glutamate application in HeLa cells in which no NMDA receptor subunits had been transfected. The increase in coupling depended on CaMKII activity, as would be predicted by a Ca 21 -mediated response, suggesting the presence of an endogenous glutamate receptor that supports Ca 21 influx in HeLa cells. Furthermore, the CaMKII activity in HeLa cells was not constitutive. The intervention routinely performed in our group of inhibiting PKA activity with Rp-8-cpt-cAMPS, which appears to increase coupling by reducing PKA-stimulated protein phosphatase 2A activity (Kothmann et al., 2009;Yoshikawa et al., 2017), increased coupling in a manner independent of CaMKII activity. Thus, coupling could be increased by constitutive activity of an as yet unidentified protein kinase as well as through glutamateinducible CaMKII activity. The latter pathway provides a useful model system to investigate activity-dependent potentiation of coupling.
In Neuro2A cells, Del Corsso et al. (2012) found CaMKII activation to be responsible for a gradual increase in Cx36-mediated coupling as a result of breaking into a cell pair with patch pipettes. The stimulus for this "run-up" is unclear, although a transient increase in intracellular Ca 21 on break-in was proposed. Thus, in both expression systems (Del Corsso et al., 2012 and this study) and natural systems (Pereda et al., 1998;Kothmann et al., 2012;Turecek et al., 2014), CaMKII activity supports potentiation of coupling. CaMKII is a potent and central regulator of synaptic activity, affecting both AMPA receptor conductance and addition of AMPA receptors to post-synaptic densities (Coultrap and Bayer, 2012). Its role at electrical synapses seems to be comparable. CaMKII phosphorylates Cx36 directly and indeed binds to Cx36 in a manner similar to its binding to NR2B (Alev et al., 2008). In addition, CaMKII is proposed to support the maintenance or insertion of Cx36 channels, since its inhibition reduced the number of gap junctions on inferior olive neurons (Bazzigaluppi et al., 2017). Thus, CaMKII has attributes that make it likely to be a component of the complex associated with electrical synapses. Figure 8. Protein kinase signaling pathways responsible for potentiation of Cx36-GCaMP tracer coupling by 100 mM glutamate in HeLa cells. The diffusion coefficient (k) for Neurobiotin tracer diffusion is shown using the same experimental paradigm as in Fig. 6. Rp = PKA inhibitor (10 mM); KN-93 = CaMKII inhibitor (10 mM); Glu = 100 mM glutamate. All data are shown from three experiments; bars show mean values; error bars show 95% confidence limits of the mean; pp , 0.05, ppppp , 0.0001; ns = not significant. Symbols above bars represent comparison versus same transfected construct in control media; symbols above brackets represent comparison of the conditions underlying the ends of the brackets.
CaMKII is comprised of four isoforms, each with several splice variants, that have distinct cell type distributions, activity and protein binding characteristics, and effects on various aspects of learning and memory (Zalcman et al., 2018). The distributions of the four major isoforms in the context of Cx36 has been examined in the retina (Tetenborg et al., 2017). Interestingly, the major neuronal isoform CaMKIIa showed virtually no association with Cx36, while the neuronal isoform CaMKIIb and the widespread isoform CaMKIId were associated with Cx36 gap junctions in a somewhat cell-type selective manner (Tetenborg et al., 2017(Tetenborg et al., , 2019. In AII amacrine cells, in which the CaMKIIdriven potentiation of Cx36 coupling depends on Cx36 phosphorylation (Kothmann et al., 2012), CaMKIId was the major isoform present (Tetenborg et al., 2017). This isoform supports the highest rate of autophosphorylation of the four (Gaertner et al., 2004), which leads to autonomous activity, albeit considerably lower in the absence of Ca 21 -calmodulin binding (Coultrap and Bayer, 2012). In this manner, transient glutamate receptor signaling at Cx36 gap junctions can activate CaMKII to produce sustained phosphorylating activity that potentiates the electrical synapse. In contrast, Bazzigaluppi et al. (2017) have found that knockout of CaMKIIb , but not CaMKIIa, reduced the total number of Cx36 gap junctions in inferior olive neurons, suggesting that the CaMKIIb isoform is important for maintenance or insertion of Cx36 in gap junctions. Thus, the means through which CaMKII enhances coupling may differ according to the isoform involved.
Potentiation is not the only outcome of neuronal activity surrounding electrical synapses. In contrast to the NMDA receptor and CaMKII-dependent potentiation of coupling in inferior olive neurons resulting from NMDA application or selective paired-pulse presynaptic stimulation (Turecek et al., 2014), low frequency presynaptic stimulation resulted in lasting depression (Mathy et al., 2014). This depression depended on NMDA receptors, intracellular Ca 21 and CaMKII activity, although it is not clear how the electrical synaptic depression would be achieved. More in keeping with current understanding is the depression of thalamic reticular neuron electrical synapses by burst activity (Sevetson et al., 2017). In these neurons, Ca 21 entry through T-type voltage-gated Ca 21 channels activates calcineurin, with the predicted reduction in Cx36-mediated coupling. The bidirectional modulation of electrical synapse strength by different Ca 21 -dependent processes is highly analogous to the differences in synaptic activity that drive long-term potentiation and long-term depression in central glutamatergic synapses. The Cx36-GCaMP biosensor we have designed will be useful to investigate these differences.