Gap junctions constitute the only form of synaptic communication between neurons in the inferior olive (IO), which gives rise to the climbing fibers innervating the cerebellar cortex. Although its exact functional role remains undetermined, electrical coupling was shown to be necessary for the transient formation of functional compartments of IO neurons and to underlie the precise timing of climbing fibers required for cerebellar learning. So far, most functional considerations assume the existence of a network of permanently and homogeneously coupled IO neurons. Contrasting this notion, our results indicate that coupling within the IO is highly variable. By combining tracer-coupling analysis and paired electrophysiological recordings, we found that individual IO neurons could be coupled to a highly variable number of neighboring neurons. Furthermore, a given neuron could be coupled at remarkably different strengths with each of its partners. Freeze-fracture analysis of IO glomeruli revealed the close proximity of glutamatergic postsynaptic densities to connexin 36-containing gap junctions, at distances comparable to separations between chemical transmitting domains and gap junctions in goldfish mixed contacts, where electrical coupling was shown to be modulated by the activity of glutamatergic synapses. On the basis of structural and molecular similarities with goldfish mixed synapses, we speculate that, rather than being hardwired, variations in coupling could result from glomerulus-specific long-term modulation of gap junctions. This striking heterogeneity of coupling might act to finely influence the synchronization of IO neurons, adding an unexpected degree of complexity to olivary networks.
- electrical synapse
- connexin 36
- gap junction
- synaptic plasticity
gap junction-mediated electrical synapses allow synchronization of subthreshold and spiking activity among groups of neurons (Connors and Long 2004; Bennett and Zukin 2004). In the Inferior Olive (IO), a structure where electrical transmission constitutes the only form of synaptic communication between its principal cells (De Zeeuw et al. 1998), clusters of synchronized neurons are dynamically sculpted as a result of transient regulation of electrical coupling (Llinas et al. 1974; Leznik and Llinas 2005). This regulation takes place at the glomerulus, an anatomical structure where coupled dendritic processes of IO neurons and afferent chemical synaptic contacts coexist (De Zeeuw et al. 1998). Here, release of GABA from terminals of axons originating in the deep cerebellar nuclei has been proposed to indirectly regulate coupling by shunting depolarizing currents in dendritic processes of IO neurons (Llinas et al. 1974; De Zeeuw et al. 1998). The time window of this “uncoupling” is determined by the duration of the shunting synaptic conductances, which, although unusually long due to desynchronized transmitter release (Best and Regehr 2009), last only tens of milliseconds. By promoting synchronous complex spikes that influence sets of Purkinje cells in the cerebellar cortex, these quickly and transiently formed compartments are thought to encode important functional parameters that influence the cerebellar cortex (Leznik and Llinas 2005), suggesting that the formation of groups of functionally interconnected neurons underlies essential aspects of IO function. These functional considerations generally assume that inhibitory inputs act on a network of permanently coupled IO neurons in which compartments can be sculpted.
In contrast to the mechanisms underlying fast regulation of coupling, which target the extrajunctional membrane, little is known regarding the contribution of the gap junction channels themselves to the formation of compartments of functionally coupled IO neurons. The dendrodendritic gap junctions that couple IO neurons contain channels formed by the gap junction protein connexin 36 (Cx36) (Condorelli et al. 1998). This ubiquitous neuronal connexin mediates electrical coupling between neurons in various structures including retina, hippocampal inhibitory interneurons, and thalamic relay cells (Connors and Long 2004; Bennett and Zukin 2004), where electrical synapses are reportedly regulated by various mechanisms (Urschel et al. 2006; Zsiros and Maccaferri 2008; Landisman and Connors 2005; Kothman et al. 2009). Electrical synapses formed by Cx35, a fish ortholog that shares a high degree of homology with Cx36 (O'Brien et al. 1998), were shown to be regulated by the activity of their glutamatergic counterparts at mixed synaptic contacts on the goldfish Mauthner cell (Yang et al. 1990; Pereda et al. 1998; Smith and Pereda 2003). These activity-dependent changes are terminal-specific, and as a result of this property, electrical synapses between neighboring mixed terminals coexist at various degrees of conductance, suggesting that modulation by chemical synapses is short-range and therefore constitutes a spatially restricted mechanism (Smith and Pereda 2003).
Recent evidence suggests that molecular mechanisms for the regulation of coupling that are similar to those operating in goldfish mixed synapses could also operate in the mammalian IO (Alev et al. 2008). This possibility is supported by the particular arrangement of the IO glomeruli, where electrical and chemical synapses coexist (De Zeeuw et al. 1990, 1998). As a step toward determining the contribution of gap junction channels to the formation and properties of groups of interconnected neurons in the IO, we asked whether, as they occur between mixed synapses and the Mauthner cell, 1) electrical synapses between IO neurons coexist at various degrees of conductance, and 2) glutamatergic synapses are located in close proximity to gap junctions in IO glomeruli. For this purpose, we combined detailed quantitative tracer-coupling analysis with paired electrophysiological recordings and freeze-fracture replica immunolabeling (FRIL). Our results indicate that, rather than forming a homogenous network, coupling in the IO is highly variable, since individual IO neurons were found coupled to a dramatically different number of other cells. Strikingly, we found that individual IO neurons are coupled at remarkably different degrees of conductance with each of their partners. FRIL analysis of IO glomeruli revealed the proximity of glutamatergic postsynaptic densities (PSDs) to Cx36-containing junctions at distances comparable to separations between chemical transmitting domains and gap junctions in goldfish mixed synapses. We speculate that these variations in coupling might result from glomerulus-specific modulation of gap junctional conductance. Heterogeneity of coupling could constitute an important regulatory property of olivary networks, determining the ability of neurons to participate in dynamically sculpted clusters and to finely influence the synchronization of their cellular activity.
MATERIALS AND METHODS
Brain stem slices (300 μm thick) were prepared from 14- to 18-day-old Sprague-Dawley rats (Charles River). Animal handling and use followed a protocol approved by the Animal Care and Use Committee of Albert Einstein College of Medicine, in accordance with National Institutes of Health guidelines. The animals were anesthetized with inhaled isoflurane, and, following decapitation, the brain stem was quickly removed and sliced parasagitally (DTK-1000 DSK Microslicer) in an ice-cold sucrose cutting solution containing (in mM) 238 sucrose, 26 NaHCO3, 11 glucose, 2.5 KCl, 1 NaH2PO4, 1.3 MgSO4, and 2.5 CaCl2. Slices were then incubated for 1.5 h at room temperature in solution containing 50% of the sucrose cutting solution and 50% artificial cerebral spinal fluid (ACSF) solution containing (in mM) 119 NaCl, 26 NaHCO3, 10 glucose, 2.5 KCl, 1 NaH2PO4, 1.3 MgSO4, and 2.5 CaCl2. This incubation solution was slowly replaced with ACSF to finally reach a 20% sucrose-80% ACSF composition. Both cutting and incubation solutions were saturated with 95% O2-5% CO2 and maintained at pH 7.4 (proper balance was obtained using small amounts of 1 M HCl or 1 M NaOH). Sliced sections were kept at room temperature until transferred to a submerged recording chamber.
Brain stem slices were placed in a recording chamber mounted on an upright microscope (Nikon E600FN), maintained at constant temperature (32°C), and perfused with oxygenated (95% O2-5% CO2) ACSF. Differential interference contrast (DIC) microcopy was used to visually identify IO neurons. Patch pipettes, 6–9 MΩ, were filled with internal solution containing (in mM) 0.01 EGTA, 4 NaCl, 0.001 CaCl2, 10 HEPES, 140 K-gluconate, 4 Mg-ATP, and 0.6 Mg-GTP (pH 7.2, 280 mosM). Sucrose was added to correct osmolarity. For tracer-coupling experiments, 0.5% neurobiotin (NB; Vector Labs) was added to the internal solution. Recordings were obtained using a Multiclamp 700A amplifier (Axon Instruments) in current-clamp mode. IGOR Pro software (WaveMetrics) was used for data acquisition and analysis. For paired recordings, adjacent cells within 50 μm (approximately the length of the diameter of 2 cell bodies) were recorded simultaneously in current clamp. To reduce hyperpolarization-activated inward current (Ih) and thus provide more accurate estimates of the coupling coefficient, 2 mM CsCl was routinely added to the ACSF. In some experiments, the gap junction blocker meclofenamic acid (MFA; 100 μM; Sigma) was added to the ACSF.
Calculation of coupling coefficient.
The strength of electrotonic coupling was estimated using measurements of steady-state coupling coefficients (CC). For this purpose, a current pulse of −120 to −400 pA and 250-ms duration was injected into one cell of a coupled pair and resulting voltage deflections were measured in both cells. The coupling coefficient of cell 1 to cell 2 (CC1–2) was defined as V2/V1 where V1 is the voltage deflection in the current-injected cell and V2 is the corresponding, although attenuated, voltage deflection in the noninjected cell. Current pulses ranged from −120 to −400 pA. Averages of 60–100 traces were used to improve the signal-to-noise ratio. Only CC values >0.005 were considered evidence of coupling between pairs of recorded IO neurons.
Estimate of junctional conductance.
Estimates of the conductance of gap junctional channels coupling IO neurons (Gj) were calculated following Bennett (1966) and Parker et al. (2009): where Rin represents the input resistance of the coupled cells, and the transfer resistance was defined as the voltage response in cell 2 (when current is injected into cell 1) divided by the amplitude of the current step injected into cell 1. These estimates assume a simple two-neuron model with passive membrane properties coupled directly by a single junction and do not account for parallel conductance pathways via adjacent coupled neurons or dendritic cable properties. Because no evidence of electrical rectification was observed between pairs of IO neurons, estimates of coupling coefficient and junctional conductance are reported from voltage drops in only one direction.
Cell Labeling and Imaging
To test tracer coupling, 0.5% NB was iontophoretically injected using 500-pA pulses of 250-ms duration every second for 13–14 min. After injection, the electrode was removed and slices were incubated at room temperature for 1–2 h to allow diffusion. Slices were then fixed overnight at 4°C in 4% paraformaldehyde in 0.2 M sodium phosphate buffer (1 part sodium phosphate monobasic and 4 parts sodium phosphate dibasic; Sigma). After being washed with PBS [10× Dulbecco's PBS containing (in g/l) 1 CaCl2, 2 KCl, 2 KH2PO4, 1 MgCl2·6H20, 80 NaCl, and 11.5 Na2HPO4 was diluted to 1× and the pH adjusted to 7.4 (Cellgro Mediatech)], slices were washed in PBS containing 5% normal goat serum and 0.3% Triton and then incubated for 1 h with AlexaFluor 568-conjugated streptavidin (7.5 μg/ml, S11226; Invitrogen) in PBS containing 3% Triton. Finally, slices were washed in PBS and mounted on microscope slides for confocal analysis. Control experiments were performed by placing a NB-containing pipette under positive pressure in the extracellular space adjacent to IO neurons. The slices were then processed for NB labeling as described above.
After exposure to AlexaFluor-conjugated streptavidin, NB-labeled cells were visualized under confocal microscopy with an Olympus microscope controlled by Fluoview software. For each injected cell, 25–45 optical z-section images were obtained 1 μm apart to include the dendritic tree. The settings for laser power and photomultiplier tube (PMT) gain were initially set to detect all tracer-coupled cells during the confocal scans. For accurate measurements of intensity at the injected cell (which usually appears saturated), a second scan at lower laser power and PMT gain settings was used. Because of the high concentration of NB in the injected cell, measurements of this cell's labeling intensity with lower laser power were comparable to saturated values obtained with high laser power. To maintain a uniform level of excitation and detection, both the high and low laser power settings were maintained constant for all experiments. Confocal images were analyzed off-line using basic features of ImageJ software (NIH). The intensity of labeling for each cell was estimated using the “measurement” feature of ImageJ, which provided a mean intensity value for a region of interest representing the outline of the somata. To express differences in tracer-coupling strength, an “intensity ratio” was calculated by dividing the labeling intensity of the tracer-coupled cells by the intensity of the directly labeled cell, estimated at low gain. A cell was considered coupled if the complete outline of its cell body was clearly recognizable from background. Thresholding was required to distinguish coupled somata. Digital isolation of IO neurons injected with NB was obtained from z-stack projection images using the “magic tool” feature of Photoshop software (Adobe). Results were expressed as average values ± standard deviation (SD) or standard error of the mean (SE). Significance of quantitative data was determined by using Fisher's exact test, the χ2 test, and Student's t-test.
Freeze-Fracture Replica Immunolabeling
Two adult male Sprague-Dawley rats (158 and 496 g) were anesthetized using ketamine (80 mg/kg) and xylazine (8 mg/kg), fixed by perfusion with 1% formaldehyde in Sorenson's phosphate buffer (pH 7.4), and decapitated. The brain stem was removed and transversely sliced at 150 μm using a refrigerated Lancer Vibratome 3000 (Technical Products International, St. Louis, MO) that maintained samples at 4°C. One goldfish was prepared for FRIL, as described in Pereda et al. (2003a), and portions of that sample were previously described. Samples of rat IO were frozen, and freeze-fracture replicas were made according to our published methods (Rash and Yasumura 1999). (However, in one replica made in 1999, during modification and testing of our freeze-fracture machine, Pt was applied continuously rather than in short bursts, resulting in slightly larger grain size and lower specimen contrast in one sample, as described below.) A gold “index” grid containing a drop of 1.5% Lexan dissolved in dichloroethane was placed on the frozen sample, and the solvent was allowed to evaporate overnight at −25°C. The replicated tissue/Lexan/gold grid “sandwiches” were thawed and photomapped by confocal microscopy using tissue autofluorescence. Bulk tissue was removed from the Lexan-stabilized replicas by washing in 2.5% SDS detergent in Tris·HCl buffer (pH 8.9) for 22 or 27 h at 40 or 47°C. This washing procedure leaves a thin film of lipid and protein molecules adhering to the highly absorptive platinum-carbon replica (Fujimoto 1995), plus additional molecules that are bound to the molecules that are directly adsorbed to the replica (Fujimoto 1995; Rash and Yasumura 1999). SDS-washed replicas were rinsed in “blocking buffer” [10% heat-inactivated goat serum and 1.5% fish gelatin in Sorenson's phosphate buffer, pH 7.4 (Dinchuk et al. 1987)] and labeled for 60–220 min using rabbit polyclonal antibodies to Cx36 [Ab298 (Rash et al. 2000; Pereda et al. 2003a); from Dr. James Nagy, University of Manitoba, Winnipeg, Canada] with or without monoclonal antibodies to glutamate receptor subunit NR1 (Ab55-6308; BD Biosciences PharMingen, San Diego, CA). Replicas of goldfish Mauthner cells were labeled using polyclonal antibody Ab298 and monoclonal antibodies to NR1 (both as described above). Labeled samples were rinsed and counterlabeled using goat anti-rabbit IgG and goat anti-mouse IgG, each coupled to a separate size of uniform-diameter gold beads [10 or 20 nm (Chemicon, Temecula, CA; secondary immunogold antibodies no longer available from Chemicon); 6, 12, and 18 nm (Jackson ImmunoResearch Laboratories); and 10 or 30 nm (BBInternational, Cardiff, UK; obtained from Ted Pella, Redding, CA)]. After immunogold labeling, each sample was air-dried and coated on the labeled side with 10–20 nm of evaporated carbon to anneal temperature-induced stress cracks in the replica, as well as to stabilize the immunogold beads. The Lexan support film was removed by immersing the grids in ethylene dichloride solvent for 1–2 h and was then air-dried.
FRIL samples were examined in a JEOL 2000 EX-II transmission electron microscope (TEM), and stereoscopic images (8° included angle; tilt range ±60°) were obtained at TEM magnifications from ×10,000 to ×100,000. Negatives were digitized using an ArtixScan 2500f digital scanner (Microtek, Carson, CA) and processed using Photoshop 7.01 (Adobe). Stereoscopic images were used to assess complex three-dimensional membrane topography, as well as to confirm that each label was on the tissue side of the replica. [Relatively rare nonspecific binding is primarily on the Lexan-coated side of the replica (Rash and Yasumura 1999)]. Neurons and glia were identified in freeze-fracture replicas on the basis of 24 inclusive and exclusive criteria described by Rash et al. (1997). For neurons, these included the presence of >25 uniform-diameter synaptic vesicles in nerve terminal cytoplasm, pre- and postsynaptic membrane specializations called active zones and PSDs, and absence of glial cell markers (glial fibrillary acidic protein filaments, aquaporin-4 square arrays, etc.). Gap junctions were identified using well-established criteria (reviewed in Rash et al. 1998a), including 1) uniform 8- to 9-nm-diameter intramembrane particles (IMPs) in replicated protoplasmic leaflets [“P faces”; the internationally recognized freeze-fracture terminology of Branton et al. (1975) is used in this report] and uniform 8- to 9-nm-diameter pits in the extraplasmic leaflet (E face); 2) close packing of IMPs/pits, often in hexagonal array; 3) narrowing of the extracellular space within the border of the gap junction; and 4) maintained alignment of IMPs and pits across the step from E to P face. In addition, immunogold labeling for Cx36, restricted to gap junctions only in neurons, was considered confirmatory evidence.
Confocal Analysis Reveals a High Incidence and Variability of Tracer Coupling
We investigated the variability of coupling between IO neurons by combining tracer coupling with fluorescence confocal microscopy analysis. Unlike whole cell recordings, which restrict the analysis to pairs of neurons, tracer-coupling analysis permits simultaneous evaluation of the degree of coupling between a given IO neuron and all its coupled partners. To this end, brain stem slices containing the inferior olivary nucleus (Fig. 1A) were prepared and IO neurons were identified with DIC optics (Fig. 1B, inset) for whole cell patch-clamp recordings with electrodes containing NB, a tracer molecule permeant to gap junctions. Most of the recorded cells (∼85%) were located in the central region of the principal olive, whereas fewer were located in the dorsal olive. IO neurons were identified by their electrophysiological properties, consisting of a characteristic Ih sag in response to hyperpolarizing current injection that triggers an action potential, immediately followed by a prolonged afterdepolarization, which is terminated by a pronounced afterhyperpolarization (Llinas and Yarom 1981; Bal and McCormick 1997) (Fig. 1B). After NB injection, the recording pipette was removed and slices were incubated to allow diffusion of NB. Slices were then processed with AlexaFluor 568-conjugated streptavidin to label NB-containing cells for their anatomical identification and analysis with confocal microscopy (Fig. 1, C, D, and F). In all cases, the injected cells were unambiguously identified as IO neurons because of their size and characteristic dendritic morphology. Most of these cells (∼90%) had anatomical features characteristic of “curly” IO neurons, whereas only 3 of the 32 injected cells showed “straight” dendritic morphology. In contrast to a previous report (Devor and Yarom 2002), our approach revealed a high incidence of coupling in the IO, where at least one tracer-coupled cell was detected for 27 of the 32 injected neurons (incidence 84%) (Fig. 1E). Notably, the number of tracer-coupled cells was highly variable, ranging from 1 to 38 cells, averaging 12.5 ± 1.8 SE (Fig. 1G). Such variability in the number of coupled cells was also observed within the same animal (Fig. 1H) and even within the same slice (not shown), where clusters of coupled cells dramatically differed in size, suggesting that the observed variability cannot be ascribable to subtle differences in experimental conditions between experiments. Although we restricted our quantitative analysis to curly neurons to avoid potential differences between cell types, tracer coupling was also observed in “straight” neurons (see below). Unfortunately, because dendritic morphology was not evident in tracer coupled cells, we were unable to determine whether coupling specifically occurs only between curly or between straight neurons or whether it could also occur heterologously between these two types of neurons. Thus our data showed that although coupling between IO neurons seems prevalent (∼85%), the number of cells in these clusters of coupled IO neurons was highly variable.
Labeling of IO Cells is Mediated by Gap Junctions
We carried out control experiments to unambiguously establish that the observed cell labeling corresponded to coupling via gap junction channels. First, to rule out the possibility that labeling of noninjected cells resulted from extracellular accumulation of NB as a result of leakage from the recording electrode during the formation of the patch, we compared intracellular and extracellular applications of NB. Both dark and faint somata labeling were only detected when NB was injected intracellularly in IO neurons (Fig. 2A). In contrast, puffing of the NB-containing internal solution into the extracellular space for about 4 min resulted in labeling of blood vessels only (Fig. 2B), which exhibited affinity for this tracer. Second, experiments using the gap junction blocker MFA (Pan et al. 2007; Veruki and Hartveit 2009) indicated that cell labeling was indeed gap junction mediated. Brain stem slices containing the IO were incubated for at least 40 min with MFA (100 μM) before NB injection into IO neurons. This pretreatment greatly reduced the incidence of tracer coupling to 36% compared with 89% in untreated slices harvested from the same animals (n = 11 and n = 9 injected cells, respectively; P = 0.024, Fisher's exact test) (Fig. 2C). MFA pretreatment also decreased the number of coupled cells. Only 5 coupled cells were identified from 11 pretreated injected neurons (averaging 0.82 ± 0.46 SE) compared with 92 from 9 untreated injected cells (averaging 10.2 ± 3.3 SE; P < 0.01, Student's t-test) (Fig. 2D). Incubation with MFA (40 min, 100 μM) also reduced the likelihood of finding pairs of electrically coupled cells during paired whole cell recordings (10%, compared with 81% of adjacent cell pairs in untreated slices, P = 0.000021, Fisher's exact test; also significant with the χ2 test, P < 0.001) (Fig. 2, E and F), independently confirming that MFA effects were mediated by acting on gap junction channels. Together, these results suggest that NB labeling of noninjected cells does not result from surface cell labeling or its uptake from the extracellular space but from diffusion through NB-permeable gap junctions.
Coupling Between a Given IO Neuron and Each of Its Partners is Highly Variable
Strikingly, we found that the intensity of labeling of the tracer-coupled cells was highly variable. The observation and quantification of this variability was facilitated by our experimental approach, which relied on confocal microscopic examination and measurement of maximum fluorescence at each coupled cell. An example of this analysis is illustrated in Fig. 3A, where a confocal reconstruction of a NB-injected IO neuron and tracer-coupled somata is shown. The injected IO neuron (curly type) was digitally isolated (Fig. 3B), and the fluorescence of each of the 13 coupled cells, represented by the outline of their somas (Fig. 3C), was measured at the confocal section that showed maximum intensity (expressed as the various shades of gray in Fig. 3). To express the variability of coupling, we created a ratio between the labeling intensity for each cell and the intensity of the directly injected neuron. This intensity ratio (see materials and methods), conceptually comparable to coupling coefficients in paired recordings (although coupling coefficients are measured at steady-state conditions), allowed us to quantify differences between tracer-coupled cells. The distribution of the intensity ratios for this particular IO neuron is illustrated in the histogram of Fig. 3D, showing that the degree of coupling of this neuron with its partners was highly variable. This variability of coupling was not specific to curly neurons, since similar variability was observed in straight neurons (Fig. 4).
Tracer-Coupling Labeling Intensity is Not Correlated With Distance
Variability of labeling intensity in tracer-coupled cells was observed in other structures, such as the retina (Mills and Massey 1998), and in cell cultures (Cusato et al. 2003), where secondary tracer coupling (transfer of tracer from a primarily coupled neuron to a secondary neuron and so forth) was documented. In these cases, labeling intensity was found to be a function of distance from the originally tracer-containing coupled cells (Mills and Massey 1998). To investigate the possibility that the observed variability of labeling results from secondary coupling between IO neurons, we correlated the labeling intensity of coupled neurons with the distance between their somata and the NB-injected neurons. We found that the labeling intensity of the coupled neurons was not correlated with the distance from the NB-injected cell. That is, both darkly and faintly labeled somata were located at widely different distances from the injected IO neurons (see Figs. 3, 4, 5, and 6A). This finding is illustrated in Fig. 5, in which star plots that graphically code for intensity (shade of gray) and distance (length) were constructed for five different injected IO neurons (A and D–G). As can be appreciated in the example illustrated in Fig. 5A, in which an IO neuron was coupled to 35 neighbors, the labeling intensities of the coupled somata were not correlated with their distance from the injected cell, despite the wide variability of labeling intensity (Fig. 5B; see also Fig. 5, D–G, for the other examples). As expected, the intensity ratio for the total population of IO neurons was also highly variable (Fig. 5C) and was not correlated with the distance between somata of the coupled cells and the NB-injected cell (Fig. 5H), suggesting that the variability of labeling intensity observed in coupled cells is unlikely to result from secondary tracer coupling. Finally, tracer coupling almost exclusively occurred within the dendritic tree of the injected neuron (where contacts are made), suggesting that all of the labeled cells were directly coupled (primary coupling). Only 4 of the 338 cells in which the tracer was detected were located outside the dendritic tree of the injected neuron. These four cells were located very close to labeled dendrites (less than ∼50 μm) and thus perhaps within the range of smaller, undetected, dendritic processes. Furthermore, this distribution was independent of the cell type and the extent of the dendritic tree, which is almost twice as large in straight neurons (compare Figs. 3A and 4).
Paired Recordings of IO Neurons Show High Variability of Electrical Coupling
Analysis of tracer coupling suggested that coupling between IO neurons was highly variable. To independently confirm this variability, we performed paired whole cell recordings from somas of adjacent neurons that were within a <50-μm-diameter area, within which we observed a high variability in tracer-coupling labeling intensity (Fig. 6, A and B). We also observed a wide variability of electrical coupling between adjacent IO neurons where pairs with low (Fig. 6C) and high (Fig. 6D) coupling coefficients coexist. This striking diversity of electrical coupling illustrated by the distribution of estimates of junctional conductance (Fig. 6E) was obtained for each pair of IO neurons (n = 91 cell pairs). Thus both tracer and electrical coupling were highly variable between neighboring IO neurons.
Although measurements of electrical coupling are consistent with the variability observed in tracer-coupling analysis, they were obtained by measuring pairs of IO neurons, an approach that does not reveal differential coupling (i.e., differences in coupling between a given neuron and its partners) because it only examines two neurons at a time. To more conclusively establish the variability of electrical coupling between IO neurons, we performed sequential paired recordings between a given IO neuron and some of its partners (Fig. 7). For this purpose, after measuring electrical coupling between a pair of adjacent IO neurons, one of the pipettes was removed and recordings from different adjacent neurons, located within a 50-μm radius, were sequentially obtained while the recording pipette was held in the other cell. This approach allowed us 1) to directly establish the variability of electrical coupling between a given IO neuron and its coupled partners and 2) to correlate the strength of electrical coupling with the distance between their somata (Fig. 7A). In all cases, recordings were stable and no significant changes in the input resistance of the holding cell were detected during the sequential recordings (input resistance measurements stayed within 15% of their original values). As observed in Fig. 7, A and B, the strength of coupling could be markedly different between immediate neighbors and was not correlated with the distance between the recorded cells. Similar results were observed in three other examples (Fig. 7, C–E). Finally, as with intensity ratios in tracer-coupling experiments, the strength of electrical coupling (expressed as junctional conductance) was not correlated with the distance between somata in these paired recordings (Fig. 7F). Thus, although a given IO neuron can be electrically coupled to multiple neurons, the strength of coupling was different in each case.
Proximity of N-Methyl-d-Aspartate Receptor-Containing Postsynaptic Densities to Gap Junctions in IO Glomeruli
The observed coupling variability could be due to differences in the size of gap junctions coupling the cell processes of a given IO neuron with each of its partners. Alternatively, it could result from differential regulation of gap junction channels at IO glomeruli. Spatially restricted N-methyl-d-aspartate receptor (NMDAR)-dependent regulation of gap junctions that requires downstream activation of Ca2+-calmodulin-dependent kinase II (CaM-KII) was shown to occur at goldfish mixed synapses (Yang et al. 1990; Pereda and Faber 1996), where NMDAR-containing PSDs are located in close proximity to Cx35-containing gap junctions (Smith and Pereda 2003) (Fig. 8A). Because modulation of electrical transmission at these mixed synapses occurs postsynaptically (Pereda et al. 1998), the interactions may not be only limited to mixed synapses but are likely found where chemical synapses are close to gap junctions formed by different presynaptic elements. Recent reports indicate that CaM-KII colocalizes with Cx36 in the IO (Alev et al. 2008) and with Cx35 in goldfish mixed synapses (Flores et al. 2010), suggesting that similar mechanisms could operate at these gap junctions. Given the convergence of chemical and electrical synapses in the IO glomerulus, we examined whether glutamatergic PSDs are located close to gap junctions using FRIL double-immunolabeling. We observed E-face particle aggregates close to dendrodendritic gap junctions containing Cx36 in IO glomeruli at distances comparable to separations between chemical transmitting domains and gap junctions in goldfish mixed synapses (∼0.2–0.3 μm; Pereda et al. 2003a,b) (Fig. 8B). Figure 8C illustrates the proximity of one of these E-face particle aggregates that is associated with a presynaptic process (probably a glutamatergic terminal) in close proximity to a gap junction formed by two other processes, which likely correspond to spines of IO neurons. These E-face particle aggregates are known to correspond to glutamatergic postsynaptic densities (Harris and Landis 1986), as confirmed by immunogold antibodies directed against the NR1 subunit of the NMDA receptor (Fig. 8, B and D). These PSDs were labeled by only a small number of immunogold beads, which is expected because NMDA receptors constitute only a small fraction of the glutamate receptors present at PSDs (Takumi et al. 1999). Thus ultrastructural analysis indicates that gap junctions and glutamatergic synapses containing NMDARs, a key element for regulation of coupling at goldfish mixed synapses (Yang et al. 1990; Pereda and Faber 1996), coexist in IO glomeruli at distances where functional interactions can take place.
IO Neurons are Heterogeneously Coupled
We report in this study that coupling between IO neurons is highly heterogeneous. The analysis of variations in coupling was facilitated by a sensitive technique that takes advantage of confocal microscopy for accurate detection of tracer coupling. By virtue of its three-dimensional nature, this confocally based approach enhanced detection and revealed a higher incidence of tracer coupling compared with a previous analysis that relied on DAB staining and conventional microscopy and was therefore restricted to two-dimensional analysis (Devor and Yarom 2002). Such enhanced detection reduced the ambiguity in the identification of coupled cells and optimized the measurement of labeling intensity. This approach allowed us to reveal an unexpected high variability of coupling between IO neurons, which ranged from apparently “noncoupled” to coupling with up to 38 other IO neurons. Furthermore, by combining quantitative tracer-coupling analysis and electrophysiological approaches, we have demonstrated that the degree of coupling between a given IO neuron and each of its partners is surprisingly heterogeneous. The observed variability of coupling cannot be ascribed to secondary coupling (from primary to secondary labeled neurons and so forth) that was detected in other structures (Mills and Massey 1998) because the intensity of tracer-coupling labeling was not correlated with distance from the injected cell, a finding that was independently confirmed during sequential recordings from pairs of IO neurons. Although when compared with other brain regions the anatomical arrangement of the IO is less favorable for detecting secondary coupling, if secondary labeling were a significant component of the observed diversity of coupling, we would have expected that the labeling intensity of coupled somata would have shown at least weak correlation with the distance from the NB-injected cell. Rather, heterogeneity of both tracer and electrical coupling almost exclusively took place within the dendritic tree of the injected neuron (where contacts are made), suggesting that all of the labeled cells were directly coupled (primary coupling). Although measurements of electrical coupling are likely to be more sensitive, since tracer coupling has been reportedly undetected in some networks of electrically coupled neurons (Gibson et al. 1999), tracer and electrical coupling observations were similar in our study, and both revealed a high variability in coupling strength. Nevertheless, although more sensitive than previous approaches, our measurements of tracer coupling are likely to represent an underestimate of the total number of coupled cells (estimated to be ∼50, based on estimates of neuronal density; Devor and Yarom 2002), because the tracer could exist at levels below our threshold of detection in some neurons. Secondary coupling, which must exist, is also likely to be below our level of detection. Heterogeneity of coupling between an IO neuron and its partners was observed in virtually all of the explored cells, both the curly and straight types, and therefore is likely to constitute a widespread property of neurons across the principal olive. Thus, although the properties of electrical coupling in the IO were explored in previous studies (Devor and Yarom 2002; Leznik and Llinas 2005), which reported the existence of some variability in tracer and electrical coupling, our study differs from those previous reports in that it specifically focused on unambiguously establishing the existence and properties of the variability of coupling between IO neurons. Since loss of electrical coupling in Cx36-deficient animals has a profound effect in the synchronization of inferior olivary cells (Long et al. 2002, Van Der Giessen et al. 2008), the variabilities in strengths of these electrical connections are likely to be functionally relevant.
On the Origin of Coupling Heterogeneity
A simple interpretation regarding the origin of coupling diversity could be that it results from differences in the number of gap junction channels linking dendritic processes within IO glomeruli. The notion that coupling heterogeneity might be hardwired constitutes a distinct possibility. An alternative interpretation, more consistent with current views regarding mechanisms of brain function (Marder and Goaillard 2006), is that this variability reflects dynamic aspects of olivary networks. The possibility that the observed heterogeneity of coupling between a given IO and its partners results from a spatially restricted regulation of gap junctions is supported by the example of goldfish mixed synapses. At those terminals, gap junctions are regulated by the activity of the glutamatergic synapses localized within the same contact, creating a wide diversity of coupling between individual terminals and the lateral dendrite of the Mauthner cell (Smith and Pereda 2003). The regulatory mechanism requires NMDAR activation and downstream activation of CaM-KII (Yang et al. 1990; Pereda and Faber 1996; Pereda et al 1998; reviewed in Pereda et al. 2004). Because induction of modulation of electrical coupling at goldfish mixed synapses occurs postsynaptically (Pereda et al. 1998), short-range functional interactions may not be limited only to mixed synapses but may be found where glutamatergic synapses are close to gap junctions formed by other presynaptic elements. Consistent with this possibility, our FRIL analysis revealed the proximity of NR1-containing PSDs to Cx36-containing gap junctions at distances comparable to separations between the chemical transmitting domains and gap junctions in goldfish mixed synapses, suggesting that similar interactions between electrical and glutamatergic synapses might occur in the IO. Several lines of evidence also support this possibility: 1) Cx36 and Cx35 are highly homologous and share regulatory sequences for CaM-KII (Flores et al. 2010); 2) CaM-KII has been shown to colocalize with both Cx36 in the IO (Alev et al. 2008) and Cx35 in goldfish mixed synapses (Flores et. al 2010); 3) CaM-KII associates with both Cx36 (Alev et al. 2008) and Cx35 (Flores et al. 2010); and 4) the association of CaM-KII with Cx35 is believed to be indicative of the degree of potentiation of electrical transmission at goldfish mixed synapses (Flores et al. 2010). Interestingly, a recent report suggests that the drug modafinil, an antinarcoleptic and mood-enhancing drug, might enhance electrical coupling between IO neurons through a mechanism that requires the activation of CaM-KII (Urbano et al. 2007). Glutamatergic transmission was also shown to promote activity-dependent long-term depression of electrical coupling between the inhibitory neurons of the rat thalamic reticular nucleus via activation of metabotropic glutamate receptors in this case (Landisman and Connors 2005), providing a potential mechanism for bidirectional control of electrical coupling. Thus, although GABAergic terminals originating in the deep cerebellar nuclei promote transient decoupling by acting on the nonjunctional membrane at IO glomeruli (Llinas et al. 1974), our work raises the possibility that excitatory terminals of mesodiencephalic origin (De Zeeuw et al. 1990, 1998) could act to promote more lasting modifications of electrical coupling via activation of glutamate receptors (Fig. 8). Future studies are necessary to obtain functional evidence for the existence of such mechanisms. Supporting this possibility, a recent report suggests that Cx36 phosphorylation can vary locally (Kothmann et al. 2009).
Although its exact function remains to be elucidated, electrical coupling is believed to play an essential role in olivary function (Llinas et al. 1974; Welsh et al. 1995; De Zeeuw et al. 1998; Placantonakis et al. 2004; Leznik and Llinas 2005; Kistler and DeZeeuw 2005; Blenkinsop and Lang 2006; Placantonakis et al. 2006; Van Der Giessen et al. 2008). For example, clusters of synchronized IO neurons were reported to be dynamically sculpted as a result of transient regulation of electrical coupling by GABAergic inputs originated in the deep cerebellar nuclei (Llinas et al. 1974; Leznik and Llinas 2005). These transiently formed compartments are thought to encode important functional parameters by promoting synchronous “complex spikes” (characteristic burstlike action potentials in climbing fibers) that instruct sets of Purkinje cells in the cerebellar cortex (Leznik and Llinas 2005; Blenkisop and Lang 2006). Our results indicate that the formation of these transient compartments takes place on a network of heterogeneously coupled IO neurons, where some might not always be coupled. Variations in the degree of coupling might not only finely determine the degree of synchronization between IO neurons within a compartment (see below) but might also determine the ability of individual IO neurons to participate in these coupled compartments.
Electrical coupling is essential for the synchronization of IO neurons that leads to proper timing of their action potentials (Long et al. 2002) and thereby for learning-dependent timing in cerebellar motor control (Van Der Giessen et al. 2008). Analysis of Cx36-deficient mice revealed impaired timing of both locomotion and eye-blink response, together with an increased variability in the latencies of spike activities in response to the unconditioned stimulus (Van Der Giessen et al. 2008). These differences in spike timing were found to result from altered interactions with subthreshold oscillations, likely as a result of the decrease of synchronization between IO neurons in Cx36-deficient animals (Van Der Giessen et al. 2008). Subthreshold oscillations (Llinas and Yarom 1986; Lampl and Yarom 1997) are considered an essential timing mechanism that regulates the output of the downstream Purkinje cells (Welsh et al. 1995; Yarom and Cohen 2002). The heterogeneity of coupling between IO neurons that we report in the present study suggests that climbing fibers might convey more complex information to the cerebellar cortex, because differences in coupling might finely influence synchronization among climbing fibers and in this manner regulate learning-dependent timing in the olivocerebellar network. Supporting this possibility, a recent report challenged the view that the climbing fibers convey an all-or-none signal to the cerebellar cortex (Mathy et al. 2009). That study elegantly demonstrated that the number of spikes in the burst of a complex spike in a climbing fiber depends on the phase of subthreshold oscillations, thereby encoding the state of the olivary network. The bursts of the climbing fibers are transmitted to the cerebellar cortex, where they have a significant impact on Purkinje cells, since they were shown to promote short-term and long-term plasticity at parallel fiber synapses in a manner dependent on the number of spikes contained in the burst (Mathy et al. 2009). Heterogeneity of coupling between IO neurons might finely influence synchronization and properties of climbing fibers and, as a result, their probability of eliciting plastic changes in the cerebellar cortex. If dynamic, heterogeneity of electrical coupling between IO neurons could represent a novel learning mechanism within the olivocerebellar system, where multiple learning mechanisms have been reported (Llinas et al. 1975; Welsh and Harvey 1998; Hansel and Linden 2000; Pugh and Raman 2006; Van Der Giessen et al. 2008; Mathy et al. 2009; for review see Hansel et al. 2001).
The essence of brain function arguably resides on the diversity of its mechanisms for intercellular communication and, perhaps more importantly, on their variability. The remarkable heterogeneity of coupling between IO neurons that we report in this study reveals an unexpected degree of complexity of olivary networks. Interestingly, a recent report has suggested that the existence of coupling heterogeneity is a critical characteristic of the Golgi cell network in the cerebellum, where electrical synapses can underlie desynchronization (Vervaeke et al. 2010; see also Connors et al. 2010). Thus heterogeneity of coupling may represent a general property of networks of extensively coupled neurons and underlie important functions in various mammalian brain structures.
This research was supported by National Institutes of Health Grants DC03186 and NS0552827 to A. E. Pereda; S10RR05831 (JEOL 2000 EXII), S10RR08329 (JEOL 9010c), NS044395, and NS044010 to J. E. Rash; and MH081935 to P. E. Castillo.
No conflicts of interest, financial or otherwise, are declared by the author(s).
We thank Ramani Balu for early contributions to tracer-coupling studies and Pamela Cabahug for assistance with the confocal microscope. We also thank Sebastian Curti for useful discussions and comments on the manuscript.
- Copyright © 2011 the American Physiological Society