Abstract
The olivo-cerebellar system is crucial for smooth and well timed execution of movements based on sensory and proprioceptive cues. The inferior olive (IO) plays a pivotal role in this process by synchronizing its activity across neurons internally through connexin36 gap junctions and providing a timing and/or learning signal to the cerebellum. Even though synchrony achieved through electrical coupling in IO cells is generally thought to be important in timing motor output, a direct relation between timing of movement and synchrony of olivary discharges has never been demonstrated within functional microcomplexes using transgenics. Here we combined in vivo, two-photon calcium imaging of complex spikes in microcomplexes of Purkinje cell (PC) dendrites with high-speed filming of tail, trunk, and limb movements in awake wild-type and connexin36-deficient mice. In wild types at rest, functional clusters of PCs were poorly defined with synchrony correlations that were relatively small and spatially limited to mediolateral distances of ∼50 μm, whereas during locomotion synchrony of the same PCs increased in strength and extended over distances spanning multiple microzones that could be correlated to specific components of sharp and well bounded movements. Instead, connexin36-deficient mice exhibited prolonged and desynchronized complex spike activity within PC microcomplexes both at rest and during behavior. Importantly, the mutants also showed concomitant abnormalities in the execution of spinocerebellar reflexes, which were significantly slower and more gradual than in wild-type littermates, particularly following sensory perturbations. Our results highlight the importance of modulation of synchronous activity within and between cerebellar microcomplexes in on-line temporal processing of motor output.
Introduction
The cerebellum's structured neural circuitry assists processing of proprioceptive and sensory cues and execution of phasic reflexes (Bloedel and Bracha, 1995). Its two main input channels, the mossy fiber and climbing fiber (CF) pathways, converge in the Purkinje cell (PC) layer, where mossy fibers modulate PC intrinsic simple spike firing indirectly and CFs directly elicit complex spikes (CSs) in PCs (Eccles et al., 1964). PCs receive input from a single CF and synapse onto cerebellar nuclei (CN) neurons that project back to the source of that CF (Andersson and Hesslow, 1987; De Zeeuw et al., 1997). Consequently, the cerebellum is said to consist of microcomplexes or microzones (Ito, 1997; Bengtsson and Hesslow, 2006): functional ensembles spanning mediolateral distances of several tens to a few hundred micrometers (Oscarsson, 1979; Ozden et al., 2009), falling well within the bounds of cerebellar zones of ∼1 mm (Oscarsson, 1979; Apps and Garwicz, 2005).
Within the inferior olive (IO) cerebellar microcomplexes are thought to be dynamically grouped (Llinás, 2009; Lefler et al., 2013). Its cells are extensively and exclusively coupled through connexin36 gap junctions (Belluardo et al., 2000; Condorelli et al., 2000; Van Der Giessen et al., 2006) located in glomeruli (Sotelo et al., 1974; De Zeeuw et al., 1995, 2003). IO glomeruli receive input from chemical synapses that may regulate the extent to which IO cells are coupled at that locus (de Zeeuw et al., 1990), since olivary gap junctions mediate synchrony (Long et al., 2002). Synchronous IO discharges are likely important (Welsh et al., 1995; Ivry, 1996; Lang, 2001; Ozden et al., 2012; Schweighofer et al., 2013) and the IO may dynamically group microcomplexes to shape-specific spatiotemporal CS patterns (Lang, 2001). Multi-electrode approaches have shown that synchronous CS discharges over distances of 250–2000 μm can be related to motor control (Welsh et al., 1995; Lang et al., 1999). However, it is unclear to what extent cerebellar microzone dynamics (i.e., within distances of 200 μm) are correlated to motor behavior. This question is relevant as synchrony and dynamic regulation are particularly important within clusters of gap-junction coupled olivary neurons feeding into smaller microcomplexes (De Zeeuw et al., 2011).
We combined in vivo two-photon calcium imaging with high-speed filming of wild-type (WT) and connexin36-deficient mice on a transparent disk treadmill to correlate spatiotemporal CS patterns with tail, trunk, and limb movements. Calcium imaging has already been used to investigate PC CS activity through the large dendritic calcium transients elicited by CF firing (Miyakawa et al., 1992; Sullivan et al., 2005; Ozden et al., 2009). In our experiments we monitored up to ∼50 PCs simultaneously in medial zone lobule V of cerebellar cortex, where receptive fields for tail, trunk, and proximal limbs are found (Provini et al., 1968; Eccles et al., 1968b; Jörntell et al., 2000; Ghosh et al., 2011; Ozden et al., 2012; Witter et al., 2013). This allowed us to investigate differences in CS firing between Cx36−/− mice and WT littermates within microzones with high spatial acuity and relate this to observed movements.
Materials and Methods
All animal and experimental procedures were approved by the Animal Experimentation and Ethics Committee of the Royal Netherlands Academy of Arts and Sciences. A total of 21 mice between 2 and 6 months of age and of either sex were used for the current study, divided into experimental groups of Cx36−/− mice (11 animals: 5 females and 6 males) and WT littermates (10 animals: 3 females and 7 males) bred on a C57BL/6 background.
Headplate surgery.
A stainless steel plate with a circular cutout was mounted on the mouse skull using dental cement (Super Bond C&B; Sun Medical). To attach the plate, mice were anesthetized with isoflurane (2%, IsoFlo; Abbott) and placed in a stereotactic frame, on a temperature-controlled heating blanket with a thermistor to maintain body temperature at 37°C. Eye ointment was applied to the eyes to keep them moist and lidocaine HCl was applied topically to the site of incision. A skin incision was made from bregma extending beyond lambda after which skin was parted and neck muscles were partially removed at the border of the interparietal and occipital bone. The plate was centered at the midline 2 mm caudal of lambda. The placement of the circular cutout was located above zone A of lobule V of the cerebellum. After surgery, mice were allowed to recover for at least 2 d, during which time their wellbeing was monitored.
Craniotomy and dye loading.
After headplate surgery and recovery, a dental drill was used to thin the skull at the headplate's circular cutout and fine tweezers (tip inserted at lambda) were used to lift the thinned portion of the skull exposing the cerebellar cortex. Tissue was then bolus loaded with OGB-1/AM (Invitrogen; O6807) as described previously (Stosiek et al., 2003; Sullivan et al., 2005; Ozden et al., 2009). OGB-1/AM was dissolved in pluronic in DMSO (4 μl) and diluted in 76 μl of aCSF (150 mm NaCl, 2.5 mm KCl, and 10 HEPES 10, pH 7.4). Agarose (2%) and a cover glass held down by a custom-designed clamp attached to the headplate were used to stabilize the brain during imaging.
Experimental setup.
A commercially available two-photon microscope (TriM Scope II; LaVision BioTec) was used in conjunction with a Chameleon Ultra II tunable, infrared pulsed laser (Coherent) set at 800 nm. A 20×, 1.0 NA objective (XLUMPLSLN; Olympus) was used in all experiments.
Mice were head fixed during the experiments, but could locomote at will on a custom-built treadmill. This treadmill consisted of a transparent Plexiglas disk (of 260 mm diameter) secured on a ball bearing, which allowed disk rotation around a central axis. We used an optical computer mouse or a magnetic encoder (both 1 kHz polling) to report disk movement as relative linear, or angular displacement, respectively. A shaft-mounted electromagnetic clutch was implemented along the rotational axis to provide sensory perturbations to the mouse when engaged for brief durations (50 ms). We used an infrared-sensitive camera (frame rate: 100 Hz) to capture images of the mouse from below, through the transparent disk treadmill.
During the experiments, white noise was continuously played through a speaker located directly behind the mouse at 68 dB in combination with ambient noise measured at the mouse's location. In addition, tone stimuli could be generated at a frequency of 8 kHz at 75 dB measured at the mouse's location.
Behavioral analysis.
Camera data were screened for small behavioral “twitches” using an approach described previously (Witter et al., 2013). This approach uses a reference frame coinciding with a trigger, establishes pixel variance levels in frames up to 200 ms before the trigger, and performs a variance-corrected comparison between the reference frame and the subsequent frames. This way, minute postural changes that would be missed using more conventional approaches such as the mean difference between frames could be detected while still correcting for camera sensor noise and movements caused by the animal at rest, e.g., due to breathing.
To test for statistical significance of responses and response differences, we used the Ansari–Bradley test, which assesses the likelihood that two samples derive from the same distribution or from distributions with differing dispersion characteristics. As differences in median values can lead to false results using this test, we performed the test only on median-corrected data. Trials in which movement was detected were tested against bootstrapped data. In addition, direct comparisons between the experimental groups were made to test for differences in twitch execution. For an indication of effect size of a Cx36 deficiency on mean twitch values, we used Glass' Δ as a measure to compare WT and Cx36−/− groups, taking the WT values as the control group for the variance estimate. A value >1 was considered a large effect.
Calcium transients, signal analysis, and hyperacuity.
The cerebellar cortex was imaged with the fast scan direction aligned transversely, to ensure that the respective timing of signals in ensembles of the sagittally oriented PC dendrites would be recorded with the highest possible fidelity. Calcium imaging data were analyzed using spatial independent component analysis (ICA) to extract dendrite masks. Dendrite masks were automatically thresholded (using mean pixel intensity and 5 SDs) and after removal of overlap of pixels used to calculate the changes in fluorescence ΔF/F for individual PC dendrites. See Figure 1 for an overview of the imaging experiment dendrite isolation and a sample of the data.
Detection of CS events from calcium transients was performed using a method previously described (Ozden et al., 2012), which had an estimated detection accuracy of >91% for our data. To obtain this estimate, we tested the algorithm on a substantial mock dataset, which was generated based on our entire original dataset. For every trace of our original data, we generated a mock counterpart of 50 min in which events were randomly inserted at a mean rate of 1 Hz using a Poisson process. To estimate the signal-to-noise levels in the mock traces we calculated for every corresponding original ΔF/F trace (1) the mean and SD of that trace and (2) the mean and SD of the portion of the trace exceeding a threshold of mean + 2 SDs as established in step 1. The event amplitudes in the mock traces were randomly taken from the peak amplitude distribution calculated in (2) and convolved with an exponential decay function, after which Gaussian noise as determined in (1) was added to the trace (Vogelstein et al., 2010). Due to the biases introduced in calculating the noise and signal variance levels, the signal-to-noise ratios were deliberately underestimated. Furthermore, the random Poisson spiking allowed for a closer temporal spacing of events than possible for CSs. This provided us with a worst-case estimate of the performance of the algorithm on our dataset. Under these adverse conditions it achieved a detection accuracy of 91%, with a higher chance of false positives (6%) than false negatives (3%). False positives tended to occur following a uniform distribution, leading us to believe that these will not have a significant effect on our averaged values. False negatives tended to occur when an event had already been detected directly prior (typically one or two frames distance), which could occur more readily in the mock data traces due to higher noise levels and randomized Poisson spiking than in our original data. We deemed the performance of the algorithm satisfactory, especially considering that it would perform better on our actual data.
To look at event correlations between dendrites within the same field of view (FOV), the Pearson correlation coefficient was calculated using the (binary) detected event matrices to prevent high correlation values due to locally correlated noise. Events between dendrites that occurred within 30 ms of each other were considered to be coactive. As the imaging frame rates we used were either 30 ms/frame or 15 ms/frame, we defined coactivity as taking place in the same frame or in the same or subsequent frame, respectively.
To be able to determine absolute latencies of response times in calcium transients to stimuli despite the relatively low imaging frame rate, a hyperacuity algorithm was used on the binary traces representing CS event occurrences during stimulation trials. The algorithm mapped traces to 1 kHz resolution, keeping the original data bins (of 15 or 30 ms duration) intact but aligning them to a stimulus trigger, which was logged at 1 kHz. Once realigned with millisecond precision, we could thus improve the estimate of time of occurrence over other methods such as binning of data through averaging multiple realigned event representation traces of low temporal resolution (Fig. 2A–C). The algorithm's reconstruction of triggered event timing was highly accurate provided that the number of traces used to generate the response delay estimate was at least equal to the number of milliseconds per imaging frame (Fig. 2D). In our experiments, the number of trials was well over the 100 required for an estimated error of 0 ms for both experimental groups (WT, n = 246; Cx36−/−, n = 317).
To measure effect size of calcium transient responses to stimuli we used Glass's Δ. A value >1 was considered a large effect. For the calculations of effect size of responses within groups, we used the baseline samples as the control condition. For the comparison between WT and Cx36−/− animals, we used the WT group's data as the control condition.
Results
Treadmill behavior comparison
Both WT and Cx36−/− mice spent most of the time on the treadmill sitting still, but also frequently exhibited bouts of walking or running. Bouts of locomotion varied substantially in duration within both groups, though on average WT animals walked for slightly longer periods of time than their Cx36-deficient littermates (WT durations of locomotion bouts, mean ± SD: 18 ± 10 s; Cx36−/− durations of locomotion bouts, mean ± SD: 15 ± 10 s, p < 0.05, two-tailed t test). The small differences between experimental groups could be caused by differences in animal size, as Cx36−/− animals are generally smaller than their WT littermates (Kistler et al., 2002). No differences in locomotion parameters such as stride length and base of support were found.
General CS firing properties
At rest and in the absence of any stimuli, CS firing rates were highly similar for both Cx36−/− and WT animals, as is apparent from their overlapping interspike interval (ISI) distributions (Fig. 3A). The cumulative distributions of ISIs did not differ significantly between the groups [Kolmogorov–Smirnov (K–S) test, p = 0.9]. Both groups had a mean ± SD CS firing rate of 0.8 ± 0.3 Hz averaged per trace (and a population median firing rate of 1.1 Hz, as derived from the ISIs).
During locomotion, CS ISI distributions were markedly different between the experimental groups. As seen in Figure 3B, the distribution of ISIs was considerably wider in Cx36−/− mice compared with that in WT littermates (K–S test, p < 0.005). The shift in ISI distributions yielded different median firing rates between the experimental groups (Cx36−/−: 1.1 Hz vs WT: 1.4 Hz), while the mean firing rates averaged per trace remained the same for both groups (mean ± SD: 0.9 ± 0.4 Hz for both groups).
The disk-based treadmill was fitted with a magnetic coupling device located along the rotational axis that, when engaged for a brief period (50 ms), produced disk vibrations of sufficient saliency to act as sensory perturbations, which triggered reflexive movements. Sensory perturbations provided us with a condition that was more temporally bound than ongoing locomotion (Gellman et al., 1983), facilitating temporal analyses of movements. ISI distributions taken from full traces during which sensory perturbations were provided at a mean rate of 0.2 Hz were highly similar to baseline (Fig. 3C). This indicated that an effect, if present, would be temporally bounded and brief. For both Cx36−/− mice and WT littermates the mean ± SD. firing rates for these traces including perturbations were 0.8 ± 0.3 Hz, with a median firing rate of 1.0 Hz derived from the ISIs from both groups (K–S test, p = 0.8).
In a previous study, the occurrence of doublets, multiple CS events within a time window of 200 ms, in Cx36−/− mice was quantified (Van Der Giessen et al., 2008). These doublets occurred at low rates and were found to have a higher chance of occurring in Cx36−/− than WT animals. The mechanism behind the increased doublet occurrence following from a Cx36 deficiency is not yet fully understood. It could be due to increased excitability and tendency to fire rebound spikes in IO cells of Cx36−/− mice (De Zeeuw et al., 2003). However, this provides only a partial explanation as we previously found that population profiles from pooled single-unit recordings show two distinct latencies for IO firing in response to a stimulus, of which the later response's likelihood is increased in Cx36−/− mice and includes cells that did not fire during the first response (Van Der Giessen et al., 2008). This indicates a network effect through the olivo-cerebellar loop, rather than a purely cell-intrinsic phenomenon in the IO, and is supported by recent findings on CS modulation through olivo-cerebellar network effects (Chaumont et al., 2013; Witter et al., 2013).
In our experiments in awake mice at rest and without external stimuli, we found no difference in doublet firing between WT and Cx36−/− mice: for both groups 7% of all ISIs was smaller than 200 ms. Interestingly, for locomotion trials the probability of doublet occurrence increased for both groups, but the WT animals showed the highest percentage of ISIs smaller than 200 ms (WT: 12% vs Cx36−/−: 9%). As could be expected when considering the similarity in ISI distributions between baseline and perturbation conditions, we found no difference between Cx36−/− and WT doublet probability when analyzing entire traces. However, when considering only calcium transients occurring up to 1000 ms after a perturbation, the fraction of doublets increased markedly for both animal groups, but was highest in Cx36−/− animals (WT: 20% vs Cx36−/−: 28%), in line with previous findings (Van Der Giessen et al., 2008).
Spatial correlations
We investigated correlations between CS activity in PC dendrites within the same FOV. Results obtained by Ozden et al. (2009) through the use of intra-olivary mefloquine injections showed a marked decrease in CS correlations between PC dendrites in anesthetized animals to the point that correlations were virtually absent. In our experiments, CS activity in awake Cx36−/− mice was similar to that of WT littermates under baseline conditions (baseline p = 0.6, repeated-measures ANOVA; Fig. 4A). Since concerted CS activity likely has behavioral correlates (Welsh et al., 1995; Ivry, 1996; Lang, 2001; Schweighofer et al., 2013) and since CS activity can be generated intrinsically by the olivo-cerebellar system (Chaumont et al., 2013; Witter et al., 2013) as well as through sensory stimulation (Eccles et al., 1968a), we investigated CS activity during locomotion as well as during sensory perturbations. Significant differences between the experimental groups became apparent, both when the animals were locomoting and when they were subjected to sensory perturbations (Fig. 4).
During locomotion, spatial correlations in WT animals reached significantly higher values over mediolateral distances than in Cx36−/− animals (p < 0.001, repeated-measures ANOVA; Fig. 4B). This difference was also present when the animals were exposed to sensory perturbations (p < 0.005, repeated-measures ANOVA; Fig. 4C) despite the similarity in ISI distributions for that experimental condition when considering full traces. Together, these findings suggest an effect on spatiotemporal patterning of CS activity (Fig. 4D–G).
To determine whether these differences in spatiotemporal activity patterns were due to stable differences in spatial grouping of active dendrites, we performed meta-k-means clustering on our data (Ozden et al., 2008) and compared the number of dendrites per cluster between groups and for all experimental conditions.
Clustering outcomes were often variable within and between FOVs for both Cx36-deficient and WT animals. The Dunn index value used as a criterion in the clustering procedure to attain good cluster separation yielded variable outcomes for values lower than 0.8, even during baseline conditions. For values of 0.8 and higher, clustering results became more stable (though never fully stable) with higher values showing a trend of larger fractions of dendrites not assigned to a cluster. Mean cluster sizes did not differ significantly per condition and within groups for Dunn index values of 0.8 and 0.9. We therefore pooled the data of these clustering runs for further analysis. The mean number of dendrites per cluster was similar for both groups under baseline conditions. During sensorimotor processing (i.e., locomotion and perturbation trials) the cluster sizes followed a trend in that they were consistently elevated over baseline levels in WT but not in Cx36−/− animals. However, only the number of dendrites per cluster during perturbation trials tested was significantly larger in WT animals than in Cx36−/− mutants (Fig. 5, mean ± SD, Cx36−/− 4.8 ± 5.6 dendrites; WT 7.7 ± 8.5 dendrites; p = 0.02, Wilcoxon rank-sum test).
In summary, we made three main observations regarding the spatial correlations of CS activity during baseline and sensorimotor processing conditions (i.e., locomotion and perturbation conditions): (1) CS correlations are similar for Cx36−/− mice and WT littermates during baseline conditions (Fig. 4A); (2) correlations over mediolateral distances are reduced in Cx36−/− mice when compared with WT littermates during sensorimotor processing (Fig. 4B,C); (3) the cluster size does not change during sensorimotor processing in Cx36−/− mice, whereas there is a significantly larger cluster size in WT animals during perturbation trials and there is an overall trend for increased cluster size in WT animals during sensorimotor processing (Fig. 5).
Triggered CS responses
A previous study has shown a substantial CS coactivation in PC dendrites at locomotion onset (Ozden et al., 2012). We also investigated CS coactivation triggered to locomotion onset (nCx = 7 animals, nWT = 6 animals). In addition to the increased spatial correlation between calcium transients and a minor increase in CS firing (both mentioned previously), CS coactivation of PCs was increased significantly in a time window starting ∼100 ms before disk motion onset with the highest coactivation attained ∼15 ms before disk motion (WT fraction of coactive dendrites 0.11; Cx36−/− fraction of coactive dendrites 0.13). No substantial coactivation was observed when triggering to termination of disk motion.
In response to sensory perturbations, all mice showed strong long-latency PC coactivations (p < 10−308, for both WT and Cx36−/− groups, two-sided K–S test on 200 ms data blocks prestimulus and poststimulus) with similar timing for the maximum response. For both groups the increase in CS firing was large when compared with baseline mean and variance levels, but largest in WT animals (effect sizes: WT Glass's Δ = 7, Cx36−/− Glass's Δ = 3). The center of the first peak in the response profile (first response latency) was 53 ms for both groups, but the CS response profiles differed substantially (p < 10−80, two-sided K–S test on 200 ms data blocks post stimulus; Fig. 6A). In Cx36-deficient mice, the perturbation-triggered coactivation was initially weaker and followed by a more pronounced second component ∼60 ms after the first peak (i.e., doublets and delayed responses) compared with WT mice, which showed a strong first response with a second component only weakly present. A millisecond to millisecond comparison using a two-sided K–S test revealed that the only significant differences between the coactivation profiles of the groups were centered on the peak responses (first peak p < 0.001; second peak p < 0.0001), with no significant differences in baseline activity. The effect size of Cx36 deficiency on the mean CS coactivation levels following a perturbation was large (Glass's Δ = 2).
Since activation of the magnetic coupling of the disk setup also generated a distinctive sound, we wanted to ascertain to what extent auditory stimulation could affect the recorded CS activity. To our knowledge, olivary projections to medial areas of lobule V have never been shown to encode for auditory input, though use of a “clap” stimulus loud enough to cause startle responses has been reported to evoke CS activity in mice in this region (Ozden et al., 2012). We did not expect a CS response to a nonstartle auditory stimulus and verified this by playing a tone stimulus at times when the mouse was not moving, after which we investigated PC coactivation using the tone onset as a trigger. The tone stimulus did not elicit any behavioral responses, nor did it elicit coactivation of PCs as shown in the averaged trace for tone-triggered coactivation (Fig. 6B).
Reflexes and neural correlates
Next, we investigated potential correlations between stimuli and resulting neural activity and small behavioral motor responses we refer to as twitches. Whereas locomotion is a cyclic pattern with some variation in cycle length upon which the olivo-cerebellar system acts via proprioceptive and sensory cues that are largely unknown, a twitch is generally an immediate spinocerebellar response to a controllable external event, thus facilitating further analysis. We used an automated detection algorithm described previously (Witter et al., 2013) to analyze the camera images of the mouse behavior and mark the presence of twitches (Fig. 6D). Careful inspection of a subset of the twitch detection data showed that twitches were predominantly detected in the tail, hind limbs, and front limbs with no significant differences between Cx36−/− and WT mice. Whereas tone stimuli elicited neither behavioral nor neuronal responses, sensory perturbations generally did. Furthermore, the mean twitch response probability for sensory perturbations was similar for both groups (Cx36−/−: 65 ± 19% vs WT: 68 ± 14%, mean ± SD).
Not only did the neural responses differ between WT and Cx36-deficient mice (Fig. 6A,C), the execution of twitches was also different. The onset of the motor response for both groups was similarly timed at ∼50 ms (WT p < 0.05, Cx36−/− p < 0.0005, Ansari–Bradley test against bootstrapped data), coinciding with the first response in the calcium transient profiles. The WT animals showed a response pattern with a sharp onset (Fig. 6E). In individual twitches of WT mice we generally observed a biphasic response profile (Fig. 6D), but this was largely lost in the population average (Fig. 6E). The consistent sharpness present in the WT response was lost in Cx36−/− animals, which showed a more diffuse onset followed by a gradual execution and more variability in the timing of the response (Fig. 6E). The difference between the groups in twitch execution was highly significant after the first 10 ms of the twitch onset (60 ms post trigger, p < 10−308, Ansari–Bradley test). The effect on the mean twitch response was small (Glass's Δ = 0.1), indicating that the timing and reproducibility of twitch responses rather than, e.g., movement amplitude was affected in Cx36−/− animals.
The CS responses could be a correlate of the maximal twitch velocity, as their profiles seemed to coincide (Fig. 7). An alternative explanation might be that the initial part of the twitch was mediated by the spinal cord, which normally also shows Cx36 expression (Matsumoto et al., 1988). After the initial part of the twitch, CS volleys fired by PCs might then modulate movement with a delay through rebound firing in the CN, which could explain the similarity in shape between the observed CS coactivations and twitch velocity profiles 80 ms later (Fig. 7).
We have demonstrated a difference in PC population firing of CSs separated by observed behavior, but so far the trigger was a sensory perturbation or onset of disk motion. Though these triggers showed an increased chance of concerted CS generation in PC dendrites correlated with behavior, the question to what extent motor events can be predicted on the basis of CS coactivations still remained. Since such synchronous CF activity is thought to be important for motor behavior (Welsh et al., 1995; Ivry, 1996; Lang, 2001; Schweighofer et al., 2013), we investigated whether substantial PC CS coactivations in zone A of lobule V had behavioral correlates under baseline conditions (i.e., when the mouse was at rest without external stimuli). Indeed, when selecting coactivations larger than 25% as a trigger, twitches could often be detected in the camera frames immediately after the frame coinciding with the onset of the two-photon imaging frame used for triggering. Even though the distributions of coactivations larger than 25% of the population of PC dendrites within a FOV were similar in both groups (meanCx = 35%, SDCx = 10%, medianCx = 31% and meanWT = 35%, SDWT = 10%, medianWT = 32%), the probability of detecting a twitch based on coactive PC dendrites was significantly different between Cx36−/− and WT animals (Cx36−/− 0.32 vs WT 0.45, p < 0.01, nCx = 323 and nWT = 198 triggers). Twitch execution shape between WT and Cx36−/− animals when triggered to CS coactivations did not test as significantly different (p = 0.3, Ansari–Bradley test).
Since the twitch detection algorithm uses a reference frame and a correction for the variance in pixel values in frames preceding the reference frame, its accuracy in detecting a twitch starting before the trigger might be compromised. To exclude this possibility, we repeated the experiment with the trigger points shifted 100 ms before coactivations exceeding 25% of the dendrites within the FOV (Fig. 8). This yielded the same twitch detection probabilities mentioned earlier with the twitch onset timed around CS coactivation of PCs.
Discussion
Neurons in the IO prominently express Cx36 (Condorelli et al., 1998, 2000), from which dendrodendritic gap junctions are formed (De Zeeuw et al., 1995, 2003). These were predicted four decades ago to be essential for the timing of motor output (Llinas et al., 1974; Sotelo et al., 1974). Remarkably, so far the phenotype of Cx36-deficient mice has appeared relatively mild with little sign of impaired motor performance (Kistler et al., 2002). Mainly when subjected to sophisticated motor tasks, such as eyeblink conditioning or the Erasmus Ladder test, did these animals show impairments in motor timing of a conditioned response (Van Der Giessen et al., 2008). Yet here, despite the redundancy generally inherent to biological systems, by using a novel transparent treadmill setup, we show that Cx36−/− mice have deficits that cannot be compensated for in the timing of their spinocerebellar reflexes. Moreover, we have been able to simultaneously visualize with high spatial acuity the CS activity of neighboring PCs in cerebellar microcomplexes involved in the coordination of these reflexes.
Spatial organization of CS activity
Previously, multi-electrode approaches revealed CS synchrony in sagittal bands along CF projections (Lou and Bloedel, 1992; Bosman et al., 2010), but these approaches lack the spatial resolution attained with two-photon calcium imaging, which allows the visualization of PC ensembles spanning mediolateral distances that fall well within a microzone (Oscarsson, 1979; Chen et al., 1996). In such PC ensembles concerted activity in pairs of dendrites was inversely correlated with distance (Ozden et al., 2009). We demonstrate that coupling through Cx36 appears essential for state-dependent changes in PC coactivation during sensorimotor processing. Our data showed significantly different response profiles between WT and Cx36−/− mice and a significant increase in correlated CS activity in WT over Cx36−/− mice immediately following a sensory perturbation. Differences in correlation values between experimental groups for perturbation trials cannot be readily explained by mismatches in firing rates (de la Rocha et al., 2007), as ISI distributions and the mean firing rates were similar for both groups. Despite the significant effects on CS correlations, the ISI distributions even remained similar to baseline conditions as sensory perturbations occurred at low frequencies during rest. The observation that correlation values increased significantly in WT animals despite the low frequency of perturbations highlights the prominent impact of Cx36 gap-junctional coupling. The correlations between dendrites over mediolateral distances in Cx36-deficient mice appeared fairly invariant with respect to our experimental conditions (Fig. 4D, bottom), whereas WT animals showed increased correlations during sensorimotor processing (Fig. 4D, top). Thus, even though CS firing rates remained similar in Cx36−/− animals when compared with WT littermates and even though there was no detectable effect of Cx36 deficiency on CS activity under baseline conditions, the spatiotemporal patterning in functional ensembles of PCs alters during sensorimotor processing in that CS synchrony increases dynamically and prominently in a system with coupled olivary neurons, but not without.
Performing meta-k-means clustering (Ozden et al., 2008) on our data provided no clear indication of stable clustering patterns of dendrites. This could be due to the fact that necessary assumptions for this method are that grouping of dendrites is fairly stable throughout the recording and that the recording is sufficiently long (Ozden et al., 2012). Locomotion and twitch responses were generally brief. Furthermore, we studied an area involved in postural motor processes and our baseline condition may not be behaviorally invariant, but rather a fluid succession of minute postural adjustments. This possibility is supported by our finding that spontaneous coactivations during baseline conditions can be used as a predictor for small movements. We surmise that in behaving animals, clusters of dendrites within microcomplexes can vary in size depending on the region of cerebellar cortex and the animal's behavioral state, similarly to interzonal clusters (Welsh et al., 1995). Interestingly, during sensorimotor processing WT animals on average have more dendrites per cluster within a FOV than Cx36−/− mutants. This may reflect the ability of WT animals to recruit and group separate microcircuits as needed through olivary gap junctions, which is lost in Cx36−/− mice.
Temporal patterns and behavior
Synchronous CS discharges have been proposed to play a functional role in the onset of motion and/or coordination of muscle activity (Welsh et al., 1995; Ivry, 1996; Lang, 2001; Schweighofer et al., 2013) and Cx36 gap junctions mediate IO synchrony (Long et al., 2002) as well as coherence of harmaline-induced muscle tremor (Placantonakis et al., 2004). To find out to what extent sensory-evoked synchronous volleys of IO discharges can elicit motor responses we studied correlates of CS coactivity and motor behavior. Using treadmill motion as a trigger we found correlated CS coactivations preceding locomotion, but no correlates for its termination.
The controllable sensory perturbations and the temporally bounded twitch response they could evoke provided us with a means to carefully investigate differences in spatiotemporal CS patterning during sensorimotor processing and movement onset. The CS response profile differences we recorded following a sensory perturbation shed light on the observed differences in spatial correlations between dendrites. Following a perturbation, WT animals showed an immediate and well timed primary CS response spread across many dendrites after which a small secondary response generally occurred. The CS responses in Cx36−/− animals, on the other hand, were more localized and poorly timed, with a strong secondary response in part due to an increased occurrence of doublets compared with that in WT littermates.
The temporal execution of a twitch also differed: whereas the twitch execution in WT animals had a sharp onset, it was gradual and less consistent in Cx36−/− mice. Olivary gap junctions synchronize CS firing (Long et al., 2002; De Zeeuw et al., 2003), likely leading to tightly timed strong inhibition in the CN, followed by rebound firing during the typical PC CS pause. Such rebound firing can cause motor responses (Witter et al., 2013). A temporally spread CS volley could theoretically lead to a reduced or slow motor response as its impact on CN neurons would be reduced (Schweighofer et al., 2013), providing a possible explanation for both the difference in twitch execution between the groups and the lower chance of detecting twitches based on coactivations in Cx36−/− animals compared with WT littermates. Further inconsistencies in late twitch execution as observed in Cx36−/− mutants with respect to WT littermates could be due to on-line adjustments, which in turn also suffer from timing inaccuracies.
Whether the timing of neural signals involved in spinocerebellar reflexes deteriorates solely in the IO cannot be deduced with absolute certainty, since Cx36 gap junctions are also expressed upstream of the IO in the spinal tract (Matsumoto et al., 1988) and downstream in interneurons of the cerebellum (Dugué et al., 2009; Vervaeke et al., 2010). Given the low levels of expression of Cx36 in the spinal cord, it seems likely that temporal shifts in the spinal signals to the IO are initially small. It should be noted that activity within the olivo-cerebellar system can theoretically be maintained through recurrent connections (Kistler and De Zeeuw, 2002), potentially leading to temporal error propagation during reverberating olivo-cerebellar processing.
Functional implications
The subthreshold oscillations (Llinás and Yarom, 1981a,b) and extensive gap-junctional coupling in networks of olivary cells have given rise to theories of clock function and synchrony for motor performance mediated by the IO in cerebellar processing (Llinás, 2009; Lefler et al., 2013). Seemingly counter to this school of thought are the motor learning hypotheses that regard the IO primarily as a nucleus providing a signal to guide cerebellar learning (Marr, 1969; Albus, 1971). Reports on connexin36 gap-junction knock-outs have so far seemed to indicate a role for olivary electrical coupling in motor learning with little effect on on-line motor control (Van Der Giessen et al., 2008; van Essen et al., 2010).
We now demonstrate that a connexin36 deficiency affects timing of CS activity and motor responses directly, as evident from CS activity within distances up to 150 μm and temporal execution of spinocerebellar reflexes. Our results indicate that learning of timed responses may be impaired not only by lack of olivary connexin36-mediated modulation of cerebellar plasticity (Mathy et al., 2009; Bazzigaluppi et al., 2012; De Gruijl et al., 2012; Rasmussen et al., 2013), but also by timing inaccuracies in CS activity and movements arising directly from a lack of connexin36 gap-junctional coupling. Recent findings also provide evidence for short- and long-term modulation of olivary coupling strength, giving rise to nuanced functional hypotheses for olivary coupling (De Gruijl et al., 2014; Lefler et al., 2014; Mathy et al., 2014; Turecek et al., 2014). Possibly, the IO forms functional ensembles through Cx36 gap junctions to execute timed reflexes via synchronized spiking, the occurrence of which is used to drive cerebellar learning. The cerebellum would aim to override or avoid network states that triggered IO reflexes (De Gruijl et al., 2014). Thus, IO synchrony could both drive cerebellar learning (Van Der Giessen et al., 2008; De Gruijl et al., 2012) and mediate reflex-like movements as we have shown in the current study.
Footnotes
This work was supported by the Dutch Organization for Medical Sciences (ZonMw; C.I.D.Z.); Life Sciences (ALW; C.I.D.Z.); Senter (Neuro-Basic); and ERC-adv, CEREBNET, and C7 programs of the European Community (C.I.D.Z.). We thank J. Plugge for her expert technical assistance.
The authors declare no competing financial interests.
- Correspondence should be addressed to Chris I. De Zeeuw, Erasmus MC, Department of Neuroscience, P.O. Box 2040, 3000 CA Rotterdam, The Netherlands. c.dezeeuw{at}erasmusmc.nl