1. Neuroscience
Download icon

Complex spike synchrony dependent modulation of rat deep cerebellar nuclear activity

  1. Tianyu Tang
  2. Timothy A Blenkinsop
  3. Eric J Lang  Is a corresponding author
  1. New York University School of Medicine, United States
  2. Mount Sinai School of Medicine, United States
  • Cited 7
  • Views 678
  • Annotations
Cite this article as: eLife 2019;8:e40101 doi: 10.7554/eLife.40101

Abstract

The rules governing cerebellar output are not fully understood, but must involve Purkinje cell (PC) activity, as PCs are the major input to deep cerebellar nuclear (DCN) cells (which form the majority of cerebellar output). Here, the influence of PC complex spikes (CSs) was investigated by simultaneously recording DCN activity with CSs from PC arrays in anesthetized rats. Crosscorrelograms were used to identify PCs that were presynaptic to recorded DCN cells (presynaptic PCs). Such PCs were located within rostrocaudal cortical strips and displayed synchronous CS activity. CS-associated modulation of DCN activity included a short-latency post-CS inhibition and long-latency excitations before and after the CS. The amplitudes of the post-CS responses correlated with the level of synchronization among presynaptic PCs. A temporal precision of ≤10 ms was generally required for CSs to be maximally effective. The results suggest that CS synchrony is a key control parameter of cerebellar output.

Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed (see decision letter).

https://doi.org/10.7554/eLife.40101.001

Introduction

The cerebellum plays a significant role in many brain functions, providing information to multiple brainstem nuclei and, via the thalamus, much of the cerebral cortex. This widespread influence makes it vital to understand the factors that shape cerebellar output, which arises primarily from the deep cerebellar nuclei (DCN). Purkinje cells (PCs) are the major synaptic input to DCN neurons, accounting for some 70–80% of the synapses onto them (Palkovits et al., 1977; De Zeeuw and Berrebi, 1995). Thus, understanding how cerebellar output is formed requires knowing the relationship between PC and DCN activity.

Although the properties of the PC-DCN synapse have been characterized in a number of studies, the rules governing the translation of PC activity into DCN firing patterns under in vivo conditions are not well understood. The GABAergic nature of PCs suggests that an inverse relationship between PC and DCN activity should exist, and indeed, manipulations that cause global changes in PC simple spike (SS) firing rates produce the predicted inverse changes in DCN activity (MacKay and Murphy, 1973; MacKay and Murphy, 1974; Colin et al., 1980; Montarolo et al., 1982; Bardin et al., 1983; Benedetti et al., 1983). However, an inverse relationship is often not seen during behavior, at least at the population level (Thach, 1970a; Thach, 1970b; Armstrong and Edgley, 1984a; Armstrong and Edgley, 1984b; Robinson and Fuchs, 2001; Sarnaik and Raman, 2018). Moreover, attempts to correlate spontaneous SS activity from individual PCs with the activity of individual DCN neurons failed to reveal any consistent correlation, despite the employment of rigorous measures to ensure that the activity of cells located in connected regions of the cortex and nuclei was being compared (McDevitt et al., 1987). Thus, the translation of PC activity into DCN firing often appears not to be a simple integration of SS activity.

The complexity of the PC-DCN transform is likely due to multiple factors, such as the existence of subpopulations of DCN neurons with distinct electrophysiological properties (Czubayko et al., 2001; Uusisaari et al., 2007; Uusisaari and Knöpfel, 2011; Najac and Raman, 2015), the convergence of many PCs (and cerebellar afferents) onto each DCN neuron, and the generation of two types of action potentials, SSs and complex spikes (CSs), by PCs. In addition, the specific patterning and timing of PC activity is likely a critical factor. For example, recent work suggests that DCN firing is more effectively phase locked by synchronized than by desynchronized PC activity (Person and Raman, 2012; Person and Raman, 2011).

Here, we investigate the importance of this last factor, the timing of PC activity, and focus specifically on the role played by CS synchrony among PCs that synapse with the same DCN cell. This focus is motivated by the characteristics of the patterns of CS synchrony. Specifically, synchronous CSs occur most often among PCs within the same rostrocaudally running strip of cortex (Sasaki et al., 1989; Sugihara et al., 1993; Lang et al., 1999), and in particular, among PCs within the same zebrin compartment (Sugihara et al., 2007). This and the fact that PCs located in the same rostrocaudally running zebrin compartment project to the same DCN region (Chung et al., 2009; Sugihara et al., 2009; Sugihara, 2011) make it plausible that the PCs that project to the same DCN cell would also have high levels of synchronous CS activity. Moreover, we previously showed that generalized increases in CS synchrony are associated with greater inhibition of DCN activity (Blenkinsop and Lang, 2011). Here, we build on this earlier work by providing evidence that high levels of CS synchrony are indeed present among PCs that synapse onto the same DCN cell and that synchrony is a major parameter of their influence on the DCN cell's activity. Furthermore, we address the issue of the temporal precision needed for CSs to affect DCN activity.

Results

Here, we investigate the relationship between CS and DCN activity in order to provide a quantitative analysis of the relationship between CS synchrony and various aspects of CS-associated modulation of DCN activity. In particular, we specifically analyze the effect of synchronous CS activity for PCs that converge onto the same DCN neuron. To do this, recordings in which at least four PCs in the electrode array could be identified as presynaptic to the DCN cell (based on analysis of their crosscorrelograms) were selected from the dataset of Blenkinsop and Lang, 2011. The specific criteria for identifying a PC as presynaptic are given in the Methods (also see, Blenkinsop and Lang, 2011). Out of the original dataset (n = 717 PCs, 100 DCN cells, 35 rats), four DCN cells were found that had synaptic input from at least four crus 2a PCs in the recording array. These four DCN cells (Figure 1A–D) were recorded along with CS activity from 59 PCs (Figure 1E), of which 24 showed evidence of a synaptic connection to the recorded DCN cell (henceforth referred to as presynaptic PCs), and 35 did not (non-synaptic PCs) (n = 3 rats). Of the four DCN cells investigated in the present report, the locations of two were localized histologically to the interpositus nucleus. The other two were not histologically localized but were likely in either the interpositus or medial dentate nuclei, given the distribution of recordings in the prior study. All four DCN cells had spike trains that consisted of a mix of tonic and bursting firing patterns (Figure 1F).

Extracellular records of DCN neurons.

(A–D) Extracellular records (10 overlapped sweeps) showing the spike waveforms of the DCN neurons whose activity is analyzed throughout the paper. The letter of each panel corresponds to the experiment name used throughout the remainder of the paper for analyses related to that cell. Horizontal calibration bars are 1 ms. Vertical calibration bars: A, 200 µV; B-D, 100 µV. (E) Example of CS activity recorded from one of the PCs identified as presynaptic to a DCN cell (10 overlapped spikes). Calibration bars are 5 ms and 50 µV. (F) Example recording of DCN neuron (same as in panel A) to show typical firing pattern consisting of both tonic and bursting activity. Calibration bars are 0.5 s and 200 µV.

https://doi.org/10.7554/eLife.40101.002

To test whether being identified in the presynaptic group reflected a difference in CS firing rates, the average firing rates of the presynaptic and non-synaptic PCs were compared. No significant difference was found (presynaptic, 0.96 ± 0.78 Hz, n = 24; non-synaptic, 0.94 ± 0.71 Hz, n = 54; p=0.91); note that 19 PCs were recorded with two DCN cells and their firing rates in each recording session are included in the totals; however, including the firing rates of those PCs from only one of the sessions did not result in significant differences between the groups (p=0.41, n = 17 and 41, and p=0.44, n = 18 and 40). The firing rates of the four DCN cells on average (34.19 ± 7.01 Hz) were lower those of our larger sample that was reported previously (45.3 ± 21.4 Hz; n = 100; p=0.037; Blenkinsop and Lang, 2011); however, they were all still well within the range of rates typical for DCN cells. In sum, the basic firing rates of the PCs and DCN cells identified as being synaptically-connected were typical of these cell types.

Examples of CS-triggered correlograms illustrating the relationship between CS activity in presynaptic PCs and DCN activity are shown in Figure 2. In every case, a sudden sharp drop in activity occurs just after the CS (dashed lines at a latency of 0 ms), which was the criterion for identifying a monosynaptic connection between the PC and DCN neuron (the precise timing of the inhibition onset is seen most clearly in the lower row of histograms, which show the central portion of the histograms in the top row with an expanded time scale). In a few cases (n = 4 out of 24), a brief excitation preceding the CS (Figure 2B) was present and likely reflected the excitation of the DCN cell by collaterals of the same olivary axon that was evoking the CSs (Blenkinsop and Lang, 2011). However, because this response was present so rarely, we did not analyze it further. In addition to these short-latency effects, the histograms often also showed longer latency changes in the DCN activity associated with CSs. In particular, there were periods of increased DCN activity roughly 200–500 ms before and after a CS. The amplitude of both the short- and long-latency modulations varied between experiments and between PCs within an experiment, as can be seen by comparing the correlograms in Figure 2. Characterizing how these modulations vary with CS synchrony is the primary focus of the remainder of this paper; however, we first describe the patterns of synchrony among presynaptic PCs, because they provide further evidence in support of the correlogram-based identification of presynaptic PCs and because these patterns define the range of synchronous activity generated by the olivocerebellar system.

CS-associated changes in DCN activity.

(A–D) CS-triggered histograms of DCN activity for four different PC-DCN pairs identified as synaptically-connected. Panels A and B show histograms from the same DCN cell with two different PCs. Panels C and D show histograms from two other DCN cells. The bottom row (A2–D2) shows the corresponding histograms of the top row (A1–D1) with an expanded time scale in order to see the timing of the onset of the inhibition in DCN activity at the time of the CS (latency = 0 ms; dashed line). Bin width = 1 ms.

https://doi.org/10.7554/eLife.40101.003

PCs that project to the same DCN neuron form narrow rostrocaudal bands

PCs connecting to the same DCN cell would be expected to be clustered in the mediolateral direction (i.e., along the longitudinal folial axis), because of the topography of the PC-DCN projection (Chung et al., 2009; Sugihara et al., 2009; Sugihara, 2011; Voogd and Bigaré, 1980). The actual spatial distributions of the presynaptic PCs for each of the four DCN neurons are shown in Figure 3A–D. (Note that for the remainder of the paper each experiment will be referred to by the letter of the panel that shows its PC array in Figure 3, and that the panels in Figure 1 showing the DCN spike waveforms also correspond to this naming convention.) In each panel, the overall recording array on crus 2a is indicated by the distribution of circles in each panel, where the filled circles represent the presynaptic PCs. The orientation of the recording arrays on crus two is shown by the schematic of the cerebellum in panel 3A. Visual inspection of the circle plots suggests that the presynaptic PCs are, in fact, restricted in the mediolateral direction. To test this statistically, we compared the average separation among PCs in the synaptic groups to that in groups that had an identical number of cells but in which cell locations were randomly chosen throughout the array (only one cell per location). One thousand such random groups were generated for each experiment, and the resulting distributions of average mediolateral separation between cells are shown by the histograms in Figure 3A–D. The dashed and dash-dotted lines indicate the 5th and 1st percentiles of the random group distributions, respectively. The arrow indicates the average separation of PCs in the presynaptic group, which falls below the 5th percentile in all cases and below the 1st in three out of four cases. Further evidence of the spatial restriction in the mediolateral direction is shown by the scatterplot of the average separation distances for the presynaptic and random groups for the four experiments, where all of the points fall above the x = y line (Figure 3E). In contrast, the average separation along the rostrocaudal axis was similar for the presynaptic and random groups (Figure 3F). In sum, the spatial extent of each presynaptic PC group is consistent with all of the PCs in the group projecting to the same DCN neuron, supporting their identification as presynaptic by the cross-correlogram analysis.

PCs identified by correlogram analysis as projecting to the same DCN neuron form spatially restricted groups.

(A–D) Schematics showing recording arrays from four experiments. Circles indicate positions of recorded PCs; filled circles indicate positions of PCs identified as presynaptic to the recorded DCN cell. The histogram below each schematic shows the distribution of the mean mediolateral separation among cells from groups having the same number of cells as the presynaptic group but whose locations on the array were randomly chosen. Arrow indicates the mean separation value for presynaptic group in each experiment. The dashed and dash-dotted lines indicate the 5% and 1% percentiles, respectively. (E) Scatter plot of the mean mediolateral separation of PCs in the presynaptic group versus the mean of the distribution for the random groups. (F) Same as (E) except that the separation in the rostrocaudal direction is plotted. Note that each experiment shown in panels A-D will be referred to throughout the paper and in other figures with reference to their panel designation in this figure.

https://doi.org/10.7554/eLife.40101.004

PCs that project to the same DCN neuron show high levels of CS synchrony

In light of the spatial organization of the presynaptic PC groups, and as previous work shows that synchronous CS activity occurs among PCs aligned in the same rostrocaudally running strip of cortex (Bell and Kawasaki, 1972; Sasaki et al., 1989; Lang et al., 1999), we investigated the patterns of synchrony for the presynaptic groups. The level of synchrony and the mediolateral separation distance were computed for each PC pair in the arrays from all four experiments. Average (mean) synchrony was then plotted as a function of separation distance for the presynaptic and non-synaptic PCs (Figure 4A; presynaptic group PCs: for 0, 250, 500, 750, and 1,000 µm, n = 16, 31, 13, 2, and 1; nonsynaptic group PCs: for 0, 250, 500, 750, 1,000, 1,250, 1,750, and 2,000 µm, n = 43, 95, 69, 51, 32, 29, 10, and 3). Comparison of the mean synchrony curves for the two cell groups shows that much higher synchrony was present at all separation distances for presynaptic PCs (0 µm, p=0.04, n = 16 and 43; 250 µm, p=0.0005, n = 31 and 95; 500 µm, p=0.02, n = 13 and 69). Tests for normality (Kolmogorov-Smirnov) showed that the distributions were normal at all separation distances for the presynaptic PC pair populations but not for the nonsynaptic PC pair populations. Thus, median synchrony values were also plotted, and they showed an essentially identical result. In addition, a non-parametric test of the presynaptic and nonsynaptic synchrony distributions showed that they were significantly different (0 µm: p=0.002, n = 16 and 43; 250 µm: p=3×10−07, n = 31 and 95; 500 µm: p=2×10−07, n = 13 and 69; Wilcoxon-Mann-Whitney two sample rank test).

DCN projection groups show high CS synchrony levels.

(A) Plot of mean synchrony as a function of the mediolateral (ML) separation between PCs from all experiments. The filled circles show the mean synchrony for pairs in which both PCs were in one of the identified presynaptic groups and the unfilled circles show the synchrony for pairs from the remaining cells in the arrays. Error bars are ±1 SEM. (B) Similar to A, except that median synchrony is plotted as a function of ML separation. (C) The ratio of mean synchrony of the synaptic and non-synaptic PCs is plotted as a function of ML separation for each of the four experiments.

https://doi.org/10.7554/eLife.40101.005

The consistency of higher synchrony among PCs in the presynaptic group across experiments is shown by curves in Figure 4C, which plot the ratio of the mean synchrony curves for the presynaptic and nonsynaptic PC cell pairs. In all cases where it is defined (n = 13; 4 at 0 µm, 4 at 250 µm, 3 at 500 µm, 1 at 750 µm, 1 at 1,000 µm), the ratio is above one.

In Figure 4, it can also be seen that for the nonsynaptic PCs, synchrony falls off with increasing ML separation, consistent with previous studies (Sasaki et al., 1989; Lang et al., 1999). In contrast, for the presynaptic PCs, synchrony does not decline, but rather stays relatively constant (0–500 µm) or rises (750–1,000 µm) with increasing ML separation, though the n is small for these larger distances. This divergence in the two synchrony curves was present in the experiments where there were separations of at least 500 µm between synaptic PCs (Exps. A, C, and D), which is demonstrated by the increase in the synchrony ratio with separation for these experiments (Figure 4C). In sum, synchrony is better maintained among PCs within the same presynaptically-connected group.

The above analyses considered all of the PCs pairwise to derive the average or median levels of synchrony between PC pairs. We next investigated the prevalence of more widespread synchronization among presynaptic PCs. For this analysis, each PC was considered as a reference in turn. Each CS fired by the reference cell was assigned a synchrony level based on the number of PCs in the presynaptic group firing CSs within a specific time window (±5 ms) centered on the time of the CS in the reference cell. The distributions of synchrony levels for each PC and the average of all PCs in the group are shown for each experiment in Figure 5A–D. The average curves for all of the experiments are replotted together in Figure 5E for direct comparison. In general, events comprising CSs in only one or a minority of the PCs in the presynaptic group were most common; however, synchrony events at all levels occurred in the experiments.

Relative prevalence of synchronous CS activity.

(A–D) Plots of the percent of CSs at each synchrony level (Percent of cells firing CSs) within the presynaptic group. Each panel shows data from one experiment. Each curve shows the distribution of CSs for one PC in the presynaptic group. The number of PCs in each group is as follows: A, 7; B, 4; C, 7; D, 6. (E) Average distribution for each experiment.

https://doi.org/10.7554/eLife.40101.006

In sum, the wider distribution and generally higher levels of synchrony, and in particular, the presence of synchrony events involving most or all cells, that was found for the presynaptic PC groups indicate that individual DCN neurons are often subjected to synchronized CS input, raising the question of what is the effect of such activity on DCN cells.

Inhibition of DCN activity varies with level of CS synchrony

CS-triggered rasters were used to begin investigating the relationship between CS synchrony and DCN activity (Figure 6A). Visual inspection of the rasters revealed the presence of CS-related modulation of DCN activity at a variety of time scales. To investigate the relationship of these modulations to CS synchrony, CSs were sorted according to the fraction of PCs in the presynaptic group that were firing simultaneously (synchrony level), as shown in Figure 6B. Comparison of the rasters across synchrony levels showed the presence of a strong relationship between synchrony and the short-latency inhibition that follows a CS. That is, when a PC fired a CS and the other PCs in the group did not, the post-CS inhibition was very weak, being barely visible in the raster display (Figure 6B, 1/7 PCs), and the histogram compiled from these CSs showed only a shallow depression following the CS (Figure 6C, 1/7). However, the inhibition of DCN activity in this period increased significantly with synchrony, which can be seen by comparing either the rasters or the histograms across synchrony levels. Indeed, at the highest synchrony levels (6/7 and 7/7), there is almost a complete silencing of the DCN cell for ~100 ms following a CS.

CS-associated modulation of DCN activity.

(A) CS-triggered raster display of DCN activity. The raster was generated from all CSs of one PC during a 20 min recording (firing rates: CS, 0.70 Hz; DCN cell, 30 Hz). Each row represents the activity surrounding one CS. The rows are arranged in chronological order (top to bottom). (B) Raster of same data as in A, but CSs are first grouped according to synchrony level, as indicated by labels on y-axis. Within each synchrony level the rows are arranged chronologically. In particular, note how the inhibition of DCN activity just after the CS (latency = 0 ms) increases with synchrony. (C) CS-triggered histograms of DCN activity for different synchrony levels. Top histogram (all) generated from all CSs and corresponds to the entire raster in A and B. The remaining histograms were generated from their corresponding groups in the raster of panel B. Note the expanded time scale compared to the rasters in order to highlight the change in the inhibition that follows the CS, and the transient excitation just preceding the CS (visible in bottom four histograms).

https://doi.org/10.7554/eLife.40101.007

CS-triggered histograms of DCN activity were used to quantify the relationship between CS synchrony and the short-latency inhibition seen in the raster displays. For each synchrony level, a histogram was made using the CSs of each presynaptic PC, as shown in Figure 6C. The individual cell histograms for each synchrony level were then averaged to get the overall group response at each level. In this analysis and in succeeding ones, a synchrony level generally corresponds to a specific fraction of PCs in the group firing a CS; however, the number of events at the highest synchrony levels was small for experiments A, C and D (Figure 5 and Figure 6B,C, levels 6 and 7), and so for these experiments, the highest 2–3 synchrony levels were collapsed into one 'high' synchrony level, such that every PC in the group had at least 10 high synchrony events (Exp. A: levels 5–7; Exp. C: levels 6–7; Exp. D: levels 4–6). The value assigned to this high synchrony level was a weighted average of the combined synchrony levels (the weights corresponded to the total number of CSs at each level summed across all PCs in the group).

The inhibitory effect of the CSs was then measured as the percent change in the average height in the histogram during the response period (defined below) from baseline. The baseline was defined as the average height of the histogram at latencies of −55 to −5 ms, except for experiment D, where the latencies −30 to −5 ms were used (because the correlograms were relatively flat over a shorter period preceding the CS in this experiment).

The response period limits were determined in two ways. In the first, the response period was determined using the CS-DCN histogram constructed from all CSs in all PCs of the group. The start of the response was defined as the time after the CS (t = 0 ms) when the histogram first fell below the baseline level (a moving average of three 1 ms bins was used, with the central bin being the time point). The end of the response was defined as the first point after the start time that the moving average crossed back above baseline. Thus, the duration and timing of the response period were held constant across synchrony levels, allowing the same time period to be compared. In all experiments, the start of the response, thus defined, occurred at a latency of 1.5 ms from the CS, consistent with a monosynaptic connection from the PCs to the DCN cell. The end of the response occurred at a latency of 68.75 ± 12.66 ms (n = 4; Figure 7A, All), giving an average response duration of 67.25 ms.

CS-associated inhibition of DCN activity is a function of synchrony.

(A) Duration of post-CS inhibition is plotted as a function of synchrony level for each experiment. The durations were determined from histograms generated from the CS activity of all PCs in a presynaptic group. The points plotted above 'All' are durations determined from the histogram generated from all CSs. The other points were generated from all CSs at a particular synchrony level. (B–E) For each experiment, the percent change in DCN activity during the post-CS inhibition period from baseline activity is plotted as a function of synchrony level. The time of the response period was set either based on the All-CS histogram (filled circles) or by measurement of the duration on the specific synchrony level histograms (unfilled circles). Dashed lines indicate the change in activity measured from the All-CS histogram. (F) Comparison of the effect of synchrony across all experiments. The 'all CSs' curves in panels B-E were normalized by dividing each by the corresponding average synchrony for that experiment (horizontal dashed line in each panel). Symbols for experiments are the same as in panel A. Solid line is least squares regression line. (G) Change in the level of the baseline activity with synchrony. Histograms were normalized for number of CSs, and then baseline activity at each synchrony level was compared to the baseline level for the All-CS histogram. In some of the experiments, the highest synchrony levels were combined because there were too few events (<10) at the highest levels to analyze. In these cases, the synchrony level was set to a weighted average of the combined levels: experiments A and C, levels 6-7/7; experiment D, levels 4-6/6. No levels were combined for experiment B. Solid line is least squares regression line.

https://doi.org/10.7554/eLife.40101.008

In the second method, the response period was determined individually for each synchrony level. The same procedure as in the first method was followed, except that the start and end of the response were defined using the individual synchrony level histograms instead of the all-CSs histogram (three-point averages were again used, except in experiment C, where the sparseness of the higher synchrony levels required a moving average of six points to obtain limits that reasonably agreed with those based on visual inspection). Although less reliable than the first method, the second allowed us to test whether the response duration varied with synchrony.

For most synchrony levels, the second method produced similar estimates of the response duration to that obtained by the first method (Figure 7A, compare the All value with those at different synchrony levels for each experiment). Indeed, in two experiments, there was no significant correlation between synchrony level and response duration. In the other two experiments (experiments A and C), the correlation was significant (p<0.05, n = 7 in both cases); however, most of the variation resulted from a large jump in duration at the highest synchrony levels for which the values are the least reliable, because the histograms corresponding to these levels are constructed from relatively few sweeps (e.g., compare the number of sweeps per level in the raster of Figure 6B). In sum, the duration of the initial inhibition of DCN activity following a CS generally does not appear to vary significantly with synchrony, except possibly at the highest levels.

We next quantified the change in the amplitude of the short-latency inhibition with synchrony. With either method for defining the response period, the inhibition of DCN activity relative to the baseline became stronger with increasing synchrony (Figure 7B–E). In all experiments, the inhibitory effect was weakest for events in which only one PC of the presynaptic PC group fired a CS and strongest at the highest synchrony level. Because no consistent variation of response duration with synchrony was found, the response duration defined by the first method is used hereafter, because of its greater reliability.

Although the amplitude of the short-latency inhibition varied with synchrony in every experiment, the average inhibition (i.e., that produced by all CSs and indicated by the dashed lines in panels 7B-E) varied between experiments. So, to compare the synchrony dependence across experiments, the inhibitory effect curve from each experiment was normalized by dividing it by the average inhibition in the experiment (Figure 7F). The combined data from the four experiments show a strong correlation between CS synchrony and the strength of the inhibition of DCN activity for the entire dataset (r = 0.93, p = 4×10−9, n = 20). Indeed, that the data points from all experiments are fit well by a single regression line suggests that the synchrony dependence was similar for all of the DCN cells recorded, despite the variations in the absolute strength of the inhibition.

Because the inhibitory effect was measured with respect to the baseline DCN activity that preceded the CS, it is possible that it could in part reflect a variation of baseline activity with synchrony. To test this possibility, the baseline activity in all histograms was normalized for (divided by) the number of CSs used to generate them and then compared (Figure 7G). In general, there was little change in the baseline across synchrony levels. Only at the highest synchrony levels did relatively large changes sometimes occur, but there was no clear directionality to these changes across experiments, and the slope of regression line was not significantly different from zero (p = 0.14, n = 20). Moreover, even if the slope had been significant, it would have accounted for only a 12.4% change in DCN firing rate across the synchrony levels, in contrast to the ~80% or greater decreases that were generally found.

In sum, the results suggest CS triggered inhibition of DCN activity is strongly dependent on the level of synchronization among the PCs projecting to a DCN cell.

Synchrony determines the inhibitory efficacy of almost all PCs

We next analyzed how the dependence of the short-latency inhibition on CS synchrony varied between individual PCs. For each PC, the inhibitory effect on DCN activity based on all of its CSs was compared to that of its high synchrony CSs as defined earlier (Figure 8). The inhibition caused by the high synchrony CSs was significantly greater than that caused by all of the CSs that a PC fired (black data points) (All: −38.4 ± 15.4%; Synchronous: −72.5 ± 13.1; p =1×10−7, n = 24, Wilcoxon signed rank test). Moreover, for all PCs but one (n = 23/24), highly synchronized CSs caused greater inhibition of DCN activity. In the remaining case, the difference was negligible. Thus, the synchrony dependence for inhibiting DCN activity holds not only on average, but essentially for all PCs.

Cell by cell comparison of the inhibitory effect on DCN activity of synchronous CS activity.

The percent change from baseline DCN activity following a CS was calculated for each PC based on all of its CSs or only those that were highly synchronous. The percent change values for each cell are connected by a line. The black circles indicate the averages of the two conditions. Error bars are SD. The highest synchrony levels were combined to ensure all cells in an experiment had at least 10 events: experiment A, levels 5-7/7; experiment C, levels 6-7/7; experiment D, levels 4-6/6. No levels were combined for experiment B.

https://doi.org/10.7554/eLife.40101.009

What is the effect of non-synchronized CS activity?

For the group analysis, the lowest synchrony level showed the presence of a substantial inhibition (Figure 7). This was also true for the individual cell analysis (data not shown). This is most likely due to the fact that one PC represented a significant fraction of the presynaptic group (0.14–0.25), and its firing a CS thus likely corresponds, on average, to the synchronous firing of a similarly significant fraction of all of the PCs that project to a DCN cell.

But does a single CS (i.e., non-synchronous CS activity) have a significant effect on DCN activity? To address this question, we estimated the inhibitory effect of an individual PC firing a CS by fitting regression lines to the data shown in Figure 7B–E (lines were fit to the solid symbols). The regression analysis produced good fits in all cases (r2: 0.85, 0.97, 0.91, and 0.99 for Exps. A-D). Next, the fraction of the population represented by a single PC was estimated to be 0.02–0.01 under the assumption that 50–100 PCs converge onto a DCN cell. Using these values, the regression equations yielded almost identical estimates of the inhibitory effect, which were not significantly different from zero (0.02: −15.3 ± 12.9%, p=0.10; 0.01: −14.6 ± 12.8%, p=0.11; n = 4 experiments). Moreover, for each experiment individually, the estimate of the inhibition from a single PC firing a CS was not significantly different from zero in three of four cases (p>0.05, n = 6, 6, 4; Exp. D was the exception, p=0.01, n = 4 synchrony levels). Thus, the action of an isolated CS appears to have at most a weak effect on DCN activity.

Effect of varying the definition of CS synchrony

The prior analyses used a ± 5 ms window to define synchronous CSs relative to the CSs in a reference PC. This duration was chosen because it is on the same order of precision as indicated by the width of the central peak in correlograms published in prior multielectrode studies of CS activity (Sasaki et al., 1989; Lang et al., 1996). Nevertheless, from the perspective of the DCN cell, a different definition (or definitions) of synchrony might be warranted. Thus, we investigated the dependence of the relationship between synchrony and the short-latency inhibition on the time window used to define synchrony.

We first assessed the number of high-synchrony events in differently sized time windows (Figure 9A, histograms). The histograms plot the number of events in a given window that did not meet the criteria to be in smaller (or larger) windows (window size was increased in steps of ±5 ms; thus, each histogram bin has a total duration of 10 ms). In all experiments, the initial bin, corresponding to the ±5 ms window, was the highest and was followed by a sharp drop in bin height. Each experiment's histogram also showed a secondary peak, which was usually small (experiment C being the exception) at windows of ~80–120 ms, which might reflect the ~10 Hz rhythmicity sometimes displayed by CS activity. Of note is that windows between 10 and 80 ms had fewer than the average number of events. The relative rarity of events for these time windows is also observable in shallower slopes of the cumulative summation (cusum) curves for these windows, particularly for experiment D (Figure 9A, gray trace). The value of the cusum curve at each time window is the sum of all events that occur in that time window plus each of the smaller time windows (i.e., the sum of the histogram bars preceding and including the one corresponding to the value of cusum time window).

Dependence of inhibitory effect on the precision of synchrony.

(A) Histograms of the number of highly synchronous events among presynaptic PCs as a function of the time window used to define synchrony for each of the experiments. Each histogram bar represents the number of events in that window but not in any shorter window. Each bar reflects a 5 ms increase in window size for both positive and negative latencies from the reference CS. Gray traces are the cumulative summation (cusum) curves. Dashed lines indicate twice the height of the first histogram bar (±5 ms time window) as plotted using the cusum (right) axis. (B) Plots of the inhibitory effect (percent change from baseline) on DCN activity of highly synchronous CS events as a function of the time window. Dashed lines indicate the overall inhibitory effect of all CSs in the presynaptic group PCs.

https://doi.org/10.7554/eLife.40101.010

The prevalence of the precise synchrony events (i.e., those with the ±5 ms window) and the relative rarity of events in the next smallest set of windows, suggests that these precisely synchronized CSs are a distinct pattern of activity that could have particular significance with regard to DCN activity. To test this possibility, we compared the inhibitory effect of the high synchrony events (i.e., those events in which all or almost all PCs in the presynaptic group fired CSs, as defined earlier for each experiment) when defined using different time windows (Figure 9B). Note that each time window includes all high synchrony CS events given the limits of that window, including those that meet the criteria for smaller windows. For large time windows, the curves in Figure 9B approximate the average inhibitory effect calculated for the entire population of CSs (dashed lines), as expected, because most CSs will meet the criteria to be included in these windows. Of more interest is the window at which the curves begin to rise, as this indicates when inhibitory effect due to synchronization of CS activity begins to dissipate. In three of the experiments (B, C, and D), the curves show an immediate or near immediate sharp upward deflection, starting to rise either between the ±5 and ±10 ms windows (exps. C and D) or between the ±10 and ±15 ms windows (Exp. B). This upward deflection is followed by a continued general rise that can be interrupted by regions where the slope of the curve is reduced or the curve even plateaus (this is easiest to see for exps. B and D). The plateaus occur because essentially no new synchrony events were added for those time windows, as is shown by the shallow to flat slopes of the cusum curves over the range of those windows (Figure 9A, gray traces). Experiment A shows a somewhat distinct pattern in that its curve doesn't start to rise consistently until the ±45 ms window (a possible explanation for this delayed rise is given in the Discussion). Nevertheless, once it does start rising, it shows the same general pattern as those of the other experiments.

The relative effectiveness of different CS events is contrasted in Figure 10A to show the effect of varying the definition of synchrony further. Comparing the values for the events involving a CS in only one presynaptic PC with those of high synchrony CS events, as defined with a ± 5 ms window, shows the much greater effectiveness of the synchronous CS activity (these are the same values as the left- and right-most points, respectively, on the filled symbol curves in Figure 7B–E). As was shown in Figure 9B, expanding the window for defining high synchrony events reduces their effectiveness. So, we expanded the time window such that the total number of synchronous events was twice that found in the ±5 ms window (Figure 10A, Cusum; note that this is the window at which the cusum curve crosses the dashed line in Figure 9A). In three of the four experiments (B, C, and D), the inhibition is reduced compared to that of the ±5 ms window. However, by definition, half of the events in this window are those from the ±5 ms window. Thus, we removed these most precisely synchronized events (i.e., those whose CSs all fell within the central portion of the cusum window) and calculated the inhibition caused by the remaining half of the events in the original Cusum window (Figure 10A, Donut cusum). For experiments B, C, and D, this caused a further reduction in the inhibitory effect. Indeed, for experiment D, the inhibition became equivalent to that of events in which only a single PC fired a CS, as shown in Figure 10B, where the inhibitory effect has been normalized by subtracting the value of these single PC events. In sum, in most cases, precisely synchronized CS activity causes a stronger inhibition of DCN activity than does more temporally dispersed activity.

Precisely synchronized CS events are responsible for inhibitory effect on DCN activity.

(A) Plot of different sets of CSs comparing their effectiveness in inhibiting DCN activity. Each symbol shows the data from a separate experiment. (B) Data in (A) replotted after subtracting the inhibitory effect of events in which only one PC fired a CS.

https://doi.org/10.7554/eLife.40101.011

CS-associated, longer latency modulation of DCN activity and CS synchrony

The rasters in Figure 6 show a slower modulation of DCN activity that is associated with CS activity in addition to the short latency effects just described. This slow,~1 Hz modulation was prominent in three experiments (exps. A, C, and D) and absent in one (Exp. B). For the experiments that showed the modulation, we investigated whether the increases in DCN activity before and after the CS were related to CS synchrony.

The periods of increased activity before and after the CS were determined from the CS-DCN correlogram generated using all CSs (Figure 11A). Here, the baseline was defined as the average level of the histogram between ±1000 ms, and the times at which each period started and ended were determined visually by when the histogram crossed this baseline (pre-CS period: −414 ± 60 ms to −136 ± 65 ms; long latency post-CS period: 146 ± 19 ms to 456 ± 40 ms; n = 3). The average change in activity during these periods relative to the baseline was plotted as a function of CS synchrony, as was done for the short-latency inhibition (Figure 11B,C). Although activity in the pre-CS period shows a trend toward increasing with synchrony, the correlation was not significant (r = 0.45, p=0.08, n = 16). However, a significant correlation was found for the long-latency post-CS period (r = 0.56, p=0.024, n = 16).

Other aspects of CS-associated modulation of DCN activity show weaker relationship to synchrony.

(A) CS-triggered correlogram of DCN activity generated from all CSs of all presynaptic PCs for experiment A. Horizontal dashed line indicates baseline level of histogram and vertical dashed lines indicate limits of long-latency modulation period. (B–C) Change from baseline activity during the pre-CS (B) and post-CS (C) periods as a function of synchrony. Each curve is the average effect for one experiment. Experiment B did not show clear long-latency modulation and thus was not plotted.

https://doi.org/10.7554/eLife.40101.012

Discussion

The present results provide new information about the relationship between PC and DCN activity, and in particular, they highlight the importance of CS activity in shaping cerebellar output. Specifically, the results show that PCs that project to the same DCN neuron are located within the same narrow, rostrocaudally-oriented strip of cortex and have relatively high levels of synchronous CS activity with each other. Moreover, the effectiveness of the CS activity of these PCs in inhibiting DCN activity was shown to be strongly correlated with the level of synchrony. Lastly, the precision of synchrony (i.e., the temporal dispersion of CSs at which their ability to inhibition DCN activity begins to decrease) was found to be ~10 ms in most cases.

Identification of synaptically-connected PC-DCN cell pairs

The significance of the present findings rests in large part on whether PC-DCN synaptic connections were correctly identified by the analysis of CS-triggered crosscorrelograms of DCN activity. Cross-correlograms are an accepted method for identifying synaptic connectivity (Perkel et al., 1967; Aertsen et al., 1989; Csicsvari et al., 1998). Here, the specific timing of the correlogram event used for identification (i.e., the start of decreased DCN activity at a latency from the onset of the CS that matches the conduction time of the PC-DCN pathway) is consistent with the CSs in the identified PC directly causing this inhibition, but there is still a possibility that this pattern reflects a spurious correlation instead of a true synaptic connection.

However, consideration of the known anatomy of the involved circuits when combined with the present findings on the spatial distribution of the presynaptic groups, provides further support that the correlograms have identified synaptic connections. Specifically, every group of presynaptic PCs had a spatially-restricted distribution in the mediolateral direction, which follows the known topography of the PC-DCN projection in which each DCN region receives input from a narrow rostrocaudally-running cortical strip (Voogd and Bigaré, 1980).

In addition, the better maintenance of synchrony with increasing mediolateral separation between PCs in the presynaptic group than in the rest of the PCs in the array is also consistent with their identification as being presynaptic to the same DCN cell, if one considers the relationship between zebrin bands, the topography of the PC-DCN projection, and CS synchrony. First, anatomical studies show that PC input to each DCN region (and hence to each DCN neuron) essentially comes from one zebrin band (Sugihara and Shinoda, 2007; Chung et al., 2009; Sugihara, 2011), so that if the presynaptic group PCs are correctly identified they must all belong to the same zebrin band. Second, PCs within the same zebrin band show high levels of synchrony that are maintained with ML separation, whereas synchrony drops more rapidly with ML separation when PCs are located in neighboring bands (Sugihara et al., 2007). Thus, the better maintained synchrony with mediolateral separation for the presynaptic PC groups is consistent with these PCs all being in the same zebrin band, and thus consistent with all of them projecting to the same DCN cell.

It is also worth considering whether DCN afferents other than PCs could have produced an inhibition of DCN activity precisely timed to the onset of a CS, as excluding them would provide further support for the identification process. In this regard, other than PCs, collaterals from the mossy fibers and olivocerebellar axons are the major sources of input to the DCN. They are both excitatory, implying that they would have to show a reduction in activity that is precisely timed to affect the DCN within several milliseconds of a CS. The olivocerebellar system is, by definition, active when a CS occurs, and so reduction in the activity of these collaterals cannot be driving the decrease in DCN activity. It is also implausible that a reduction in mossy fiber activity precisely timed to within a few milliseconds of a CS underlies this reduction.

Another possibility is that changes in SS activity are responsible for the short-latency inhibition that we used to identify PCs as presynaptic to a DCN cell. This possibility can also be excluded, as it is well-known that there is a pause in SSs following a CS, which would be expected to result in an increase in DCN activity, contrary to what was observed. Note that the various types of modulation of SS activity that can follow the SS pause would have latencies that would make it impossible for them to cause this initial inhibition of DCN activity, but they could contribute to the overall response of the DCN to CSs, as is discussed later.

In sum, it seems highly likely that the decrease in DCN activity immediately following the CS is due to CS activity and thus identifies the action of PCs on the DCN cells. However, we need to consider whether this decrease specifically identifies a synaptic connection between the PC and DCN cell being recorded or whether it could instead reflect the connections from other PCs to the DCN cell. This, in turn, requires consideration of two cases. First, olivary axons branch to innervate about 7 PCs in rats (Schild, 1970), and so, in theory, the correlogram pattern used to identify PC-DCN connections could reflect the connection to the DCN cell by any one of the PCs contacted by the same olivary axon as the PC being recorded, in addition to (or instead of) a connection by the recorded PC itself. However, in this case, the correlogram would still be identifying a PC-DCN synaptic connection, and the CSs of the two PCs (the one being recorded and the hypothetical one innervated by another branch of the olivary axon) would be nearly perfectly correlated. Moreover, olivary axon branches are generally confined to the same narrow cortical strip (Sugihara et al., 2001), but cf. Ekerot and Larson, 1982), and so the finding of a constricted spatial distribution of the presynaptic group along the mediolateral axis would also remain valid. In sum, even if this scenario were to occur, it would have little impact on the interpretation of the results.

The second possibility reflects the fact that CSs are correlated between PCs because of gap junction coupling between inferior olivary (IO) neurons. However, this correlation, while statistically highly significant, is far less than perfect (e.g., r-values are mostly <0.1 when PCs are in the same rostrocaudal band and much less otherwise, see Figure 4A). More importantly, the central peak of the crosscorrelogram of CSs in two PCs can be 5–10 ms or wider. Given these characteristics, the inhibition of DCN activity should start before the CS and no precisely timed increase in it should be observed following the CS if it were due to correlated CS activity from other PCs instead of being directly due to the CSs of the PC being recorded. Indeed, correlograms with this pattern of activity did occur often and were excluded.

In sum, the evidence supports interpreting CS-triggered correlograms that show an onset of decreased DCN activity within several milliseconds following the onset of a CS as evidence for a synaptic connection from the recorded PC (or essentially equivalently from a PC receiving input from another branch of the olivary axon that innervates the recorded PC) to the DCN cell.

Technical limitations

The major limitation of the present study is the low number of DCN cells for which a group of PCs in the recording array could be identified as presynaptic. This limits the generalizability of the conclusions in terms of whether the results apply to all categories of DCN neurons. Nevertheless, that all four DCN cells for which such a group of PCs were identified showed a similar, strong relationship between CS synchrony and the amplitude of the inhibition of DCN activity suggests that this relationship is not rare. It should be noted that the analyses of each of these cases rests on the analysis of spike trains consisting of thousands of CSs (~2500–12,000) and tens of thousands of DCN spikes (~30,000–50,000), and thus rests on a large amount of data.

A second issue that should be addressed is the use of anesthesia, which was required, given the recording methodology and the need to record both PCs and DCN cells for long periods. Differences in cerebellar activity between ketamine anesthetized and awake animals have been reported (e.g., Schonewille et al., 2006). This necessitates caution when drawing conclusions from the anesthetized preparation to the awake animal. However, the critical issue is not whether the patterns of activity in the two conditions match exactly, but rather whether the system's operating characteristics in the two states are similar. Indeed, in many ways, PCs and DCN neurons have similar firing characteristics in the two states, though there are some differences, as just mentioned. For example, CS, SS and DCN firing rates are similar in both conditions. In particular, the reported spontaneous firing rates for DCN cells in alert animals vary widely, both within and between studies; overall, rates range from 10 Hz to greater than 100 Hz (e.g., Thach, 1968; Burton and Onoda, 1978; Harvey et al., 1979; Chapman et al., 1986; Fortier et al., 1989; Gruart et al., 2000; Sarnaik and Raman, 2018). The data set from our previous report showed a similar range for the overall DCN population, as did the population of DCN cells for which a presynaptic PC was identified (see Figure 2A,B of Blenkinsop and Lang, 2011).

More importantly, the patterns of CS synchrony are similar (Lang et al., 1999), and the PC-DCN synapse is GABAergic, and thus should not be greatly influenced by ketamine, whose primary action is as an NMDA blocker. Note that ketamine has a negligible effect on GABA-A receptors, except those containing the alpha6 subunit (Hevers et al., 2008), but such receptors are found only on granule cells in the rat brain (Wisden et al., 1992). In sum, under ketamine, PC CS activity is similar to that in the awake rodent, as likely is the efficacy of the PC-DCN synapse.

Finally, several studies have demonstrated that changes in CS synchrony patterns occur during movement (Welsh et al., 1995; Mukamel et al., 2009), and CS activity has been inferred to influence interpositus firing patterns during conditioned eyeblinks (Ten Brinke et al., 2017), suggesting that synchrony is a physiologically relevant parameter for determining cerebellar output (i.e., DCN activity).

Synchrony may be needed to modulate DCN activity

Our results indicate that CSs need to be synchronized to be able to affect DCN activity. They are consistent with previous work showing that the tonic SS activity of individual PCs has a ‘near negligible’ effect on DCN cells (Bengtsson et al., 2011), and with in vitro studies that suggest that DCN firing is more effectively phase locked by synchronized PC activity (Person and Raman, 2012; Person and Raman, 2011). Nevertheless, it is possible that the CS activity, unlike the SS activity, of an individual PC could significantly influence DCN firing, because a CS often triggers a high frequency burst of axonal spikes (Ito and Simpson, 1971; Khaliq and Raman, 2005; Monsivais et al., 2005). However, at best, this mechanism appears to produce a minimal effect, based on our extrapolation of the relationship between synchronous CS activity and its inhibitory effect on DCN firing down to the effect caused by an individual PC firing a CS (Figure 7).

The inability of asynchronous activity from an individual PC to influence the DCN cells it contacts is understandable given the high firing rates of PCs and the convergence of at least dozens of PCs onto each DCN neuron. The need for synchrony is also consistent with the electrophysiology of most classes of DCN cells and the characteristics of their synaptic input. In particular, under in vitro conditions, the membrane time constant is reported to be ~35 ms for nonGABAergic DCN cells (Uusisaari et al., 2007); however, most DCN cells are subjected to an intense ongoing barrage of synaptic activity from the dozens of PCs that converge onto each one, which would reduce this value. Indeed, in decerebrate cats, a value of ~7 ms was reported (Bengtsson et al., 2011). Furthermore, the GABAergic IPSCs of most classes of DCN cells are extremely brief, lasting only ~2 ms on average (Person and Raman, 2011). In sum, the presynaptic and intrinsic properties of most classes of DCN cells suggest that they should function as coincidence detectors, consistent with our finding that precisely synchronized (within 5–10 ms) CSs are most effective in inhibiting DCN activity.

Although the above discussion generally applies to most classes of DCN neurons, there are exceptions. For example, DCN neurons that project to the IO have very long IPSC time constants (Najac and Raman, 2015), and likely have longer membrane time constants than most other DCN cells as well (Uusisaari et al., 2007). Thus, modulation of the activity of these cells would not be expected to require precisely synchronized synaptic input. In this regard, it is worth noting that one experiment (Exp. A) was distinct from the others, in that far less precise CS synchrony was needed to inhibit DCN activity strongly. Indeed, in this case, CSs dispersed over about 50 ms still produced as strong DCN inhibition as precisely synchronized CSs. It is tempting to speculate that in this case we were recording a DCN-IO projection neuron. Consistent with this, the firing rate of the DCN cell in Exp. A was 29.6 Hz, which is relatively low for a DCN neuron in vivo and consistent with the lower firing rates reported for DCN-IO neurons, and small GABAergic DCN neurons generally, recorded in vitro (Uusisaari et al., 2007; Najac and Raman, 2015). However, additional experiments with larger numbers of DCN recordings and where IO projecting neurons are identified as such (e.g., by antidromic stimulation) are needed to address this possibility.

CS synchrony: direct and indirect actions on the DCN

Exactly how synchronous CS activity translates into changes in DCN activity is not straightforward, in part because CSs are associated with changes in SS activity and therefore could influence the DCN both directly, because of the axonal spikes triggered by the CS, and indirectly, via changes in SS activity. The latter includes the pause in SSs following a CS and the further modulation of SS activity following the pause (Granit and Phillips, 1956; Bell and Grimm, 1969; Murphy and Sabah, 1970; Bloedel and Roberts, 1971; Latham and Paul, 1971; Burg and Rubia, 1972; Ebner and Bloedel, 1981a; Ebner and Bloedel, 1981b; McDevitt et al., 1982; Bloedel et al., 1983; Ebner et al., 1983; Mano et al., 1986; Sato et al., 1992; Burroughs et al., 2017; Tang et al., 2017). Moreover, these direct and indirect effects are not necessarily mutually exclusive, and indeed, we would argue that both are likely needed to explain fully the modulation of DCN activity we observed.

To begin, the short-latency inhibition indicates that a direct effect of CS activity dominates the initial response of the DCN cell following a CS. In contrast, as mentioned earlier, if the pause in SSs, which occurs during the initial 20–30 ms of the short-latency inhibition, were the dominant mechanism during this period, the resulting disinhibition would have caused an increase in DCN activity, contrary to what was observed. Moreover, a positive correlation between synchrony and DCN activity, if anything, should have been present rather than the strong negative correlation that was observed. Thus, at least the initial portion of the short-latency inhibition appears to be due to the axonal spikes triggered by the CSs.

The duration of the short-latency inhibition (50–100 ms), however, raises the question of whether it can be attributed solely to the direct effect of CS activity, given the short time course of GABAergic IPSCs and IPSPs in DCN neurons (Person and Raman, 2011). The direct effect of the CS, however, is likely to be prolonged by several factors. A CS can have as many as eight spikelets in addition to its initial spike, and they occur over a period of ~20 ms (Tang et al., 2017). Although not all spikelets are associated with axonal spikes (Ito and Simpson, 1971; Khaliq and Raman, 2005; Monsivais et al., 2005), the extended nature of this activity means that a DCN cell could be subjected to recurring IPSCs for a period of ~20 ms. Furthermore, we need to consider the precision of the synchrony, as the width of the central peak in CS-CS correlograms typically is on the order of 5–10 ms, and in the current analysis, the narrowest time window (±5 ms) allowed a scatter of up to 10 ms. Combining these factors suggests that a synchronous CS event might lead to IPSCs bombarding a DCN cell over some 30 ms, albeit with decreasing numbers toward the ends of this range. In addition, although the strongest inhibition would occur while the IPSCs were ongoing, the longer duration of the IPSPs would lengthen the inhibitory period for perhaps another 5–10 ms, giving a total duration of ~40 ms, which is close to that of the lower limit of the short-latency inhibition. It is also worth noting that DCN neurons have plateau potentials and can exhibit bistability (Jahnsen, 1986; Llinás and Mühlethaler, 1988; Han et al., 2014), and thus the powerful inhibition caused by synchronous CS activity could shift the DCN cell to a less excitable state that exceeds the duration of the IPSPs.

Even taking all of these factors into account, the duration of the short-latency inhibition appears to exceed what is easily explained by a direct action of CS activity. Thus, we propose that the later stages of the short-latency inhibition of DCN activity also reflect an indirect effect of CS activity to synchronize the modulation of SS activity. Specifically, following its post-CS pause, SS activity often rebounds to rates that are significantly higher than baseline for 50–100 ms (Tang et al., 2017). This period of increased SS activity coincides with the second half of the short-latency inhibition period and would thus contribute to maintaining the inhibition. This rebound period is followed by a longer period of mildly decreased SS activity that lasts until ~600 ms after the CS (Tang et al., 2017), a period that is similar to that of the long-latency excitation in DCN cells. Moreover, as was the case for the short-latency inhibition, the amplitude of the post-CS long-latency excitation was correlated with the level of CS synchrony, as it should be if it depends on synchronization of PC activity.

CS synchrony may be key to modulating cerebellar output

The present results combined with those of previous studies (Bengtsson et al., 2011; Person and Raman, 2012; Person and Raman, 2011) suggest that synchrony is a critical parameter of PC influence on the DCN. This in turn points to a central role for the olivocerebellar system in the control of cerebellar output, as it is organized to generate synchronous CS activity across large and varied ensembles of PCs (Sasaki et al., 1989; Sugihara et al., 1993; Lang et al., 1996; Lang, 2001; Lang, 2002). By contrast, spontaneous SS activity shows comparatively low levels of synchronization between small numbers of neighboring PCs (Bell and Grimm, 1969; Heck et al., 2007; Wise et al., 2010). Moreover, the most obvious mechanism for generating synchronous SS activity (that doesn't involve the olivocerebellar system), shared parallel fiber input, would induce synchronous activity among relatively few PCs that would converge to the same DCN cell, given the spatial alignment of the parallel fibers. In contrast, the olivocerebellar system causes synchronous CS activity in rostrocaudal strips that can extend across multiple cerebellar lobules (Sugihara et al., 1993; De Zeeuw et al., 1996; Yamamoto et al., 2001), and thus among PCs likely to converge onto the same DCN cells. That olivocerebellar synchrony may be the key control parameter of DCN activity does not, however, exclude a role for SSs. Rather, as discussed above, synchronous CS activity may act both directly and by synchronizing the modulation of SS activity.

Materials and methods

Experiments were performed in accordance with the National Institute of Health Guide for the Care and Use of Laboratory Animals. Experimental protocols were approved by the Institutional Animal Care and Use Committee of New York University School of Medicine. All surgical and recording procedures were performed under general anesthesia, and every effort was made to minimize suffering.

General surgical and recording procedures

Request a detailed protocol

Female Sprague-Dawley rats (225–300 g) were initially anesthetized with ketamine (100 mg/kg, ip) and xylazine (8 mg/kg, ip). Supplemental anesthetic was supplied continuously via a femoral catheter to maintain a constant depth of anesthesia. The depth of anesthesia was assessed by paw pinch and by the absence of spontaneous movements. Rectal temperature was maintained at 37°C using a heating pad connected to a temperature control system. To gain access to the cerebellum, animals were placed in a stereotaxic frame, and a craniotomy was performed to expose the posterior lobe of the cerebellum. The overlying dura mater was then removed, and the cortical surface was stabilized and protected by covering it with a platform constructed from an electron microscope grid that was pre-embedded in a thin sheet of silicone rubber and supported by tungsten rods. Gelfoam pieces soaked in Ringers solution were used to cover any remaining brain surfaces. Dental cement was then used to fix the platform to the skull and to seal the craniotomy.

Extracellular recordings of CS activity were made using a multiple electrode technique in which glass microelectrodes (50/50 mixture of 2M NaCl solution and glycerol) were implanted by driving them through the rubber of the spaces in the grid using a micromanipulator. Electrodes were implanted into the molecular layer to a depth of 100–150 µm below the folial surface, where CSs but not SSs can be observed. To record SSs, the electrodes need to be implanted to a depth of ~225–250 µm, close to the PC layer in rats. While also obtaining SS recordings is desirable, there are several reasons why this was not done. These include a much longer implantation process, greater overall damage to the cortex, more difficult spike sorting for analyses, and less stability during recording (i.e., the electrode is much more likely to damage or kill the cell being recorded because of its closer proximity to the cell body).

Electrodes were implanted with a spacing of ~250 µm using the rows and columns of the electron microscope grid as a guide. For further details on the multielectrode technique, see (Sasaki et al., 1989; Lang, 2018). Once an array of electrodes was implanted to crus 2a, a single glass microelectrode (2M NaCl) was lowered under stereotaxic guidance through either the crus 1b or the paramedian lobule to the DCN, where recordings of well-isolated single units were obtained. Upon good isolation of a DCN cell, the activity was typically recorded for 20 min. At the end of the recording session, the animal was perfused under deep anesthesia and the cerebellum was removed for histological processing.

All neuronal activity was recorded using a multichannel recording system (MultiChannel Systems MCS GmbH, Reutlingen, Germany) with a 25 kHz/channel sampling rate, gain of 1000x, and band pass filters set at 0.2–8.0 kHz. Spike sorting of CSs and DCN activity was done offline using routines written in Igor Pro (Wavemetrics, Lake Oswego, OR, USA).

Data analysis

In the text, all mean values are given ±one standard deviation. Populations were tested for normality using the Kolmogorov-Smirnov (KS) goodness-of-fit test. For normally distributed populations, paired or unpaired t-tests, as appropriate, were used to compare means. When differently sized populations were compared, t-values and degrees of freedom were calculated using formulas that account for differences in population size (Welch's approximate t) (Zar, 1999). For non-normally distributed populations a nonparametric test was used and is indicated in the text. Correlation coefficients (Pearson's r) were tested for significance using the formulas given in (Zar, 1999).

Identification of monosynaptically-connected PC-DCN cell pairs

Request a detailed protocol

CS-triggered histograms of DCN activity were used to identify PCs that had a monosynaptic connection to the DCN cell being recorded, as described previously (Blenkinsop and Lang, 2011). Identification of a PC-DCN connection rested on the existence of a significant negative deflection in the CS-triggered correlograms within 5 ms after CS onset, consistent with the conduction time of the PC-DCN pathway and inhibitory nature of PCs. Significance was assessed by testing whether the slopes of the histogram just prior and following the CS were different (Zar, 1999). Specifically, the period preceding the CS was −10 to 0 ms in almost all cases, but was −15 to −5 ms if an excitatory response was visible in the histogram at the time of the CS. The post-CS period was 3 to 13 ms. On this basis all PCs in a recording array were divided into presynaptic and non-synaptic groups (i.e., PCs that were connected or were not connected to the DCN cell being recorded).

Synchrony analysis

Request a detailed protocol

To determine the average CS synchrony level of a pair of PCs for the entire recording session, a cross-correlation coefficient was calculated by treating the spike trains as zero-one variables (Gerstein and Kiang, 1960; Sasaki et al., 1989). The spike train of a cell was represented by X(i), where i represented the time step (i = 1, 2,. .., N). X(i)=1 if a CS onset occurred in the ith time bin, otherwise X(i)=0. Y(i) represented the spike train of the reference cell and was defined the same as X(i). The synchrony level, C(0), was then calculated as

C(0)=[i=1NV(i)*W(i)]/i=1NV(i)2*i=1NW(i)2

where V(i) and W(i) were

V(i)=X(i)j=1NX(j)N,W(i)=Y(i)j=1NY(j)N

Note that C(0) ranges between ±1, having a value of 1 when the two spike trains are perfectly synchronized and a value of 0 when the spikes of the two cells are independent of each other. In previous work we have used simulated, randomized and time-reversed spike trains to assess the statistical significance of the experimentally-observed C(0) values and in all cases the values for the experimental data are well above the 99th percentiles for the various test data (Sugihara et al., 1993; Lang et al., 1996).

It was also necessary to determine the synchronization level in the presynaptic group at the times of individual CSs in each PC of the group. For this purpose, each PC served as the reference cell, and the time of the onset of each of its CSs marked the center of a time window that defined the event. The synchrony level for each such event was then defined as the fraction of PCs in the presynaptic group that fired CSs within the specific time window. A ± 5 ms window was used except where stated otherwise. Thus, all CSs could be assigned a synchrony level ranging between 1/x (no CSs occurred in other PCs) and 1 (all PCs fired a CS within the time window), where x is the number of PCs in the presynaptic group.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
    Reliability and state dependence of pyramidal cell-interneuron synapses in the hippocampus: an ensemble approach in the behaving rat
    1. J Csicsvari
    2. H Hirase
    3. A Czurko
    4. G Buzsáki
    (1998)
    Neuron 21:179–189.
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
    Multielectrode arrays for recording complex spike activity
    1. EJ Lang
    (2018)
    In: R. V Sillitoe, editors. Extracellular Recording Approaches - 2018. Humana Press. pp. 73–85.
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
    Topographical distribution of olivary and corticonuclear fibers in the cerebellum
    1. J Voogd
    2. F Bigaré
    (1980)
    In: J Courville, C de Montigny, Y Lamarre, editors. The Inferior Olivary Nucleus: Anatomy and Physiology. New York: Raven Press. pp. 207–234.
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
    Biostatistical Analysis (Fourth)
    1. JH Zar
    (1999)
    Pearson Education.

Decision letter

  1. Vatsala Thirumalai
    Reviewing Editor; National Centre for Biological Sciences, India
  2. Richard B Ivry
    Senior Editor; University of California, Berkeley, United States

In the interests of transparency, eLife includes the editorial decision letter, peer reviews, and accompanying author responses.

[Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed.]

Thank you for submitting your article "Complex spike synchrony dependent modulation of rat deep cerebellar nuclear activity" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Vatsala Thirumalai as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Richard Ivry as the Senior Editor. The following individual involved in review of your submission has agreed to reveal her identity: Chris I De Zeeuw (Reviewer #3). Reviewer #2 remains anonymous.

The Reviewing Editor has highlighted the concerns that require revision and/or responses, and we have included the separate reviews below for your consideration. If you have any questions, please do not hesitate to contact us.

Summary:

This manuscript presents a nice meta-analysis of synaptically connected Purkinje neurons and deep cerebellar nuclear cells. Importantly, it presents the idea that inhibition of spiking in nuclear cells is controlled by the synchrony of complex spikes in the presynaptic Purkinje neuron population. All three reviewers agreed that this is a significant finding and appreciate the authors for this insight. However, all three reviewers also had concerns that need to be addressed and suggestions for improving the manuscript.

Major concerns:

One of the concerns is the low number of DCN cells recorded from, especially because of the heterogeneity of DCN cells. While the efforts required to find PC-DCN pairs are akin to finding a needle in a haystack (x4), it is important that the authors address the weakness of low n's in their discussion of the results. There are also some other caveats (such as indirect synaptic connections, the use of anaesthesia, etc.,), which should be addressed by the authors in the discussion.

In general, reviewers felt that the manuscript could be rewritten such that results are presented thoughtfully and even-handedly, with an awareness of the strengths and limitations that is conveyed to the expert and non-expert reader. Reviewers also had several suggestions to improve the presentation of figures and overall lucidity of the manuscript and these specific suggestions can be found in the individual reviews appended below.

Separate reviews (please respond to each point):

Reviewer #1:

This manuscript argues that the synchrony of complex spikes in the Purkinje neuron cohort projecting to the same DCN neuron might hold the key for controlling cerebellar output. The manuscript is written in clear style and the data are analyzed and presented well. My biggest worry is that the data presented are from only four or five DCN cells (Results paragraph one says five cells, but four experiments are presented. Authors need to clarify how data from two DCN cells in the same animal were combined). In their dataset, authors could identify 4-7 PCs to be synaptically connected to the DCN cell being recorded. This combined with the low number of highly synchronous CS inputs in each experiment (Figure 6B), places the findings on not-so-solid ground. In addition, of the four, one DCN cell (A) did not show the tight synchrony-dependent modulation seen in the other three cells (Figure 9B and 10). The authors speculate that this cell could be an IO projecting DCN but this idea is not tested. With this in mind, I am not certain that the main point the authors push, i.e. that CS synchrony in a 10ms window is the most important parameter for DCN output, can be accepted. Some specific points:

1) Figure 1 is not useful. Either remove or include PC recordings and a Venn diagram of cell numbers discussed at the end of paragraph one of the Results.

2) Will be useful to indicate the long latency modulations of DCN activity in Figure 2, with an arrow or grey bar.

3) In Figure 3, authors show lines to indicate 95th and 99th percentile ML separation and state that the synaptic group value is above the 99th percentile. This should be the 5th and 1st percentile and "below the 1st percentile" given that the synaptic group is placed more closely than the others.

4) Figure 4A: SD or SEM? Prefer that the authors would show the individual synchrony values of all PC's in that experiment as scatter with an indication of the mean. Similar plots can be made for all four experiments. Figure 4B does not make much sense to me. What does it mean that the ratio is several-fold over very long distances?

5) Figure 11B: Authors argue that the long latency modulation of DCN activity is also synchrony dependent but the effect is weak and variable. One of the four cells recorded did not show this modulation. This point appears weak and not related to the main point of the manuscript.

6) Results paragraph two, final sentence: What is the larger sample used in this comparison?

7) Subsection “Inhibition of DCN activity varies with level of CS synchrony” paragraph five: insert "Figure 7F"

8) Sixth paragraph of the same section: Could singular CS's still be high synchrony events synchronized with PCs not being recorded from?

9) Subsection “Synchrony may be needed to modulate DCN activity”: "Our results indicate that CSs need to be synchronized to be able to affect DCN activity". Given that singular CSs also show about 20% inhibition, one would argue that they are also able to affect DCN activity, albeit at a lower level relative to the synchronous ones. Again, as noted above, it is likely that the singular CSs are in fact a mixed population of events of widely varying synchronicity. This point needs to be brought out in the discussion.

10) Subsection “CS synchrony: direct and indirect actions on the DCN” paragraph four: later

Reviewer #2:

This manuscript is a meta-analysis of multielectrode data from recordings of Purkinje cells (PCs) and deep cerebellar nuclei (DCN) in anesthetized rats. The authors correlate activity of PC complex spikes and DCN spikes to find likely connected cells and then explore the effects of synchronized PC complex spikes on 4 DCN cells. The greater the degree of complex spike synchrony, the greater the inhibition of DCN cells. The manuscript has useful information and the authors are careful to report all the exceptions and special cases, but the text doesn't always make it clear what the relevant points are or what aspects of the data the authors wish to stress. There are relatively few points along the way through 11 figures where the results and interpretations are synthesized. It would be helpful if the text were re-written to include a few more summary sentences. Likewise, there are many figures with relatively few panels, but sometimes the transformations of the raw data aren't immediately obvious. Possibly some further illustration would be helpful. Specifics are given below.

Comments.

1) Figure 1 has example traces of 4 DCN cells, but nothing else, and it is hard to know what we are looking at them for. Are these the 4 cells that are identified in Figure 3 and carried through the paper? Figure 2 also shows 4 histograms from DCN cells but, in this case, 2 plots are from one DCN cell. PC spikes and PC synchrony are never actually illustrated. It could be helpful to see some of the PC traces (and maybe combine Figures 1 and 2 and relate them to 3) so that the raw data are clearer.

2) Figure 3 is used to make the argument that the likelihood of finding converging PCs decreases with mediolateral (M-L) but not rostrocaudal (R-C) distance. This argument is convincing in the M-L dimension but less so in the R-C dimension, since the multielectrode array has 7-9 electrodes M-L but only 3-4 R-C, and in Figure 3E and F the converging PCs have about equivalent mean M-L and R-C separations. Should this analysis be curtailed to talk about clustered PCs rather that PCs in an R-C plane?

3) The Synchrony Index described in the Materials and methods and used first in Figure 4 could be explained more clearly and intuitively, possibly even illustrated. The equation indicates that it is the summed product of time-matched deviations from the mean for each of two cells, divided by the root of the product of the summed-square-deviations. It is clear that when both signals go high simultaneously (i.e., synchronize), The Synchrony Index value will increase, but it is not immediately obvious what the range of values is expected to be, how this will vary with different levels of background firing, etc. It would be useful if the authors could help calibrate the reader, so that the values of Figure 4 (0.12, 0.06) can be placed in perspective. This figure might be a useful place to illustrate the successive transformations of the data.

4) Subsection “Inhibition of DCN activity varies with level of CS synchrony”; “Although singular CSs had the weakest inhibitory effect, they still reflect a significant level of synchronization….” It is not clear on what basis the authors say that the “singular” CSs, which seems to mean non-synchronized CSs, reflect synchrony. The argument about the 50-100 converging PCs is hard to follow, especially when the concluding sentence indicates that the effect is not different from no synchrony. Please clarify.

5) As far as this reader can tell, the central, most important finding of the manuscript is Figure 7F, which shows that for all four DCN cells, there is a strong correlation with steep slope of the amount of inhibition as a function of synchrony level. This key panel is not called in the text and hardly stands out in the illustrations. With so many figures and a relatively monotonic (without emphasis) presentation, it is possible that it will be missed. Some rewriting would help make this result more salient.

6) The 7F panel is complemented by Figure 8. It switches to illustrating the inhibitory efficacy of PCs based on whether they are in a synchronous group, but the authors just refer to this as a “cell-by-cell” comparison, so it takes several readings before one realizes that this is a demonstration from the “perspective” of the PCs, which is quite important. Please rephrase for clarity. (Perhaps the authors could consider putting this key panel with Figure 7F somehow so that they stand out as a central figure?)

7) Discussion; “these [convergent] PCs have highly synchronized CS activity…” Is what is meant here that they have a high probability of synchronizing (which does not appear to have been tested; or was it?) or that when they synchronize they do so with <5 ms jitter (which is stated again in the last clause)?

Minor points.

1) Results; “In a few cases (n=4)” Please indicate n=4 out of how many cases.

2) Figure 2, please indicate bin width in legend.

3) Figure 4B, perhaps different line patterns (solid, dashed, dotted) could help distinguish the four curves.

4) The use of the word “singular” to describe non-synchronous CS's is rather non-standard and unintuitive. Perhaps a definition could be given at first use.

5) Figure 5, it would be useful to indicate the number of PCs in each group in the legend or on the plots.

6) Subsection “Inhibition of DCN activity varies with level of CS synchrony”; (Figure 7) “the baseline activity in all histograms was normalized for sweep number…” It is not clear what analysis was done. Please clarify.

7) Please define “cusum.”

8) The word “axonic” should probably be “axonal.”

Reviewer #3:

The authors of the manuscript entitled "Complex spike synchrony dependent modulation of rat deep cerebellar nuclear activity" investigated how Purkinje cell complex spike activity and synchrony affects activity in their main downstream target, the cerebellar nuclei (CN). By doing simultaneously recordings of Purkinje cells (PCs) in the cerebellar cortex and of CN neurons, they describe that PCs that are putatively presynaptic to recorded CN cells are located in rostrocaudal strips. In addition, they describe that CN cells show specific patterns of short-latency responses to complex spike activity of which the strength depended on the degree of complex spike synchrony. The paper addresses an important and timely topic using a series of detailed in depth analyses.

I have several suggestions to improve the paper.

1) The manuscript largely lacks raw data. Only in Figure 1, traces are shown but at a very compressed time scale. Please expand.

2) The number of CN cells from which they recorded in a paired format with PCs is low (n = 4/5). Given that there are many different cell types in the nuclei (e.g., Dulac and Uusisaari studies), it may be worthwhile to increase the number of CN cells. In addition, the number of high-synchrony events per cell is low. This makes the conclusions, although likely to be accurate, somewhat preliminary.

3) The authors discriminate between directly coupled Purkinje cell – DCN neuron pairs and non-synaptic pairs. Their criteria for establishing synaptic connections are not mentioned in the current paper, but a reference is made to a previous publication (Blenkinsop and Lang, 2011). Also in that publication, the criteria are not overly clear. In the latter study, the impact of indirectly coupled pairs has also been described (see Figure 4C of Blenkinsop and Lang, 2011). As the authors discuss (subsection “Identification of synaptically-connected PC-DCN cell pairs”), identification of synaptic connections is difficult and could be compromised by synchrony between Purkinje cells. As Purkinje cells with a high level of synchrony might be considered functionally equivalent, the authors rightfully remark that it is questionable whether this would affect their conclusions. However, the abovementioned, indirect connections (see Figure 4C of Blenkisop and Lang, 2011), could also affect the patterns. Thus, it is crucial to understand which criteria were used to define mono-synaptic connections and to what extent indirect connections could contribute to the findings.

4) Regarding the indirect connections: the authors focus on the synaptically connected pairs, but was there also an impact of complex spike firing in the non-synaptic group? According to Figure 4A (of the present manuscript), also these cells do show synchronous complex spike firing, albeit at a lower level, so that an effect would be predicted. Please discuss.

5) The experiments were performed in anesthetized animals (with ketamine/xylazine). Since this anesthesia can affect cerebellar firing patterns including patterns of rhythmicity and synchrony (e.g., Schonewille et al., 2006), it will be worthwhile to replicate some recordings in awake behaving animals. This will increase the functional relevance of the study.

6) In addition, or alternatively, the authors may want to discuss the Ten Brinke studies (2015 and 2017) in which complex spikes and CN activity were recorded during eyeblink conditioning. In these studies the same patterns were detected with separated single unit recordings of PCs and CN cells in awake behaving animals as described with simultaneous dual recordings in the current study. This would highlight the achievement of the current dual recordings of the current Lang study, while clarifying the putative relevance of it during natural behavior. In fact, the CS synchrony and impact thereof in the CN appear to be enhanced during motor learning.

7) The authors relied on 1 ms binning of their data. The data underlying Figure 7 depend on the duration of the inhibition, which is defined as ending when the 3-point average reaches baseline levels again. This seems a very noise-prone analysis, especially for relatively rare occurrences of high-synchrony (see Figure 6). The open symbols in Figure 7B-E are therefore questionable; the closed ones a bit more reliable, but even here an estimate of the baseline firing during only 50 or 25 ms (the choice of either which value being not really justified), might be quite arbitrarily. Overall, the conclusion that more synchrony leads to more inhibition seems plausible, but the statistical evaluation could be improved (see also below). One should notice, again, that all these conclusions are based upon only four cells.

Additional data files and statistical comments:

The authors report using paired and unpaired t-tests in addition to Pearson correlation analyses. However, their reasoning for choosing these tests is unclear. Based on the low n, inequality of sample sizes (e.g. the difference in number of low vs high synchrony events) and questionable normality of distributions, one may also have to consider non-parametric testing; this reviewer is not a statistician, but please ask advice from a professional to make sure the current approach is correct or not. For sure, the authors should include a detailed description of tests used (and for what reasons they were used) and also include the specific outcomes like t-values, degrees of freedom etc. Evaluation of the statistics may also indicate how many pairs of PC-CN cells need to be illustrated in the current paper to back up the conclusions. So this reviewer has no real doubts about the conclusions of the current manuscript, but they need to be properly backed with solid statistics to stand the test of time.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your article "Complex spike synchrony dependent modulation of rat deep cerebellar nuclear activity" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Vatsala Thirumalai as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Richard Ivry as the Senior Editor. The following individual involved in review of your submission has agreed to reveal his identity: Chris I De Zeeuw (Reviewer #3).

The manuscript is indeed written more accessibly than in the previous submission, but there are some remaining issues that we feel should be addressed before your paper is published, as outlined below:

1) The authors claim (subsection “Technical limitations”) that DCN firing rates are the same with and without anesthesia. No citation is given. The 34 Hz DCN firing rate seen here seems relatively low. The main line of the authors' arguments doesn't rely on claiming that there is no effect of anesthesia, and so unless there is explicit evidence, it is probably better not to say that anesthesia does not affect the firing rates.

2) Figure 1: Reviewers wanted PC recordings to be shown here and this has not been done or addressed in the response.

3) Figure 4A: Individual PC synchrony values – not been done or addressed in the response (reviewer #1, point 4).

4) Figure 4: No illustration/schematic for data transformation used in this and subsequent figures, as suggested by reviewer 2.

5) No response provided for Point #6 from reviewer 2.

6) Figure 4: No statistical tests seem to have been done?

7) Figure 5: Will be better to plot the different lines in different colors, especially for Figure 5E where the spread around the mean can also be shown.

Results paragraph three: Were SS firing rates compared?

Subsection “Inhibition of DCN activity varies with level of CS synchrony” paragraph nine and elsewhere: Wherever p values are given, please state the n's and the test used.

Subsection “What is the effect of non-synchronized CS activity?” first paragraph: "This is not…" Odd sentence construction

Subsection “Identification of synaptically-connected PC-DCN cell pairs” and response letter: Authors argue that earlier work (Sugihara, 2007) shows that CS synchrony is higher within a zebrin compartment than between. However, that paper showed that this was true only for central and lateral crus2a. In medial crus2a, wide synchrony bands covering more than one zebrin compartment were observed. In the present study, it is not clear from which region of crus 2a neurons were recorded from.

Subsection “Identification of synaptically-connected PC-DCN cell pairs” paragraph five and elsewhere in the Discussion: "…as it is well-known that there is a pause in SSs following a CS," – A CS may have diverse effects on SS firing such as facilitation, pausing or suppression (De Zeeuw et al., 2011) or state switches (Loewenstein et al., 2005; Sengupta and Thirumalai, 2015). In this and other places in the manuscript, the writing seems to suggest that CS leads only to SS pauses.

https://doi.org/10.7554/eLife.40101.017

Author response

Major concerns:

One of the concerns is the low number of DCN cells recorded from, especially because of the heterogeneity of DCN cells. While the efforts required to find PC-DCN pairs are akin to finding a needle in a haystack (x4), it is important that the authors address the weakness of low n's in their discussion of the results. There are also some other caveats (such as indirect synaptic connections, the use of anaesthesia, etc.,), which should be addressed by the authors in the discussion.

In general, reviewers felt that the manuscript could be rewritten such that results are presented thoughtfully and even-handedly, with an awareness of the strengths and limitations that is conveyed to the expert and non-expert reader. Reviewers also had several suggestions to improve the presentation of figures and overall lucidity of the manuscript and these specific suggestions can be found in the individual reviews appended below.

The major concern was that the discussion should address the low n's regarding the number of DCN cells in the discussion as well as other caveats such as the use of anesthesia and indirect synaptic connections.

The issue of indirect synaptic connections was addressed in the original discussion in the 'Identification of synaptically-connected PC-DCN cells pairs' subsection, however, the discussion of this issue has been expanded. In addition, a new section entitled, 'Technical Limitations' has been added to the Discussion to address the other issues raised by the reviewers and editor.

With regard to the number of DCN cells, it is true that the number of DCN cells that were analyzed in detail is small because only a few examples out of a large database could meet the criteria to be included, and we agree that this limits how much we can generalize from our results.

However, it is important to realize that the general relationship between the level of CS synchrony and the strength of inhibition of DCN was observed for a larger population of DCN cells as reported in our previous work (Blenkinsop and Lang, 2011) and that the smaller subset of DCN cells whose activity is analyzed here in greater detail rests on this initial finding. Moreover, we believe the relationships that were found for each DCN neuron whose activity was analyzed in the current paper are based on strong evidence from large amounts of data. Specifically, each such DCN cell, the analyses are based on thousands (~2500-12,000) of CSs in its presynaptic PCs and tens of thousands (~30,000-50,000) of its own spikes. Thus, we feel confident of the relationships that are described in the paper, and though we agree that the data are insufficient to say whether these are the only types of relationship that exist between DCN and PC CS activity, the fact that a significant synchrony-dependence was found for all four DCN cells that were analyzed suggests that synchrony is likely to be an important parameter in many cases.

Separate reviews (please respond to each point):

Reviewer #1:

This manuscript argues that the synchrony of complex spikes in the Purkinje neuron cohort projecting to the same DCN neuron might hold the key for controlling cerebellar output. The manuscript is written in clear style and the data are analyzed and presented well. My biggest worry is that the data presented are from only four or five DCN cells (Results paragraph one says five cells, but four experiments are presented. Authors need to clarify how data from two DCN cells in the same animal were combined). In their dataset, authors could identify 4-7 PCs to be synaptically connected to the DCN cell being recorded. This combined with the low number of highly synchronous CS inputs in each experiment (Figure 6B), places the findings on not-so-solid ground. In addition, of the four, one DCN cell (A) did not show the tight synchrony-dependent modulation seen in the other three cells (Figure 9B and 10). The authors speculate that this cell could be an IO projecting DCN but this idea is not tested. With this in mind, I am not certain that the main point the authors push, i.e. that CS synchrony in a 10ms window is the most important parameter for DCN output, can be accepted. Some specific points:

The reviewer raises a number of points related to the low number of DCN cells that were ultimately analyzed, including how data from two DCN cells recorded from the same animal were combined.

The issue of the number of DCN cells that could meet the criteria to be analyzed is addressed above.

The two DCN cells referred to by the reviewer were recorded separately. Thus, they were treated as separate cells and analyzed individually.

1) Figure 1 is not useful. Either remove or include PC recordings and a Venn diagram of cell numbers discussed at the end of paragraph one of the Results.

We feel it is generally important to show examples of the actual recordings to let the readers judge the quality of the raw data that was used for the subsequent analyses. Usually one or more 'representative' examples are given, however, given the limited number of DCN cells that are the focus of the paper, we thought it better to show all of them.

We have, however, modified the figure based on the suggestions of several of the reviewers. First, we now show each of the DCN cells using an expanded time scale so that the spike waveform of each can be seen. In addition, we now show only one example over a longer (5 s) period to demonstrate the typical firing pattern. Lastly, we have added an example of the CS activity recorded from one of the PCs.

In addition, the initial two paragraphs of the results, which describe the database, have been revised to make clear the relationship between the original dataset, which was first described in our previous paper (Blenkinsop and Lang, 2011), and the subset of cells that are analyzed in detail in the current paper.

2) Will be useful to indicate the long latency modulations of DCN activity in Figure 2, with an arrow or grey bar.

The histograms shown in Figure 2 are for individual cells, but the long-latency analysis was based on an analysis of the group activity. Thus, instead of showing the modulation periods on the histograms in Figure 2, a population histogram has been added to Figure 11 and the modulation periods are shown on it.

3) In Figure 3, authors show lines to indicate 95th and 99th percentile ML separation and state that the synaptic group value is above the 99th percentile. This should be the 5th and 1st percentile and "below the 1st percentile" given that the synaptic group is placed more closely than the others.

The changes were made to Figure 3.

4) Figure 4A: SD or SEM? Prefer that the authors would show the individual synchrony values of all PC's in that experiment as scatter with an indication of the mean. Similar plots can be made for all four experiments. Figure 4B does not make much sense to me. What does it mean that the ratio is several-fold over very long distances?

The error bars in Figure 4A indicate SEM. This is now stated in the figure legend.

We could not determine exactly what reviewer was referring to with the phrase 'the ratio is several-fold over very long distances', but Panel 4B shows the ratio of the two curves shown in 4A for each of the four experiments, and is simply a way to show how the presynaptic cells display higher levels of synchrony among themselves than do the remaining PCs at all separation distances. It is true that this information (the relative difference in synchrony) is implicit in 4A, but we think that showing it explicitly is also useful. In particular, doing this facilitates comparison of the relative synchrony (presynaptic vs other PCs) distributions across experiments. In addition, it more clearly shows that synchrony is better maintained with distance for the presynaptic PC group (indicated by the increase in the ratio with increasing separation).

This last result is important because it leads to another line of evidence supporting the correlogram-based identification of presynaptic PCs. This follows from two findings: that synchrony is better maintained within, as compared to between, zebrin bands (Sugihara et al., 2007) and that each DCN region receives PC input from only one zebrin band. Thus, the better maintained synchrony with distance suggests that the presynaptic PCs were more likely to located within the same zebrin band (otherwise their synchrony would drop off similarly to the other PCs, which cannot all lie within the same band), which fulfills the requirement of the second finding, which implies that PCs that project to the same DCN cell are from the same zebrin band. This argument has been added to the Discussion.

5) Figure 11B: Authors argue that the long latency modulation of DCN activity is also synchrony dependent but the effect is weak and variable. One of the four cells recorded did not show this modulation. This point appears weak and not related to the main point of the manuscript.

It is true that long latency modulation is not a major point of the paper, but this modulation was present in most cases, which raises the question of what its relationship to synchrony is, and so for completeness we believe it is worth including an analysis of it.

6) Results paragraph two, final sentence: What is the larger sample used in this comparison?

The larger sample refers to the DCN cells that were recorded in Blenkinsop and Lang, 2011. The citation to this work is listed following the firing rates that were given. We added text to the sentence to make this reference clearer.

7) Subsection “Inhibition of DCN activity varies with level of CS synchrony” paragraph five: insert "Figure 7F"

Figure reference was inserted.

8) Sixth paragraph of the same section: Could singular CS's still be high synchrony events synchronized with PCs not being recorded from?

Our presynaptic PC group is a sample of the population of PCs that project to the DCN cell. So, it is conceivable (and even likely) that some singular CS events in our sample population are part of high synchrony events in the whole population. However, the assumption is that the activity of the PC sample group represents that of the population (i.e., is a random sample), a standard, if not wholly testable, assumption of recording studies. To the extent that the presynaptic group is representative, the level of synchrony in the sample should reflect that of the population as a whole, on average.

9) Subsection “Synchrony may be needed to modulate DCN activity”: "Our results indicate that CSs need to be synchronized to be able to affect DCN activity". Given that singular CSs also show about 20% inhibition, one would argue that they are also able to affect DCN activity, albeit at a lower level relative to the synchronous ones. Again, as noted above, it is likely that the singular CSs are in fact a mixed population of events of widely varying synchronicity. This point needs to be brought out in the discussion.

The reviewer is correct that the singular CS events represent a mixed population of events (as mentioned in response to the previous point), and this may contribute to their showing an inhibitory effect. However, we do not believe this is a major factor for why a significant inhibition is observed with singular CS events.

The key point is that one the firing of PC in the group represents the firing of a significant fraction of our synaptic group firing (between 14% and 25% of the group depending on the experiment). In the Results section we show that if you extrapolate the relationship between synchrony and inhibition of the DCN cell back to a single PC firing a CS (to do this we assumed 50-100 PCs converge onto a single DCN cell), the inhibitory effect is negligible (i.e., not significantly different from zero).

Also, because our use of the word 'singular' (as a single member of the presynaptic group) could be misleading (i.e., taken to mean one PC out of the entire presynaptic population) we have replaced it with text that removes this ambiguity.

10) Subsection “CS synchrony: direct and indirect actions on the DCN” paragraph four: later

Fixed

Reviewer #2:

This manuscript is a meta-analysis of multielectrode data from recordings of Purkinje cells (PCs) and deep cerebellar nuclei (DCN) in anesthetized rats. The authors correlate activity of PC complex spikes and DCN spikes to find likely connected cells and then explore the effects of synchronized PC complex spikes on 4 DCN cells. The greater the degree of complex spike synchrony, the greater the inhibition of DCN cells. The manuscript has useful information and the authors are careful to report all the exceptions and special cases, but the text doesn't always make it clear what the relevant points are or what aspects of the data the authors wish to stress. There are relatively few points along the way through 11 figures where the results and interpretations are synthesized. It would be helpful if the text were re-written to include a few more summary sentences. Likewise, there are many figures with relatively few panels, but sometimes the transformations of the raw data aren't immediately obvious. Possibly some further illustration would be helpful. Specifics are given below.

Summary sentences were added to the Results subsections.

Comments.

1) Figure 1 has example traces of 4 DCN cells, but nothing else, and it is hard to know what we are looking at them for. Are these the 4 cells that are identified in Figure 3 and carried through the paper? Figure 2 also shows 4 histograms from DCN cells but, in this case, 2 plots are from one DCN cell. PC spikes and PC synchrony are never actually illustrated. It could be helpful to see some of the PC traces (and maybe combine Figures 1 and 2 and relate them to 3) so that the raw data are clearer.

To clarify, the four DCN cells in Figure 1 are the ones analyzed throughout the paper. This is now stated. Examples of CS activity were not included originally because such examples were shown in several prior publications. However, we have now included an example.

The reason for showing each correlogram in Figure 2 with both a compressed (top row) and expanded (bottom row) x-axis is to show both the long-latency modulation (top row) and the precise timing of the CS-triggered inhibition (bottom row) that was used to define a synaptic relationship.

2) Figure 3 is used to make the argument that the likelihood of finding converging PCs decreases with mediolateral (M-L) but not rostrocaudal (R-C) distance. This argument is convincing in the M-L dimension but less so in the R-C dimension, since the multielectrode array has 7-9 electrodes M-L but only 3-4 R-C, and in Figure 3E and F the converging PCs have about equivalent mean M-L and R-C separations. Should this analysis be curtailed to talk about clustered PCs rather that PCs in an R-C plane?

We agree that the dimensions of crus2 and the recording array allow detection of synchrony changes over larger distances in the mediolateral direction than the RC direction. Nevertheless, we prefer to keep the separate analysis of the RC distance for several reasons. Despite the limited range in the RC direction, the resolution in both planes is the same (250 µm). The maximal separation in the RC direction that be detected was 750 µm, and so we should be able to detect a drop off in synchrony that occurs over shorter distances. Indeed, such a rapid drop off in synchrony occurs in the ML direction as shown in previous publications and in Figure 4, and so the lack of finding a difference in the RC direction is a useful contrast, and furthers the argument that convergent PCs form a RC oriented band.

3) The Synchrony Index described in the Materials and methods and used first in Figure 4 could be explained more clearly and intuitively, possibly even illustrated. The equation indicates that it is the summed product of time-matched deviations from the mean for each of two cells, divided by the root of the product of the summed-square-deviations. It is clear that when both signals go high simultaneously (i.e., synchronize), The Synchrony Index value will increase, but it is not immediately obvious what the range of values is expected to be, how this will vary with different levels of background firing, etc. It would be useful if the authors could help calibrate the reader, so that the values of Figure 4 (0.12, 0.06) can be placed in perspective. This figure might be a useful place to illustrate the successive transformations of the data.

We have added a brief description in the methods to provide the information the reviewer requested. The synchrony level, C(0), defined by the formulas in the Materials and methods is simply the cross-correlation coefficient between the two spike trains at 0 ms latency and this is stated in the Materials and methods. As such, it is bounded with a range of 1 to -1, where these limits reflect perfect correlation and inverse correlation, respectively, and a value of 0, represents no correlation. A detailed description of the statistical significance of particular values of C(0) has been presented previously, and this work is now summarized and cited. Essentially, values that would be expected by chance are one to two orders of magnitude smaller than those observed among the presynaptic PCs.

C(0) is normalized for firing rate. That is, more synchronous events will occur by chance when the neurons have higher firing rates, but the value of the terms reflecting each synchronous event are smaller (because the mean firing rates being subtracted are greater), which compensates for this effect. However, the issue of changing firing rates doesn't seem to be of major concern here as C(0) is not being compared across conditions (e.g., after giving a drug like harmaline or picrotoxin) in which CS rates have changed significantly.

4) Subsection “Inhibition of DCN activity varies with level of CS synchrony”; “Although singular CSs had the weakest inhibitory effect, they still reflect a significant level of synchronization….” It is not clear on what basis the authors say that the “singular” CSs, which seems to mean non-synchronized CSs, reflect synchrony. The argument about the 50-100 converging PCs is hard to follow, especially when the concluding sentence indicates that the effect is not different from no synchrony. Please clarify.

Since the term 'singular' could be misleading and taken to mean a complete absence of synchrony (i.e., the firing of only one PC in the entire population of PCs that project to a DCN cell) we have removed it. Instead, we now use a phrase like 'events in which only one PC in the group fires a CS'. This should make it clearer that we are referring to a single PC of the group of recorded PCs, which actually represents a significant fraction of the total population of PCs that are presynaptic to the DCN cell.

5) As far as this reader can tell, the central, most important finding of the manuscript is Figure 7F, which shows that for all four DCN cells, there is a strong correlation with steep slope of the amount of inhibition as a function of synchrony level. This key panel is not called in the text and hardly stands out in the illustrations. With so many figures and a relatively monotonic (without emphasis) presentation, it is possible that it will be missed. Some rewriting would help make this result more salient.

The figure reference has been added to the text. Also, the text in the section related to this finding has been revised.

6) The 7F panel is complemented by Figure 8. It switches to illustrating the inhibitory efficacy of PCs based on whether they are in a synchronous group, but the authors just refer to this as a “cell-by-cell” comparison, so it takes several readings before one realizes that this is a demonstration from the “perspective” of the PCs, which is quite important. Please rephrase for clarity. (Perhaps the authors could consider putting this key panel with Figure 7F somehow so that they stand out as a central figure?)

As the reviewer says, Figures 7 and 8 address distinct points, we believe they should remain as separate figures. However, we have revised the text to clarify the different points being made by Figures 7F and 8.

7) Discussion; “these [convergent] PCs have highly synchronized CS activity…” Is what is meant here that they have a high probability of synchronizing (which does not appear to have been tested; or was it?) or that when they synchronize they do so with <5 ms jitter (which is stated again in the last clause)?

What was meant is that they show more synchronous activity, which is demonstrated by the results related to Figure 4. The sentence was revised to eliminate the ambiguity related to the meaning of 'highly'.

Minor points.

1) Results; “In a few cases (n=4)” Please indicate n=4 out of how many cases.

The number of cases (24) was added to the text.

2) Figure 2, please indicate bin width in legend.

The bin width (1 ms) was added to the legend.

3) Figure 4B, perhaps different line patterns (solid, dashed, dotted) could help distinguish the four curves.

We don't believe this is necessary, as the curves are distinct when they are separated from each other at the larger separations, and at the smaller separations the data points are virtually identical to each other, making it unnecessary to distinguish them.

4) The use of the word “singular” to describe non-synchronous CS's is rather non-standard and unintuitive. Perhaps a definition could be given at first use.

We eliminated the word and used a more precise phrase to define the event of one PC in the group firing a CS, as described in the responses to earlier comments.

5) Figure 5, it would be useful to indicate the number of PCs in each group in the legend or on the plots.

The numbers are now given in the legend.

6) Subsection “Inhibition of DCN activity varies with level of CS synchrony”; (Figure 7) “the baseline activity in all histograms was normalized for sweep number…” It is not clear what analysis was done. Please clarify.

7) Please define “cusum.”

Cusum is now defined in the text.

8) The word “axonic” should probably be “axonal.”

Changed

Reviewer #3:

The authors of the manuscript entitled "Complex spike synchrony dependent modulation of rat deep cerebellar nuclear activity" investigated how Purkinje cell complex spike activity and synchrony affects activity in their main downstream target, the cerebellar nuclei (CN). By doing simultaneously recordings of Purkinje cells (PCs) in the cerebellar cortex and of CN neurons, they describe that PCs that are putatively presynaptic to recorded CN cells are located in rostrocaudal strips. In addition, they describe that CN cells show specific patterns of short-latency responses to complex spike activity of which the strength depended on the degree of complex spike synchrony. The paper addresses an important and timely topic using a series of detailed in depth analyses.

I have several suggestions to improve the paper.

1) The manuscript largely lacks raw data. Only in Figure 1, traces are shown but at a very compressed time scale. Please expand.

As requested, we have added records at an expanded time scale to show the individual DCN spike waveform.

2) The number of CN cells from which they recorded in a paired format with PCs is low (n = 4/5). Given that there are many different cell types in the nuclei (e.g., Dulac and Uusisaari studies), it may be worthwhile to increase the number of CN cells. In addition, the number of high-synchrony events per cell is low. This makes the conclusions, although likely to be accurate, somewhat preliminary.

We agree that the number of DCN cells is too low to allow strong statements about how the results may generalize to the various DCN cell types (in the discussion only a short speculation about DCN-IO neurons is made, and we clearly state that further evidence is needed to support this speculation). However, as discussed earlier, we believe that relationships that were found for each of the DCN cells that were analyzed rest on a strong foundation. Moreover, that all four cases showed a strong relationship between CS synchrony and the strength of DCN inhibition makes it unlikely that this is a rare phenomenon.

3) The authors discriminate between directly coupled Purkinje cell – DCN neuron pairs and non-synaptic pairs. Their criteria for establishing synaptic connections are not mentioned in the current paper, but a reference is made to a previous publication (Blenkinsop and Lang, 2011). Also in that publication, the criteria are not overly clear. In the latter study, the impact of indirectly coupled pairs has also been described (see Figure 4C of Blenkinsop and Lang, 2011). As the authors discuss (subsection “Identification of synaptically-connected PC-DCN cell pairs”), identification of synaptic connections is difficult and could be compromised by synchrony between Purkinje cells. As Purkinje cells with a high level of synchrony might be considered functionally equivalent, the authors rightfully remark that it is questionable whether this would affect their conclusions. However, the abovementioned, indirect connections (see Figure 4C of Blenkisop and Lang, 2011), could also affect the patterns. Thus, it is crucial to understand which criteria were used to define mono-synaptic connections and to what extent indirect connections could contribute to the findings.

The identification criteria of a monosynaptic connection from PC to DCN cell that we used are based on an accepted methodology (cross-correlogram analysis) for identifying such connections (e.g., Perkel et al., 1967; Aertsen et al., 1989; Csicvari et al., 1998). In our specific case there had to be a change in the correlogram at a latency that matched the conduction time of the pathway from the presynaptic (PC) to the postsynaptic cell (DCN). In the Materials and methods we state this criterion as, "a significant negative deflection in the CS-triggered correlograms within 5 ms after CS onset, consistent with the conduction time of the PC-DCN pathway and inhibitory nature of PCs". A more detailed discussion of the validity of this criteria and other evidence supporting identification of synaptic connections is now in the Discussion.

Regarding the issue of correlations among PCs, it is necessary to consider two distinct causes of these correlations: PCs contacted by branches of the same olivocerebellar axon and electrotonic coupling of olivary neurons. The first case would produce a near perfect correlation (r~1) with very little jitter in the timing of the spikes in the two PCs, and essentially would not affect the interpretation of the results for reasons elaborated in the discussion.

In contrast, CS synchrony due to the gap junction coupling between IO neurons, a 'high correlation' would typically result in correlations of <.2 and a central peak of the correlogram that spreads over 5-10 milliseconds. Thus, CS-triggered correlograms of DCN activity from PCs that do not synapse onto the DCN cell in question but whose CSs are correlated with those of a PC that does synapse onto the DCN cell, should show an onset of inhibition at negative latencies without any new effect at a latency matching the PC-DCN conduction time. Indeed, such correlograms are found (and were rejected).

4) Regarding the indirect connections: the authors focus on the synaptically connected pairs, but was there also an impact of complex spike firing in the non-synaptic group? According to Figure 4A (of the present manuscript), also these cells do show synchronous complex spike firing, albeit at a lower level, so that an effect would be predicted. Please discuss.

Testing whether the effect of 'indirect connections' is due to the correlation of the CSs of those PCs with the CSs of the presynaptic PCs is not easy to do, because the activity of both groups of PCs is correlated, as the reviewer notes. Indeed, we previously reported that the level of synchronization of PCs throughout the recording array could modify the inhibition of DCN activity related to the activity in presynaptic PCs. For the current paper, we attempted an analysis to test whether the effect seen with the cells not identified as presynaptic to the DCN cell was due to their activity being correlated with that of the presynaptic cells. Specifically, we treated the non-synaptic cells as a group and performed the same analysis as we did for the synaptic group. The first problem that arose is that this larger group has few or none of the higher synchrony events involving most or all PCs. This is not surprising, given the lower synchrony overall amongst these cells. In fact, it points to an immediate difference from the synaptic group. However, for the synchrony levels that did occur, one sees a good correlation between synchrony and the strength of inhibition of the DCN cell. Again, this is not surprising given that the non-synaptic PCs do show correlated activity with the synaptic ones. We then tried to decorrelate the activity of the synaptic and non-synaptic PCs by removing any events in which synaptic cells participated. When we did this the relationship between synchrony and DCN inhibition was weakened, often considerably. However, this procedure resulted in very few synchrony events being present at any level other than the lowest few levels, so that it was impossible to test the effect of this procedure at the high synchrony levels, which are the most critical to test.

5) The experiments were performed in anesthetized animals (with ketamine/xylazine). Since this anesthesia can affect cerebellar firing patterns including patterns of rhythmicity and synchrony (e.g., Schonewille et al., 2006), it will be worthwhile to replicate some recordings in awake behaving animals. This will increase the functional relevance of the study.

While we agree recordings from awake animals would be worthwhile, we believe obtaining them would be exceedingly hard technically, particularly given the difficulty in finding synaptically-connected pairs even in an anesthetized preparation. In the discussion we acknowledge this limitation but also provide reasons why it is plausible that the results should be relevant to understanding cerebellar function under more physiological conditions.

In particular, although some higher order statistics of SS, CS, and DCN activity may vary between ketamine/xylazine and awake states, on many measures their activity is similar. For example, their firing rates under ketamine/xylazine anesthesia are essentially the same as in the awake animal (note that the study of Schonewille, which only reports SS rates, specifically confirms this for SSs). Moreover, the phenomenon of long pauses of SSs under anesthesia that is a focus of the Schonewille report is not generally seen using our recording technique (see, Tang et al., 2017). Moreover, a more recent report from Schonewille and colleagues provided evidence that this activity pattern may actually be due to damage or cooling of the cortex (Zhou et al., 2015 J Neurophysiol 113: 2524-2536), and so raises the possibility that it is not directly relate to anesthesia. Finally, Schonewille, 2006, does not address differences in synchrony patterns, and CS synchrony patterns were shown to be essentially the same in ketamine/xylazine anesthetized and awake animals (Lang et al., 1999). In sum, the many similarities of PC and DCN activity in the anesthetized and the awake animal suggest that the present results will be relevant for physiological states.

6) In addition, or alternatively, the authors may want to discuss the Ten Brinke studies (2015 and 2017) in which complex spikes and CN activity were recorded during eyeblink conditioning. In these studies the same patterns were detected with separated single unit recordings of PCs and CN cells in awake behaving animals as described with simultaneous dual recordings in the current study. This would highlight the achievement of the current dual recordings of the current Lang study, while clarifying the putative relevance of it during natural behavior. In fact, the CS synchrony and impact thereof in the CN appear to be enhanced during motor learning.

The ten Brinke, 2017, paper is cited in the Discussion.

7) The authors relied on 1 ms binning of their data. The data underlying Figure 7 depend on the duration of the inhibition, which is defined as ending when the 3-point average reaches baseline levels again. This seems a very noise-prone analysis, especially for relatively rare occurrences of high-synchrony (see Figure 6). The open symbols in Figure 7B-E are therefore questionable; the closed ones a bit more reliable, but even here an estimate of the baseline firing during only 50 or 25 ms (the choice of either which value being not really justified), might be quite arbitrarily. Overall, the conclusion that more synchrony leads to more inhibition seems plausible, but the statistical evaluation could be improved (see also below). One should notice, again, that all these conclusions are based upon only four cells.

The periods defined by the various methods might vary somewhat because of noise; however, the overall results are not particularly sensitive to the precise limits, as shown by the overlap and similarity of the curves using different methods for defining these limits in Figure 7. We also tried using a fixed window of 5 to 60 ms for all experiments (so that the same time period could be compared across experiments), which again produced similar curves (data not shown). In sum, the relationships shown in Figure 7 did not greatly depend on the exact limits to do the analysis.

Additional data files and statistical comments:

The authors report using paired and unpaired t-tests in addition to Pearson correlation analyses. However, their reasoning for choosing these tests is unclear. Based on the low n, inequality of sample sizes (e.g. the difference in number of low vs high synchrony events) and questionable normality of distributions, one may also have to consider non-parametric testing; this reviewer is not a statistician, but please ask advice from a professional to make sure the current approach is correct or not. For sure, the authors should include a detailed description of tests used (and for what reasons they were used) and also include the specific outcomes like t-values, degrees of freedom etc. Evaluation of the statistics may also indicate how many pairs of PC-CN cells need to be illustrated in the current paper to back up the conclusions. So this reviewer has no real doubts about the conclusions of the current manuscript, but they need to be properly backed with solid statistics to stand the test of time.

The reviewer is concerned about using t-tests because of low n, unequal sample sizes and questions about the normality of the distributions. The first two of reasons are by themselves not an issue, as differences in sample size and variance between populations can be taken into account in the formula used to compute the t-statistic and degrees of freedom for the t-test (sometimes referred to as the Welch test). This was done in all tests where differently sized populations were compared, and this is now stated in the Materials and methods.

Regarding the low n, a t-test was used in a handful of cases, and in only one case was the n particularly low (n=4) The n's of the other populations were: 24, 54, and 100. It is worth noting that t-tests were specifically designed for testing small samples, so a low n is not a problem in itself, but only may become an issue if the population deviates strongly from normality.

Nevertheless, to address the concern about normality, each of sample populations was tested for normality using the Kolmogorov-Smirnov (KS) goodness-of-fit test. If the populations were not normally distributed, a non-parametric test was used to compare the populations.

With regard to Pearson's correlation (r), Zar, 1999, states that the test for whether r is different from zero is robust with regard to violations of normality so that t-tests are appropriate and that transforms to normalize the data are necessary only if one wants to test whether r differs from some non-zero value (see section 19.2, Zar, 1999).

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript is indeed written more accessibly than in the previous submission, but there are some remaining issues that we feel should be addressed before your paper is published, as outlined below:

1) The authors claim (subsection “Technical limitations”) that DCN firing rates are the same with and without anesthesia. No citation is given. The 34 Hz DCN firing rate seen here seems relatively low. The main line of the authors' arguments doesn't rely on claiming that there is no effect of anesthesia, and so unless there is explicit evidence, it is probably better not to say that anesthesia does not affect the firing rates.

We agree that this point is not vital to the argument, but we nevertheless believe it supports the argument in that it provides some evidence that DCN cells are in a similar state in awake and anesthetized animals, and thus makes it more likely the operational principles of the system reported here are also valid for the system in the awake state.

To support our statement about DCN firing rates we have now given citations for the DCN firing rates in awake animals (see list below). Based on these references, 34Hz is well within the ranges of values reported for DCN neurons in awake animals, even if it is toward the lower end of the distribution. It is perhaps worth noting that DCN cells in mice seem to have very high rates, e.g., 85 Hz (Sarnaik and Raman, 2018); however, examples from other species often report much lower average values or a large range that extends well below 34 Hz.

The Discussion was modified to state the following:

“In particular, the reported spontaneous firing rates for DCN cells in alert animals varies widely, both within and between studies; overall, rates range from 10 Hz to greater than 100 Hz (e.g., Thach, 1968; Burton and Onoda, 1978; Harvey et al., 1979; Chapman et al., 1986; Fortier et al., 1989; Gruart et al., 2000; Sarnaik and Raman, 2018). The data set from our previous report showed a similar range for the overall DCN population as did the population of DCN cells for which a presynaptic PC was identified (see Figure 2A,B of Blenkinsop and Lang, 2011).”

Examples of spontaneous DCN firing rates reported for awake animals

monkeys

37 Hz, Thach 1968.

30-50 Hz, Harvey et al., 1979.

52 Hz, Chapman et al., 1986.

30-35 Hz Fortier et al., 1989

10-60 Hz and 30-80 Hz for cell types A and B, Gruart et al., 2000.

21 Hz, 16 Hz, and 29 Hz, depending on cell type, Burton and Onoda, 1978.

2) Figure 1: Reviewers wanted PC recordings to be shown here and this has not been done or addressed in the response.

Actually, this point was addressed in the previous response to reviewer 1 (the following was stated, "we have added an example of the [complex spike] CS activity recorded from one of the PCs") and panel E of the revised Figure 1 does show an example of CS activity from one of the PCs analyzed in the paper. Moreover, examples of CSs from the same parent data set can be seen in our previous publication (Figure 1 of Blenkinsop and Lang, 2011).

Perhaps the reviewers were equating PC activity with SS activity, but SSs were not recorded (see response to point 7 below).

3) Figure 4A: Individual PC synchrony values – not been done or addressed in the response (reviewer #1, point 4).

To include all of the individual synchrony values would make the plots in the figure difficult to read (i.e., in several cases 50-100 points of very similar values would be plotted for one x-axis value). So, we would prefer not to plot the data this way. However, to address the reviewer's concerns about the distribution of the values, we have described the statistics of the population in the text and added a new panel to the figure (new panel B), which plots the median synchrony values. See answer to point 6 below for details on the statistics used to describe the distribution and to test the significance of the differences between the synaptic and non-synaptic groups.

4) Figure 4: No illustration/schematic for data transformation used in this and subsequent figures, as suggested by Reviewer 2.

The synchrony index is simply the cross-correlation coefficient of two spike trains at zero time lag, calculated by standard formulas that have been published many times. The papers that first developed it and that first used it with regard to CS activity are cited for reference. We also added a description of its properties in the Materials and methods (this follows the equations). Given this information, we do not see what would be added by showing a schematic of the calculations.

5) No response provided for Point #6 from reviewer 2.

We have modified the sentence to read, "To test this possibility, the baseline activity in all histograms was normalized for (divided by) the number of CSs used to generate them and then compared (Figure 7G). "

6) Figure 4: No statistical tests seem to have been done?

In the previous version of Figure 4A, the results from a single experiment were shown. Although statistical tests could be done to test the significance of the differences in the example that was shown, in other cases the number of cell pairs in a particular group from an individual experiment was too small to perform a statistical test. So, we adopted a different approach that would allow testing of all the data, which is shown in the new panels A and B of Figure 4. We combined the cell pairs from all four experiments, which then allowed statistical comparisons for the population at separation distances of 0, 250, and 500 µm, which are the distances for the vast majority of presynaptic cell pairs (~95%, 60/63 pairs). In the revised figure, mean synchrony as a function of ML separation is compared in 4A for the populations of presynaptic and non-presynaptic cell pairs. t-tests showed that the means were significantly different at these three distances. The p-values and n's are given in the text. Tests for normality showed that the synchrony values were normally distributed at all three distances for the presynaptic population, but not for non-synaptic population. Thus, we added panel B, which shows the median synchrony values corresponding to the means shown in 4A and performed nonparametric tests (Wilcoxon-Mann-Whitney) on the distributions. These tests also indicated that the presynaptic and nonsynaptic distributions were significantly different at each distance. The p-values and n's are given in the text.

7) Figure 5: Will be better to plot the different lines in different colors, especially for Figure 5E where the spread around the mean can also be shown.

The curves in Figure 5E are now plotted in different colors and error bars indicating the SD are now plotted.

Results paragraph three: Were SS firing rates compared?

No, SSs were not recorded in this study. To make clear why only CSs were recorded, we now state the following in the Materials and methods section.

“Electrodes were implanted into the molecular layer to a depth of 100-150 µm below the folial surface, where CSs but not SSs can be observed. To record SSs, the electrodes need to be implanted to a depth of ~225-250 µm, close to the PC layer in rats. While also obtaining SS recordings is desirable, there are several reasons why this was not done. These include a much longer implantation process, greater overall damage to the cortex, more difficult spike sorting for analyses, and less stability during recording (i.e., the electrode is much more likely to damage or kill the cell being recorded because of its closer proximity to the cell body).”

Subsection “Inhibition of DCN activity varies with level of CS synchrony” paragraph nine and elsewhere: Wherever p values are given, please state the n's and the test used.

The n's are now stated with the p values. The default type of test used for each statistical comparison is stated in the Materials and methods rather than repeated for each instance. Where ever a non-default test is used, the specific test is named with the p-value.

Subsection “What is the effect of non-synchronized CS activity?” first paragraph: "This is not…" Odd sentence construction

The sentence has been modified to read as follows:

“This is simply because one PC represented a significant fraction of the presynaptic group (0.14 – 0.25) and its firing a CS thus likely corresponds, on average, to the synchronous firing of a similarly significant fraction of all of the PCs that project to a DCN cell fire CSs.”

Subsection “Identification of synaptically-connected PC-DCN cell pairs” and response letter: Authors argue that earlier work (Sugihara, 2007) shows that CS synchrony is higher within a zebrin compartment than between. However, that paper showed that this was true only for central and lateral crus2a. In medial crus2a, wide synchrony bands covering more than one zebrin compartment were observed. In the present study, it is not clear from which region of crus 2a neurons were recorded from.

Regardless of the exaction locations that PCs were recorded from (but see below for likely locations) we believe that the finding of high synchrony does provide supporting evidence for the cross-correlation based identification of PC-DCN connections. However, it is not absolute proof of their validity, and wasn't claimed to be, because of the possibility of false positives.

The reviewer is correct in stating that in the Sugihara et al. (2007) paper compartmental restriction of high synchrony was demonstrated for much of crus 2, but not for its most medial aspect. However, the implication that compartmental restriction was shown not to occur on medial crus2 is not correct. It is more accurate to say that paper showed it to be true for central and lateral crus 2a but was unable to determine whether it was also true for the medial crus 2a because of technical limitations (i.e., the medial compartments are too narrow to test adequately). Indeed, in the example that was shown (Figure 3 of that paper) one can see that no electrodes were definitively implanted to zebrin positive bands that neighbor the negative bands in the medial region. Also, note that two of the three examples in that figure do show that the highest synchrony levels are found among PCs within the same band, but because of the technical issues, the conclusions were conservative.

The key point is that PCs within the same compartment showed high synchrony where it was adequately tested. Of course, this pattern may not hold elsewhere, but given the similarity of cerebellar and olivocerebellar organization throughout, the most plausible assumption is that it does hold. This, when combined with the fact that each region of the DCN receives virtually all of its input from either a zebrin positive or negative compartment, implies that PCs that project to the same DCN cell should show high synchrony. Thus, testing for synchrony levels becomes a reasonable test for the validity of the cross-correlation based identification of the presynaptic PC group, and had the presynaptic PCs not shown high synchrony it would have questioned the validity of their identification. Conversely, since they did show high synchrony it supports it (i.e., is consistent with their identification as all connecting to the same DCN cell).

High synchrony could represent a false positive if there were cases in which a group of PCs show a pattern of high synchrony unrelated to the zebrin bands. This may be what the reviewer is concerned about, as there are generally high synchrony levels on medial crus 2a. However, as just discussed, this doesn't preclude a banding pattern being superimposed on the baseline synchrony level.

Regardless, although the exact location of the recording array on crus2 was not determined in the present study, it is that the array was mainly or entirely over the central and lateral portions of crus 2. Moreover, it is even more likely that the synaptic PCs were located in these portions of crus 2a (zebrin bands 4a- through 5a-, as defined in (Sugihara et al., 2007b). Medial crus 2a projects to various parts of the medial DCN. In contrast, neurons with histologically identified locations in the original data set were almost exclusively (one exception was on the lateral border of the medial DCN) located in the interposed and lateral DCN (see Figure 9 of Blenkinsop and Lang, 2011). Furthermore, DCN cells (including two of the cells analyzed here) that had synaptic connections with PCs were all in the middle to lateral interposed nuclei, which indicates that the synaptic PCs would be in the central and lateral crus 2a (zebrin bands 5+ through 7+, as defined in (Sugihara et al., 2007b).

Subsection “Identification of synaptically-connected PC-DCN cell pairs” paragraph five and elsewhere in discussion: "…as it is well-known that there is a pause in SSs following a CS," – A CS may have diverse effects on SS firing such as facilitation, pausing or suppression (De Zeeuw et al., 2011) or state switches (Loewenstein et al., 2005; Sengupta and Thirumalai, 2015). In this and other places in the manuscript, the writing seems to suggest that CS leads only to SS pauses.

We disagree with this characterization of what was written. For example, in the Discussion we state, "The latter includes the pause in SSs following a CS and the further modulation of SS activity following the pause" and "following its post-CS pause, SS activity often rebounds to rates that are significantly higher than baseline for 50 – 100 ms". Thus, we clearly discuss other aspects of how CSs affect SS activity. It is worth noting that the SS modulations mentioned by the reviewer always follow the initial pause in SS activity that follows a CS. In fact, if to definitively assign spikes as PC SSs you need to see this pause, otherwise it is possible that they are spikes from another cell type (e.g., basket or stellate).

In the specific line referenced by the reviewer, possible causes of the changes in activity of the DCN cells during the time immediately following the CS are being discussed. The post CS SS pause is the focus there, because it occurs during that time period. The other types of modulation that the reviewer raises largely occur after this time period, and so are not relevant to that particular discussion. They are discussed at other points in the discussion, as shown by the example quotations above.

https://doi.org/10.7554/eLife.40101.018

Article and author information

Author details

  1. Tianyu Tang

    Department of Neuroscience and Physiology, New York University School of Medicine, New York, United States
    Contribution
    Formal analysis, Writing—original draft, Writing—review and editing
    Contributed equally with
    Timothy A Blenkinsop
    Competing interests
    No competing interests declared
  2. Timothy A Blenkinsop

    Department of Developmental and Regenerative Biology, Mount Sinai School of Medicine, New York, United States
    Contribution
    Formal analysis, Investigation, Methodology, Writing—review and editing
    Contributed equally with
    Tianyu Tang
    Competing interests
    No competing interests declared
  3. Eric J Lang

    Department of Neuroscience and Physiology, New York University School of Medicine, New York, United States
    Contribution
    Conceptualization, Resources, Software, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    Eric.Lang@nyumc.org
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8524-9854

Funding

National Institute of Neurological Disorders and Stroke (NS-37028)

  • Eric J Lang

National Institute of Neurological Disorders and Stroke (NS-95089)

  • Eric J Lang

National Science Foundation (IOS-1051858)

  • Eric J Lang

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by grants to EJL from the National Science Foundation (IOS-1051858) and the National Institute of Neurological Disorders (NS-37028; NS-95089).

Ethics

Animal experimentation: Experiments were performed in accordance with the National Institute of Health Guide for the Care and Use of Laboratory Animals. Experimental protocols (030202-02, 030202-03) were approved by the Institutional Animal Care and Use Committee of New York University School of Medicine. All surgical and recording procedures were performed under general anesthesia, and every effort was made to minimize suffering.

Senior Editor

  1. Richard B Ivry, University of California, Berkeley, United States

Reviewing Editor

  1. Vatsala Thirumalai, National Centre for Biological Sciences, India

Publication history

  1. Received: July 13, 2018
  2. Accepted: December 20, 2018
  3. Version of Record published: January 9, 2019 (version 1)

Copyright

© 2019, Tang et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 678
    Page views
  • 121
    Downloads
  • 7
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)