Neuronal Activity in the Cerebellum During the Sleep-Wakefulness Transition in Mice

Abstract

Cerebellar malfunction can lead to sleep disturbance such as excessive daytime sleepiness, suggesting that the cerebellum may be involved in regulating sleep and/or wakefulness. However, understanding the features of cerebellar regulation in sleep and wakefulness states requires a detailed characterization of neuronal activity within this area. By performing multiple-unit recordings in mice, we showed that Purkinje cells (PCs) in the cerebellar cortex exhibited increased firing activity prior to the transition from sleep to wakefulness. Notably, the increased PC activity resulted from the inputs of low-frequency non-PC units in the cerebellar cortex. Moreover, the increased PC activity was accompanied by decreased activity in neurons of the deep cerebellar nuclei at the non-rapid eye-movement sleep-wakefulness transition. Our results provide in vivo electrophysiological evidence that the cerebellum has the potential to actively regulate the sleep-wakefulness transition.

Electronic supplementary material

The online version of this article (10.1007/s12264-020-00511-9) contains supplementary material, which is available to authorized users.

Introduction

The cerebellum is a critical component of the central nervous system that is involved in motor planning [1], motor execution [2–4], and motor learning [5–7]. However, patients suffering from cerebellar disorders have not only motor deficits, but also non-motor symptoms. For example, patients with spinocerebellar ataxias, which are characterized by degeneration of the cerebellum, often show excessive daytime sleepiness [8–10]. In addition, cerebellectomized cats show decreased wakefulness, and increased rapid eye-movement (REM) sleep [11]. Vice versa, patients suffering from primary sleep disorders often have a decreased cerebellar volume [12, 13]. Altogether, the current findings support a hypothesis that the cerebellum is involved in regulating the sleep and/or wakefulness states [14].

In support of this hypothesis, neuroanatomical studies have revealed that deep cerebellar nuclei (DCN), the final integration and output nuclei of the cerebellum, innervate several brain areas critical for maintaining or promoting wakefulness. As such, the DCN have been reported to innervate the ventral thalamus [15–17], which participates in the transition from non-REM (NREM) sleep to wakefulness [18]. In addition, the DCN have been revealed to interconnect with the hypothalamus [19–21], where several types of wakefulness-promoting and sleep-promoting neurons are located [22, 23]. Consequently, the cerebellum lies in a good position to communicate with arousal neurons, so as to regulate the sleep and/or wakefulness state [14, 24].

Nevertheless, the cerebellum can regulate the sleep and/or wakefulness state only if its activity changes preceding the transition between various sleep and wakefulness states. Consequently, monitoring neuronal activity in the cerebellum is the first step to uncover the specific roles and features of the cerebellum in the regulation of sleep and wakefulness. Extending early studies reported that the activity of Purkinje cells (PCs) and of DCN neurons both increase during REM sleep [25–28]. Along this line, functional imaging studies have revealed changes in cerebellar activity across various sleep stages. For instance, it has been reported that human subjects show lower cerebellar activity during NREM sleep than during wakefulness [29], and show increased activity during REM sleep [30, 31]. However, neuronal activity in the cerebellum during the transition between various sleep and wakefulness states remains largely unclear.

Consequently, in the current study, we performed multiple-unit recording in the mouse cerebellum across sleep-wakefulness cycles, in order to investigate the temporal features of cerebellar neuronal activity during this process.

Materials and Methods

Animals

All experiments were approved by the Animal Care Committee of Army Medical University. Male C57/BL6 mice (n = 10, 3–5 months old, 20 g–25 g) were individually housed in a 12-h light-dark cycle with lights on at 08:00, with free access to food and water. All experiments were conducted during the light phase of the cycle.

Surgery

The mice were implanted with a tetrode assembly attached to a microdrive, under isoflurane anesthesia (0.6%–1.0% by volume in O₂). The detailed implantation procedures have been described recently [32]. Each tetrode was composed of four tungsten wires (bare diameter, 20 μm; insulated diameter, 25 μm; California Fine Wire, Grover Beach, CA); the impedance measured before implantation ranged from 200 kΩ to 400 kΩ. One stainless-steel screw was inserted through the skull above the right hemisphere of the cerebellum to serve as ground, and another screw was inserted above the right hemisphere of the neocortex to serve as reference. These screws were secured to the skull using Metbond cement (Parkell, Shiga, Japan). Thereafter, a stainless-steel wire (bare diameter, 254 μm; insulated diameter, 330 μm; #792300, A-M Systems Inc., Sequim, WA) was implanted into the right frontal cortex (AP, + 1.7 mm; ML, + 0.5 mm; DV, + 1.0 mm). In addition, a craniotomy was made above the left cerebellum. The tetrode assembly was implanted perpendicular to the longitudinal axis with its center at the coordinates AP, − 6.3 mm; ML, − 1.7 mm, and lowered into the cerebellar cortex (depth, 1400 μm–1500 μm) with the help of a brain stereotaxic apparatus (model 68006, RWD, China). Low-viscosity silicone (Kwik-Cast^TM; WPI, Sarasota, FL) was then applied to cover the craniotomy. During postoperative recovery, the weight of each mouse was measured daily until it recovered to the pre-surgical level. The tips of tetrodes were moved down (~ 70 μm/day) until the firing of putative PC was monitored.

Sleep Monitoring

Postoperative mice were individually housed in transparent plastic cages (28 × 13 × 18 cm³) for the duration of the experiment. After ~ 1 week of postoperative recovery and tetrode adjustment, the freely-moving mice started habituation and sleep monitoring at the beginning of the early light phase (08:00–11:00).

In Vivo Electrophysiology

As we described recently [33], in this study, we performed data acquisition with an Intan interface board (RHD2000, Intan Technologies, Los Angeles, CA). The neuronal signals were amplified, digitized at 20 kHz using a multiplexed preamplifier (C3334, Intan Technologies), and stored for off-line analysis in a 16-bit resolution format. The wide-band signal was sampled to 1250 Hz and used as the local field potential (LFP). During tetrode adjustment and recording, neuronal activity was continuously visualized using Neuroscope software (http://neurosuite.sourceforge.net) [34]. In 8 mice, initial recording was performed in the vicinity of the Purkinje cell layer (1500–2000 μm ventral to the surface) to monitor cerebellar cortical neuronal activity across sleep-wakefulness cycles. After the last PC recording in the cerebellar cortex, the tetrodes were moved down at least 280–490 μm until DCN neuronal activity was recorded. On the next day, the DCN neuronal activity across sleep-wakefulness cycles was recorded. This measure was used to minimize the possibility of recording cerebellar cortical units in the DCN, considering that a tetrode covers only ~ 140 μm from its tip [35].

Data Analysis

Spike Sorting

The detailed procedures for unit clustering have been described [36]. Briefly, the spikes were extracted from high-pass filtered signals off-line, and their waveforms were projected onto a common basis obtained by principal component analysis of the filtered data. Single-unit spikes were isolated off-line using both semi-automatic clustering using KlustaKwik software [37] (http://klustakwik.sourceforge.net/) and manual clustering using Klusters software (http://neurosuite.sourceforge.net/) [34]. The accuracy of unit clustering was verified by confirming the existence of a 2-ms refractory period devoid of spikes in the autocorrelogram of a given single unit.

Sleep Analysis

In this study, we combined LFP oscillation in the frontal cortex and head movement of the mice (reflecting the activity of their neck muscles) to define recordings associated with distinct brain states. This method has been used to define brain states in our recent studies [32, 36] and by other research groups [38, 39]. Moreover, frequency spectrum changes in the LFP oscillation have been demonstrated to be sensitive, providing an opportunity to define state transitions with high temporal resolution (i.e., 1 s) [36, 38, 39]. Consequently, the sleep stages (NREM and REM) were determined by LFP oscillation in the frontal cortex and head movement of the mouse, according to the criteria described by the previous studies [36, 38, 39]. K-means clustering of the theta/delta ratio was automatically extracted from the power spectrogram. NREM sleep was defined as periods in the sleep epoch that met the following criteria: (1) the mouse was immobile (velocity < 0.5 cm/s for at least 10 s) and (2) slow-wave oscillations were detected. REM sleep was defined based on an elevated ratio of cortical theta (5–10 Hz) to delta (1–4 Hz) power with the requirement that a single REM bout lasted at least 10 s.

In Vivo Electrophysiological Isolation of Putative PCs

PCs were identified by their high firing activity and distinctive complex spikes in the vicinity of PC layer, according to the criteria described by ten Brinke et al. [40]. Putative PCs were initially identified by their intense spontaneous activity and later confirmed by the presence of both simple and complex spikes during unit clustering. The complex spikes were manually discriminated on the basis of the presence of a 1–3 ms slow component together with their stereotypical shape (i.e., board monophasic negative waves). For a putative PC, the presence of a pause (> 10 ms) in simple spike firing after a complex spike was verified.

Quantification of the Cross-Correlogram (CCG)

As described by Bartho et al. [41], the putative excitatory monosynaptic connection was identified by a short-latency peak in the CCG. Significant peaks (1-ms bin width) within 3 ms of the center bin were considered to be putative excitatory monosynaptic connections. A peak in a CCG was defined as significant when at least one of the bins exceeded the 99.9th percentile of the mean. The mean (control number of spikes) was calculated between − 50 ms and − 10 ms to control for potential low-frequency fluctuations of firing rate. Normalized CCG was obtained by dividing the observed CCG by the expected number of spikes per bin.

Statistics

Data are expressed as the mean ± S.E.M. The firing rate difference between sleep and wakefulness states was determined by paired t tests. The mean firing rate was calculated between − 10 and − 2 s, which served as control. The changes in neuronal activity across sleep-wakefulness transitions were then determined by paired t tests. Parametric or non-parametric tests were used as appropriate. A value of p < 0.05 was considered to be significant for all tests.

Histology

To visualize the tip locations of tetrodes, electrolytic lesions (30 μA for 10 s, DC current) were made at the end of all behavioral and recording experiments. The mice were anesthetized with pentobarbital (100 mg/kg intraperitoneal) and perfused with saline and 4% paraformaldehyde (PFA; prepared in 0.1 M phosphate buffer, pH 7.4). The brain was removed and post-fixed in 4% PFA for 8 h. Afterwards, the tissue was transferred to 30% sucrose for 48 h. Coronal sections (20 μm thick) were cut on a freezing microtome (CM1900, Leica, Germany) and collected in phosphate buffered saline (0.01 mol/L, pH 7.4) for later staining. After three washes (5 min each), the sections were mounted in Fluoromount medium with DAPI fluorescence (F6057, Sigma-Aldrich, St. Louis, MO). The tip placements of tetrodes were checked and the images were acquired using a fluorescence microscope (BX53, Olympus, Tokyo, Japan). In 8 mice, the recording sites in the cerebellar cortex were reconstructed on the basis of tetrode movement increments after the PC recording.

Results

Purkinje Cell Activity in the Cerebellar Cortex across the Sleep-Wakefulness Cycle

To explore cerebellar neuronal activity at the single-unit level across sleep-wakefulness cycles, we performed tetrode recordings in the cerebellar cortex. Histological results revealed valid recording sites mostly in the vicinity of the PC layer (Fig. 1). All in vivo tetrode recordings were performed during the light phase, and the duration of each recording was 139.5 ± 4.9 min on average (mean ± SEM, n = 10 mice).

Fig. 1.

Locations of recording sites in the cerebellar cortex. A A DAPI-stained coronal section illustrating a representative recording site in the cerebellar cortex (dashed circle) (scale bar, 200 μm). B Schematic of the recording sites in the cerebellar cortex (blue circles, n = 10). Note that, in 8 of the mice, the recording site was estimated by the tetrode movement distance between the initial cortical recording and final DCN recording.

As the sole output neurons of the cerebellar cortex, PCs integrate information in this area, and thereafter strongly control neuronal activity in the downstream DCN. Consequently, we first investigated the activity of PCs over many episodes of distinct sleep-wakefulness states. The putative PCs were initially determined in vivo by their spontaneous intense cellular activity. Typically, a putative PC showed an autocorrelogram with a profound central trough (Fig. 2A) and a unimodal inter-spike interval distribution with a peak at 12–30 ms (Fig. 2A). Afterwards, the putative PCs were further confirmed by the presence of both simple and complex spikes during unit clustering (Fig. 2B–D). The complex spikes were manually discriminated on the basis of the presence of a 1–3 ms slow component together with their stereotypical shape (i.e., board monophasic negative waves, Fig. 2D). For each putative PC, the presence of a pause (> 10 ms) in simple spike firing after a complex spike was verified (Fig. 2C, E). A total of 33 putative PCs were determined, showing an average firing rate of 38.7 ± 3.9 Hz.

Fig. 2.

Electrophysiological classification of putative Purkinje cells in vivo. A Autocorrelograms, average filtered waveforms, and inter-spike interval (ISI) distributions of two putative Purkinje cells. B Trace of continuous recording from the Purkinje cell layer. C Two simple spikes (red asterisks) and four complex spikes (green asterisk) magnified from the recording in panel B (red triangle). Note the pause of simple spikes after complex spikes. D Superimposed waveforms of 100 complex spikes (CS, upper) and simple spikes (SS, lower) in B. E Autocorrelograms of the complex spikes (CS, upper) and simple spikes (SS, lower) in B. The cross-correlogram between the complex spikes (CS) and simple spikes (SS) shows a decrease of simple spike firing after the occurrence of a complex spike.

We then calculated the average firing rates of putative PCs during each state, and compared their quantitative features across states. The brain state was determined by oscillation of the LFP in the frontal cortex and head/neck movements (Fig. 3A). The proportion of each state in our recordings was similar to that in previous reports [42, 43] (Fig. 3B). We found that the average firing rates of putative PCs were highest during wakefulness (Fig. 3A, C). Furthermore, the high PC activity was always associated with head/neck movements (Fig. 3A, upper panel). Statistical analysis revealed that the firing rates of PCs during wakefulness were significantly higher than those in NREM sleep (t₍₃₂₎ = − 3.8725, P < 0.001, paired t test, n = 33, Fig. 3C). In addition, we found a significant difference in the firing rates of PCs between REM sleep and wakefulness (t₍₃₂₎ = − 2.7429, P = 0.0099, Fig. 3C). However, there was no significant difference in the firing rates of PCs between NREM and REM sleep (t₍₃₂₎ = 0.8262, P = 0.4148, Fig. 3C). Considering that the PCs showed several firing patterns, such as a regular pattern during NREM and a burst-like pattern during awake and REM (Fig. 3A), we further calculated the coefficient of variance (CV) of PC spikes during each state. We found that the CV of PC spikes showed changes similar to the firing rates across distinct states (awake vs NREM: t₍₉₎ = 3.937, P = 0.004; NREM vs REM: t₍₉₎ = 0.094, P = 0.927; awake vs REM: t₍₉₎ = 2.333, P = 0.048, paired t test, n = 10 mice; Fig. S1A).

Fig. 3.

Firing of putative Purkinje cells across distinct states. A High-pass filtered traces showing representative firing of single putative Purkinje cell across the wakefulness (upper), NREM sleep (middle), and REM sleep (lower) states. The states were determined by the local field potential (LFP) in the frontal cortex and head/neck movements. B Percentage of time in awake, NREM sleep, or REM sleep state during recordings, averaged from 10 mice. Data are expressed as the mean ± SEM. C Averaged firing rate of putative Purkinje cells (n = 33) in each state. **P < 0.01; ***P < 0.001; n.s. Not significant.

In addition, similar to previous in vivo electrophysiological studies [44, 45], we recorded complex spikes with an average rate of 1.05 ± 0.09 Hz in putative PCs (n = 33). However, there were no significant changes in complex spike rates across states (awake vs NREM: t₍₃₂₎ = 0.7363, P = 0.5168; NREM vs REM: t₍₃₂₎ = –0.7117, P = 0.6333; awake vs REM: t₍₃₂₎ = 0.2357, P = 0.8358), indicating that climbing fiber input is unlikely to be the main signal source affecting state transitions.

Changes in PC activity have been demonstrated to be fast and temporally precise, participating in both the accurate execution and error-correction of motor behavior in the awake state [2, 4]. Therefore, we further investigated whether a fast and dynamic change of PC activity occurs during state transition. Based on the criteria listed in a recent study [23], we found that the firing activity of putative PCs increased significantly from NREM sleep to wakefulness (t₍₃₂₎ = – 2.8529, P = 0.0075, paired t test, n = 33, Fig. 4A, B), with 15/33 cells showing greater activity (> baseline + 1.5 SD) during wakefulness than NREM sleep (Fig. 4C). Notably, in those PCs with greater activity, the average latency between increased PC firing and wakefulness onset was 2.2 ± 0.5 s (n = 15). Conversely, the firing activity of putative PCs decreased when mice fell asleep (t₍₃₂₎ = 2.547, P = 0.0159, paired t test, n = 33, Fig. 4D, E), with 9/33 cells showing lower activity (< baseline − 1.5 SD) during NREM sleep than wakefulness (Fig. 4F). The latency between decreased PC firing and NREM sleep onset was 3.4 ± 0.4 s (n = 9). In addition, the putative PCs tended to increase their activity at the transition from REM sleep to wakefulness (t₍₃₂₎ = –2.5262, P = 0.0167, paired t test, n = 33, Fig. 4H, I) with a mean latency of 1.6 ± 0.4 s. We also plotted the NREM–wakefulness modulation index (F_NREM − F_wake)/(F_NREM + F_wake), where F is the average firing rate within each state versus the REM–NREM modulation index (F_REM − F_NREM)/(F_REM + F_NREM) for each PC. The NREM–wakefulness modulation index of PCs showed a unimodal distribution centered on 0.06 (Fig. S1B), which corresponded to the awake state. In contrast, the REM–NREM modulation of PCs exhibited a bimodal distribution centered on − 0.08 (Fig. S1B), with the peak corresponding to the REM–off state. These results thus suggested that the PCs are both wake–on and REM–off neurons. In contrast, there was no significant change in activity at the NREM-to-REM (t₍₃₂₎ =− 0.1442, P = 0.8862, paired t test, n = 33, Fig. S2A) or REM-to-NREM transition (t₍₃₂₎ = − 1.5295, P = 0.136, Fig. S2B). Taken together, our results provide evidence that endogenous putative PC activity is specifically correlated with and occurs prior to the state transition from sleep to wakefulness.

Fig. 4.

Dynamics of putative Purkinje cell activity at state transitions. A Representative LFP power spectrum and head/neck movement traces illustrating the NREM–wakefulness transition. B Increased firing activity of putative Purkinje cells preceding the transition from NREM sleep to wakefulness (n = 33). C Proportions of Purkinje cells showing increased, decreased, or minimal responses to the NREM–wakefulness transition. D Representative LFP power spectrum and head/neck movement traces illustrating the wakefulness–NREM transition. E Decreased firing activity of putative Purkinje cells preceding the transition from wakefulness to NREM sleep (n = 33). F Proportions of Purkinje cells showing increased, decreased, or minimal responses to the wakefulness–NREM transition. G Representative LFP power spectrum and head/neck movement traces illustrating the REM–wakefulness transition. H Increased firing activity of putative Purkinje cells preceding the transition from REM sleep to wakefulness (n = 33). I Proportions of recorded Purkinje cells showing increased, decreased, or minimal responses to the REM–wakefulness transition. Color scale indicates the power (mV²) of raw power spectral density in A, D, and G; data are expressed as the mean ± SEM; shaded areas indicate SEM in B, E, and H.

Non-PC Unit Activity in the Cerebellar Cortex Across the Sleep–Wakefulness Cycle

The firing activity of PCs is modulated by inputs from non-PC units such as granule cells and interneurons in the cerebellar cortex [46]. Therefore, we evaluated the firing activity of non-PC units during each state. In 10 mice, we recorded and clustered 42 non-PC units, which showed a relatively low average firing rate of 3.9 ± 0.1 Hz. The firing rates of non-PC units were significantly lower than those of the putative PCs (Z = 7.2312, P < 0.001, Wilcoxon rank sum test). In addition, no complex spikes were found in these non-PC units. Similar to the PCs, the non-PC units also exhibited brain-state-dependent activity (Fig. 5A). Notably, the lowest firing activity of non-PC units occurred during NREM sleep (NREM vs wakefulness: t₍₄₁₎ = − 3.1404, P = 0.0031; NREM vs REM: t₍₄₁₎ = − 3.7565, P < 0.001, paired t test, n = 42, Fig. 5A). In contrast, there was no significant difference in firing rates between wakefulness and REM sleep (t₍₄₁₎ = − 1.4469, P = 0.1555, Fig. 5A).

Fig. 5.

Dynamics of non-PC unit activity at state transitions. A Averaged firing rates of non-PC units in each state (n = 42). Data are expressed as the mean ± SEM (**P < 0.01, ***P < 0.001, n.s., Not significant). B REM–NREM activity difference versus wakefulness–NREM activity difference. Each symbol represents one neuron (n = 42 units from 10 mice). C Increased firing activity of non-PC units preceding the transition from NREM sleep to wakefulness (n = 42). D Proportions of non-PC units showing increased, decreased, or minimal responses to the NREM–wakefulness transition. E Decreased firing activity of non-PC units preceding the transition from wakefulness to NREM sleep (n = 42). F Proportions of non-PC units showing increased, decreased, and minimal responses to the wakefulness–NREM transition. G Firing activity of non-PC units at the transition from REM sleep to wakefulness (n = 42). H Proportions of non-PC units showing increased, decreased, or minimal responses to the REM-wakefulness transition. Data are expressed as the mean ± SEM (shaded areas indicate SEM) in C, E, and G.

Again, we plotted the NREM–wakefulness modulation index (F_NREM − F_wake)/(F_NREM + F_wake), where F is the average firing rate in each state versus the REM–NREM modulation index (F_REM − F_NREM)/(F_REM + F_NREM). The NREM–wakefulness modulation index of non-PC units showed a unimodal distribution centered on 0.18 (Fig. 5B), which corresponded to the awake state. Moreover, the REM–NREM modulation of non-PC units also exhibited a unimodal distribution centered on 0.17 (Fig. 5B), with the peak corresponding to the REM sleep state. These results thus suggested that the non-PC units in the cerebellar cortex are both wake-on and REM-on active.

Next, we recorded the activity of non-PC units during various state transitions, and found that their activity increased significantly at the NREM sleep-to-wakefulness transition (t₍₄₁₎ =  − 2.4886, P = 0.0170, paired t test, n = 42, Fig. 5C), with 24/42 cells showing greater activity during wakefulness than NREM sleep (Fig. 5D). For the non-PC units with greater activity, the mean latency between increased non-PC firing and wakefulness onset was 0.0 ± 0.6 s (n = 24). Conversely, the firing activity of non-PC units decreased when the mice fell asleep (t₍₄₁₎ = 3.9863, P < 0.001, paired t test, n = 42, Fig. 5E), with 24/42 cells showing lower activity during NREM sleep than wakefulness (Fig. 5F). The mean latency between decreased non-PC firing and NREM sleep onset was 1.4 ± 0.9 s (n = 24). In contrast to the putative PCs, the non-PC units did not increase their activity at the transition from REM sleep to wakefulness (t₍₄₁₎ =  − 0.0319, P = 0.9747, paired t test, n = 42, Fig. 5G, H). Likewise, there was no significant change in activity at the NREM-to-REM or REM-to-NREM transition (both P > 0.05, paired t test, n = 42). In sum, these results suggested that the activity of non-PC units varies with brain states, and they are especially active in the wakefulness and REM sleep states.

The PCs and non-PCs exhibited similar brain-state-dependent activities, suggesting that they interacted with each other. To test this possibility, we calculated the excitatory monosynaptic connection between non-PCs and PCs (non-PC × PC pairs) using spike cross-correlation between simultaneously-recorded non-PCs and PCs (with non-PC spikes at t = 0). We detected monosynaptically-connected pairs from the neuronal pools recorded by the same tetrode (Fig. 6A). Notably, putative PCs tended to fire after the firing of a non-PC (Fig. 6A). Of 19 cross-correlograms, 7 (36.8%) had a short-latency (< 3 ms) and unimodal peak (Fig. 6B), indicating monosynaptic inputs from non-PCs to PCs. Indeed, the probability of short-latency (< 3 ms) PC firing after a non-PC spike was significantly higher than that among non-PCs (Z = 2.113, P = 0.0346, Wilcoxon rank sum test, Fig. 6C, D). These results suggested that there are significant interactions between the non-PCs and PCs across sleep-wakefulness cycles.

Fig. 6.

Monosynaptic connections between non-PC units and PCs. A Raw traces showing tetrode recording (channels 1–4) in the cerebellar cortex. The high-pass filtered trace (below) shows simultaneously-recorded non-PC and PC units in channel 2. Notably, the putative PC (red) tended to fire after the firing of a non-PC unit (blue) within a 3-ms window. B Autocorrelograms and cross-correlogram between the non-PC unit (blue) and the PC unit (red) illustrated in A. C Normalized cross-correlograms (CCG) for non-PC × PC pairs (n = 19) and non-PC × non-PC pairs (n = 40). D Probability of excitatory monosynaptic connection between non-PCs and PCs was significantly greater than that between non-PC units. Data are expressed as the mean ± SEM (*P < 0.05).

Neuronal Activity in the DCN Across the Sleep-Wakefulness Cycle

The DCN are the final outputs of the cerebellum, innervating several areas critical for promoting or maintaining wakefulness [15–17, 19]. Moreover, their firing is strongly controlled by inhibitory PC inputs [46, 47]. Consequently, we further investigated whether the firing of DCN units is affected by the changes in firing activity of PCs across distinct states. To this end, we moved the tetrodes from the vicinity of the PC layer into the DCN in 8 mice. The recording sites in the DCN were verified post hoc (Fig. 7A, B). We recorded from and clustered a total of 21 DCN units. The mean firing rate of the DCN units was 25.8 Hz ± 5.1 Hz, ranging from 2.7 to 94.9 Hz. On average, the firing rates of DCN neurons were significantly lower than those of PCs in both wakefulness and NREM sleep (awake: Z = − 2.3067, P = 0.0211; NREM: Z = − 2.2002, P = 0.0278, Wilcoxon rank sum test). Although there were no significant differences in the averaged firing rates among distinct states (P > 0.05, Fig. S3A), statistical analysis revealed decreased firing activity of DCN units at the transition from NREM sleep to wakefulness (t₍₂₀₎ = 2.1770, P = 0.0416, paired t test, n = 21, Fig. 7C), with 9/21 cells showing lower activity during wakefulness than NREM sleep (Fig. 7D). In these units, the mean latency between the decreased firing and wakefulness onset was 2.0 ± 0.7 s (n = 9). Likewise, the firing activity of DCN units tended to decrease at the transition from REM sleep to wakefulness, although it failed to reach a significant level (t₍₂₀₎ = 1.8572, P = 0.0781, Fig. 7E, F). Moreover, there were no significant changes in firing activity during the wakefulness–NREM (t₍₂₀₎ = − 0.8008, P = 0.4326, paired t test, n = 21, Fig. 7G, H), NREM–REM (P > 0.05, Fig. S3B, C), and REM–NREM transitions (P > 0.05, Fig. S3D, E). These results suggested that, in line with the change of PC firing, the downstream DCN units also exhibit significant change in firing activity, especially during the transition from NREM sleep to wakefulness.

Fig. 7.

Dynamics of DCN unit activity at state transitions. A A DAPI-stained coronal section showing a representative recording site in the DCN (arrow; scale bar, 200 μm). B Schematic of the recording sites in the DCN (red circles, n = 8 mice). C Decreased firing activity of DCN units preceding the transition from NREM sleep to wakefulness (n = 21). D Proportions of the DCN units showing increased, decreased, or minimal responses to the NREM–wakefulness transition. E Firing activity of DCN units at the transition from REM sleep to wakefulness (n = 21). F Proportions of DCN units showing increased, decreased, or minimal responses to the REM–wakefulness transition. G Firing activity of DCN units at the transition from wakefulness to NREM sleep (n = 21). H Proportions of DCN units showing increased, decreased, or minimal responses to the wakefulness–NREM transition. Data are expressed as the mean ± SEM (shaded areas indicate SEM in C, E, and G).

Discussion

In order to determine the specific role of the cerebellum in the regulation of sleep and wakefulness, it is necessary to detect the activity patterns of cerebellar units during this process. To this end, we performed in vivo multi-unit recordings in naturally sleeping mice, and found that PCs in the cerebellar cortex exhibited significantly increased firing activity prior to the transition from sleep to wakefulness. Moreover, the increased PC activity resulted from inputs from low-frequency firing non-PC units in the cerebellar cortex. Downstream of the inhibitory PCs, neurons in the DCN manifested decreased firing during the transition from NREM sleep to wakefulness. Our results, together with information about cerebellar–hypothalamic and cerebellar–ventral thalamic circuitry, highlight the temporal features by which the cerebellum is actively involved in regulating sleep and wakefulness.

The cerebellum has long been known to be critical for motor planning [1], motor execution [2–4, 48], and motor learning [5–7]. Recently, it has also been implicated in several non-motor functions such as cognition [49], reward [50], social behavior [51], spatial learning [52, 53], and fear conditioning [54]. Aside from these non-motor functions, we showed here that neurons in the mouse cerebellum exhibited state-dependent activities across sleep-wakefulness cycles (Figs. 4, 5). In particular, the changes of cerebellar neuronal firing preceded the transitions between states. Therefore, the cerebellum might be a novel candidate for regulating sleep and/or wakefulness states via its interaction with arousal neurons in the ventral thalamus and hypothalamus [14]. In support of this hypothesis, sleep disturbance is common in patients suffering from cerebellar degeneration [8–10]. Likewise, increased sleep often occurs in cerebellectomized cats [11].

Previous studies combining fMRI and EEG recordings have indicated that cerebellar activity is lower during NREM than during wakefulness [29, 55]. Relative to the imaging technologies, invasive multi-unit recordings allow recordings of single unit activity at higher temporal and cellular specificity in the cerebellum [56–58]. In this study, we classified cerebellar cortical units into putative PCs and non-PCs according to their firing characteristics (Fig. 2). Notably, we found that both the putative PCs and non-PCs increased their activities prior to the transition from NREM sleep to wakefulness (Figs. 4, 5). Moreover, our results revealed that the increased PC activity resulted from the inputs of non-PCs. Along this line, both types of neurons decreased their firing activity during the transition from pre-sleep wakefulness to NREM sleep (Figs. 4, 5). Consequently, our results not only resemble the functional imaging findings in human subjects [29, 55], but also demonstrate the cellular mechanisms underlying cerebellar cortical activation during the NREM–wakefulness transition. Interestingly, significant firing changes of cerebellar cortical units did not occur during the transition between the NREM–REM or REM–NREM sleep states (Fig. S2), indicating that the cerebellar cortex might be involved in the transition between specific states. However, future experiments using Cre-dependent optogenetic or chemogenetic manipulation are needed. These experiments can reveal the control of specific neuronal activity in the cerebellar cortex [4, 59, 60], and thus determine its role in regulation of the sleep-wakefulness transition [24, 61].

Previous studies showed that the activities of PCs and DCN neurons increase during REM sleep [25–28]. In contrast, we found here that the activity of PCs decreased in REM sleep. Considering that the firing activity of PCs is strongly correlated with movement and muscle tone [2–4, 48], it is reasonable to conclude that our current results are compatible with the loss of muscle tone during REM sleep. Together with temporal dynamics of the PC activity [2–4], we speculate that PCs might be involved in the state transition by means of the change of firing activity, rather than the average firing rate. However, this speculation needs to be tested to see whether the state transition can be induced by the rapid optogenetic manipulation of PCs.

The averaged firing rates of non-PCs recorded in the cerebellar cortex are similar to the findings reported in previous in vivo electrophysiological studies [57, 62]. Considering that the overwhelming majority of neurons in the cerebellar cortex are excitatory granule cells [46, 47], our results most likely reveal a change in granule cell activity during the state transition. However, we cannot exclude the possibility that distinct firing patterns occur in other non-PCs such as inhibitory Golgi cells and/or basket cells [62]. Indeed, as shown in Fig. 5C, D, the non-PCs showed both increased (57%) and decreased (17%) firing activity during the NREM-to-awake transition. This result thus leads to the interpretation that both types of non-PCs (excitatory granule cells and inhibitory Golgi cells) may be involved in the state transition.

In this study, there was a difference in the latency to the state transition between the putative PCs (2.2 ± 1.8 s) and the non-PCs (0.0 ± 2.7 s). This result seemed to contradict our proposal that there are monosynaptic inputs from the non-PCs to the PCs as shown in Fig. 6. Nevertheless, it should be noted that only 36.8% of the non-PCs had monosynaptic inputs to the PCs. Moreover, we showed that non-PCs had both increased and decreased firing activity during the NREM-to-awake transition, indicating the firing complexity of non-PCs. Previously, the cerebellar cortex has been revealed to contain a huge number of granule cells, which are inhibited by their neighboring Golgi cells [46, 47, 62]. Moreover, the activity of Golgi cells can be indirectly influenced by PCs [63]. Therefore, it is reasonable to expect that common inputs to both granule cells and Golgi cells may result in complex firing patterns in non-PCs, as evidenced by their greater variance of latency (± 2.7 s) to the state transition.

Responding to the increased PC firing, the downstream DCN units exhibited significantly decreased firing activity during the transition from NREM sleep to wakefulness. Anatomically, as the final integration and output of the cerebellum, the DCN are strongly and reciprocally connected to several regions associated with sleep-wakefulness [15–17, 19–21]. Therefore, the change in neuronal activity enables the DCN to participate in the NREM-wakefulness transition. This notion is further supported by indirect evidence that activation of the ventral thalamus, one of the downstream areas of the DCN, induces a rapid state transition from NREM sleep to wakefulness [18]. Nevertheless, future research is needed to test whether direct activation of the DCN terminals in the ventral thalamus has a similar effect on the NREM-wakefulness transition.

It should also be noted that there was no significant change in the DCN neuronal activity during the transition from REM sleep to wakefulness (Fig. 7), implying that the cerebellum as a whole is unlikely to participate in the regulation of the REM-wakefulness transition. In support of this idea, activation of the ventral thalamus does not promote the transition from REM sleep to wakefulness [18]. As noted, however, the firing activity of PCs increases during the transition from REM sleep to wakefulness (Fig. 4), which is associated with the movements. Therefore, the question remained as to why the neuronal activity in the downstream DCN remained relatively stable. Evidence has accumulated that the DCN consists of at least three subgroups: glutamatergic, GABAergic, and glycinergic cells [64, 65]. In particular, the glutamatergic DCN neurons are inhibited by their GABAergic neighbors [65]. In this study, we expected that the increased PC activity would inhibit the downstream neurons in the DCN. Meanwhile, the inhibited GABAergic DCN cells can disinhibit the glutamatergic DCN neurons. We speculate that these two contradictory effects on the glutamatergic DCN neurons, to some extent might result in overall unchanged activity in the DCN. In addition, it has been demonstrated that the excitatory neuromodulator acetylcholine is released in the cerebellum during REM sleep [66, 67], which might antagonize the inhibitory PC effects on DCN cells during the transition from REM sleep to wakefulness.

The purpose of this study was to provide electrophysiological evidence that the cerebellum has the potential to regulate the sleep-wakefulness transition. We revealed here the temporal features of single-unit activity in the mouse cerebellum, which are essential for determining whether and how the cerebellum is actively involved in regulating the sleep-wakefulness transition. Evidence has accumulated that the cerebellum is involved in various non-motor functions by means of its connections with the cerebral cortex [1, 3, 4], hippocampus [52, 53], pontine nuclei [68], and other subcortical areas [50, 51, 54]. In this study, we focused on the cerebrocerebellum because sleep and wakefulness have been strongly associated with activity in the cerebral cortex [69, 70]. However, future experiments are needed to test whether the firing patterns of each cerebellar region show a similar spatiotemporal pattern in the various sleep stages.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Conflict of interest

The authors declare that they have no conflict of interest.