REM Sleep Twitches Rouse Nascent Cerebellar Circuits: Implications for Sensorimotor Development

Abstract

The cerebellum is critical for sensorimotor integration and undergoes extensive postnatal development. During the first postnatal week in rats, climbing fibers polyinnervate Purkinje cells and, before granule cell migration, mossy fibers make transient, direct connections with Purkinje cells. Activity-dependent processes are assumed to play a critical role in the development and refinement of these and other aspects of cerebellar circuitry. However, the sources and patterning of activity have not been described. We hypothesize that sensory feedback (i.e., reafference) from myoclonic twitches in sleeping newborn rats is a prominent driver of activity for the developing cerebellum. Here, in 6-day-old rats, we show that Purkinje cells exhibit substantial state-dependent changes in complex and simple spike activity—primarily during active sleep. In addition, this activity increases significantly during bouts of twitching. Moreover, the surprising observation of twitch-dependent increases in simple spike activity at this age suggests a functional engagement of mossy fibers before the parallel fiber system has developed. Based on these and other results, we propose that twitching comprises a unique class of self-produced movement that drives critical aspects of activity-dependent development in the cerebellum and other sensorimotor systems.

INTRODUCTION

During active (or REM) sleep, skeletal muscles throughout the body twitch, causing jerky movements of the limbs (Blumberg et al., 2013a), eyes (Seelke et al., 2005), and even whiskers (Tiriac et al., 2012). Across many species of mammals, twitching is one of the most conspicuous of all infant behavior (Gramsbergen et al., 1970; Jouvet-Mounier et al., 1970). The numbers tell the story: Each day, across all skeletal muscles, infant rats produce hundreds of thousands, if not millions, of twitches (Blumberg et al., 2013b).

Similar to other forms of spontaneous activity in the developing nervous system (Katz and Shatz, 1996; O'Donovan, 1999), sensory feedback (or reafference) from twitching limbs has been implicated in the self-organization of spinal circuits and somatotopic maps (Blumberg et al., 2013b; Khazipov et al., 2004; Petersson et al., 2003). This implication rests in part on the quantity and patterning of neural activity associated with twitch-related reafference. First, using novel methods developed for in vivo recording in rat pups, Khazipov and colleagues (2004) discovered that twitches and other movements trigger oscillatory events, referred to as spindle bursts, in primary somatosensory cortex. Second, subsequent work has demonstrated that, at least under some testing conditions, twitch-related reafference potently increases neural activity, whereas wake-related reafference does not (Mohns and Blumberg, 2010; Tiriac et al., 2012). These and other observations are building a case for considering twitching as a unique form of spontaneous activity that contributes to the development of sensorimotor systems (Blumberg et al., 2013b).

Any discussion of sensorimotor integration must include the cerebellum, a hindbrain structure that receives sensory and motor signals from spinal cord, brainstem, and cerebral cortex (Apps and Garwicz, 2005; Huang et al., 2013; Odeh et al., 2005). Inputs to the cerebellum arrive primarily through one of two pathways (Eccles et al., 1967): climbing fibers that arise exclusively from the inferior olive and mossy fibers that arise from several brainstem nuclei (e.g., pontine gray, external cuneate). Climbing fibers and mossy fibers directly or indirectly modulate the activity and plasticity of Purkinje cells (Ito, 1989), the sole output of the cerebellar cortex.

Cerebellar circuitry develops substantially over the first three postnatal weeks in rats (Altman, 1972; Shimono et al., 1976; Wang and Zoghbi, 2001). Although molecular cues play important roles in shaping cerebellar development (Sillitoe and Joyner, 2007; Wang and Zoghbi, 2001), activity-dependent mechanisms may be equally important. For example, neural activity contributes to synapse elimination and climbing fiber translocation at Purkinje cells (Andjus et al., 2003; Kakizawa et al., 2000; Kano and Hashimoto, 2011; Mikuni et al., 2013; Watanabe and Kano, 2011), as well as spinocerebellar afferent topography (Tolbert et al., 1994). It is also likely that activity-dependent processes guide the convergence of somatotopically related climbing fiber and mossy fiber inputs onto Purkinje cells (Huang et al., 2013; Odeh et al., 2005; Pijpers et al., 2006).

Little is currently known about the sources of neural activity that drive activity-dependent development of the infant cerebellum (Hashimoto and Kano, 2013). Spontaneous neural activity may contribute. Also, sensory experiences arising from interactions with the mother and littermates could drive neural activity in the cerebellum. Indeed, despite their immaturity in newborns, climbing fibers and mossy fibers transmit sensory information from the periphery to Purkinje cells (Puro and Woodward, 1977a; 1977b).

Here we test the novel hypothesis that sensory feedback from myoclonic twitching drives activity in nascent cerebellar circuits. Recording from cerebellar cortex in 6-day-old rats, we report that Purkinje cell activity increases substantially and predominantly during periods of twitching. These results reveal for the first time a prodigious source of spatiotemporally structured activity to this critical sensorimotor structure during a period of extreme immaturity and rapid developmental change.

MATERIALS AND METHODS

Subjects

Six-day-old male and female Sprague-Dawley Norway rats (Rattus norvegicus) from nine litters were used. Litters were culled to eight pups within three days of birth. Mothers and their litters were housed and raised in standard laboratory cages (48 × 20 × 26 cm). Food and water were available ad libitum. All animals were maintained on a 12-h light-dark schedule with lights on at 0700 h. All experiments were carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication No. 80-23) and were approved by the Institutional Animal Care and Use Committee of the University of Iowa.

Surgery

A pup with a visible milk band was removed from the home cage. Under 2–5% isoflurane anesthesia, 2–4 bipolar stainless steel hook electrodes (50 µm diameter; California Fine Wire, Grover Beach, CA) were implanted bilaterally into the nuchal muscle and in one or both of the hindlimbs and secured with collodion. A ground wire on the back was looped transdermally and secured with collodion. As described previously (Karlsson et al., 2005), a custom-built head-fix apparatus, secured to the skull with cyanoacrylate adhesive, allowed for attachment to the earbars of a stereotaxic apparatus (David Kopf Instruments, Tujunga, CA). At the end of the surgery, which lasted approximately 10 min, the pup’s trunk was lightly wrapped in gauze and the pup was placed in a humidified incubator maintained at thermoneutrality (35 °C) to recover for at least 1 h. Before placing the subject into the testing apparatus, the subject was anesthetized with 2–5% isoflurane anesthesia and three small holes were drilled in the skull for insertion of the electrode, ground wire, and thermocouple (the last for measuring brain temperature). After 1–2 hours of acclimation in the testing apparatus, pups typically cycle regularly between sleep and wake. Despite prolonged isolation from their mother, pups exhibit regular sleep-wake cycles in these temperature-controlled conditions for many hours (Seelke and Blumberg, 2005).

Cerebellar neurophysiology

A total of nine 6-day-old rats were used to produce 30 units for analysis. To record from Purkinje cells, 16-channel silicon depth electrodes (NeuroNexus, Ann Arbor, MI; Models A1×16, A1×16-Poly2), with impedances ranging from 1–4 MΩ, were connected to a data acquisition system (Tucker-Davis Technologies, Alachua, FL) that amplified (10,000×) and filtered (500–5000 Hz band-pass) the neural signals. Prior to insertion, the electrode surface was coated with fluorescent DiI solution (Life Technologies, Grand Island, NY) for subsequent histological verification of electrode placement. A Ag/AgCl ground electrode (Medwire, Mt. Vernon, NY, 0.25 mm diameter) was placed into the cerebral cortex. EMG electrodes were connected to a differential amplifier (A-M Systems, Carlsborg, WA) that amplified (10,000×) and filtered (300–5000 Hz band-pass; 60 Hz notch filter) the EMG signal. Neural and EMG signals were sampled at 12.5 kHz and 1 kHz, respectively, using a digital interface and Spike2 software (Cambridge Electronic Design, Cambridge, UK). Brain temperature, maintained at 36–37°C, was measured using a fine-wire thermocouple (Omega Engineering, Stamford, CT) placed in the cerebral cortex contralateral to the ground wire.

Using a stereotaxic apparatus, the electrode was positioned above cerebellar cortex (coordinates: 1.0–1.5 mm caudal to lambda, 1.5–2.0 mm lateral to midline, 10–12° angle) and was slowly lowered until action potentials were detected on one or more recording channels. Purkinje cell activity was identified by the occurrence of large-amplitude action potentials (signal-to-noise ≥ 2:1) and the presence of complex spike waveforms as described by others (Crepel, 1971; Puro and Woodward, 1977a). Using procedures similar to those described previously (Karlsson et al., 2005; Mohns and Blumberg, 2010; Tiriac et al., 2012), data acquisition began after Purkinje cell unit activity was identified and had stabilized for at least 10 min. Each recording session lasted for 15 min and comprised continuous collection of neural and EMG data as the pup cycled freely between sleep and wake. During acquisition, an experimenter monitored the subject’s behavior and digitally marked the occurrence of wake and active sleep movements in synchrony with the physiological data. The experimenter was always blind to the physiological data while scoring behavior. Following the first recording session, the electrode was lowered further and additional recordings were acquired if new Purkinje cell unit activity was identified.

Histology

After the final recording session, the pup was overdosed with sodium pentobarbital (1.5 mg/g) and perfused transcardially with phosphate buffered saline followed by 4% paraformaldehyde. Recording sites were verified by visualizing the DiI tract with fluorescence illumination at 5–10× magnification using a Leica microscope (Leica Microsystems, Buffalo Grove, IL). Using a calibrated reticle, the location of the unit recordings was determined.

Data Analysis

Spike sorting & burst analysis

Spike sorting was performed using template matching in Spike2 (Cambridge Electronic Design, Cambridge, UK). The distributions of waveforms comprising each template were analyzed using the principal component analysis tool; waveforms greater than 3.5 standard deviations outside of each distribution were excluded. After spikes were sorted into units and converted into events, a second analysis was performed to separate complex spikes from simple spikes using a custom-written burst script for Spike2. Bursts were identified as multispike events with interspike intervals ≤15 ms. All unit activity comprising a burst was subsequently extracted from the raw record of unit activity, resulting in separate simple spike and complex spike records for each unit.

Identification of behavioral state

Sleep and wake periods were defined by calculating mean EMG activity, based on methods described previously (Karlsson et al., 2005; Mohns and Blumberg, 2010; Tiriac et al., 2012). Briefly, each nuchal or hindlimb EMG record was rectified and smoothed (tau = 0.001 s). The mean amplitude of high muscle tone and atonia was calculated from five representative 1-s EMG segments, and the midpoint between the two was used to establish a threshold for determining state. The EMG signal was then dichotomized into periods of high tone (indicative of wake) and atonia (indicative of sleep). Sleep bouts were further divided into periods of quiet sleep and active sleep: quiet sleep was characterized by hypotonia/atonia and behavioral quiescence, and active sleep was characterized by the appearance of myoclonic twitches against a background of muscle atonia (Blumberg and Seelke, 2010; Seelke and Blumberg, 2008). For three pups with more than one useable EMG record, nuchal and hindlimb EMG events were merged into one event channel.

Analysis of state-dependency

Mean unit firing rates were determined across all periods of wake, quiet sleep, and active sleep for each pup. For any single period, firing rates that exceeded three times the standard deviation were excluded as outliers. To assess whether units were significantly more active during different behavioral states, independent pairwise comparisons of the mean firing rates during wake, quiet sleep, and active sleep (AS) were assessed using the Wilcoxon matched-pair signed-rank test (SPSS, IBM, Armonk, NY). After characterizing units as AS-On, Sleep-On, Wake-On, AS/Wake-On, or State-Independent, mean firing rates for each unit were used in a summary analysis. Mean firing rates were also assessed for all units during each state. For all tests, including those described below, alpha was set at 0.05 and a Bonferroni correction was used when appropriate.

Event correlations

Using methods similar to those described previously (Mohns and Blumberg, 2010; Tiriac et al., 2012), twitches were identified as discrete EMG events with amplitudes exceeding three times the mean EMG baseline during atonia. For pooled data across all subjects, a twitch-triggered cross-correlogram of complex spike activity was constructed for a 1-s window (10-ms bins) around twitch onset. We tested statistical significance for the cross-correlogram by jittering twitch events 1000 times within a 400-ms window using PatternJitter (Amarasingham et al., 2012; Harrison and Geman, 2009). Finally, using a custom-written Matlab program, a 95% probability threshold (p < 0.05) was determined for the bins spanning 250 ms on either side of a twitch. For presentation purposes, cross-correlograms were smoothed using a 3-bin moving window. An identical method was used to test the relationship between simple spikes and twitching.

Window analysis

As a population, complex spikes exhibited rhythmic activity after a twitch. For each unit, a “separation procedure” (Bloedel and Ebner, 1984; Nicholson and Freeman, 2003) was used to quantify the number of complex spikes within each of three time windows after a twitch: Window 1 at 0–50 ms, Window 2 at 100–150 ms, and Window 3 at 200–250 ms. If 45% or more of the complex spikes for that unit were associated with only one window, it was designated as either a W1 unit (n=16), W2 unit (n=4), or W3 unit (n=3). Seven units exhibited no consistent window preference (NP). For each type of unit, cross-correlational analyses were performed to determine their relationship with twitching. Simple spike activity was assessed based on that unit’s complex spike window assignment and cross-correlational analyses were performed. Again, significance was determined using the jitter procedure described above.

Complex spike autocorrelations

Autocorrelations of complex spikes were analyzed in 5 ms bins using the time series command in JMP (SAS, Cary, NC). For each cerebellar unit, autocorrelograms and spectral density plots were created. Using a white noise test, we tested against the null hypothesis that the production of complex spikes had no periodic component. Both Fisher’s Kappa and Bartlett’s Kolmogorov-Smirnov statistic were relied upon to establish significant periodicity.

RESULTS

Identifying Complex and Simple Spikes

We used 16-channel silicon electrodes to record from the cerebellar cortex of head-fixed, unanesthetized 6-day-old rats (n=30 units across nine subjects) as they cycled between sleep and wake. Electrodes spanned multiple cortical layers in the medial cerebellar cortex, including vermis, simplex, and crus I and II (Figure 1A).

Figure 1.

Purkinje cell activity in 6-day-old rats during sleep and wake. (A) The location of the recording sites (white dots) in this study. All sites were located in the Purkinje cell layer and typically yielded more than one unit. The yellow arrow indicates the track of a silicon electrode in cerebellar cortex, traversing multiple cell layers in simplex. Inset: an expanded view of the location of that track. The track was visualized by coating the electrode with fluorescent DiI. dcn, deep cerebellar nuclei; pl, Purkinje layer; egl, external granular layer. (B) Representative sample of data showing occurrences of behaviorally scored twitches (vertical green ticks), nuchal EMG, and Purkinje cell activity during sleep and wake. The neural activity (recorded from vermis) is expanded at bottom to highlight instances of complex spike (CS) and simple spike (SS) activity. (C) Frequency distribution of the number of individual spikes comprising complex spikes across all 30 Purkinje cells. Most, but not all, complex spikes were “doublets,” consistent with previous reports in infant rats (Crepel, 1971; Puro and Woodward, 1977a). Inset: Complex spikes were defined on the basis of action potentials exhibiting interspike intervals (ISIs) ≤15 ms. The example shown here has an ISI of 9 ms.

Consistent with previous studies in newborn rats (Crepel, 1971; Puro and Woodward, 1977b; 1977a; Woodward et al., 1969), we observed multi-spike bursts of Purkinje cell activity with intervening periods of regularly firing single action potentials (Figure 1B). Importantly, stimulation of the inferior olive early in development elicits complex spikes, predominantly in the form of “doublets” comprising two individual spikes (Crepel, 1971; Puro and Woodward, 1977a). Doublets were also the predominant form of complex spikes observed here (Figure 1C).

All non-complex spikes were designated as simple spikes (Figure 2A). Overall, simple spikes exhibited suppression after complex spikes, as well as higher tonic firing rates than complex spikes (Figures 2B and 2C and Table 1). Consistent with a previous report (Woodward et al., 1969), the firing rates reported in Table 1 are much lower than those seen in adults; preliminary analyses of Purkinje cell activity in older pups indicate that firing rates increase substantially by the end of the second postnatal week.

Figure 2.

(A) Interspike intervals (ISIs) for all action potentials across all pups. Complex spikes were defined as 2 or more action potentials with ISIs ≤ 15 ms (red bars). All other unit activity was classified as simple spikes (black bars). (B) Representative Purkinje cell recording from a 6-day-old rat to show a train of simple spikes. (C) Perievent histogram plotting simple spike firing rate per 10-ms bin in relation to complex spikes in a sleeping 6-day-old rat. The red bar designates the period when, by our definition of complex spikes, a simple spike could not occur. Note the suppression of simple spike activity after complex spikes as well as the presence of significantly increased simple spike activity before, but not after, complex spikes. * significant increase in relation to jittered values, p < 0.05; † significant decrease in relation to jittered values, p < 0.05.

Table 1.

Descriptive statistics of complex spike and simple spike activity across all 30 units in 6-day-old rats. Mean (SEM)

Unit Activity	Mean Inter-Spike Interval (ms)	Mean Firing Rate (Hz)	Mean Burst Duration (ms)
Complex Spikes	4.03 (.10)*	0.36 (.11)	9.15 (.26)
Simple Spikes	240.98 (14.1)	2.62 (.49)	NA

^*Inter-spike interval within a complex spike

State-Dependency of Purkinje Cell Activity

For each Purkinje cell recording, we determined the relations among complex spikes, simple spikes, and behavioral state. For many units, it was readily apparent that complex and simple spikes occurred more often during sleep than during wake, and also more often during the bursts of twitching that characterize active sleep. To quantify and test these relationships, the firing rates of complex and simple spikes for each recorded unit across behavioral states was determined; for example, AS-On units exhibited significantly higher firing rates during active sleep than during either quiet sleep or wake (Figure 3A). On the basis of these comparisons, each unit was categorized as Sleep-On, Wake-On, AS-On, AS/Wake-On, or State-Independent. For both complex spikes (73%) and simple spikes (57%), the majority of Purkinje cells exhibited some form of state-dependent activity, and the vast majority of these state-dependent units were AS-On or AS/Wake-On (Figure 3B). Of the 9 subjects, 8 contributed state-dependent units. Importantly, across all units and regardless of state-dependency, complex and simple spike firing rates were significantly greater during active sleep than during wake or quiet sleep (Figure 3C). Although various forms of state-dependent cerebellar activity have been reported in adult rats, cats, and monkeys (Andre and Arrighi, 2001; Hobson and McCarley, 1974; Mano, 1970), this is the first demonstration of state-dependent activity in the newborn cerebellum and the first demonstration of a relationship between cerebellar activity and twitching.

Figure 3.

Complex and simple spike activity is state-dependent. (A) Firing rate (Hz) of complex spikes (top) and simple spikes (bottom) during wake, quiet sleep (QS), and active sleep (AS) in Purkinje cells characterized as AS-On and AS/Wake-On (data for Sleep-On, Wake-On, and State-Independent categories not shown). * significantly different from the other two groups (p < 0.017). (B) Percentage of complex and simple spikes (n=30 in each group) characterized as AS-On, AS/Wake-On, Sleep-On, Wake-On, and State-Independent. (C) Firing rate (Hz) of complex spikes (top) and simple spikes (bottom) during wake, quiet sleep, and active sleep in Purkinje cells regardless of state. * significantly different from the other two groups (p < 0.017).

Purkinje Cells Increase Activity in Close Association with Twitching

In light of evidence that neural activity in the thalamus, cerebral cortex, and hippocampus is activated in response to sensory feedback from twitching (Khazipov et al., 2004; Mohns and Blumberg, 2010; Tiriac et al., 2012), we next assessed twitch-dependent activity of all recorded Purkinje cells. Complex spikes exhibited a significant peak in activity immediately after twitching and a second peak approximately 100 ms after the first, along with a third, weak peak approximately 100 ms after the second (Figure 4A). Simple spikes also exhibited a significant peak in activity immediately after twitching; this peak was broader than that of complex spikes and additional peaks were not evident (Figure 4B).

Figure 4.

Complex and simple spike activity increases immediately after twitches. (A) Perievent histogram plotting the number of complex spikes per 10-ms bin in relation to myoclonic twitching in sleeping 6-day-old rats. Data were pooled across all 30 units. The vertical black bar is the last bin before a twitch and the horizontal red bars indicate statistical significance (p < 0.05), as determined using a jitter protocol (see Materials and Methods). (B) Same as in (A) except for simple spikes (p < 0.05). (C) Perievent line histograms for complex spikes broken down for the W1 (red), W2 (green), and W3 (blue) units, as well as the units that showed no window preference (NP, black). For a small subset (10%) of all twitches analyzed, a unit that fired in one window fired again in one or both of the other windows. There was no clear relationship between a unit’s window assignment and its state-dependency. The data were smoothed using a 3-bin moving window. * significant peak within window (p < 0.0125). (D) Same as in (C) except for simple spikes. * significant peak within window (p < 0.0125).

Next, using a “separation procedure” used previously in adult cats and weanling rats (Bloedel and Ebner, 1984; Nicholson and Freeman, 2003), we tested the possibility that the multiple peaks in complex spike activity were a population-level feature of ensembles of Purkinje cells, arising when Purkinje cells fire during the first peak or the second peak or the third, but rarely more than one. To do this, we noted the occurrence of complex spikes within each of three 50-ms windows defined in relation to twitch-dependent activity: 0–50 ms (Figure 4A, red), 100–150 ms (green), and 200–250 ms (blue). Units that fired preferentially during Windows 1, 2, or 3 were designated as W1, W2, or W3 units, respectively. Cross-correlational analyses showed that W1, W2, and W3 units increased their complex spike activity exclusively within their associated window (Figure 4C). Units exhibiting no preference (NP) did not exhibit any twitch-related increase in complex spike activity. A similar result was obtained for simple spikes, but with only the W1 units exhibiting a significant increase in activity after a twitch (Figure 4D). Thus, twitching triggers short-latency complex and simple spike responses (W1), as well as a separate longer-latency complex spike response (W2) that may result from resetting of the intrinsic olivary rhythm (Bloedel and Ebner, 1984; Nicholson and Freeman, 2003).

Based on the complex spike data presented in Figure 4A, we repeated the twitch-triggered cross-correlational analyses separately for the AS-On units. Complex spikes again exhibited pronounced and significant W1 and W2 peaks after twitches (Figure 5A), and simple spikes again exhibited a significant W1 peak (Figure 5B). In contrast, neither AS/Wake-On nor State-Independent units exhibited pronounced peaks, thus indicating that AS-On units contributed predominantly to the overall twitch-related activity presented in Figure 4. For one representative AS-On unit recorded from crus II (Figure 5C), the tight coupling between Purkinje cell activity and active sleep is apparent. Importantly, Purkinje cell activity decreased markedly during periods of active wake when the pup was moving its limbs.

Figure 5.

Twitch-related complex and simple spike activity for AS-On units. Perievent line histograms plotting complex (A) and simple spike (B) counts per 10-ms bin in relation to myoclonic twitching in sleeping 6-day-old rats. Each plot is broken down for the W1 (red), W2 (green), and units that showed no window preference (NP, black). All data were smoothed using a 3-bin moving window. * significant peak within window (p < 0.017). (C) Representative data showing occurrences of behaviorally scored twitches from all visible limbs and tail (vertical green ticks), complex spikes (vertical purple ticks), nuchal EMG, and Purkinje cell activity (recorded from crus I) during periods of active sleep and active wake. Note the increased Purkinje cell and complex spike activity during periods of active sleep.

Complex Spike Autorhythmicity

Autorhythmicity in climbing fiber activity, at a frequency of about 10 Hz, has been reported in adult rats during wakefulness, but its occurrence is variable both within and across species (Kitazawa and Wolpert, 2005; Llinas and Sasaki, 1989). To assess whether such autorhythmicity occurred in our subjects, we performed auto-correlation analyses. Of the 30 units examined, six exhibited statistically significant autorhythmicity. Of these, three exhibited strong rhythms at frequencies of 8–12 Hz (Figure 6). Importantly, this analysis shows that the inferior olive is capable of rhythmic activity as early as 6 days of age. However, as the window analysis suggests, complex spike autorhythmicity is not responsible for the twitch-related multiple peaks in complex spike activity.

Figure 6.

Complex spike auto-rhythmicity in sleeping 6-day-old rats. For three individual units, autocorrelograms are shown at left and spectral density plots are shown at right. (A) A unit exhibiting strong 10-Hz rhythmicity. This unit was Sleep-On and did not exhibit a window preference. (B) A unit exhibiting strong 8-Hz rhythmicity. This unit was State-Independent and exhibited a preference for Window 2 (W2). (C) A unit exhibiting weaker rhythmicity at 11.5 Hz. This unit was As-On and exhibited a preference for Window 1 (W1 unit). These three units were recorded from two pups.

DISCUSSION

We found that the Purkinje cells of 6-day-old rats exhibit substantial state-dependent increases in complex and simple spike activity in close temporal association with twitching. These results are striking given the lack of several key cerebellar features considered integral for its mature function, including the granule cell-parallel fiber network and innervation of Purkinje cells by inhibitory interneurons and basket cells (Shimono et al., 1976). In addition, before the granule cell-parallel fiber network develops, mossy fibers form direct, transient connections onto Purkinje cell somas (Kalinovsky et al., 2011; Mason and Gregory, 1984; Takeda and Maekawa, 1989). Consequently, the observed twitch-dependent increases in simple spike activity at the age tested here indicates, for the first time in unanesthetized animals, functional engagement of these direct mossy fiber connections. All together, the current findings introduce a new and unanticipated mechanism for driving activity-dependent development of mossy and climbing fiber connectivity (Kalinovsky et al., 2011; Kano and Hashimoto, 2011; Takeda and Maekawa, 1989).

Twitching, Reafference, and Corollary Discharge

Although the function of the cerebellum continues to elude simple description, most current theories tend to focus on its role in motor timing or motor learning (Kitazawa and Wolpert, 2005). Of critical importance to these theories are the spatial and temporal details regarding the transmission of sensory and motor signals through mossy and climbing fibers. Toward that end, investigators seek to understand how the cerebellum processes the two established classes of sensory input: reafference, arising from self-produced movements, and exafference, arising from other-produced movements (Crapse and Sommer, 2008; Holst and Mittelstaedt, 1950; Poulet and Hedwig, 2007; Sperry, 1950). In addition to reafference, motor commands trigger corollary discharge (or efference copy), which serve the function of helping animals distinguish between expected (i.e., reafferent) and unexpected (i.e., exafferent) signals (Crapse and Sommer, 2008). Importantly, both reafference and corollary discharge signals are conveyed to the adult cerebellum via climbing and mossy fibers (Huang et al., 2013; van Kan et al., 1993; Wolpert et al., 1998). Further investigation of the infant cerebellum may provide insight into the development of both signals in this essential sensorimotor structure.

The green cells in Figure 7 denote the two conventional classes of sensory input: reafference from self-produced movements that is modified by corollary discharge, and exafference that is not. There is, however, a potential third class of input that is biologically plausible but rarely discussed: Self-produced movements that are not accompanied by corollary discharge (red cell in Figure 7). Here we propose the working hypothesis that twitches embody this novel class of self-produced movement.

Figure 7.

Hypothesized framework relating self-produced and other-produced movements with the accompaniment of corollary discharge. Conventionally, self-produced movements generate reafference that is accompanied by corollary discharge, whereas other-produced movements generate exafference that is not accompanied by corollary discharge (green cells). For obvious reasons, other-produced movements cannot be accompanied by corollary discharge (n/a: not applicable). Here we propose a third possible category whereby sleep-related twitches are self-produced movements that are not accompanied by corollary discharge (red cell).

We were led to this hypothesis for several reasons. Foremost among them is the observation that, similar to previous findings in hippocampus (Mohns and Blumberg, 2010; 2008) and sensory thalamus (Tiriac et al., 2012), Purkinje cell activity was prominent during active sleep in close temporal proximity to twitching, whereas wake-related activity was weak or absent. It should be stressed that we cannot be certain that the twitch-related Purkinje cell activity observed here reflects reafference alone. Nonetheless, the totality of evidence from cerebellum, hippocampus, and thalamus strongly suggests that—at least in rat pups under these testing conditions—reafferent signals associated with twitches are processed differently from those associated with wake movements. And one possible explanation for this differential processing is that corollary discharge is absent during twitching. This notion of state-dependent modulation of corollary discharge has not, to our knowledge, been considered previously.

There is intriguing but indirect support for the hypothesis that twitches are not accompanied by corollary discharge. First, human adults awaking from REM-related dreams display deficits in their ability to distinguish self- from other-produced stimulation, suggesting suspended corollary discharge (Blagrove et al., 2006). Second, REM-dream narratives exhibit qualities suggestive of suspended corollary discharge, similar to the hallucinations of schizophrenics (Feinberg and Guazzelli, 1999; Ford et al., 2008). Accordingly, a better understanding of differential processing of corollary discharge during sleep and wakefulness could greatly aid our understanding of how we learn to distinguish self from other—and how such processes go awry.

Limitations of this Study

The present results are clear in showing that Purkinje cell activity is strongly sleep- and twitch-dependent in rats at 6 days of age. We are confident that we recorded from Purkinje cells because of the location of the electrode sites within the Purkinje layer and the detection of doublets in the neurophysiological record. Doublets are considered an immature form of complex spike based on evidence that they are produced by climbing fibers arising from the inferior olive (Crepel, 1971; Puro and Woodward, 1977a). However, a fully accurate characterization of the activity triggered by climbing and mossy fibers early in development will require selective inactivation of the inferior olive and other structures that project to the cerebellum.

As mentioned in the previous section, another concern relates to the precise timing of the Purkinje cell activity in relation to twitching. For the analyses of twitch-related neural activity reported here and elsewhere, we rely on the fact that, because of high cross-correlations in the twitch activity of skeletal muscle groups, any given muscle can act as a reasonable proxy for another one (Mohns and Blumberg, 2010; Tiriac et al., 2012). This reliance on statistical association has not presented interpretative problems when recording from forebrain sites where the latencies between twitches and reafferent-related activity are approximately 200 ms. In the cerebellum, however, the first peak in twitch-related activity was 50 ms or less, which may or may not be too short for a reafferent signal at this age (Puro and Woodward, 1977a; Takeda and Maekawa, 1989). To resolve this issue, it will be important to record from locations in the cerebellum that can be definitively related to twitching in specific muscle groups. Once that is accomplished, it may also then be possible to assess how twitch-related activity guides the development of convergent and somatotopically organized mossy and climbing fiber inputs onto Purkinje cells (Huang et al., 2013; Odeh et al., 2005; Pijpers et al., 2006).

Finally, we recorded here from 6-day-olds because this is an age when Purkinje cells are functional but granule cells have not yet migrated to the internal granular layer. It is also an age when the Purkinje cells themselves are undergoing rapid developmental change (Watanabe and Kano, 2011). Of great interest will be documenting changes in state- and twitch-dependent activity in the cerebellum at younger and older ages and relating that activity to known features of cerebellar circuitry (Shimono et al., 1976). Such a developmental study is currently underway.

A New Era in Developmental Neurophysiology

As a contribution to this special issue on “Neural Mechanisms of Behavioral Maturation,” it seems worthwhile to place the present work in a broader historical context. Before 2004, there were only a handful of papers devoted to state-dependent neural activity in infant animals, including recordings at the cortical surface and multiunit recordings in the brainstem (Gramsbergen, 1976; Mirmiran and Corner, 1982; Tamásy et al., 1980; Tamásy and Korányi, 1980). For a variety of technical and methodological reasons, these studies were unable to document clear and convincing relationships between neural activity and established behavioral and EMG measures of sleep in early infancy. Moreover, with regard to the cerebral cortex, the conclusion drawn from these earlier studies was that this structure does not exhibit state-dependent activity before the emergence of slow waves at P11 (for review, see Blumberg and Seelke, 2010).

Then, beginning only a decade ago, new in vivo recording methods were introduced and refined for recording hippocampal (Karlsson and Blumberg, 2003; Leinekugel et al., 2002) and cortical (Khazipov et al., 2004) activity in freely moving and head-fixed infant rats. These methods have made it possible to investigate state-dependent neural activity in the infant brainstem (Karlsson et al., 2005; Karlsson and Blumberg, 2005) and forebrain (Mohns et al., 2006), discover previously undetected state-dependent cortical activity patterns and trace their early development (Seelke and Blumberg, 2010; Yang et al., 2009), assess the contributions of subplate neurons (Hanganu et al., 2008; Tolner et al., 2012) and the corpus callosum (Marcano-Reik and Blumberg, 2008) to infant cortical activity, and investigate the consequences of twitch-related reafference for neural activity throughout the neuraxis (Khazipov et al., 2004; Mohns and Blumberg, 2010; 2008; Tiriac et al., 2012). Imaging techniques, including the use of voltage-sensitive dyes, can now capture broader features of twitch-related cortical activity (McVea et al., 2012; Tiriac et al., 2012).

Although similar in vivo recording and imaging methods are providing new insights into sensory processing in the visual system (e.g., (Ackman et al., 2012; Hanganu et al., 2006; 2007), these methods can and should be expanded to the study of other behavioral contexts and the interactions among behavioral systems. For example, even in head-fixed pups, it should now be possible to investigate neural mechanisms associated with ingestion (Hall and Williams, 1983), thermoregulation (Blumberg, 2001), and learning and memory (Campolattaro and Freeman, 2008; Johanson and Hall, 1979), among other behaviors and processes. Such advances, when fully realized, hold the promise to reshape our understanding of brain-behavior relations and how they emerge across early development.

Conclusion

Spontaneous activity is a ubiquitous feature of developing nervous systems (O'Donovan, 1999). For example, spontaneous activity in the developing retina (Galli and Maffei, 1988; Wong, 1999) and cochlea (Tritsch et al., 2007) is thought to contribute to the development of the visual and auditory systems, respectively. Twitching is different from spontaneous retinal and cochlear activity in that it entails motor as well as sensory components. As such, it is ideally suited to provide information to the developing sensorimotor system about the functional properties of limbs and their neural control (Blumberg et al., 2013b). Previous findings have highlighted how reafference from twitching profoundly and selectively activates forebrain structures, including thalamus, cerebral cortex, and hippocampus (Khazipov et al., 2004; Mohns and Blumberg, 2010; Tiriac et al., 2012). Here we have shown for the first time that twitch-related processes modulate the activity of the cerebellum, a hindbrain structure that is critical for sensorimotor integration and motor learning. This finding introduces a potent source of structured activity that can plausibly account for activity-dependent development in the cerebellar system. In doing so, these findings lend additional support to the counterintuitive notion that, contrary to the still-popular view of twitching as a functionless by-product of dreams (Blumberg, 2010), twitching is a vital source of neural activity for the developing infant.