It ’ s not you , it ’ s me : corollary discharge in the precerebellar nuclei of sleeping infant rats

In week-old rats, somatosensory input arises predominantly from stimuli in the external environment or from sensory feedback associated with myoclonic twitches during active (REM) sleep. A previous study of neural activity in cerebellar cortex raised the possibility that the brainstem motor structures that produce twitches also send copies of motor commands (or corollary discharge, CD) to the cerebellum. Here, by recording from two precerebellar nuclei—the inferior olive and lateral reticular nucleus—we demonstrate that CD does indeed accompany the production of twitches. Within both structures, the CD signal comprises a surprisingly sharp activity peak within 10 ms of twitch onset. In the inferior olive, this sharp peak is attributable to the opening of slow potassium channels. We conclude that a diversity of neural activity is conveyed to the developing cerebellum preferentially during sleep-related twitching, enabling cerebellar processing of convergent input from CD and reafferent signals.

The sensorimotor systems of diverse vertebrate and invertebrate species distinguish 2 signals arising from self-generated movements (i.e., reafference) from those arising from 3 other-generated movements (i.e., exafference; Cullen, 2004). To make this distinction, 4 motor structures generate copies of motor commands, referred to as corollary discharge 5 (CD; Crapse and Sommer, 2008;Poulet and Hedwig, 2007). CD is conveyed to non-motor 6 structures to inform them of the imminent arrival of reafference arising from self-generated 7 movements (Crapse and Sommer, 2008). By comparing the two signals, animals are able 8 to distinguish between self-produced and other-produced movements. 9 Self-produced movements are not restricted to periods of wakefulness, especially 10 during development. Infants produce brief, discrete, jerky movements of skeletal muscles 11 during active sleep (AS or REM sleep), which is the predominant behavioral state during 12 early infancy (Jouvet-Mounier et al., 1970;Roffwarg et al., 1966). These spontaneous 13 movements are known as myoclonic twitches, and are among the most conspicuous 14 behaviors during development in a diversity of species (Blumberg et al., 2013; behavioral state and detect sleep and wake movements Figure 1 1A). 2 Electrode placement within the IO was confirmed histologically ( Figure 1B). Recording 3 sites were located within the dorsal accessory olive (DAO; n=19 units across 12 pups) or 4 the medial accessory olive (MAO) and the principal olive (PO; n=18 units across 8 pups; 5 Figure 1C). Overall, unit activity was phasic and largely restricted to periods of AS; also, 6 activity appeared to be suppressed immediately after the onset of active wake (AW). 7 Sparse activity was observed during behavioral quiescence (BQ), which is a period of low 8 muscle tone interposed between AW and AS ( Figure 1D). The majority (24/35 units, 69%) 9 of IO units were AS-on ( Figure 1E) and the mean firing rate of the AS-on units (2.9 ± 0.4 10 Hz) was approximately three times higher during AS than for the other two states 11 (p<0.0005; Figure 1F). Two IO units were excluded from state analysis due to movement 12 artifact during AW. 13 14 IO neurons exhibit sharp activity peaks at twitch onset 15 The phasic IO activity clustered around periods of myoclonic twitching; therefore, we 16 examined the temporal relationship between twitches and unit activity by creating 17 perievent histograms (5-ms bins, 1-s window) with unit activity triggered on twitch onset. 18 Previous studies have revealed two distinct patterns of twitch-triggered perievent 19 histograms in sensorimotor structures ( Figure 2A). First, in a motor structure like the RN 20 (green trace in Figure 2A), unit activity increases 20-40 ms before the onset of a twitch 21 . Second, in a sensory structure like the ECN (blue trace 22 in Figure 2A), unit activity increases at least 10-50 ms after the onset of a twitch (Tiriac 23 and . In the IO, however, we found that the majority of units (23/37 units, 1 62%) were active within +10 ms of the onset of a twitch ( Figure 2B, C). These IO activity 2 profiles were very sharp and thus strikingly different from those observed in any of the 3 motor and sensory structures from which we have previously recorded (e.g., see Figure   4 2D). Also, the IO units that exhibited this profile were responsive primarily to nuchal and/or 5 forelimb twitches and rarely to hindlimb twitches (Figure 2-figure supplement 1A-D). 6 Finally, the characteristics of the neural responses recorded in the IO did not appear to 7 differ across anatomical subdivisions. 8 There are three possible explanations for these sharply peaked activation patterns 9 observed in IO units: (a) the IO is part of the motor pathway, (b) the IO receives reafference 10 from twitches, and (c) the IO receives CD from a motor structure that produces twitches. 11 With respect to (a), the IO, despite being implicated in the precise timing of motor 12 behaviors (De Zeeuw et al., 1998), is not directly involved in the generation of movements 13 (Horn et al., 2004;Lang et al., 2017). Although it receives afferent projections from motor 14 areas (Saint-Cyr, 1983;Saint-Cyr and Courville, 1981), there are no efferent projections 15 from the IO to spinal motor neurons. In fact, the sole efferent projection from the IO 16 comprises the climbing fibers that innervate cerebellar Purkinje cells (Ruigrok et al., 2014). 17 Consequently, in adults, stimulation of the IO does not evoke or modulate movements 18 (Gellman et al., 1985). 19 With respect to (b), although the IO can receive short-latency reafferent signals 20 (Gellman et al., 1983;Sedgwick and Williams, 1967), it is unlikely that reafference can 21 account for the short-latency peaks observed here. Consider that for the structures in 22 which we have seen clear evidence of twitch-related reafference (e.g., ECN), we have also 23 seen clear evidence of exafferent responses (Tiriac et al., 2014;Tiriac and Blumberg, 1 2016). In contrast, of the IO units that exhibited sharp peaks with a latency of +10 ms, 2 none responded to exafferent stimulation. Moreover, of the 7 IO units that exhibited twitch-3 triggered responses at latencies consistent with reafferent processing (i.e., > 10 ms; Figure   4 2-figure supplement 1E), 3 responded to exafferent stimulation. Thus, the signature 5 feature of the majority of IO activity-the sharp peak centered on twitch onset-is 6 consistent with the notion that the IO receives CD associated with the production of a 7 twitch. 8 9 LRN neurons exhibit two kinds of twitch-related activity 10 Based on research in adults (Alstermark and Ekerot, 2015;Arshavsky et al., 1978), we 11 predicted that the LRN, like the IO, would exhibit CD-related activity. Moreover, because 12 the LRN also receives sensory inputs from the limbs ( Figure 3A), we expected to see 13 evidence of reafference in that structure. To test these two possibilities, we next recorded 14 spontaneous LRN activity in P8 rats across sleep and wake. 15 We confirmed electrode placements in the LRN (n=27 units across 9 pups, 1-6 16 units/pup; Figure 3B). Similar to the IO, the unit activity in the LRN was phasic and 17 restricted to periods of AS, particularly around twitches. LRN activity was sparse during 18 BQ and was suppressed after AW onset ( Figure 3C). The majority (16/27 units, 59%) of 19 LRN units were AS-on ( Figure 3D) and the mean firing rate of the AS-on units (1.4 ± 0.2 20 Hz) was approximately three times higher during AS than during any of the other two states 21 (p=<0.0005; Figure 3E). 22 Next, we assessed the temporal relationship between unit activity and twitches by 1 creating perievent histograms (5-ms bins, 1-s window). Regardless of state dependency, 2 the majority of LRN units (24/27 units, 89%) showed significant increases in firing rate in 3 response to a twitch ( Figure 3F). As predicted, we observed two different neural 4 populations the exhibited distinct patterns of twitch-triggered activity. First, we found a 5 subpopulation of LRN units (12/27 units, 44%) that, like the majority of IO units, exhibited 6 a sharp peak within ±10 ms of a twitch ( Figure 3G, left); also, none of these LRN units 7 responded to exafferent stimulation of the limbs (data not shown). 8 Second, the remaining LRN units (12/27 units, 44%) exhibited broader twitch-related 9 activity profiles consisting of a peak in activity around twitch onset (+10 ms) and/or a peak 10 with a latency of >10 ms ( Figure 3G, right). The latter peak is what is expected from a 11 short-latency reafferent responses (Tiriac and Blumberg, 2016;Tiriac et al., 2014). Indeed, 12 6 of these 12 units also responded to exafferent stimulation of either the forelimb or 13 hindlimb with an average latency of 40 ms ( Figure 3H).  Zeeuw et al., 1998;Saint-Cyr and Courville, 1981;Lakke, 1997;Onodera and Hicks, 2009;19 Zuk et al., 1983) and are therefore directly involved in the generation of movements 20 (Fukushima, 1991;Morris et al., 2015;Onodera and Hicks, 1996;Williams et al., 2014). 21 To determine whether MDJ neurons also project to the IO and LRN at P8, we performed 22 retrograde tracing from each structure. 23 Wheat germ agglutinin (WGA) conjugated with Alexa Fluor 488 or 555 was 1 microinjected into the IO or LRN and retrogradely labelled cell bodies were imaged using 2 a fluorescent microscope. Retrograde tracing from the IO (n=5; Figure 4A) revealed robust 3 labeling of cell bodies in the mesodiencephalic junction (MDJ), including the rostral 4 interstitial nucleus of Cajal (RI), nucleus of Darkschewitsch (Dk), accessory oculomotor 5 nuclei (MA3), and nearby diffuse areas in the MDJ. Little or no labelling was observed in 6 the RN. In contrast, retrograde tracing from the LRN (n=4; Figure 4B) revealed robust 7 labeling in the contralateral RN but not elsewhere in the MDJ. Finally, when the two 8 different tracers were injected separately into the IO and LRN of the same pup (n=2; Figure   9 4C), we found that the LRN-projecting cell bodies were located within the RN and the IO-10 projecting cell bodies were located outside the RN. These findings are consistent with 11 those in adult rats (Ruigrok et al., 2014) and suggest that the major afferent connections 12 to the IO and LRN arise from non-overlapping structures in the MDJ at this age. . 19 Subsequently, we performed immunohistochemistry to determine the expression of the c-20 Fos protein, a marker of neural activity (Chung, 2015), in the IO and LRN. connected to the IO and LRN at these ages. 5 It is possible that the c-Fos activation in the IO and LRN was due to sensory feedback 6 arising from the stimulated movements. However, we observed little or no c-Fos 7 expression in sensory areas like the cuneate nucleus and ECN (data not shown). if MDJ neurons outside of the RN convey twitch-related CD to the IO, we would also expect 14 these neurons to be involved in the production of twitches ( Figure 5A). Therefore, we 15 characterized the spontaneous activity of non-RN MDJ neurons in P8 rats during sleep 16 and wake. We aimed to record from MDJ neurons located in regions implicated earlier as 17 projecting to the IO and, upon stimulation, producing limb movements (see Electrode placements in the MDJ immediately medial to the RN were confirmed (n=7 20 pups, 17 units, 1-5 units/pup; Figure 5B). The spontaneous activity of neurons in this 21 region appeared mostly around twitches and wake movements ( Figure 5C). When twitch-22 triggered perievent histograms (10-ms bins, 1-s window) were created, we found that the 23 majority of the recorded units (15/17 units, 88%) showed significant twitch-dependent 1 activity ( Figure 5D). 2 The temporal relationship between neural activity and twitches revealed two primary 3 subpopulations of units ( Figure 5D): There were units that significantly increased their 4 firing rates before the onset of a twitch (twitch-preceding units) and units that significantly 5 increased their firing rates after the onset of a twitch (twitch-following units). The twitch-6 preceding units (10/17 units across 5 pups, 59%) showed an increase in firing rate 10-70 7 ms before a twitch ( Figure 5E, left). Interestingly, the majority (7/10 units, 70%) of these 8 twitch-preceding units also exhibited an increase in firing rate 10-50 ms after a twitch, 9 indicative of reafference. The twitch-following units (5/17 units across 3 pups, 29%) 10 showed an increase in firing rate 20-40 ms after a twitch ( Figure 5E, right), suggesting they 11 only receive reafference from a twitch. In terms of pattern and latency of twitch-triggered 12 activity, these neurons behave similarly to those described previously in the RN ( shown). Therefore, the IO and LRN receive CD from the MDJ associated with both types 20 of movements, although wake-related activity is less abundant and pronounced than 21 twitch-related activity. Calcium-activated slow-potassium (SK) channels contribute to the 1 sharp peak in IO activity 2 Having identified motor structures that send CD to the IO and LRN, we next sought to 3 determine how a motor command with a broad twitch-preceding peak (see Figure 5F) is 4 transformed into a sharp, precise peak around a twitch (see Figure 2B). We focused on 5 the IO to address this question because of the reliably high percentage of units that exhibit 6 twitch-related CD. 7 In the adult IO, SK channels prevent temporal summation of excitatory presynaptic 8 inputs (Garden et al., 2017). SK channels are also expressed early in development 9 (Gymnopoulos et al., 2014). Because afferent projections from the MDJ to the IO are We confirmed drug or vehicle diffusion and recording sites within the IO (n=18 units 1 across 10 pups in the saline group; n= 21 units across 8 pups in the apamin group; 1-3 2 units/pup; Figure 6B). There was no difference in the amount of time spent in AS (p=0.78) 3 or in the number of twitches produced per unit time of AS between the apamin and saline 4 groups (ps>0.15 for nuchal, contralateral and ipsilateral forelimb twitches; Figure  There was no significant difference in the overall firing rate during AS between groups 8 (p=0.59; Figure 6-figure supplement 1D). 9 Perievent histograms (10-ms bins, 1-s window) were created for each individual unit in 10 both groups ( Figure 6C). As predicted, whereas twitch-triggered activity in the saline group 11 exhibited the expected sharp peak around twitches, the activity in the apamin group was 12 broader during the period after twitch onset. 13 The number of units exhibiting significant twitch-related activity did not differ between 14 the two groups (n=13/18 in saline and n=11/21 in apamin groups; X 2 (1, N=39) = 1.6, p=0.2; 15 Figure 6D). In contrast, the number of units exhibiting sharp peaks within ±10 ms of a 16 twitch was significantly lower in the apamin group (5/21 units, 24%) than in the saline group 17 (11/18 units, 61%; X 2 (1, N=39) = 5.6, p=0.02; Figure 6D). To illustrate the effect of apamin 18 on twitch-related activity, we pooled the data for the significant units to create perievent 19 histograms of IO activity. As shown in Figure 6E, the activity in the apamin group, unlike 20 that in saline group, persisted beyond the 10-ms window after a twitch. To quantify the 21 difference, we calculated the area under the curve for each unit during two time windows: 22 ±10 ms around a twitch and 20-200 ms after a twitch ( Figure 6F). As expected, we found 23 no significant difference between the two groups in the ±10-ms window (U=56.5, Z=-0.87, 1 p=0.4), but did find a significant difference for the 20-200-ms window, with the apamin 2 group being significantly larger (U=30.5, Z=-2.2, p=0.03). In fact, the pattern of twitch-3 triggered neural activity in the apamin group was similar to that recorded in the MDJ 4 ( Figure 6G). Based on these results, we conclude that SK channels are involved in 5 sharpening the CD signal arriving from the MDJ. Several criteria have been proposed for identifying CD signals (Poulet and Hedwig, 2007;9 Sommer and Wurtz, 2008). First, a CD should originate in a structure that is demonstrably 10 involved in the production of movement; as shown here, the twitch-related activity in the 11 IO and LRN originates from several independent motor structures in the MDJ that are 12 involved in the production of twitches and wake movements (Del Rio-Bermudez et al., 13 2015). Second, areas receiving CD should themselves play no direct role in the production 14 of movement; this is clearly true of the IO and LRN (Gellman et al., 1985;Ruigrok et al., 15 2014). Third, neurons receiving CD should increase their activity at the onset of a 16 movement; as shown here, the activity of IO and LRN neurons occurs precisely at the 17 onset of twitches, exhibiting a temporal profile that clearly distinguishes it from twitch-18 preceding activity in the MDJ nuclei and twitch-following activity in the ECN. Thus, the 19 twitch-related activity in the IO and LRN satisfies the key criteria of CD. Below we discuss 20 the implications of this finding and its significance for sensorimotor development.

Neurophysiological identification of CD signals in behaving animals
Neural pathways conveying CD have been delineated in a diverse array of species (Dale 1 and Cullen, 2017;Davis et al., 1973;Fee et al., 1997;Schneider et al., 2014;Sommer and 2 Wurtz, 2002;Yang et al., 2008). Neural recordings of the CD signal itself, however, have 3 mostly been performed in non-mammalian species, including crickets, sea slugs, crayfish, 4 tadpoles, and electric fish (Evans et al., 2003;Kirk and Wine, 1984;Li et al., 2004;Poulet 5 and Hedwig, 2006;Requarth and Sawtell, 2014). The relatively small and simple nervous 6 systems of these species have allowed for the isolation of neurons that carry or receive 7 CD signals and identify their relationship to behavior. In contrast, CD signals have thus far 8 only been recorded in the mediodorsal thalamus of non-human primates during eye 9 movements (Sommer and Wurtz, 2004) and the auditory cortex of mice (Schneider et al.,10 2014). 11 The current findings provide the first direct neurophysiological evidence of CD in a 12 developing mammal. Moreover, this is the first direct evidence of CD in the IO and LRN, 13 consistent with what has been proposed for these two structures (Alstermark and Ekerot, 14 2013;Arshavsky et al., 1978;De Zeeuw et al., 1998;Devor, 2002). Also, with this 15 discovery of a unique neural CD signature-comprising a short-latency onset and sharp 16 activity peak-we have a clear template to guide future neurophysiological investigations 17 of CD signals in other species and neural systems across the lifespan.

19
A neural mechanism for sharpening the CD signal 20 As mentioned above, one of the signature features of the twitch-related CD signal is the 21 sharp peak. This is surprising because, as shown here and in a previous study (Del  Bermudez et al., 2015), twitch-related motor activity in MDJ neurons exhibits broad peaks 23 (see Figure 5F). How does a broad presynaptic signal in the MDJ get converted into a found that pharmacological inactivation of the DCN had no effect on IO activity at P8, 12 consistent with a previously published report (Nicholson and Freeman, 2003). 13 Consequently, we hypothesized that inhibition in the IO is mediated by SK channels. 14 In the IO of adult rats, these channels prevent summation of excitatory inputs (Garden et   15 al., 2017). Here, using pharmacological inactivation, we demonstrate that SK channels are 16 responsible for sharpening the olivary CD signal. A similar mechanism could be functional 17 in the LRN as SK channels are also expressed in that structure in adult rats (Xu et al.,18 2013).
19 20 Differential actions of CD signals at precerebellar nuclei 21 In a previous study (Tiriac and Blumberg, 2016), it was demonstrated that wake-related 22 reafference is blocked within the ECN, an effect that we attributed to modulation by a 23 wake-related CD signal. That CD-mediated blockade was lifted during twitching, thereby 1 allowing twitch-related reafference to be conveyed to downstream motor structures, 2 including the cerebellum. In contrast, focusing here on the IO and LRN, we found that the 3 twitch-related CD signals themselves-not reafference-are conveyed to the cerebellum 4 ( Figure 7A). Therefore, within this broader context, we see that CD accompanies sleep 5 and wake behavior in infants, but its effects are not monolithic: It can modulate the action 6 of a comparator to gate reafference (as in the ECN) or be transmitted sequentially to 7 multiple downstream structures (as in the IO or LRN à cerebellum). Such diverse effects 8 of CD have been described elsewhere (Crapse and Sommer, 2008). 9 Although the three precerebellar nuclei-IO, LRN, and ECN-process CD and 10 reafference differently, the common denominator of all this activity is the inundation of the 11 developing cerebellum with twitch-related information ( Figure 7A). In contrast, wake-12 related activity has been shown to be blocked at the level of the ECN (Tiriac and Blumberg,13 2016), thereby accounting in part for the relatively low levels of activity in the cerebellum 14 at this age (Sokoloff et al., 2015a;Sokoloff et al., 2015b). In addition, we found here that 15 the IO and LRN are also considerably less active during periods of wake. Although wake opportunities for wake movements to activate downstream cerebellar structures ( The developing sensorimotor system receives substantial sensory input from self-5 generated twitches and from external stimulation arising from the mother and littermates. 6 It has been suggested that the infant brain does not distinguish between these two sources 7 of input and that twitch-related reafference serves merely as a "proxy" for exafferent 8 stimulation (Akhmetshina et al., 2016;McVea et al., 2016). This suggestion rests in part 9 on the observation that both forms of stimulation, despite their very different origins, trigger 10 similar patterns of cortical activity (Akhmetshina et al., 2016;Tiriac et al., 2012;Yang et 11 al., 2013). Thus, with our finding that CD accompanies the production of twitches, it is now 12 clear that there exists a mechanism with which the infant brain can distinguish self-13 generated from other-generated movements; the ability to make this distinction is thought 14 to rely in part on the cerebellum (Blakemore et al., 2000;Wolpert et al., 1998). 15 There are a number of ways in which twitch-related CD could contribute to cerebellar 16 development and function. For example, in the adult cerebellum, CD and reafference are 17 known to converge via climbing and mossy fibers (Blakemore et al., 2001;Huang et al., 18 2013;van Kan et al., 1993;Wolpert et al., 1998). In this way, it is thought that the 19 cerebellum instantiates a forward model that receives sensory predictions and computes 20 prediction errors (by comparing CD with reafference) in order to facilitate motor learning 21 (Blakemore et al., 2000;Brooks et al., 2015;Requarth and Sawtell, 2014;Wolpert et al., 22 1998). 23 Twitches could contribute to the process by which forward models are instantiated and 1 updated, especially in the context of a rapidly growing body. To appreciate this possibility, 2 consider this description of cerebellar function: "After much trial and error during infancy 3 and throughout life, the cerebellum learns to associate actual movements with intended 4 movements. Many of our motor memories are movements that we have repeated millions 5 or billions of times..." (p. 538, Mason, 2011). In that context, the millions of twitches 6 produced in early infancy could be a critical source of repeated convergent input to the 7 developing cerebellum. This convergence, illustrated in Figure 7, would provide the 8 developing cerebellum with abundant opportunities to align prediction and feedback 9 signals in a topographically organized fashion. 10 Cerebellar circuitry undergoes substantial development over the first three postnatal 11 weeks in rats (Altman, 1972a;Shimono et al., 1976;Wang and Zoghbi, 2001). Many week as one climbing fiber is selectively strengthened over others. Importantly, spike 18 timing-dependent plasticity (STDP) has been implicated in this process (Kawamura et al.,19 2013); STDP depends on the repetitive and sequential firing of pre-and post-synaptic cells 20 within a short and precise time window (Feldman, 2012;Kawamura et al., 2013;Sgritta et 21 al., 2017). The present findings in precerebellar nuclei, in which twitch-related CD reliably 22 preceded reafference by approximately 10-30 ms, are consistent with twitches playing a 23 role in cerebellar development via STDP. Moreover, recording from Purkinje cells at P8, 1 we previously found that complex and simple spikes were highly likely to occur within 0-2 50 ms after twitches (Sokoloff et al., 2015a). Accumulating evidence also suggests that CD-related processing is dysfunctional in 17 patients with schizophrenia. Specifically, failure to disambiguate "self-generated" from 18 "other-generated" sensory input may underlie hallucinations and delusions of control 19 (Feinberg and Guazzelli, 1999;Ford et al., 2008). If twitches help to instruct the developing 20 brain to distinguish self from other, disruption of sleep and sleep-related sensorimotor 21 processing may have later-emerging negative consequences for the processing of CD.

22
Conclusion 1 It has been argued that the discreteness of twitches makes them ideally suited to provide 2 high-fidelity sensory information at ages when activity-dependent development is so 3 important for the developing nervous system (Blumberg et al., 2013;Tiriac et al., 2015). 4 The present results go further to suggest that the convergence of twitch-related CD and 5 reafference associated with millions of twitches over the early developmental period 6 provides ample opportunity for assimilating growing limbs into the infant's emerging body 7 schema (Blumberg and Dooley, 2017). x 20 x 26 cm). Food and water were available ad libitum. The animals were maintained on 20 a 12-h light-dark cycle with lights on at 0700 h. Littermates were never assigned to the 21 same experimental group. 1 Surgery 2 A pup with a visible milk band was removed from the home cage. Under isoflurane (3-5%) 3 anesthesia, bipolar hook electrodes (50 µm diameter, California Fine Wire, Grover Beach, 4 CA) were inserted into the nuchal, forelimb, and hindlimb muscles for electromyography 5 (EMG) and secured with collodion. A stainless steel ground wire was secured transdermally 6 on the back. A custom-built head-fix device was then secured to the exposed skull with 7 cyanoacrylate adhesive . The local anesthetic, Bupivicaine (0.25%) 8 was applied topically to the site of incision and some subjects were also injected 9 subcutaneously with the analgesic agent carprofen (0.005 mg/g). The pup was lightly 10 wrapped in gauze and allowed to recover in a humidified, temperature-controlled (35 ºC) 11 incubator for at least one hour. After recovery, the pup was briefly (<15 min) re-anesthetized 12 with isoflurane (2-3%) and secured in a stereotaxic apparatus. A hole was drilled in the The head-fix device was secured to the stereotaxic apparatus housed within the recording 3 chamber and the pup was positioned with its body prone on a narrow platform with limbs 4 dangling freely on both sides . Care was taken to regulate air 5 temperature and humidity such that the pup's brain temperature was maintained at 36-37 6 ºC. Adequate time (1-2 h) was allowed for the pup to acclimate to the recording 7 environment and testing began only when it started cycling normally between sleep and 8 wake. Pups rarely exhibited abnormal behavior or any signs of discomfort or distress; 9 when they did, the experiment was terminated. The bipolar EMG electrodes were 10 connected to a differential amplifier (A-M Systems, Carlsborg, WA; amplification: 10,000x; 11 filter setting: 300-5000 Hz). A ground wire (Ag/AgCl, 0.25 mm diameter, Medwire, Mt.

17
In 18 P8 rats, pups were prepared for electrophysiological recording as described above 18 and transferred to the recording rig. Once a pup started cycling between sleep and wake, 19 a 0.5 µl microsyringe was lowered stereotaxically into the IO and 100 nl of apamin. 20 (Abcam, Cambridge, MA; 1 µM, dissolved in 0.9% saline, n=8) or saline (n=10) was 21 injected over 1 min. During preparation of the drug or vehicle, fluorogold (4%, 22 Fluorochrome, Denver, CO) was added to the solutions for subsequent assessment of the extent of drug diffusion. After a 15-min period to allow for diffusion, the microsyringe was 1 withdrawn and a recording electrode was lowered in its place into the IO and activity was 2 recorded for 30 min. At the end of the experiment, the pup was sacrificed and its brain was 3 prepared for histology as described above. 4 5 Data analysis 6 Spike sorting. As described previously (Mukherjee et al., 2017;Sokoloff et al., 2015a), 7 action potentials (signal-to-noise ≥ 2:1) were sorted from MUA records using template 8 matching and principal component analysis in Spike2 (Cambridge Electronic Design). 9 Waveforms exceeding 3.5 SD from the mean of a given template were excluded from 10 analysis.

11
Identification of behavioral states. EMG activity and behavioral scoring were used to 12 identify behavioral state . To establish an EMG threshold for 13 distinguishing sleep from wake, EMG signals were rectified and smoothed (tau = 0.001 s). 14 The mean amplitude of high muscle tone and atonia were calculated from five 15 representative 1-s segments and the midpoint between the two was used to establish the 16 threshold for defining periods of wake (defined as muscle tone being above the threshold 17 for at least 1 s) and sleep (defined as muscle tone being below the threshold for at least 1 18 s). Active wake (AW) was identified by high-amplitude limb movements (e.g., stepping, 19 stretching) against a background of high muscle tone and was confirmed using behavioral 20 scoring. The onset of a wake movement was defined on the basis of EMG amplitude 21 surpassing the established threshold. Active sleep (AS) was characterized by the 22 presence of myoclonic twitches of the limbs against a background of muscle atonia. 23 Twitches were identified as sharp EMG events that exceeded by ≥ 3x the mean EMG 1 baseline during atonia; twitches were also confirmed by behavioral scoring (Seelke and 2 Blumberg, 2010). Additionally, behavioral quiescence (BQ) was characterized as periods 3 of low muscle tone interposed between AW and AS. 4 State-dependent neural activity. For each unit, average firing rate across all behavioral 5 states was determined. Bouts of AS, AW, and BQ were excluded when firing rates 6 exceeded 3 SD of the firing rate for that behavioral state; this happened rarely (0-2 per 7 unit). Next, pairwise comparison of firing rates across states was performed using the  Twitch-triggered neural activity. To determine the relationship between unit activity and 13 twitching, we triggered unit activity on twitch onsets and generated perievent histograms 14 over a 1-s window using 5-or 10-ms bins. We performed these analyses on each individual 15 unit using twitches from nuchal, forelimb, and hindlimb muscles. We tested statistical 16 significance by jittering twitch events 1000 times over a 500-ms window using PatternJitter 17 (Amarasingham et al., 2012;Harrison and Geman, 2009). Then using a custom-written 18 Matlab program (MathWorks, Natick, MA), we generated upper and lower confidence 19 bands (p<0.05 or 0.01 for each confidence band) using a method that corrects for multiple 20 comparisons (Amarasingham et al., 2012). For each unit, after histograms were separately 21 constructed for nuchal, forelimb, or hindlimb twitches, we identified and activity that was 22 significant in response to a twitch. When more than one muscle yielded a significant 23 change in neural activity, we further analyzed the data only for the muscle that showed the 1 strongest relationship (determined by the highest firing rate) between twitches and unit 2 activity. We then pooled these data to create perievent histograms composed of significant 3 units and performed jitter analyses on the pooled data. 4 Wake-triggered neural activity. To determine the relationship between neural activity 5 and wake movements, we triggered unit activity on wake-movement onset and created 6 perievent histograms on the pooled data (20-ms bins, 1-s window). We then performed 7 jitter analysis as described above. 8 Evoked response to exafferent stimulation. We identified MUA in which evoked 9 responses were observed and then sorted the units. Those units were then pooled and 10 triggered on stimulus onset (determined using EMG artifact) to create perievent 11 histograms. The jitter analysis was performed on the pooled data, as described above.

12
Intra-olivary injection of apamin. First, we identified if apamin affected sleep-wake 13 behavior. We assessed the amount of time spent in AS and the number of twitches per 14 min of AS in each pup. Differences across groups were tested using the Mann-Whitney U 15 test. Next, we determined if apamin altered the overall firing rate. We calculated the firing 16 rate of each unit during AS and compared that across groups using the Mann-Whitney U 17 test. One value exceeding 3 SD was excluded as an outlier. 18 We then assessed whether apamin altered the shape of twitch-triggered perievent 19 histogram. First, we created perievent histograms (10-ms bins, 1-s window) for each unit 20 as described above. For each unit, firing rate was normalized to the peak firing rate and 21 the average normalized firing rate across all units in each group was calculated. Perievent 22 histograms were then created with the average (+SEM) normalized firing rates triggered 23 on twitches for each group. Next, we assessed how apamin altered the pattern of twitch-1 triggered activity of individual units. To do that, we identified significant units by performing 2 jitter analysis on individual units as described above. We counted the percentage of units 3 that showed precise peak within ±10 ms around a twitch and compared that across groups 4 using a Chi-squared test. Finally, we pooled significant units in each group and pooled 5 them to create perievent histograms consisting of significant units only. To assess the 6 difference in the shape of perievent histograms, we calculated the area under the curve 7 by adding the histogram counts within a particular time window and compared that across 8 groups using the Mann-Whitney U test. One value exceeding 3 SD was excluded as an 9 outlier.