Journal Pre-proofs Diazepam effects on local cortical neural activity during sleep in mice

GABA-ergic neurotransmission plays a key role in sleep regulatory mechanisms and in brain oscillations during sleep. Benzodiazepines such as diazepam are known to induce sedation and promote sleep, however, EEG spectral power in slow frequencies is typically reduced after benzodiazepines or similar compounds. EEG slow waves arise from a synchronous alternation between periods of cortical network activity (ON) and silence (OFF), and represent a sensitive marker of preceding sleep-wake history. Yet it remains unclear how benzodiazepines act on cortical neural activity during sleep. To address this, we obtained chronic recordings of local field potentials and multiunit activity (MUA) from deep cortical layers of the primary motor cortex in freely behaving mice after diazepam injection. We found that the amplitude of individual LFP slow waves was significantly reduced after diazepam injection and was accompanied by a lower incidence and duration of the corresponding neuronal OFF periods. Further investigation suggested that this is due to a disruption in the synchronisation of cortical neurons. Our data suggest that the state of global sleep and local cortical synchrony can be dissociated, and that the brain state induced by benzodiazepines is qualitatively different from spontaneous physiological sleep.


Introduction
Benzodiazepines are frequently prescribed as sedatives, anxiolytics and anticonvulsants, with 5.2% US adults aged 18-80 reported to be using benzodiazepines in 2008 (Olfson et al., 2015). Benzodiazepines are commonly used for the management of epileptic seizures including status epilepticus (Ochoa and Kilgo, 2016;Riss et al., 2008), and anxiety disorders such as panic disorder and generalised anxiety disorder (Baldwin et al., 2005;Martin et al., 2007;Offidani et al., 2013). Benzodiazepines also have notable sedative and sleep-inducing effects, which have led to them being used as hypnotics for the treatment of insomnia (Buscemi et al., 2007;Kripke, 2000;Schutte-Rodin et al., 2008). However it was noted that benzodiazepines can impair the daytime functioning as reported in individuals experiencing sleep disturbances, leading to further impairments in performance, cognition and memory, as well as increasing the risk of accidents, particularly with long-term use (Barker et al., 2004;Johnson and Chernik, 1982;Kripke, 2000;Roehrs et al., 1994;Stewart, 2005). Benzodiazepines are also highly addictive, and patients taking them on a regular basis develop tolerance and prominent withdrawal symptoms, which can ultimately lead to their misuse (Nutt et al., 2007), while chronic use is often ineffective or can lead to impairments in cognition (Barker et al., 2004;Kripke, 2000;Riemann et al., 2020). For these reasons, benzodiazepines and other hypnotic drugs are not generally recommended for the treatment of insomnia and should be limited to short-term use (Riemann et al., 2017;Riemann and Perlis, 2009;Sateia et al., 2017).
Despite these risks, benzodiazepines remain a prevalent treatment for sleep disruptions, with pharmacological interventions resulting in similar short-term treatment outcomes for persistent insomnia as compared to behavioural interventions 4 (Smith et al., 2002), and even low doses reported to produce significant hypnotic effects (Shannon and Herling, 1983).
Many of the pharmacological interventions for the treatment of sleep disorders including insomnia, act by enhancing gamma-amino butyric acid (GABA) transmission. Multiple populations of GABAergic neurons have been identified as having important roles in the induction and maintenance of sleep throughout the brain, and have been extensively reviewed elsewhere (Luppi et al., 2017). Briefly these include non-rapid eye movement (NREM) sleep inducing populations in the ventrolateral preoptic area, the parafacial zone in the brainstem, the nucleus accumbens and the cortex, and populations associated with rapid eye movement (REM) sleep in ventral medullary reticular formation, lateral hypothalamic area and ventrolateral periaqueductal gray. Of particular interest to this study, benzodiazepines potentiate inhibitory GABAergic neurotransmission by increasing the affinity for the inhibitory neurotransmitter gamma-amino butyric acid (GABA) to bind to GABA type A receptors (GABAA), including containing alpha1, alpha2 and gamma2 subunits which are distributed throughout the brain (Kopp et al., 2004(Kopp et al., , 2003Tan et al., 2011;Tobler et al., 2001;Visser et al., 2003). It is thought that benzodiazepines act mainly by potentiating GABAergic transmission in subcortical structures controlling the thalamocortical networks responsible for the generation of delta oscillations during slow wave sleep, rather than having direct effects (Luppi et al., 2017).
Benzodiazepines, such as diazepam, reduce overall network excitability and lead to characteristic alterations in rhythmic activity. In fact there is recent evidence that diazepam administration in mice induces a slowing of global electroencephalogram (EEG) oscillations in a frequency-and behavioural-state dependent manner (Scheffzük et al., 2013). Studies in both humans and animals have 5 shown that benzodiazepines administration shortens the latency to sleep, increases sleep continuity, inhibits REM sleep and at the level of EEG spectra leads to increases in fast frequencies such as sleep spindles and stage 2 NREM sleep at the expense of stage 3 slow wave sleep (Aeschbach et al., 1994;Bastien et al., 2003;Borbély et al., 1991Borbély et al., , 1983Holbrook et al., 2000;Kopp et al., 2004Kopp et al., , 2003Lancel et al., 1996;Lancel and Steiger, 1999;Luppi et al., 2017;Scheffzük et al., 2013;Tobler et al., 2001).
Evidence in humans has shown that the reduction in sleep slow wave activity (EEG power density between 0.5-4 Hz, SWA) persists into the following drug-free night after injection with various benzodiazepines, whereas the effects on the spindle frequency range (10-15 Hz) were recovered by the following day (Borbély et al., 1983).
Interestingly benzodiazepines do not seem to affect the homeostatic regulation of sleep, as evidence suggests they do not greatly affect the time course of SWA, despite reducing absolute EEG power (Aeschbach et al., 1994). Theta frequency activity (6-9 Hz) during waking and REM sleep has also been shown to be enhanced in mice injected with diazepam (Kopp et al., 2003;Tobler et al., 2001). Diazepam also shifted the theta peak to slower frequencies, especially in the occipital derivation where theta activity is known to predominate due to its proximity to the hippocampus. It is possible that the shift in theta peak may reflect changes in body temperature, as has previously been shown (Deboer, 2002(Deboer, , 1998. This is supported by evidence that diazepam decreases body temperature (Mailliet et al., 2001).
Thus, current evidence suggests seemingly paradoxical effects of diazepam on sleep, in that it is sleep promoting yet reduces sleep-related global oscillatory EEG activities. An explanation for this could lie in potential local effects of diazepam on neural activity which are not well understood, and are likely to be complex.
Electrophysiological characteristics, connectivity patterns, ongoing behaviour and 6 preceding sleep-wake history may all together determine the firing phenotype of specific cortical neurons (Ascoli et al., 2008;Fisher et al., 2016;Kropff et al., 2015;McGinley et al., 2015;O'Keefe and Dostrovsky, 1971;Poulet and Petersen, 2008). It is crucial to further understand the physiological mechanisms underlying the efficacy of these drugs, which are not currently well understood. In this study we investigated the effects of the commonly used hypnotic, diazepam on characteristics of the electroencephalogram, local field potentials and cortical neural activity. This study aimed to address whether reduced 'global' EEG slow-wave activity after diazepam was associated with changes at the local network level.

Experimental animals
This study was carried out in n=8 male C57Bl/6J mice aged 5.23-12.83 months old.

Surgical implantation of recording electrodes
Surgical implantation of recording electrodes were carried out as previously described (Cui et al., 2014;Fisher et al., 2016;McKillop et al., 2018), under aseptic conditions. Briefly, animals were deeply anaesthetised using isoflurane anesthesia (3%-5% induction, 1%-2% maintenance). One day before surgery, animals received dexamethasone (0.2 mg/kg, p.o.). Metacam (1-2 mg/kg, s.c., Boehringer Ingelheim, UK), dexamethasone (0.2 mg/kg, s.c. Aspen Pharmacare, UK), and vetergesic (0.08 mg/kg, s.c, Ceva Animal Health ltd., UK) were administered preoperatively. Before implantation, EEG screw electrodes were soldered to custom-made head-mount connectors (Pinnacle Technology) and unilaterally implanted into frontal (motor area: anteroposterior 2 mm, mediolateral 2 mm) and occipital (visual area, V1: anteroposterior 3.5-4 mm, mediolateral 2.5 mm) cortical areas. Reference and ground screw electrodes were implanted above the cerebellum and contralaterally to the occipital screw, respectively. Two single-stranded, stainless-steel wires were inserted on either side of the nuchal muscle to record electromyography. All screws and wires were secured to the skull using dental acrylic. Mice were also implanted with a polymide-insulated tungsten microwire array (Tucker-Davis Technologies), consisting of 16 channels (2 rows each of 8 wires), with a wire diameter of 33 μm, electrode spacing 250 μm, row separation L-R 375 μm, a tip angle of 45 degrees and one row of electrodes 250 μm longer than the other to account for the curvature of the brain.
A two-component silicon gel (KwikSil; World Precision Instruments) was used to seal the craniotomy and protect the surface of the brain from the dental acrylic used to fix the array to the skull. Animals were all monitored closely after surgery and provided with analgesia as necessary (0.2 mg/kg dexamethasone (Aspen Pharmacare, UK) for 2 days and 1-2 mg/kg metacam (Boehringer Ingelheim, UK) for a minimum of 3 days).

Signal processing
A Tucker-Davis Technologies Multichannel Neurophysiology Recording System was used for data acquisition. Cortical EEG was recorded from frontal and occipital derivations. EEG, EMG, and LFP data were filtered between 0.1 and 100 Hz, amplified (PZ5 NeuroDigitizer preamplifier, Tucker-Davis Technologies), and stored on a local computer at a sampling rate of 256.9 Hz. Extracellular neuronal spike data were recorded from the microwire array at a sampling rate of 25 kHz (filtered between 300 Hz and 5 kHz). OpenEx software (Tucker-Davis Technologies) was used to manually apply amplitude thresholds for online spike detection and to eliminate artifactual waveforms caused by electrical or mechanical noise. Spikes that exceeded this predefined threshold (>2× noise level, at least −25 μV) were stored as 46 samples Recordings were subdivided into 4 second epochs and vigilance states scored offline by manual inspection of the signal (SleepSign, Kissei Comtec). Vigilance states were classified as waking (low-voltage, high-frequency EEG with a high level or phasic EMG activity), NREM sleep (presence of EEG slow waves, a signal of a high amplitude and low frequency), or REM sleep (low-voltage, high-frequency EEG with a low level of EMG activity), as shown in the representative traces shown in Figure 1A. Vigilance state artifacts in at least one EEG or MUA recording channel, were scored as artifacts so that they may be removed from appropriate analyses. After the data were scored, EEG and LFP power spectra were computed by an FFT routine for 4 s epochs (Hanning window), with a 0.25 Hz resolution (SleepSign, Kissei Comtec).
In order to detect slow waves in the LFP, firstly the signal was band pass filtered between 0.5-4 Hz (stopband edge frequencies 0.3-8 Hz) with MATLAB filtfilt function exploiting a Chebyshev Type II filter design (The MathWorks Inc, Natick, Massachusetts, USA) (Achermann and Borbély, 1997;Fisher et al., 2016;McKillop et al., 2018;Vyazovskiy et al., 2009). Slow waves were then defined as positive deflections of the filtered LFP signal between two consecutive negative deflections below the zero-crossing, in which the peak amplitude of the wave was larger than the median amplitude detected across all waves. OFF periods were defined as complete cessation of spiking activity across all channels for at least 50 ms. For some of the analyses only the largest slow waves or longest OFF periods were included (>median + one standard deviation across vehicle and diazepam conditions).

Experimental Design
This study was a cross-over design with all animals receiving both an injection of diazepam (3mg/kg, Hameln Pharmaceuticals ltd, UK) and vehicle (saline with 0.3% Tween 80, Sigma Aldrich (now Merck, Darmstadt, Germany)) in a randomised order, with 96 hours between each injection. Diazepam was dissolved in the vehicle and injected at a concentration of 10mg/ml. Injections were performed at light onset. The 24-hours prior to each injection was used as a baseline, while the 24 hours after injection was used to assess the effect of injections. Animals were undisturbed throughout the duration of the study and electrophysiological signals were recorded continuously. Example wake to NREM sleep and NREM sleep to REM sleep transitions are shown in Figure 2A Figure 5D, data were first log-transformed prior to statistical analysis. In some cases, animals were excluded from specific analyses for technical reasons, as stated in the figure legends. At the end of the study, electrode recording sites were confirmed using previously described histological methodology (Fisher et al., 2016).

Results
Firstly, we investigated the 24 hours baseline period prior to injection. All animals had at least one good EEG channel and EMG that allowed for the identification and scoring of the three vigilance states, as shown in Figure 1A. In this study, we also looked at the neural multiunit activity associated with local field potentials, which showed classical distinctions between vigilance states. For example, the periodic occurrence of OFF periods of neuronal silence was characteristic during NREM sleep, while neurons showed tonic firing during both wake and REM sleep ( Figure 1D). As p=0.18, paired t-test), suggesting that there is no major increase in sleep fragmentation after diazepam injection.
14 As NREM sleep SWA is strongly associated with sleep homeostatic processes, we next investigated the time course of SWA across the 12-hour light period. During the baseline day ( Figure 3F, left), and after vehicle injection ( Figure 3F, right), animals showed the typically observed high initial SWA levels at the beginning of the light period, which dissipated across time ( Figure 3E and F). In contrast, after diazepam injection SWA was consistently reduced compared to baseline and was significantly Next, we investigated the impact of diazepam on cortical neuronal activity and associated local field potentials recorded from deep layers of the motor cortex. This allows for the effects of diazepam on localised cortical regions to be disentangled from EEG recordings, which reflect the global behavioural state. Figure 4A shows representative LFP traces during wake, NREM sleep and REM sleep, which can be defined using the classical criteria used for scoring the EEG into vigilance states (for example the presence of slow waves is indicative of NREM sleep). LFP power spectra replicated the effects of diazepam injection observed in the EEG ( Figure 4B) Individual slow waves were identified and their incidence, amplitude and duration were quantified (schematic of an individual slow wave showing how amplitude and duration were quantified shown in Figure 4C). Diazepam injection resulted in a larger proportion of low amplitude slow waves, compared to vehicle injection (p<0.0001, paired t-test, Figure 4D). On average, across the 12 hour light period slow waves were similar in duration but smaller in amplitude after diazepam injection (dur: p=0.45; amp: p<0.0001, paired t-test; Figure 4E).
Slow waves are well established to be underpinned by regularly occurring periods of neuronal silence called OFF periods (Buzsáki et al., 2012;Chauvette et al., 2010;Okun and Lampl, 2008;Poulet and Petersen, 2008). After diazepam injection, this association was less distinct and the signals became asynchronous with one another as shown in the representative traces in Figure 4F. To investigate this further we calculated the average LFP signal and aligned this to the midpoint of all OFF periods longer than 50 ms ( Figure 4G). As expected, we observed that OFF periods were associated with LFP slow waves; however, these were reduced after diazepam.
Furthermore, while the duration of an OFF period was strongly positively correlated with the amplitude of slow waves, after both vehicle and diazepam injections ( Figure   4H), OFF periods were less frequent (repeated measures ANOVA (Greenhouse-  Figure 4I and J).
As OFF periods are the result of a synchronous silence in neuronal activity across all recording channels, one possibility is that the neocortex is overall more active after diazepam treatment, or synchronicity between channels is reduced. We first calculated overall average firing rates across all recording channels expressed a percentage of the mean during the vehicle injection light period. Contrary to our expectation, we found that neuronal activity was somewhat lower during NREM sleep after diazepam, as compared to vehicle (repeated measures ANOVA (Greenhouse-  Figure 5A). Then, we addressed the second possibility of a reduced spatial synchrony between neurons recorded from the microwire array. Visual inspection of raw multiunit traces after diazepam revealed that it was not uncommon that neighbouring channels showed very different levels of activity at the same time, such that one channel may show intense neuronal activity while another just a few hundred microns away is in complete silence ( Figure 5B). We therefore quantified the number of recording channels (out of a total of 16 recording channels) that showed multiunit activity at the same time in 20ms bins, to assess the degree of synchrony of neuronal activity across the network ( Figure 5C and D). The key observation was that on fewer occasions all recording channels were simultaneously silent after diazepam as compared to vehicle injection (p=0.03; Wilcoxon sign rank test., Figure 5D, inset).

Discussion
The profound reduction in EEG spectral power in the SWA range with diazepam injection reported in this study is consistent with previous literature (Kopp et al., 2004(Kopp et al., , 2003Lancel et al., 1996;Lancel and Steiger, 1999;Tobler et al., 2001). The influence of diazepam on EEG spectra was more apparent in the frontal region compared to the occipital, corresponding well with similar previous reports (Kopp et al., 2004(Kopp et al., , 2003. The reduction in SWA with diazepam injection was maintained across the 12 hours after diazepam. This is consistent with evidence in humans that showed the reduction in SWA persisted into the following drug-free night after injection of the benzodiazepines flunitrazepam, flurazepam or triazolam, whereas the effects on the spindle frequency range were recovered (Borbély et al., 1983). Diazepam also shifted the theta peak to slower frequencies, especially in the occipital derivation where theta activity is known to predominate due to its proximity to the hippocampus. It is possible that the shift in theta peak may reflect changes in body temperature, which have been previously associated with changes in EEG frequencies in animal and human studies (Deboer, 2002(Deboer, , 1998. This is supported by evidence that diazepam decreases body temperature (Mailliet et al., 2001), and so hypothermia may have influence on overall EEG frequencies and power spectra. However, the effects we observed in this study were state-and frequency-specific rather than resulting in a shift in overall power across all frequencies. Consistently, evidence in humans suggests that under certain conditions physiological (e.g. circadian) changes in EEG power spectra may be dissociated from unspecific effects of body temperature (Dijk, 1999). We therefore believe that the effects we report here primarily reflect changes in neural activity induced by diazepam, rather than just being an indirect effect of hypothermia. Future studies should investigate further the association between diazepam induced hypothermia and EEG power spectra. We would like to add that when average NREM sleep spectra were subdivided into 3 hour intervals, this revealed enhanced power in higher frequencies, including spindle-frequency range, during the first 3 hours after diazepam injection only, while the effects on slow frequencies persisted at least until the end of the day, in line with previous studies (Kopp et al., 2004(Kopp et al., , 2003Lancel et al., 1996;Lancel and Steiger, 1999;Tobler et al., 2001).
While the molecular mechanisms of benzodiazepines action have been elucidated in great detail, previous studies have not looked at the cortical neural activity underlying the response to diazepam in the neocortex. To our knowledge, this is the first study to investigate the neuronal activity underlying the effects of diazepam on localised cortical activity, as determined by local field potential and multiunit activity recordings from deeper layers of the primary motor cortex. Local field potential slow waves during NREM sleep were of smaller amplitude after diazepam injection, while the OFF periods associated with these slow waves were less frequent and shorter in duration. It is possible that this may be due to a reduced synchronicity across the cortex, which leads to fewer occasions where all channels are silent (OFF period) at the same time. This corresponds well with the known association between slow waves and their neuronal counterparts, where lower SWA spectral power is a result of reduced occurrence of slow waves and population OFF periods Panagiotou et al., 2017;Riedner et al., 2007;Steriade, 2006;Steriade et al., 1993;Vyazovskiy et al., 2009Vyazovskiy et al., , 2007. Finally, although overall firing rates were lower after diazepam injection compared to vehicle injection, the more striking finding of this study was that on fewer occasions all recording channels were synchronously silent after diazepam injections. This indicates that diazepam may disrupt the synchronicity of neuronal activity across the cortex and therefore localised cortical activity may be differentially affected by diazepam, as compared to more global EEG mechanisms. This is supported by recent evidence that diazepam administration in mice induces a slowing of global electroencephalogram (EEG) oscillations in a frequency-and behavioural-state dependent manner (Scheffzük et al., 2013).
It is now widely accepted that sleep is a cortical circuit phenomenon initiated locally, and gradually encompassing more networks across sleep periods (Hinard et al., 2012;Krueger et al., 2016Krueger et al., , 2008Lemieux et al., 2014;Pigarev et al., 1997;Sanchez-Vives and McCormick, 2000;Vyazovskiy and Harris, 2013). Localised occurrence of sleep-like activity in distinct cortical regions during wakefulness (so called local sleep) has been shown both in rats and humans (Cajochen et al., 1999;Finelli et al., 2001;Naitoh et al., 1969), and has been previously associated with deficits in cognitive performance in both animal and human studies (Cirelli and Tononi, 2008;Huber et al., 2004;Vyazovskiy et al., 2011). Human intracortical recordings, performed simultaneously from multiple brain regions, have shown that slow waves during sleep are mostly confined to local regions, with typically ~30% of the brain showing simultaneous slow waves . A number of species have adapted their sleep-wake cycle to express more localised forms of sleep to overcome evolutionary pressures such as migratory birds showing microsleeps (Rattenborg, 2006;Rattenborg et al., 2000) or aquatic species sleeping one hemisphere at a time (Mukhametov, 1987;Mukhametov et al., 1977;Ridgway, 2002). As adaptations such as these are fairly uncommon, it may be that the benefits associated with sleeping one hemisphere at a time are only relevant under specific extreme circumstances 20 (Rattenborg et al., 2000). It is possible therefore, that local sleep does not provide the same benefits as regular, more global sleep (Vyazovskiy and Harris, 2013). Therefore, the effect of diazepam in disrupting the overall synchronisation of cortical activity may explain the observed effects of benzodiazepines on behaviour and cognitive function (Johnson and Chernik, 1982;Kripke, 2000;Roehrs et al., 1994).
Despite these major effects of diazepam on EEG spectra and neuronal activity, this study and others (Kopp et al., 2004) have reported little or no effect of diazepam on the total amount of NREM sleep in laboratory mice. This suggests that overall vigilance states may be differentially affected by diazepam, as compared to more localised cortical activity and highlights the necessity to consider all levels of organisation when investigating the effects of sedative drugs on sleep quality. It remains to be established, whether the differences, or lack thereof, reported here may be generalised across other benzodiazepines and other cortical and subcortical brain areas, such as associative and sensory areas. As slow waves have been observed in every cortical region recorded to date, it is possible that the observations in the motor cortex could generalise to other cortical areas.
In conclusion, this study replicated the well-established effect of diazepam of reducing EEG SWA power during NREM sleep and enhancing power in high frequency ranges. In addition, our findings suggest that these effects may be the result of a disruption in the synchronicity of activity across the cortex during NREM sleep which ultimately leads to a reduction in slow wave amplitude (and therefore reduced SWA) and a reduced incidence and duration of OFF periods associated with slow