Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/springer-journals/detection-of-neuronal-off-periods-as-low-amplitude-neural-activity-IZ3EHa46ZY
- Publisher
- Springer Journals
- Copyright
- Copyright © The Author(s) 2023. corrected publication 2023
- eISSN
- 1471-2202
- DOI
- 10.1186/s12868-023-00780-w
- Publisher site
- See Article on Publisher Site
Abstract
Background During non‑rapid eye movement sleep (NREM), alternating periods of synchronised high (ON period) and low (OFF period) neuronal activity are associated with high amplitude delta band (0.5–4 Hz) oscillations in neo‑ cortical electrophysiological signals termed slow waves. As this oscillation is dependent crucially on hyperpolarisation of cortical cells, there is an interest in understanding how neuronal silencing during OFF periods leads to the genera‑ tion of slow waves and whether this relationship changes between cortical layers. A formal, widely adopted definition of OFF periods is absent, complicating their detection. Here, we grouped segments of high frequency neural activity containing spikes, recorded as multiunit activity from the neocortex of freely behaving mice, on the basis of ampli‑ tude and asked whether the population of low amplitude (LA) segments displayed the expected characteristics of OFF periods. Results Average LA segment length was comparable to previous reports for OFF periods but varied considerably, from as short as 8 ms to > 1 s. LA segments were longer and occurred more frequently in NREM but shorter LA seg‑ ments also occurred in half of rapid eye movement sleep (REM) epochs and occasionally during wakefulness. LA segments in all states were associated with a local field potential (LFP) slow wave that increased in amplitude with LA segment duration. We found that LA segments > 50 ms displayed a homeostatic rebound in incidence following sleep deprivation whereas short LA segments (< 50 ms) did not. The temporal organisation of LA segments was more coherent between channels located at a similar cortical depth. Conclusion We corroborate previous studies showing neural activity signals contain uniquely identifiable periods of low amplitude with distinct characteristics from the surrounding signal known as OFF periods and attribute the new characteristics of vigilance‑state ‑ dependent duration and duration‑ dependent homeostatic response to this phe‑ nomenon. This suggests that ON/OFF periods are currently underdefined and that their appearance is less binary than previously considered, instead representing a continuum. Keywords Sleep, ON/OFF periods, Homeostasis *Correspondence: Christian D. Harding christian.harding@sjc.ox.ac.uk Full list of author information is available at the end of the article © The Author(s) 2023, corrected publication 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Harding et al. BMC Neuroscience (2023) 24:13 Page 2 of 15 of large data sets. Additionally, such methods have ena- Introduction bled the development of real time slow wave modulation, Slow waves, high amplitude oscillations in the delta fre- the therapeutic applications of which are currently being quency range (0.5–4 Hz) observed in electrophysiological explored [16, 17]. signals are a characteristic of non-rapid eye movement Similarly, a range of methods have been devised to sleep (NREM). A role of slow waves has been suggested solve the problem of detecting ON/OFF period transi- in processes that are dependent on NREM sleep time, tions. The simplest method is to apply amplitude and such as immune function and restoration of cognitive duration thresholds to spike trains: binary traces of function [1]. Slow wave activity (SWA, spectral power high amplitude deflections (spikes) in single or multiu - in the delta frequency range) is a reliable index of sleep nit activity signals [5, 18–21]. Though widely used both homeostasis [2] and local SWA homeostasis is correlated for offline and online detection of ON/OFF periods, this with improved motor learning task performance follow- method is sensitive to the threshold used to define spikes ing sleep [3]. To help clarify the proposed involvement which is often set using visual inspection and binariza- of slow waves in restorative processes and memory func- tion results in the loss of a large amount of data. More tion, and to understand whether they are causally linked sophisticated methods for detecting ON/OFF periods in to these functions or rather simply a measurable out- continuous neuronal activity signals can broadly be clas- put of an underlying process, it is vitally important that sified as ‘threshold-crossing’ or ‘predictive’ algorithms. methods are developed which improve our understand- ‘Threshold-crossing’ algorithms work by processing the ing of the neurophysiology of slow waves. data until a bimodal distribution is obtained upon which Early studies, pioneered by Steriade et al. [4], showed a threshold is applied to separate data between ON and that cortical slow waves during sleep depend on the syn- OFF periods [13, 22, 23]. ‘Predictive’ algorithms assume chronous hyperpolarization of cortical cells. The result - bimodality and assign data to one of either state based ing periods of reduced neuronal activity, often referred on the probability of a predictive model fitted to the data to as OFF periods, are associated with depth positive/ [24–26]. surface negative potentials in electrophysiological signals A common trend observed in the design of these detec- and alternate with periods of enhanced neuronal activ- tion algorithms is that they are built and tested upon a ity referred to as ON periods which are associated with subset of data displaying clear slow waves [13] or ON/ depth negative/surface positive potentials [5]. Each slow OFF oscillations [26] or acquired from anaesthetised ani- wave cycle therefore corresponds to a single ON/OFF mals where slow wave activity is generally more regular period alternation. It has been hypothesised that some than during sleep [13, 22, 23, 25, 26]. Whilst it is likely of the functions ascribed to NREM sleep are in fact the that these methods will generalise to the case of sleep result of neuronal inactivity during OFF periods [6]. For recordings, there may be advantages to designing a detec- example, increases in neuronal activity resulting from tion method on ‘noisy’ data characteristic of in vivo free- sensory and optogenetic stimulation cause DNA double- moving sleep recordings and that can be applied to the strand breaks in neurons of mice [7]. These are repaired entire duration of chronic recordings used to study sleep more rapidly during sleep than wakefulness, suggesting behaviour. Another trend is to judge the performance of a neuronal inactivity in NREM OFF periods may provide detection method by comparing the output when applied an opportunity for housekeeping functions such as repair to different signals recording the same neural activity of minor cellular damage [6, 9]. (e.g. intra- vs extra- cellular, [13]) or with the output of If it is indeed the dynamics of the synchronised neu- other methods [25, 26]. As no method can yet be called ronal activity that we are primarily interested in under- the ‘gold standard’, this makes it challenging to judge the standing, ON/OFF periods in local neuronal networks relative merit of methods and to assess the effect of opti - may provide a more direct measure of this behaviour misation steps within the pipeline (e.g. to remove short than the slow waves it gives rise to. Despite this, far duration state transitions/interruptions). more attention has been given to detecting slow waves An alternative design approach is to detect OFF peri- than ON/OFF periods. Slow wave detection methods ods that match the established characteristics of slow have been developed based on amplitude and dura- waves during spontaneous sleep and then infer ON tion thresholds [10, 11], spectral frequency [12, 13] and periods retrospectively. OFF periods should occur pre- using a neural network approach [14]. Whilst the num- dominantly, but not exclusively, in NREM sleep. SWA ber and quality of events detected as slow waves vary is highest and neuronal firing rate lowest in NREM with method and implementation [15], these can at least sleep compared to wakefulness and rapid eye move- be validated against a ‘gold standard’ manual scoring by ment sleep (REM), which suggests that OFF periods experienced observers using tried and test criteria. Auto- are more likely to occur in this state [5]. However, slow mated slow wave detection facilitates rapid processing Har ding et al. BMC Neuroscience (2023) 24:13 Page 3 of 15 waves are known to occur regularly during REM sleep Results [27] and occasionally during both inactive and active Temporal features and state dependency wakefulness [28–30]. By definition, OFF periods should LA segments were detected in all vigilance states in be consistently associated with depth positive/sur- different cortical layers (Fig. 1). To characterise the face negative deflections in the electroencephalogram temporal features of LA segments and compare their (EEG) corresponding to the initiation of slow waves properties across vigilance states, we detected LA seg- [5], and their duration should be positively correlated ments in MUA recordings from a representative layer with slow wave amplitude [5]. Changes in the recruit- 5 channel of motor cortex on the baseline (BL) day. We ment and decruitment of cortical neurons to ON/OFF first looked at whether LA segments are more com - periods respectively are associated with homeostatic mon in NREM sleep than other vigilance states. The dynamics and infraslow fluctuations of SWA, which majority of LA segments were detected in NREM sleep may be a mechanism for synchronising neuronal activ- (72547.00 ± 9164.22, 91.70 ± 1.44%) and REM sleep ity across the cortex [5, 31]. Synchronization of OFF (5675.43 ± 1405.64, 7.11 ± 1.46%) with a small propor- periods between brain regions should therefore be vari- tion detected in WAKE (1130.29 ± 437.50, 1.20 ± 0.34%) able with widespread (global) OFF periods reflecting (Fig. 2A). LA segment onset in all states is associated high decruitment of neurons and localised OFF peri- with a characteristic sharp drop in MUA amplitude ods reflecting low decruitment of neurons [32]. Finally, (Fig. 2A). There was a significant effect of state on LA OFF periods should respond to changes in sleep pres- segment incidence (Fig. 2B, one-way repeated measures sure and circadian drive as a result of their homeostatic ANOVA, F(2,12) = 68.92, p < 0.05) with LA segments regulation [2]. occurring most frequently in NREM sleep and least fre- The aim of this study was to determine if low activity quently in WAKE. There was also a significant effect of is sufficient as a criterion to detect OFF periods in neu - state on the proportion of 4-s epochs containing at least ronal signals from freely behaving mice. To achieve this one LA segment (Fig. 2C, one-way repeated measures we designed a simple ‘threshold-crossing’ algorithm to ANOVA, F(1.02,6.11) = 286.79, p < 0.05) with the vast identify a population of low amplitude (LA) segments majority of NREM epochs (97.37 ± 0.56%), the occasional in multiunit activity (MUA) recordings from freely WAKE epoch (3.20 ± 0.87%) and over half of REM epochs behaving mice informed by the distribution of spiking (61.19 ± 5.35%) containing an LA segment. These find - amplitudes during NREM sleep and assessed whether ings show that LA segments are preferentially, but not these segments recapitulated the characteristics of OFF exclusively, associated with NREM sleep as has previ- periods expected from the latter’s association with slow ously been reported for OFF periods [5, 27–30]. We then waves. looked at the duration of LA segments to see how this Fig. 1 LA segments detected in all vigilance states. (top panel) 24‑h hypnogram of vigilance states in a freely behaving mouse on a baseline recording day. Black = NREM sleep, green = REM sleep, blue = wakefulness. (bottom panels) Extracts of multiunit activity (MUA, vertical blue bars) and local field potential (LFP, horizontal orange line) from each vigilance state in four adjacent channels along a laminar probe implanted in motor cortex. Detected LA segments denoted in red. The top two channels are located in layer 3 whilst the bottom two channels are located in layer 5 Harding et al. BMC Neuroscience (2023) 24:13 Page 4 of 15 Fig. 2 Temporal features of LA segments. A Global distribution of LA segments between vigilance states across study with examples of MUA amplitude during LA segments displayed for each state. Each figure shows 400 randomly selected LA segments sorted by duration. Segment time range = onset ‑100 ms to onset + 300 ms. Colour of each segment scaled to MUA amplitude (dark = low, colour = high). B The effect of vigilance state on the incidence of LA segments. Incidence values reported as number per minute of vigilance state. C The effect of vigilance state on the proportion of 4‑s epochs containing an LA segment. D Frequency of LA segments as a function of duration for each vigilance state. Inset y‑axis scaled to increase resolution of REM and WAKE states. E The effect of vigilance state on the duration of LA segments. Black/grey = NREM, green = REM, blue = WAKE. N = 7. Mean ± SEM. Significance of effects assessed using one ‑ way repeated measures ANOVA followed by post‑hoc pairwise t‑tests with Bonferroni correction (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001) compared with previously reported OFF period descrip- Whilst the relative duration of OFF periods in different tions. LA segment durations ranged from 8 ms to greater states has not been reported, this does mirror the find - than 1 s but on average lasted 116.26 ± 2.16 ms. This is ing that REM sleep slow waves are described as smaller slightly shorter than a previous description of OFF peri- than those occurring in NREM sleep in reference to their ods in mice of a comparable age (134 ms, [18]), however lower amplitude, which is correlated with OFF period this was not unexpected considering the comparison is duration [5, 27]. with a detection method targeting the long duration OFF periods. u Th s, this finding does not preclude the possibil - Association with LFP ity that LA segments are OFF periods. The distribution Evidence suggests that OFF periods coincide with local of LA segments was strongly positively skewed during field potential (LFP) slow waves [4, 5]. If LA segments all states (1.54 ± 0.18, Fig. 2D). The distributions were represent OFF periods, we would expect a similar asso- leptokurtic for WAKE (kurtosis = 5.16) and REM (3.68) ciation. We first extracted a 400 ms window of LFP for LA segments, such that long LA segments occurred each LA segment in a representative layer 5 channel dur- more frequently than in a standard Gaussian distribu- ing baseline sleep from onset -100 ms to onset + 300 ms tion, but there was no excess kurtosis in the NREM and calculated the average LFP signal during LA seg- sleep distribution (3.00). LA segments were longer in ments during each vigilance state (Fig. 3A). In each state, NREM sleep (116.89 ± 6.11 ms) than either REM sleep LA segments were associated with a positive deflection (102.46 ± 6.06 ms) or WAKE (99.18 ± 4.77 ms) states in of the LFP which coincides with segment onset. This which LA segment durations were similar (Fig. 2E, one- deflection lasts around 150 ms, which if considered a way repeated measures ANOVA, F(2,12) = 34.48, p < 0.05) half wave suggests a frequency of ~ 3 Hz, within the though the absolute difference was small (ca. 15 ms). delta frequency range. This deflection was bounded by Har ding et al. BMC Neuroscience (2023) 24:13 Page 5 of 15 Fig. 3 MUA LA segments associated with slow‑ wave activity in LFP. A Layer 5 LFP corresponding with LA segments in each vigilance state. Composite image of LFP traces converted to heatmap (higher colour saturation = higher density) overlaid with mean ± SEM. Segment time range = onset ‑100 ms to onset + 300 ms. B Global distributions of the LFP (2–6 Hz) phase corresponding to LA segment ONSET (bottom) and END (top) in each vigilance state. Example of a sinusoid wave overlaid for visualization purposes only. C Distribution of preferred LFP (2–6 Hz) phase corresponding to LA segment ONSET and END in each vigilance state (proportion) with mean resultant vector. D Eec ff t of state on peak amplitude in LFP corresponding to LA segments. E Relationship between LA segment duration and peak amplitude of corresponding LFP. Least‑squares regression line and significant Pearson correlation coefficient shown. Black/grey = NREM, green = REM, blue = WAKE. N = 7. Mean ± SEM. Significance of effects assessed using one ‑ way repeated measures ANOVA followed by post‑hoc pairwise t ‑tests with Bonferroni correction (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001) slight negative deflections. These features are consist - duration. We found a significant effect of vigilance state ent with LA segments being time locked to LFP slow on peak LFP amplitude (F(2,12) = 3.97, p = 0.047) how- waves. To confirm this, we performed a phase analysis ever post-hoc pairwise t-tests did not reveal a significant to determine the preferred phase of LA segment onset difference in mean LFP amplitude between any states and end in the delta frequency range of the LFP (Fig. 3B, (Fig. 3D). This suggests that LA segments in all states C). LA segment onset preferentially occurred at ~ 310° were associated with similar amplitude slow waves. It has (NREM: 306.26 ± 1.49°, REM: 315.31 ± 1.60°, WAKE: been reported that slow wave amplitude increases as a 316.84 ± 1.52°), coinciding with the rising limb of delta function of OFF period duration [32]. In agreement with oscillations. LA segment end preferentially occurred this, we found that LA segment duration was positively at ~ 50° (NREM: 53.50 ± 0.97°, REM: 50.60 ± 1.24°, correlated with peak LFP amplitude (Fig. 3E, r = 0.52, WAKE: 42.17 ± 1.43°), coinciding with the falling limb of p < 0.05). delta oscillations. In all states, onset and end phase distri- butions were significantly different (NREM: Watson-Wil - Homeostatic regulation liams, F(1,12) = 487.39, p < 0.05, REM: Watson-Williams, The build up and subsequent release of sleep pressure F(1,12) = 286.79, p < 0.05, WAKE: Watson-Williams, resulting from homeostatic regulation has two expected F(1,12) = 238.83, p < 0.05). outcomes on slow waves and presumably OFF peri- Finally, we looked at how time locked peak LFP ods: slow wave activity should decrease over the course amplitude changed with vigilance state or LA segment of the inactive period and slow wave activity should be Harding et al. BMC Neuroscience (2023) 24:13 Page 6 of 15 higher during sleep after sleep deprivation [2]. To assess was a significant effect of prior sleep–wake history on whether these features apply to LA segments, we ana- occupancy time (Fig. 4A, two-way repeated measures lysed 6 h of spontaneous activity during the second half ANOVA, F(1,5) = 27.60, p < 0.05) and duration (Fig. 4B, of the light phase (ZT6–ZT12) prior to which animals two-way repeated measures ANOVA, F(1,4) = 48.14, were undisturbed (baseline spontaneous activity, BL) or p < 0.05) of LA segments with post-hoc pairwise t-tests were kept awake by providing novel objects to explore confirming a significant increase in the both metrics (sleep deprivation, SD) for 6 h between ZT0–ZT6 on between ZT6-ZT8.5 after SD. There was a significant consecutive days. As in the previous sections, a single effect of zeitgeber time on occupancy time (Fig. 4A, layer 5 channel was used to represent each animal. There two-way repeated measures ANOVA, F(11,55) = 11.80, Fig. 4 LA segments are homeostatically regulated. A Relationship between LA segment occupancy and time during NREM sleep between ZT6 and ZT12 on baseline (BL) and sleep deprivation (SD) days (30 min bins). Occupancy values reported as seconds of LA segments per minute of NREM. B Change in LA segment duration during NREM sleep between ZT6 and ZT12 on BL and SD days (30 min bins). C Relationship between LA segment duration and the change in LA segment incidence between BL and SD days during the 1st hour after sleep deprivation. A change > 0 describes an increase in incidence with the sleep deprivation treatment. D Relationship between LA segment incidence and time during NREM sleep between ZT6 and ZT12 on the sleep deprivation day as a function of LA segment duration. Circles denote mean incidence and lines show least‑square regression for each duration category. N = 7 . Mean ± SEM. Significance of effects assessed using two ‑ way repeated measures ANOVA followed by post‑hoc pairwise t ‑tests with Bonferroni correction (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001). Only significant Pearson correlation coefficients shown. One animal lacked LA segments < 50 ms for at least one time bin so was not included in the analysis for this group of segments in D Har ding et al. BMC Neuroscience (2023) 24:13 Page 7 of 15 p < 0.05) and duration (Fig. 4B, two-way repeated meas- time appeared to increase with LA segment duration ures ANOVA, F(11,44) = 7.04, p < 0.05) of LA segments (Fig. 4D). To confirm this, we fit a linear regression model with both metrics decreasing as a function of time. to the distribution of relative LA segment incidence to Finally, there was a significant interaction between prior zeitgeber time for each LA segment duration category. sleep–wake history and zeitgeber time on occupancy Except for segments < 50 ms in duration, the regression time (Fig. 4A, two-way repeated measures ANOVA, of zeitgeber time against LA segment incidence was sig- F(11,55) = 3.73, p < 0.05) and duration (Fig. 4B, two-way nificant across all time bins (see Additional file 1: Figure repeated measures ANOVA, F(1,44) = 6.85, p < 0.05). To S2). Furthermore, the estimate for the predictor vari- understand this further we performed post-hoc linear able (i.e. zeitgeber time) became increasingly negative regressions for each condition separately. The regression with increasing LA segment duration (β = − 3.11 (50– of zeitgeber time against LA segment occupancy time was 100 ms) < − 3.99 (100–150 ms) < − 7.33 (150–200 ms), significant for the SD condition (linear regression model, − 13.41 (200–250 ms) < − 19.75 (250 + ms)). Overall, the F(1,82) = 71.756, p < 0.05) but not for the BL condition homeostatic response of LA segments does appear to (linear regression model, F(1,81) = 3.74, p > 0.05). Simi- depend on their duration, contrary to the expectation of larly the regression of zeitgeber time against LA segment OFF periods being an all or none phenomenon. duration was significant for the SD condition (regression model, F(1,82) = 23.91, p < 0.05) but not for the BL condi- Interchannel coherence tion (linear regression model, F(1,80) = 2.54, p > 0.05). To OFF periods are both a global and a local phenomenon. establish whether the absence of a homeostatic decrease To investigate whether this was also true of our LA seg- in LA segment metrics for the BL condition was simply ments, we determined the temporal coherence between due to the animals having already paid the majority of channels separated by different distances along the lami - their sleep debt by ZT6, we repeated regression analysis nar probe (i.e. different depths within motor cortex) and on data from the entire light period (ZT0-12) and indeed compared coherence at different interchannel distances found significant regressions of zeitgeber time against LA (Fig. 5A, Additional file 1: Figures S3, S4). Coherence segment occupancy (Additional file 1: Figure S1A, linear was calculated between all LA segments detected on the regression model, F(1165) = 14.28, p < 0.05) and duration baseline day. Coherence was generally highest along the (Additional file 1: Figure S1B, linear regression model, diagonal, that is, between adjacent channels. Although F(1164) = 10.97, p < 0.05). only channels with neural activity (i.e. spikes) as deter- OFF periods are usually portrayed as an all or none mined by visual inspection were retained, OFF periods phenomenon. Thus OFF period duration would not be in superficial channels were not always coherent with expected to have a significant effect on a feature such as other channels (Fig. 5A, Ch1). Thus whilst our method homeostatic regulation. To assess whether this is true for still detects OFF periods, reliability may be lower for LA segments, we binned segments in 5 groups based on superficial channels. To determine whether any coher - duration and looked at incidence as a marker of homeo- ence found was greater than chance, we generated sur- static regulation. We found that there was a significant rogate channels by randomly shuffling LA and non-LA increase in LA segment incidence during the first hour segments from the original channels (see Methods). To after sleep deprivation as compared to spontaneous avoid vigilance state-dependent effects we only used data activity for segments of 100-150 ms (Fig. 4C, Bonferroni from NREM sleep for coherence analysis and shuffling. adjusted paired T-test, t(6) = − 2.81, p < 0.05), 150-200 ms There was a significant effect of channel type (Fig. 5B, (Fig. 4C, Bonferroni adjusted paired T-test, t(6) = − 6.67, two-way repeated measures ANOVA, F(1,6) = 109.51, p < 0.05), 200-250 ms (Fig. 4C, Bonferroni adjusted paired p < 0.05), which suggests the LA segments in original T-test, t(6) = − 8.06, p < 0.05) and 250 + ms (Fig. 4C, Bon- channels, which have a higher mean coherence across all ferroni adjusted paired T-test, t(6) = − 6.69, p < 0.05) interchannel distances, are more synchronous than LA in duration. There was no significant change in inci - segments in surrogate channels. This finding highlights dence for LA segments < 50 ms or 50-100 ms in dura- the generalised synchronicity of OFF periods across lam- tion, though incidence of 50–100 ms LA segments varied inar layers. Furthermore, there were significant effects of considerably between subjects with 5/7 animals showing interchannel distance (Fig. 5B, two-way repeated meas- an increase of > 5 segments per minute following sleep ures ANOVA, F(9,54) = 26.75, p < 0.05) and interchannel deprivation. Furthermore, whilst the incidence of LA distance * type interaction (Fig. 5B, two-way repeated segments > 50 ms in duration was negatively correlated measures ANOVA, F(9,54) = 31.01, p < 0.05). We used with time (Fig. 4D) in the 6 h following sleep deprivation, Pearson correlation to interrogate the interaction term there was no correlation for segments < 50 ms. We noted further and found that coherence was negatively cor- that the magnitude of relative decrease in incidence over related with original interchannel distance (Fig. 5B, Harding et al. BMC Neuroscience (2023) 24:13 Page 8 of 15 Fig. 5 LA segments have both local and global dynamics. A Example temporal coherence matrix between 12 channels arranged by depth for one animal (green = low coherence, yellow = high coherence). Temporal coherence reported as the average proportion of time spent in the low amplitude state during which both channels are synchronously in a low amplitude state (see Methods). B Relationship between the temporal coherence of LA segments in pairs of channels and the distance between those channels along the laminar probe for both original and surrogate channels. N = 7. Mean ± SEM. Least‑squares regression lines and significant Pearson correlation coefficients shown. There were insufficient data points to include the maximum interchannel distance (11) r = − 0.49, p < 0.05) but was not correlated with surro- occur. The duration and total occupancy time per min - gate interchannel distance. This suggests that in terms of ute of NREM sleep of LA segments decreases throughout LA segment incidence, neighbouring channels are more the inactive period of mice during spontaneous activity synchronous than distant channels for original chan- (Additional file 1: Figure S1) and after sleep deprivation nels but not for surrogate channels where channel pairs (Fig. 4) whilst the absolute magnitude of both metrics is are equally synchronous across the range of interchannel increased by sleep deprivation, consistent with expecta- distances. This is consistent with global neocortical OFF tions of a phenomenon regulated by sleep homeostasis periods displaying subtle layer dependent effects (i.e. [33]. Finally, LA segments are temporally synchronised localised dynamics). across neocortical laminar layers at both a global and local scale. Together, these findings strongly suggest that LA segments represent OFF periods. Discussion This finding is important for two key reasons. First, we Previous attempts to identify OFF periods in high fre- extend the work of previous studies showing that OFF quency neural activity have used methods of varying periods can be detected based on high frequency neural complexity and levels of processing to fit their results to activity amplitude alone without reference to slow waves co-occur with slow waves. Here we show that a popula- (e.g. [23]) by applying this methodology to recordings tion of low amplitude segments can be extracted from from freely behaving mice and by showing that it can be high frequency neural activity without prior knowledge leveraged to detect OFF periods in all vigilance states. of concomitant LFP activity which fit the expected char - This permits detection of OFF periods in cases where the acteristics of OFF periods. LFP is uninformative and opens up the possibility of stud- LA segments present as brief reductions in MUA ying LFP slow waves and MUA OFF periods as separate amplitude, primarily during NREM sleep but also measures of the phenomenon of synchronised neuronal appear in REM sleep and WAKE. LA segments usually silencing events, each providing unique information at last approximately 100 ms but duration is variable with different scales of integration across brain regions. The many segments upwards of 1 s occurring. LA segments MUA-based local assessment of OFF periods will be of are associated with a positive deflection in deep neocor - particular importance for advancing the understanding tical LFP (layer 5) characteristic of slow wave initiation. of layer-specific dynamics in neocortex because LFP sig - Furthermore, LA segment onset and end times are phase nals are influenced by volume conductance from adjacent locked to the LFP in the delta range where slow waves Har ding et al. BMC Neuroscience (2023) 24:13 Page 9 of 15 layers and neighbouring brain regions. Second, we pro- however we suggest that the latter may be more likely con- vide a simple method for detecting OFF periods that can sidering the abundant evidence of other OFF period-like be implemented as part of an easily accessible toolbox. behaviour for LA segments. This will allow for greater accessibility of OFF period analyses in electrophysiological sleep research and will LA segment characteristics show state dependency help foster the growing interest in ‘local sleep’ effects in As part of our validation of LA segments as OFF periods, behavioural neuroscience. we chose to differentiate between LA segments occurring in different vigilance states. As expected, LA segments occurred most frequently in NREM sleep but also occurred Homeostatic regulation of LA segments depends in the majority of REM sleep epochs and occasionally dur- on segment duration ing wakefulness. Furthermore, LA segments had a similar Unexpectedly, we found that the duration of LA seg- LFP profile independent of vigilance states suggesting all ments had an effect on their homeostatic response. LA were associated with slow waves. Less expected was the segments < 100 ms in duration showed a different home - finding that LA segments during NREM sleep were longer ostatic response to segments > 100 ms in duration. LA than LA segments in REM sleep and wakefulness. This segments of 50–100 ms occurred more frequently imme- could suggest that OFF periods during wakefulness and diately following sleep deprivation than after a few hours REM sleep do not achieve the same recruitment of neurons of recovery sleep, consistent with homeostatic regulation. to the OFF state than in NREM sleep. Wakefulness and However, they showed no increase in incidence after REM OFF periods may be more localised as wakefulness sleep deprivation compared with the same subjective and REM sleep are more ‘active’ states and as such the neu- hour on a day without prior sleep deprivation. LA seg- romodulatory milieu disrupts the formation of synchro- ments < 50 ms showed neither a homeostatic decrease nous activity. in incidence after sleep nor an increase after sleep dep- rivation. Most importantly, longer segments showed a Conclusion steeper decrease in incidence following sleep deprivation We provide strong evidence that OFF periods can be than shorter segments, evidence of a greater response to detected by clustering together cortical multiunit activ- a build-up of sleep pressure. ity segments of similarly low amplitude of extracellularly One explanation of these results is that short LA seg- recorded neuronal spiking. These low amplitude segments ments do not represent OFF periods as they are classi- show many characteristics expected of OFF periods, cally described. This may best explain the absence of a including NREM predominance, a strong association with clear homeostatic response in the shortest LA segments LFP slow waves, sleep homeostasis and temporal coher- (< 50 ms). Therefore, we suggest that these short LA seg - ence across cortical layers. Furthermore, we find that the ments are not consistent with current descriptions of incidence of longer LA segments respond more strongly OFF periods and that, as others have done previously to sleep pressure than short LA segments and that LA seg- [24], a minimum duration of 50 ms should be introduced ments are longer in NREM sleep than REM sleep or wake- to remove brief decreases in MUA amplitude that do not fulness. These vigilance-state- and duration-dependent behave in a similar way to longer decreases commonly effects were not previously described for OFF periods but identified as OFF periods. However, this explanation does these findings may represent additional OFF period fea - not account for the mixed results for 50–100 ms LA seg- tures that have either been overlooked or are only revealed ments and the graded intra-day homeostatic response of with multiunit activity amplitude-only detection methods. LA segments by duration. Two hypotheses would explain this result. First, the proportion of LA segments that rep- Methods resent functionally relevant OFF periods as opposed to All experiments were carried out in accordance with the transient decreases in MUA amplitude may increase with UK Animals (Scientific Procedures) Act of 1986 and in OFF period duration. As such, the longest LA segments compliance with the Animal Research: Reporting In Vivo may show the strongest homeostatic response as the effect Experiments (ARRIVE) guidelines. is less diluted by non-homeostatically regulated noise. Sec- ond, our findings are evidence that OFF periods and their Surgery and electrode implantation associated dynamics lie on a continuum depending on the Adult male wild type C57BL/6 mice (n = 7, internally strength of neuronal recruitment and therefore duration. sourced from Biomedical Services at the University of Greater recruitment leads to longer OFF periods which Oxford, 125 ± 8 d old at baseline recording) underwent display a stronger homeostatic response to sleep pres- cranial surgery to record electroencephalography (EEG), sure, either spontaneously generated or via sleep depriva- electromyography (EMG), local field potential (LFP) and tion. We recognise that both hypotheses may fit the data, Harding et al. BMC Neuroscience (2023) 24:13 Page 10 of 15 multiunit activity (MUA) as previously described [18]. Multiunit activity is the high frequency component of Briefly, under ~ 2–3% isoflurane anaesthesia and asep - neural activity that contains the spiking of multiple neu- tic conditions, stainless steel screws were implanted rons within the vicinity of an electrode. Decimation is epidurally over frontal and occipital cortical areas and a process for downsampling the MUA whilst retaining referenced to a third screw implanted over the cerebel- spiking activity by storing only the highest amplitude lum. Stainless steel wires were implanted into the nuchal value, either negative or positive, recorded during a set muscle to record EMG. A 16-channel laminar probe time period. This means that if multiple neurons spike (NeuroNexus Technologies Inc., Ann Arbor, MI, USA; during that period, only the largest is stored, thus the model: A1 × 16–3 mm-100-703-Z16) was implanted in majority of spikes in the decimated signal will originate left primary motor cortex (AP + 1.1 mm; ML − 1.75 mm; from nearby neurons. The resulting signal will therefore rotated 15° in the AP axis towards the side of the implant) have a high amplitude when nearby neurons are spik- to perform intracortical recordings (LFP and MUA as ing and a low amplitude during periods of quiescence described in [19]). The entire 3 mm length was inserted or when distant neurons are spiking. MUA was gener- gradually into the tissue under both stereotactic and ated by bandpass filtering the laminar signals between microscopic control until the most superficial electrode 300 Hz and 5 kHz then decimating to 498 Hz by split- was approximately 50 µm under the cortical surface. ting the signal into segments of ~ 50 samples and stor- The implantation site was then sealed with the silicone ing the maximum/minimum amplitude of alternating elastomer Kwik-Sil (World Precision Instruments Inc., segments as integers. LFP was generated by zero-phase Sarasota, FL, USA) and the probe was referenced to the distortion bandpass filtering the laminar signal between cerebellar skull screw. Depth and cortical layer of chan- 0.1 and 100 Hz and downsampling to 256 Hz via spline nels were subsequently determined by histology assess- interpolation. All offline manipulations and analyses ment (for histological methodology see [19]). Briefly, the were performed using MATLAB (version R2020a; The position of the laminar implant was determined using a MathWorks Inc, Natick, MA, USA). Prior to vigilance DiL (Thermo Fisher Scientific) fluorescence membrane state scoring, signals were transformed into European stain and the depth of the laminar implant was assessed Data Format as previously reported (see [30]). by measuring the distance between the cortical surface and tissue microlesions generated by applying 10 mA of Experimental design and recording procedure direct current for 25 s to each respective channel using For sleep recordings, animals were individually housed a NanoZ device (White Matter LLC). Each animal was in sound-attenuated and light-controlled Faraday cham- also implanted with a bipolar concentric electrode (Plas- ber cages (Campden Instruments, Loughborough, UK) ticsOne Inc., Roanoke, VA, USA) in the right primary with ad libitum food and water. A 12:12 h light/dark motor cortex, anterior to the frontal EEG screw in rela- cycle (lights on at 9 am = ZT0, light levels 120–180 lx) tion to a separate study (as described in [20]). Screw was implemented, temperature maintained at around electrodes were attached to an 8-pin surface mount con- 22 ± 2 °C, and humidity kept around 50 ± 20%. Animal nector (8415-SM, Pinnacle Technology Inc, KS) whilst were given at least three days post-surgery to acclima- the laminar probe was attached to a ZIF-Clip 16 chan- tize before two recording days starting at ZT0: a baseline nel headstage (Tucker-Davis Technologies Inc., Alachua, day with spontaneous sleep permitted and a sleep dep- FL, USA) and both affixed to the skull with dental cement rivation day. On the sleep deprivation day, animals were (Associated Dental Products Ltd, Swindon, UK). prevented from sleeping from ZT0-ZT6 through gentle handling and the presentation of novel objects to encour- Electrophysiological signal acquisition age naturalistic exploration behaviour [34]. Each animal All signals were first passed to a PZ-5 pre-amplifier served as its own control for the effect of sleep depriva - (Tucker-Davis Technologies Inc., Alachua, FL, USA). tion and therefore the experimental unit in this study is A 128-channel RZ-2 Neurophysiological Recording an animal per recording day (n = 7). System (Tucker-Davis Technologies Inc., Alachua, FL, USA) was then used to acquire tethered electrophysio- Vigilance state scoring and channel selection logical recordings. EEG and EMG signals were continu- EEG, LFP and EMG signals were used to score vigilance ously sampled at 305 Hz and bandpass filtered between states in the Sleep Sign for Animals scoring environ- 0.1–100 Hz. Signals were then downsampled offline to ment (version 3.3.6.1602, SleepSign Kissei Comtec Co., 256 Hz via spline interpolation. Laminar probe chan- Ltd., Nagano, Japan). Four second epochs were scored nel signals were sampled at 25 kHz. Two signals were as WAKE, NREM or REM. Epochs with high frequency extracted from the laminar probe channels: decimated EEG and high amplitude EMG activity were scored as multiunit activity (MUA) and local field potential (LFP). WAKE, epochs with a low frequency EEG characterised Har ding et al. BMC Neuroscience (2023) 24:13 Page 11 of 15 by delta band (0.5–4 Hz) slow waves and sigma band (11– protracted periods of synchronised low amplitude activ- 15 Hz) spindles and a quiet EMG were scored as NREM ity (Fig. 6A), we smoothed the absolute values of MUA sleep. Epochs with a wake-like EEG dominated by theta extracted from NREM sleep by convolution with a 62 ms band activity and a quiet EMG were scored as REM sleep. Gaussian window (width factor = 2.5, sum of weights = 1) Epochs with recording artefacts related to movement and plotted a 1D histogram of amplitudes. This gen - or electrostatic noise were rejected from further analy- erated the bimodal distribution upon which previous ses in all channels (5.06 ± 0.19% of total recording time). ‘threshold-crossing’ algorithms have been based [13, 22, Only vigilance states lasting ≥ 3 epochs were retained for 23] (Fig. 6B, top row). However, upon comparing differ - further analysis to ensure clear differentiation of states. ent channels and recording periods we found it was not Channels with low MUA amplitude variation (i.e. with- always clear where the distributions diverged (Fig. 6B, out spiking activity) were rejected by visual inspection bottom row). As MUA amplitude during ON periods is as were channels located in the corpus callosum (33/112 more varied as a result of spiking events, we theorised channels). For the purpose of inter-animal comparisons, that ON periods should be more sensitive to smoothing each animal was represented by a single layer 5 channel. window length and that this property could be leveraged to facilitate differentiation of the distributions. We com - Low amplitude segment extraction pared MUA amplitude smoothed with a 62 ms Gauss- The concept of distinct ON/OFF states necessitates ian window (width factor = 2.5, sum of weights = 1) and that MUA recorded during NREM should be bimo- a shorter 22 ms Gaussian window (width factor = 2.5, dally distributed. Assuming that OFF periods represent sum of weights = 1) using a 2D histogram. Indeed, we AB C Layer 4 0.5 s Layer 6 Fig. 6 OFF period detection rationale. Top row depicts data from cortical layer 4, bottom row from layer 6. A MUA and LFP signal from different cortical layers from the same NREM sleep interval. OFF periods can be distinguished by a reduced MUA amplitude and often by the presence of LFP slow waves. The appearance of OFF periods, in terms of amplitude and duration, differs between and within layers. B 1D Histogram of NREM MUA amplitudes after Gaussian smoothing. L = length of smoothing window in ms, width factor = 2.5. The histogram of layer 6 has a bimodal distribution with a narrow low amplitude peak (blue arrow), which we call low amplitude (LA) data points, and a broad high amplitude peak (green arrow), which we interpret as non‑LA period data points. The histogram of layer 4 is also bimodal but the peaks are closer together and have similar heights. In both cases, no obvious threshold exists at which to separate the peaks. C 2D histogram of MUA amplitude after Gaussian smoothing with two different window lengths. The histogram is unimodal with only the low amplitude peak retained (blue arrow). Rather than setting an amplitude threshold, LA data points can now be detected by finding points belonging to this high density region Harding et al. BMC Neuroscience (2023) 24:13 Page 12 of 15 consistently observed a dense region of low amplitude the dense concentration of points in the MUA amplitude MUA points which we theorised may be reflecting OFF heatmap previously identified as the likely OFF period periods (Fig. 6C). region (Fig. 6C). In the absence of clear differences in To explore this further, we sought to find the dimen - performance, we randomly selected the Calinsky-Hara- sions of this low amplitude region using Gaussian mix- basz index as our default. ture modelling (GMM). A Gaussian mixture model is a Although smoothing means that the data used for probabilistic model that assumes all the data points are clustering is dependent on surrounding timepoints, generated from a mixture of a finite number of multi - the clustering step itself is independent of time. This variate Gaussian distributions or components. Unlike contrasts with all existing descriptions of OFF peri- k-means clustering, these components and therefore the ods which are understood as a feature observed in a resulting clusters do not need to be spherical in shape. To linear time course of neuronal activity. To recapitulate find the parameters of the Gaussian components which the time domain, we decided to group low amplitude maximize the likelihood of the model given the data, the points into consolidated segments (Fig. 7A–E). First, two-step iterative Expectation–Maximization (EM) algo- we defined a population of time segments with below rithm is employed. In the expectation step, the algorithm average MUA amplitude. This population was identified computes posterior probabilities of component member- by taking all time segments in which the standardised ships for each observation given the current parameter MUA was below zero (Fig. 7C), where the standardised estimates. In the maximization step, the posterior prob- MUA is calculated by taking the absolute values of the abilities from the previous step are used to re-estimate MUA then subtracting the mean of these absolute val- the model parameters by applying maximum likelihood. ues during WAKE epochs. The waking average was cho - These steps are repeated until the change in loglikelihood sen to represent baseline MUA so that it would not be −5 function is less than the tolerance (10 ). Once the fitted dependent on the number and duration of OFF periods GMM has been obtained, new points can be assigned to in the signal. We then isolated those which coincided the component yielding the highest posterior probability with at least one time point belonging to the low ampli- (hard clustering). tude cluster (Fig. 7D). This final population of segments The primary variable that must be input for GMM represents the low amplitude (LA) segments used dur- is the number of components (k) to fit from the data. ing this analysis (Fig. 7E). When using k = 2 components, we unexpectedly found LA segment duration will inevitably depend to some that the solution consistently overestimated the size of degree on where the MUA amplitude threshold is set, the low amplitude component. If this is indeed the data increasing as the threshold is raised. There is a trade- from OFF periods, this could suggest that the varia- off between setting the threshold low enough to reg - tion between types of ON period is greater than varia- ister spiking activity but not so low as to register noise tion between OFF and ON periods. To resolve this, we fluctuations in the MUA signal during neuronal silence. decided to allow k to vary then select the resulting con- We looked at the sensitivity of LA segment duration to figuration that provides the optimal clustering solution. changes in this threshold and found that it is most sta- First, the smoothed MUA signals (L = 62 ms and 22 ms) ble when set at the mean amplitude of the MUA during from a subset of NREM episodes is clustered using wakefulness (Additional file 1: Figure S5), validating our GMMs with k = 1:8 components. Then, the optimal threshold selection. model is selected using a clustering evaluation index. Due to their independent nature, a unique cluster- Clustering evaluation indices are used to assess cluster- ing solution and mean MUA amplitude was generated ing performance when there is no ground truth, as is the for each channel in each animal to detect LA segments. case for binary OFF/ON period alternation. We investi- Where the same channel was measured over multiple gated two such indices: the Calinsky-Harabasz index and days, we reapplied the configuration generated from the the Davies-Bouldin index. The Calinsky-Harabasz index initial day under the assumption that these signals would compares the dispersion within clusters with the disper- be dependent. sion between clusters whereas the Davies-Bouldin index Low amplitude segment extraction pipeline: compares the distance between clusters with the size of the clusters themselves. As expected, we found that the 1. Cluster MUA signal smoothed at two window optimal clustering solution suggested by both methods lengths (62 ms and 22 ms, NREM sleep only) produced more than 2 clusters. The lowest amplitude 2. Detect lowest amplitude component cluster tended to be much smaller than the equivalent 3. Assign smoothed MUA data to low amplitude cluster with a 2-component model and more closely resembled (all states) Har ding et al. BMC Neuroscience (2023) 24:13 Page 13 of 15 WAKE NREM 0.3 mV LFP 0.5 s MUA 0.1 mV 50 μV Standardised MUA (with zero crossings) Standardised MUA (with clustered points) MUA (with LA segments) Fig. 7 LA segment detection pipeline. Example of LA segment detection stages from a 3 s segment of layer 6 baseline day recording. A Local field potential (LFP) trace (0.5–100 Hz filtered). Note the appearance of slow waves following the transition from wake to non‑REM sleep. B Raw decimated multi unit activity (MUA). C Standardised MUA whereby the raw signal is converted to absolute amplitude and then the mean of the absolute MUA during clean wake epochs from this 24 h recording is subtracted. The horizontal black line marks the new mean amplitude (0 μV ). Each contiguous sequence of points below this line has been recoloured alternate shades of pink. These zero crossings represent the population of possible LA segments to be investigated. A black dot demarcates the centre of each zero crossing. D Standardised MUA with low amplitude cluster points recoloured red. E Raw MUA with final LA segments recoloured red. Final LA segments represent zero crossings which intersect with low amplitude cluster points. Black dots represent the centre of each LA segment 4. Find negative zero-crossing half waves across all then calculated using the Hilbert transform. The instan - states (all states) taneous phase angle in the interval [− π,π] for each ele- 5. Re-assign clustered points to negative zero-crossings ment of the complex array was calculated by finding the to find low amplitude segments inverse tangent and converted from radians to degrees. A phase of 0° corresponds to the peak of the oscillation and a phase of 180° to the trough of the oscillation. To meas- ure the temporal coherence of LA segments between Temporal, LFP phase and channel coherence analysis pairs of channels we generated the following statistic: The following temporal parameters of LA segments were Pairwise channel coherence = (intercept(A|B)/ assessed: incidence, average duration and total occu- sum(B) + intercept (B|A)/sum(A)) /2. pancy time per minute of each vigilance state (NREM/ Where A and B are the time points of MUA signal REM/WAKE) calculated from whole recordings or within LA segments for two unique channels. A value of 30 min time bins on each recordings day. 0 denotes a situation in which all LA segments in both For phase analyses, LFP signal was zero-phase distor- channels occur independently and a value of 1 denotes tion filtered in the delta band (0.5–4 Hz) to extract slow a situation in which all LA segments in both channels wave activity. The complex-valued analytic signal was Harding et al. BMC Neuroscience (2023) 24:13 Page 14 of 15 occur simultaneously. To estimate the random chance Supplementary Information coherence between two channels we generated a ran- The online version contains supplementary material available at https:// doi. org/ 10. 1186/ s12868‑ 023‑ 00780‑w. domly shuffled surrogate signal for each channel with an identical distribution of LA and non-LA segments. To Additional file 1: Figure S1. LA segment occupancy (A) and duration (B) achieve this, we fit a normally smoothed Kernel distribu - on baseline day across light period (ZT0‑ZT12). N=7. Mean ± SEM. Figure tion to the distribution of LA and non-LA segment dura- S2. Summary statistics for linear regression of LA incidence (relative to ZT 6) against zeitgeber time for each LA segment duration category. Df = tions for each channel during the first hour of NREM degrees of freedom. Figure S3. All original channel coherence matrices sleep. We then randomly sampled these distributions in (N=7). Figure S4. All surrogate channel coherence matrices (N=7). Figure an alternating fashion to generate a shuffled sequence S5. Sensitivity of LA segment duration to MUA amplitude threshold. Aver‑ age LA segment duration detected in a layer 5 channel from a baseline of LA and non-LA segments one hour in length. Only recording day in a single mouse as a function of MUA amplitude threshold channels 1 to 12 were used in this analysis as in multiple at step 4 in the LA segment detection pipeline where 100% (denote with animals the deepest 4 channels extended into the corpus a vertical dashed line) corresponds to the average MUA amplitude meas‑ ured during wakefulness (blue). As the threshold is raised, LA segment callosum. In addition, due to channel rejection the maxi- duration increases. The first derivative of this curve is plotted on a separate mum interchannel distance (11) was not available for all axis (orange). Note that other than when the threshold is set at such a low animals and was therefore omitted. point as to miss most LA segments, the first derivative is lowest at 100%, suggesting that LA segment duration is least sensitive when the threshold is set to the average MUA amplitude measured during wakefulness. Statistics Acknowledgements Statistical analyses were performed in MATLAB and None. R. Values are reported as mean ± standard error (SEM). The normality assumption of underlying distributions Author contributions CDH, MCCG, LBK and VVV designed the study. CDH analysed the data with was assessed for each factor level by computing a Shap- input from CM and developed the MATLAB GUI. MCK, LBK, CBD and MCCG iro-Wilks test. Unless stated otherwise, significance of conducted the experiments. LBK and MCK performed histology. CDH and effects was tested using one- or two-way repeated meas - VVV wrote the manuscript with input from all authors. All authors read and approved the final manuscript. ure ANOVAs (within-subject factors “Vigilance state [NREM, REM, WAKE]”, “Day [baseline, SD]” and/or Funding “Time”) with animal ID as a factor followed by post-hoc LBK was supported by a Wellcome Trust PhD studentship (203971/Z/16/Z) and by a Mann Senior Scholarship in medical sciences at Hertford College, pairwise t-tests with Bonferroni correction. Circular sta- Oxford. CDH was supported by funding from the Engineering and Physical tistics for phase analysis were performed using the Circ- Sciences Research Council (EPSRC, EP/S515541/1). CBD was supported by Stat toolbox [35]. Non-uniformity of each distribution a Wellcome Trust PhD studentship (109059/Z/15/Z) and a Clarendon Fund Scholarship from the University of Oxford. MCCG was supported by a BBSRC against the von Mises distribution was confirmed using a DTP grant (BB/J014427/1), a Swiss National Science Foundation grant (no. Rayleigh test for circular data. Differences in mean direc - 310030_189110) and a Clarendon Fund Scholarship from the University of tion were tested using a parametric Watson-Williams Oxford. CM was supported by an ESRS Travel Grant for Early Career Researchers undertaking a Short Term fellowship. VVV is supported by Medical Research multi-sample test for equal means with Bonferroni cor- Council (UK) Grant MR/S01134X/1. rection. Statistical significance in all tests was considered as p < 0.05. For box plots, the middle, bottom, and top Availability of data and materials Code for OFF period detection (OFFAD) is available on GitHub (https:// github. lines correspond to the median, bottom, and top quartile, com/ sjoh4 302/ OFFAD). The datasets during and/or analysed during the cur‑ and whiskers to lower and upper extremes minus bottom rent study available from the corresponding author on reasonable request. quartile and top quartile, respectively. Declarations GUI design Ethics approval and consent to participate All experiments were carried out in accordance with the UK Animals (Scientific The LA segment detection algorithm was incorporated Procedures) Act of 1986 and in compliance with the Animal Research: Report‑ into a MATLAB program with a user-friendly GUI that ing In Vivo Experiments (ARRIVE) guidelines. Ethical approval was provided by allows the detection of LA segments in new data sets, the Ethical Review Panel at the University of Oxford. provides a visual representation of the results and gener- Consent for publication ates a range of useful summary statistics (https:// github. Not applicable. com/ sjoh4 302/ OFFAD). Furthermore, this GUI allows Competing interests for post-processing of LA segments, such as removal of The authors declare no competing interests. brief interruptions, to fit individual user expectations of OFF periods. Har ding et al. BMC Neuroscience (2023) 24:13 Page 15 of 15 Author details review of sleep slow wave modulation in humans using non‑invasive Department of Physiology Anatomy and Genetics, Sir Jules Thorn Sleep brain stimulation. Sleep Med Rev. 2021;58: 101438. and Circadian Neuroscience Institute, University of Oxford, Oxford, UK. The 18. McKillop LE, Fisher SP, Cui N, Peirson SN, Foster RG, Wafford KA, et al. Kavli Institute for Nanoscience Discovery, Oxford, UK. Nuffield Depar tment Eec ff ts of aging on cortical neural dynamics and local sleep homeostasis of Clinical Neurosciences, Sir Jules Thorn Sleep and Circadian Neurosci‑ in mice. J Neurosci. 2018;38:3911–28. ence Institute, University of Oxford, Oxford, UK. Institute for Neuroscience, 19. Krone LB, Yamagata T, Blanco‑Duque C, Guillaumin MCC, Kahn MC, van Department of Health Sciences and Technology, ETH Zürich, Schwerzenbach, der Vinne V, et al. A role for the cortex in sleep‑ wake regulation. Nat Switzerland. University Hospital of Psychiatry and Psychotherapy, University Neurosci. 2021;24:1210. of Bern, Bern, Switzerland. Centre for Experimental Neurology, University 20. Kahn M, Krone LB, Blanco‑Duque C, Guillaumin MCC, Mann EO, Vyazovs‑ of Bern, Bern, Switzerland. Picower Institute for Learning and Memory, Mas‑ kiy VV, et al. Neuronal‑spiking‑based closed‑loop stimulation during corti‑ sachusetts Institute of Technology, Cambridge, USA. Department of Brain cal ON‑ and OFF‑states in freely moving mice. J Sleep Res. 2022. https:// and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, doi. org/ 10. 1111/ JSR. 13603. USA. Translational Research Center, University Hospital of Psychiatry and Psy‑ 21. Karalis N, Sirota A. Breathing coordinates cortico‑hippocampal dynamics chotherapy, University of Bern, Bern, Switzerland. Private Clinic Meiringen, in mice during offline states. Nat Commun. 2022;13:1–20. Meiringen, Switzerland. 22. Dasilva M, Camassa A, Navarro‑ Guzman A, Pazienti A, Perez‑Mendez L, Zamora‑López G, et al. Modulation of cortical slow oscillations and Received: 30 November 2022 Accepted: 27 January 2023 complexity across anesthesia levels. Neuroimage. 2021;224:117415. 23. de Bonis G, Dasilva M, Pazienti A, Sanchez‑ Vives MV, Mattia M, Paolucci PS. Analysis pipeline for extracting features of cortical slow oscillations. Front Syst Neurosci. 2019;13:70. 24. Jercog D, Roxin A, Barthó P, Luczak A, Compte A, de La Rocha J. UP‑ DOWN cortical dynamics reflect state transitions in a bistable network. References Elife. 2017. https:// doi. org/ 10. 7554/ eLife. 22425. 1. Krueger JM, Frank MG, Wisor JP, Roy S. Sleep function: toward elucidating 25. Seamari Y, Narváez JA, Vico FJ, Lobo D, Sanchez‑ Vives MV. Robust Off‑and an enigma. Sleep Med Rev. 2016;28:46–54. online separation of intracellularly recorded up and down cortical states. 2. Achermann P, Borbély AA. Sleep Homeostasis and Models of Sleep Regu‑ PLoS One. 2007;2:888. lation. In: Principles and Practice of Sleep Medicine. Fifth Edition. Elsevier 26. Mcfarland JM, Hahn TTG, Mehta MR. Explicit‑ duration hidden markov Inc; 2011. p. 431–44. model inference of UP‑DOWN states from continuous signals. PLoS ONE. 3. Huber R, Ghilardi MF, Massimini M, Tononi G. Local sleep and learning. 2011;6:21606. Nature. 2004;430:78–81. 27. Funk CM, Honjoh S, Rodriguez AV, Cirelli C, Tononi G. Local slow waves 4. Steriade M, Dossi RC, Nunez A. Network modulation of a slow intrinsic in superficial layers of primary cortical areas during REM sleep. Curr Biol. oscillation of cat thalamocortical neurons implicated in sleep delta 2016;26:396–403. waves: cortically induced synchronization and brainstem cholinergic sup‑ 28. Vyazovskiy VV, Olcese U, Hanlon EC, Nir Y, Cirelli C, Tononi G. Local sleep in pression. J Neurosci. 1991;11:3200–17. awake rats. Nature. 2011;472:443. 5. Vyazovskiy VV, Olcese U, Lazimy YM, Faraguna U, Esser SK, Williams JC, 29. Nir Y, Staba RJ, Andrillon T, Vyazovskiy VV, Cirelli C, Fried I, et al. Regional et al. Cortical firing and sleep homeostasis. Neuron. 2009;63:865–78. slow waves and spindles in human sleep. Neuron. 2011;70:153–69. 6. Vyazovskiy VV, Harris KD. Sleep and the single neuron: The role of global 30. Fisher SP, Cui N, McKillop LE, Gemignani J, Bannerman DM, Oliver PL, et al. slow oscillations in individual cell rest. Nat Rev Neurosci. 2013;14:443–51. Stereotypic wheel running decreases cortical activity in mice. Nat Com‑ 7. Suberbielle E, Sanchez PE, Kravitz AV, Wang X, Ho K, Eilertson K, et al. mun. 2016. https:// doi. org/ 10. 1038/ ncomm s13138. Physiological brain activity causes DNA double strand breaks in neu‑ 31. Dash MB. Infraslow coordination of slow wave activity through altered rons—exacerbation by amyloid‑β. Nat Neurosci. 2013;16:613. neuronal synchrony. Sleep. 2019. https:// doi. org/ 10. 1093/ sleep/ zsz170. 8. Naidoo N, Giang W, Galante RJ, Pack AI. Sleep deprivation induces the 32. Vyazovskiy VV, Riedner BA, Cirelli C, Tononi G. Sleep homeostasis and cor‑ unfolded protein response in mouse cerebral cortex. J Neurochem. tical synchronization: II. A local field potential study of sleep slow waves 2005;92:1150–7. in the rat. Sleep. 2007;30:1631–42. 9. Bellesi M, Bushey D, Chini M, Tononi G, Cirelli C. Contribution of sleep to 33. Achermann P, Borbély AA. Mathematical models of sleep regulation. the repair of neuronal DNA double‑strand breaks: evidence from flies and Front Biosci. 2003;8:s683. mice. Sci Rep. 2016;6:1–13. 34. Vyazovskiy VV, Deboer T, Rudy B, Lau D, Borbély AA, Tobler I. Sleep EEG 10. McKillop LE, Fisher SP, Milinski L, Krone LB, Vyazovskiy VV. Diazepam in mice that are deficient in the potassium channel subunit K.v.3.2. Brain effects on local cortical neural activity during sleep in mice. Biochem Res. 2002;947:204–11. Pharmacol. 2021;191:114515. 35. Berens P. CircStat: a MATLAB toolbox for circular statistics. J Stat Softw. 11. Mensen A, Riedner B, Tononi G. Optimizing detection and analysis of slow 2009;31:1–21. waves in sleep EEG. J Neurosci Methods. 2016;274:1–12. 12. Lajnef T, Chaibi S, Eichenlaub JB, Ruby PM, Aguera PE, Samet M, et al. Sleep spindle and K‑ complex detection using tunable Q‑factor wavelet Publisher’s Note transform and morphological component analysis. Front Hum Neurosci. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ 2015;9:1–17. lished maps and institutional affiliations. 13. Mukovski M, Chauvette S, Timofeev I, Volgushev M. Detection of active and silent states in neocortical neurons from the field potential signal during slow‑ wave sleep. Cereb Cortex. 2007;17:400–14. 14. Bukhtiyarova O, Soltani S, Chauvette S, Timofeev I. Supervised semi‑auto ‑ matic detection of slow waves in non‑anaesthetized mice with the use of neural network approach. Transl Brain Rhythm. 2016. https:// doi. org/ 10. 15761/ TBR. 10001 04. 15. Bukhtiyarova O, Soltani S, Chauvette S, Timofeev I. Slow wave detection in sleeping mice: comparison of traditional and machine learning methods. J Neurosci Methods. 2019;316:35–45. 16. Geiser T, Hertenstein E, Fehér K, Maier JG, Schneider CL, Züst MA, et al. Targeting arousal and sleep through noninvasive brain stimulation to improve mental health. Neuropsychobiology. 2020;79:284–92. 17. Fehér KD, Wunderlin M, Maier JG, Hertenstein E, Schneider CL, Mikutta C, et al. Shaping the slow waves of sleep: a systematic and integrative
Journal
BMC Neuroscience – Springer Journals
Published: Feb 21, 2023
Keywords: Sleep; ON/OFF periods; Homeostasis