Author Correction: Neuronal firing rates diverge during REM and homogenize during non-REM

An amendment to this paper has been published and can be accessed via a link at the top of the paper.


Introduction
Firing rates vary among neurons and across time. The dynamic range of a neuron's firing is determined by a combination of membrane geometry, distribution and subtypes of ion channels, and synaptic efficacy (Koulakov et al., 2009;Yassin et al., 2010;Roxin et al., 2011;Lim et al., 2015;Stuart and Spruston, 2015;Nigam et al., 2016). Changes in these properties can potentially alter a neuron's gain function or "excitability," altering the neuron's encoding properties (Cheng and Frank, 2008;Lee et al., 2012). Recent evidence suggests that a neuron's firing rate is also homeostatically regulated (Vyazovskiy et al., 2009;Hengen et al., 2013;Miyawaki and Diba, 2016;Watson et al., 2016), and that modifications in membranes and synapses can work to maintain the neuron's dynamic range (Marder and Goaillard, 2006;Turrigiano, 2011). Several studies from different labs indicate that these modifications are at least partially state-dependent; the emerging picture is that firing rates of neurons increase during waking (Vyazovskiy et al., 2009;Hengen et al., 2016;Miyawaki and Diba, 2016;Watson et al., 2016) and decrease during sleep (Vyazovskiy et al., 2009;Grosmark et al., 2012;Miyawaki and Diba, 2016;Watson et al., 2016), in a perpetual dance around a dynamic range.
The various waking and sleep states feature different activity levels of the neuromodulatory systems, which contribute uniquely to the excitability of neuronal circuits, network firing patterns, and the plasticity of their synapses (Hobson and Pace-Schott, 2002;Brown et al., 2012).
For example, REM is characterized by high acetylcholine and low noradrenaline, serotonin and histamine levels, while waking and NREM respectively feature high and low levels of these neuromodulators (Hobson and Pace-Schott, 2002;Brown et al., 2012). Unique brainstem and thalamocortical networks are also active within each state, producing state-specific oscillatory firing patterns (Saper et al., 2010;Brown et al., 2012;Weber and Dan, 2016). The differing neuromodulatory and network backgrounds lead to different overall firing rates in REM, NREM, and waking (Vyazovskiy et al., 2009;Miyawaki and Diba, 2016;Watson et al., 2016), but averaging can also mask significant variations within each state (Grosmark et al., 2012;Miyawaki and Diba, 2016;Watson et al., 2016).
It was recently shown that sleep yields a net decrease in the firing rates of both hippocampal (Miyawaki and Diba, 2016) and frontal cortex neurons (Watson et al., 2016). These changes were likely explained by synaptic downscaling (Tononi and Cirelli, 2014), triggered in the hippocampus by sharp-wave ripples and sleep spindles during NREM sleep, and incorporated over the course of REM sleep (Miyawaki and Diba, 2016) and in the neocortex, triggered by alternating cycles of UP/DOWN states (Bartram et al., 2017;Gulati et al., 2017). In Miyawaki and Diba (2016) and Watson et al. (2016) , we took trouble to evaluate firing rate changes between different epochs of the same state (e.g. NREM i and NREM i+1 epochs in sleep) to avoid confounds of state-dependent neuromodulation. However, some questions remain regarding how firing patterns of neurons of differing excitabilities change within each of these states and on transitions between these states, and how these compare between hippocampal and neocortical neurons. In particular, low and high firing neurons, with presumed low and high levels of excitability, are expected to be affected differently by activity-driven homeostasis and bear differing levels of plasticity (Koulakov et al., 2009;Hengen et al., 2013;Lim et al., 2015;Grosmark and Buzsaki, 2016). While this question was addressed to some extent in our previous work, understanding such effects is complicated by regression to the mean (RTM), for which the null hypothesis allows that firing rates of low-firing neurons should increase and those of highfiring neurons should decrease across any two comparative periods. In this report, we investigate changes in firing rates of neurons within different stages of sleep and the effects of transitions between sleep stages in both hippocampus and frontal cortex, while carefully controlling for RTM. We find that transitions to REM and NREM sleep states differentially affect low-firing and high-firing neurons in each state. In both hippocampus and frontal cortex, we find that REM sleep is marked by inhibition, and the spread between low-and high-firing neurons increases, while NREM results in a more homogenized and narrowed distribution of rates. These observations may help provide insights into the function and effects of sleep states on cortical networks of neurons.

Materials and Methods
We re-analyzed data previously recorded from hippocampal CA1 region of four male rats (Miyawaki and Diba, 2016;Miyawaki et al., 2017) and frontal cortex of 11 male rats (Watson et al., 2016). Details of the experimental protocols, including animals, surgery, electrophysiological recoding, spike detection and clustering, and sleep detection can be found in these refs (Miyawaki and Diba, 2016;Watson et al., 2016;Miyawaki et al., 2017) and are summarized below. EMG was obtained from either the nuchal muscles or from correlated high-frequency (300 -600 Hz) signals from brain electrodes (Schomburg et al., 2014;Watson et al., 2016  since NREM epochs are generally longer. Additionally, cells were sorted into quintiles within each epoch and firing rates of cells in each quintile were calculated in each bin. Importantly, sorting was based on their mean firing rates over the entirety of the windows of interests depicted in each panel, to avoid RTM effects which are particularly evident when tracking ranked groups of units such as quintiles. While ranking was based on the entire epoch in these analyses, to allow for comparisons across quintiles, firing rates were normalized by the mean firing rate in the last one-third of the first state. Change index and deflection index. For a second set of analyses, change index (CI) for a quintile was defined as  Fig 6F), control periods were randomly selected from 1-min periods of the same wake epoch (since wake periods separated by sleep display significantly different firing rates (Vyazovskiy et al., 2009;Miyawaki and Diba, 2016)). Shuffled mean and 95% confidence intervals of CI were obtained from this surrogate data. The deflection index (DI) was defined as difference of CI from the surrogate mean.

Simulations.
To better understand the behavior of CI and DI we generated three random datasets involving noise combined with no-change, additive firing rate increase, and multiplicative firing rate increase. Each dataset has 5000 cells and 3 epochs, corresponding to epochs i, j, and k. To mimic the variability of real data, first we set a baseline firing rate for each cell based on a lognormal distribution obtained from hippocampal pyramidal cells during NREM (mean = -0.4726 log 10 Hz, std= 0.4577 log 10 Hz) and then added random ("multiplicative") noise proportional to each cell's firing rate in each epoch (std=0.1403 log 10 Hz). For the no-change simulation, each cell kept the same baseline firing rates across epochs with only random noise producing fluctuation across epochs. For additive and multiplicative increase simulations, baseline firing rates in epoch j were increased (by addition of 0.05 Hz or multiplication by 1.1 for additive and multiplicative increases, respectively).

Experimental Design and Statistical analyses.
In this work we analyzed previously obtained data and no additional experiments were performed. Diversity of firing rates was evaluated by coefficient of variance (CV) and significance of difference was tested with Wilcoxon rank sum test. P-values of DIs were calculated relative to shuffled surrogates. Differences in DI and firing rate changes among quintiles were tested with one-way ANOVA. Firing rate distributions were compared by a Kolmogorov-Smirnov test. All analyses were performed with custom-written scripts running on MATLAB with statistics and machine learning tool boxes. Code is available upon request.

Differential effects of REM and NREM on higher-and lower-firing rate hippocampal neurons
We previously recorded from populations of CA1 pyramidal cells and interneurons over multiple sleep and awake cycles (Miyawaki and Diba, 2016). Based on these data, we showed that mean firing rates in the hippocampal pyramidal cells increased within NREM but decreased through transitions between NREM and REM, and such zig-zag change resulted in net decrease across sleep (Grosmark et al., 2012;Miyawaki and Diba, 2016). However, it was not clear signed-rank test), consistent with a more competition-driven network (Xie et al., 2002). Within REM, firing rate changes were similar across quintiles and the CV did not change significantly

Differential effects of REM and NREM on higher-and lower-firing rate neocortical neurons
To examine whether these or similar state effects are also present in the neocortex, we extended these same analyses to neuronal spiking data recorded from frontal cortex of rats (Watson et al., 2016) and available on crcns.org (Fig. 2). Unlike in the hippocampus, firing rates increased at the transition from NREM to REM (Evarts, 1964;McCarley and Hobson, 1971;Vyazovskiy et al., 2009;Renouard et al., 2015; but also see Niethard et al., 2016

Regression-to-the-mean and sorting effects on firing rate changes
We wanted to better quantify these observations and to further evaluate how distributions of firing rates change within and across different sleep states. However, we first needed to better understand the relationship between variability and RTM in a population of neurons with lognormally distributed firing rates; ordering based on a part of analyzed data may bias the results due to RTM (Fig. 3A). We examined a simulated population of neurons where the source of variability is "multiplicative noise" proportional to each neuron's firing rate (also see Materials and Methods) (Fig. 3B). Despite the absence of any change in our model population, lower-firing neurons show an apparent increased firing, while higher-firing neurons show an apparent decreased firing (Fig. 3C) when the quintiles were based on rank-ordering during the first epoch (i) of a sequence i-j. This is RTM and it can confound evaluations of true effects (e.g. from sleep). To control for RTM, we need to either rank-order cells by their mean firing rates over the entire sequence, as we did for analyses in Fig. 1-2, or else instate an appropriate correction. We introduced a shuffle correction in which we randomly flipped indices for epochs of the same state (e.g. i and k for a NREM i /REM j /NREM k ) and repeated the analysis multiple times to obtain a surrogate distribution for the change index of quintiles (Miyawaki and Diba, 2016). This surrogate data provided us with valuable "control" shuffle means and confidence intervals for each quintile. We defined the "deflection index (DI)" as the difference between the observed change index (CI) and the surrogate mean within each quintile. These DIs were not significantly different from zero when changes were due only to noise (Fig. 3C).
We then examined DIs under two scenarios with a simulated change in addition to noise: when firing rates increased across the population, either additively by a fixed amount for all cells

Firing rate spread in REM and homogenization in NREM
We next used the RTM correction methods described above to enable comparisons of firing rates in pairs of epochs. We did these analyses either within each state of sleep or across different stages of sleep. This approach is complementary to that shown in Figures 1 and 2 and can serve as an independent verification of the observations shown there.
During NREM sleep, the average firing rates of hippocampal pyramidal neurons increased (mean CI = 0.058 ± 0.005, p=5.9×10 -27 , Wilcoxon signed-rank test). While the CI appeared to show the largest increase in low-firing cells, much of this apparent effect was due to RTM: in the shuffle corrected DIs, in fact the higher-firing quintiles showed the greatest relative increases steady state in the balance between network excitation and inhibition (but see Tsodyks et al., 1996). This increased inhibitory activity could potentially drive some of the decrease firing in pyramidal neurons (Niethard et al., 2016) and allow for a winner-take-all mechanism whereby some high-firing cells dominate REM dynamics at the expense of lower-firing cells. These dynamics were somewhat different for the frontal cortex, however; at the onset of REM, principal neurons in the frontal cortex increased firing across quintiles, while interneurons showed little change (DI = 0.094 ± 0.029, 0.187 ± 0.023, 0.181 ± 0.019, 0.188 ± 0.020 and 0.162 ± 0.016 for each quintile of principal neurons, p values < 0.001 relative to shuffles, CI = 0.031 ± 0.033 for interneurons, p=0.03, Wilcoxon signed-rank test). It is interesting to note however that the increased firing of hippocampal interneurons mirrored the overall increase in neocortical principal cell activity, consistent with neocortical control of hippocampal inhibition (Hahn et al., 2006;Basu et al., 2016). As a result of these changes, firing rate distributions became wider upon REM in both the hippocampus and the frontal cortex ( Fig. 4B, right), consistent with our earlier analysis. Over the course of REM, we saw decreased firing across quintiles and interneurons in the hippocampus and in some quintiles of the neocortex (Fig. 4C). However, comparisons of the overall firing rate distributions did not reach statistical significance in the frontal cortex. The overall balance between excitation and inhibition therefore did not appear to change significantly within the course of REM states (Dehghani et al., 2016).
In contrast, when REM transitioned to NREM sleep, lower-firing quintiles showed increased firing while higher-firing quintiles and interneurons showed a firing decrease both in the hippocampus and in the frontal cortex (Fig. 4D). Interestingly, among the various dynamics we investigated only this transition from REM to NREM was marked by a renormalizing effect on firing rates across quintiles even after correction for RTM, and it was the only one we investigated that was marked by decreased firing of inhibitory cells in the hippocampus. NREM sleep therefore provided for the most uniform firing among the population of cells, potentially because of lower effective inhibition, whereas REM was marked by a widened distribution of firing activity.

Neuronal firing changes at transitions to and from wake
Our results thus far have outlined the effects of transitions and continuation of REM and NREM sleep states on neurons at different levels of excitability. We next applied these same methods to analyze the effects of transitions between sleep and waking on different quintiles and focused on our corrected DI analysis. Immediately upon transitions from waking to NREM sleep (direct transitions from wake to REM are rarely observed), the hippocampus showed increases in the middle of the distribution (Fig. 5A) whereas the frontal cortex showed a decrease in highfiring cells. Nevertheless, the wake-to-sleep transition was accompanied by decreased inhibition in the hippocampus (CI = -0.076 ± 0.016, p=1.5×10 -6 , Wilcoxon signed-rank test) but not in the frontal cortex (-0.020 ± 0.032, p =0.60, Wilcoxon signed-rank test) and a narrowing of the distribution of firing rates in both regions (ΔCV = -0.708 ± 0.076, p = 2.4×10 -11 , and Δ CV = -0.239 ± 0.035, p = 6.1×10 -7 ) for the hippocampus and the frontal cortex, respectively, Wilcoxon signed-rank test). The distribution narrowing in the hippocampus again indicates a new steady state in the balance between excitation and inhibition, with increased activity in the three middle quintiles (Fig. 5A), whereas the frontal cortex narrowing was a result of decreased firing in the highest-firing quintile and a trend towards more increase in progressively lower firing cells.
In contrast, the distribution of firing rates widened at the onset of wake. At transitions from NREM to wake (Fig. 5B), hippocampal firing decreased significantly particularly among the lowest firing quintiles. These changes resulted in a leftward shift in the firing rate distribution (p = 2.9×10 -18 , Kolmogorov-Smirnov test) and an increase in the CV (ΔCV=0.915 ± 0.137, p = 1.9×10 -7 , Wilcoxon signed-rank test). In the neocortex, on the other hand, higher firing principle neurons increased firing at the transitions from NREM to wake, essentially reversing the change from wake to NREM (Fig. 5A), and producing a significant increase in the CV of the distributions (ΔCV=0.243 ± 0.082, p = 0.019, Wilcoxon signed-rank test). The transitions from REM to wake showed slightly different effects across quintiles (Fig. 5C). In sum, wake and sleep have contrasting effects on the activity of neurons in different quintiles, with sleep states displaying a more homogeneous distribution of firing rates and greater variation among the population during wake.

Lasting effects of sleep and sleep states on firing rate distributions
These analyses describe a perpetually fluctuating pattern of neuronal activity across sleep and wake transitions, with alternating narrowing and widening of firing rate distributions. We next asked which of these effects persists across longer sleep sequences composed of multiple NREM and REM episodes. First, we analyzed state triplets composed of NREM i -REM-NREM i+1 or REM i -NREM-REM i+1 (Fig. 6 A-D) (Grosmark et al., 2012;Miyawaki and Diba, 2016;Watson et al., 2016). Time normalized firing rates and CVs in the triplets further illustrated and confirmed the distribution narrowing and widening effects of NREM and REM epochs, respectively ( Fig. 6 A, B), in both brain regions. However, based on these plots it appeared that these distribution changes largely offset and cancelled one another. To better quantify these impressions, we again calculated DIs for quintiles in both regions and compared firing rate distributions and CVs. All hippocampal quintiles showed decreased firing between consecutive NREMs interleaved by REM ( Fig. 6C; note also that these decreases were more uniform across quintiles than those reported in Miyawaki and Diba (2016)  triplets, significant changes were not detected in DIs, distributions, or CVs from either region.
Extending these analyses to first and last NREM in the sleep (separated by longer sequences of alternating REM and NREM), we observed significant decreases across pyramidal cell quintiles in the hippocampus (Fig. 6E). In the frontal cortex, DIs were significantly negative only in the middle and highest firing quintiles. These indicate overall firing rate decreases resulting from sleep, consistent with previous reports (Vyazovskiy et al., 2009;Grosmark et al., 2012;Miyawaki and Diba, 2016). Interestingly, the lowest firing quintile decreased most in the hippocampus while the highest firing quintiles decreased most in the frontal cortex. But while firing rate distributions were slightly shifted leftward in the hippocampus (p = 6.1×10 -6 , Kolmogorov-Smirnov test), the difference did not reach significance in the frontal cortex (p = 0.22, Kolmogorov-Smirnov test). Importantly, pairwise comparisons of the CVs did not detect significant changes either in variability in either the hippocampus or the neocortex (ΔCV = 0.039 ± 0.018 (p = 0.093) for hippocampus and -0.010 ± 0.019, (p= 0.731) for frontal cortex, Wilcoxon signed-rank test). These results therefore indicate that distribution changes through multiple sequential REM and NREM states, alternately dispersing and homogenizing firing rates, were counter-balanced throughout sleep in both the hippocampus and the neocortex, despite excitability decreases in both regions in both the population as a whole and in specific quintiles.
Lastly, we compared the last minute of wake before sleep to the first minute of wake following sleep (Fig. 6F). In the hippocampus, DIs were significantly negative across quintiles, with lower firing quintiles were more negative than for higher firing quintiles, and firing rate distributions and CVs were significantly different (p = 0.007, Kolmogorov-Smirnov test, and Δ CV= 0.36 ± 0.10, p = 2.3×10 -4 , Wilcoxon signed-rank test). On the other hand, principal neurons in the neocortex showed a significant increase in the lowest firing quintile and a significant decrease in the highest firing quintile, consistent with a narrowed distributed.
However, neither firing rate distributions nor CVs were found to be significantly different across sleep (p = 0.97, Kolmogorov-Smirnov test, and Δ CV= -0.044 ± 0.052, p = 0.71, Wilcoxon signed-rank test). These results contradict our expectations based on previous analyses (Watson et al., 2016), which we will address in the Discussion section.

Discussion
In this work, we aimed to understand how low and high firing hippocampal and frontal cortex neurons are affected during REM and NREM stages of sleep, and upon transitions between these states, to arrive at a better understanding of the function(s) of sleep states in mammals. Since many of our analyses depended on rank-ordering of firing rates and because low-firing and highfiring neurons regress to the mean by chance alone, in this study we designed our analyses to either prevent or correct for this effect. We used two methods; either we based the ordering on the mean firing rates over the entire period being considered, or else we measured all changes relative to a surrogate distribution obtained by random shuffles of the real data. These steps were necessary, because ordering in any selected period produces illusionary normalization in a complementary period and any real changes must be evaluated in contrast to these RTM effects.
We found that, in general, sleep states and state transitions do not affect neurons uniformly, but that the changes depend on both the brain region and the relative activity of cells, which likely reflect a combination of neuromodulation of membrane excitability (Graves et al., 2012;Nadim and Bucher, 2014) and sleep-dependent network dynamics involving excitatory and inhibitory synaptic inputs to neurons Timofeev et al., 2001;Dehghani et al., 2016;Niethard et al., 2016;Stringer et al., 2016).
Among the different state dynamics we investigated, NREM sleep was notable in homogenizing excitability across neurons. The transition from REM to NREM produced a greater relative decrease in high-firing cells in both hippocampus and neocortex, with an increase activity of low-firing cells in the hippocampus and a relatively smaller decrease in the neocortex.
These changes at the onset of NREM serve to partially homogenize firing across both populations. In the frontal cortex, normalization continued during the NREM episode, and in both regions the coefficient of variation decreased at the onset and further throughout NREM.
The onset of NREM was also marked by decreased firing in interneurons in the hippocampus, indicating a shift in the excitation/inhibition balance (see also Stringer et al., 2016). These dynamics across two states characterized by a major shift in cholinergic tone are consistent with the greater relative effect of muscarine on several classes of inhibitory cortical interneurons (Kuchibhotla et al., 2016). Interestingly, atropine, a muscarinic acetylcholine antagonist, also produces increased bursting in hippocampal CA1 pyramidal neurons of lower excitability ("regular spiking") but decreased bursting in higher excitability ("bursting") cells (Graves et al., 2012). This suggests that the decreased levels of neuromodulators along with the release from active inhibition allow for a rebalancing of pyramidal cell excitability during NREM sleep.
In contrast, the NREM to REM transition led to greater interneuron spiking and a greater separation of firing between low-firing and high-firing cells, increasing the CV in both regions.
These winner-take-all type changes may be implemented in a recurrently connected circuit endowed with inhibition (Yuille and Geiger, 1995;Lee et al., 1999;Rutishauser et al., 2011), such as region CA3, one synapse upstream from our CA1 recordings, or in layer 4 of the neocortex. The shift towards further competition may be supported by increased cholinergic levels during REM sleep that favor feedforward connections, such as from entorhinal cortex to region CA1 (Mizuseki et al., 2011;Schomburg et al., 2014), while neuromodulatory tone in NREM instead favors recurrently-generated activity (Hasselmo, 2006). It is also worth noting that hippocampal interneuron firing patterns across different sleep states closely mirrored those of cortical principal neurons (e.g. see Fig 6A,B), consistent with neocortical control of hippocampal inhibition (Hahn et al., 2006). We also noticed that principal neurons showed relatively dramatic changes (e.g. see Fig. 6A, B) at the transitions between NREM and REM.
These transition points may have unique properties: the transitionary period from NREM to REM sleep may in fact be a unique period of "intermediate sleep" that is inundated with both thalamacortical sleep spindles and theta oscillations (Gottesmann, 1992;Watson et al., 2016), while the transitions from REM to NREM are often followed by LOW states and microarousals (Miyawaki et al., 2017).
The net effects of these state transitions, from the first to the last NREM epochs during extended sleep sequences were mostly consistent with our previous reports (Miyawaki and Diba, 2016;Watson et al., 2016), with some notable differences. Here, we find that distribution of the firing rates spread during REM and the homogenization during NREM largely cancel out in both hippocampus and neocortex, yielding a net effect of decreased firing rates in both regions over sleep (Cirelli, 2017). These decreases were seen across all hippocampal quintiles over sleep, but preferentially in lower-firing neurons (Miyawaki and Diba, 2016). In the neocortex, decreases were less pronounced and were specific to high-firing cells, whereas Watson et al. (2016) reported an additional parallel increase in firing of lower-firing neurons. This discrepancy between the present study and Watson et al. (2016)  We and others have conjectured that the slower firing rate decreases over sleep, on the other hand, are produced by the downscaling of synaptic connections (Vyazovskiy et al., 2009;Grosmark et al., 2012;Tononi and Cirelli, 2014;Miyawaki and Diba, 2016). Network modeling also supports the notion of a strong link between the strength a neuron's connectivity and its firing rate (Olcese et al., 2010;Lim et al., 2015). Hippocampal changes across sleep (Fig. 6E,F) are consistent with an additive change (e.g. Fig 3C), which indicates hippocampal firing decreased by a similar amount across cells. If the conjecture between synaptic connection and firing rate is correct, the uniform decrease of firing rates could imply a uniform weakening of synaptic connections which effectively improves signal-to-noise in higher-firing cells (Tononi and Cirelli, 2014; see also Cirelli, 2017). A recent study employing scanning electron microscopy of synaptic connections in the cortex supports this analysis; following sleep but not waking, smaller axonal-spine interfaces were observed in the four lower quintiles, with a lesser or no effect in the highest quintile (de Vivo et al., 2017). Higher-firing neurons appear to show the least plasticity, perhaps as a consequence of rigidity or saturated synapses (Grosmark and Buzsaki, 2016). These distinctions may also reflect differences in neuronal subtypes within the CA1 pyramidal layer (Mizuseki et al., 2011;Danielson et al., 2016) that exist throughout the cortex (Molyneaux et al., 2007), though surprisingly in the frontal cortex we saw the greatest decrease in firing across sleep in the higher-firing quintile (but see discussion points below).
Overall, these observations demonstrate a remarkable degree of agreement about the effects of wake and sleep states on neuronal firing in the hippocampus and the frontal neocortex. However, some inter-regional differences were also evident in the responses of quintiles, particularly during the course of NREM episodes and at the transition from NREM to REM. A possible source of differences in hippocampal versus cortical profiles is that the cortex has DOWN states-periods of temporary network silence during NREM-which are not as clearly defined in the hippocampus (Steriade et al., 1993;Isomura et al., 2006). The predominance of DOWN states can potentially account for the relatively decreased firing activity in the neocortex during NREM (see also Vyazovskiy et al., 2009), particularly in the highest firing rate groups as slow waves in NREM develop, and the strong rebound in firing in these quintiles at the onset of REM sleep. On the other hand, LOW states and microarousals at the onset of NREM seem to have stronger suppressive effects on the firing of hippocampal neurons (Miyawaki et al., 2017). These apparent inter-regional differences may also arise because recordings and unit and state detection were performed by different experimenters in different labs. It is worth noting that overall firing rates were higher for the frontal cortex recordings than for the hippocampus recordings, so that lowest quintiles in the frontal cortex fire at similar rates to the middle quintiles in the hippocampus. Hence, sleep states may have effects that depend on absolute rather than relative firing rates and the normalizing and dispersing effects of NREM and REM sleep, respectively, represent broad effects of the neuromodulatory tones under different brain states. for each analyzed NREM-REM doublet prior to averaging for presentation (see Table 1

Number of cells and states/transitions
Numbers for time normalized analyses (Fig. 1, 2 and 6A,B) and for CI/DI analyses (Fig. 4