Long-Lasting Novelty-Induced Neuronal Reverberation during Slow-Wave Sleep in Multiple Forebrain Areas

1Sleep is important for the consolidation of newly acquired memories. The discovery of experience-dependent neuronal reactivation during sleep corroborated the notion that novel memory traces, after successful encoding, must be replayed in their supporting neuronal networks until synaptic plasticity can effect trace consolidation. Postacquisition neuronal reactivation during sleep or quiet waking (WK) was found to preserve the temporal relationships of alert, exploratory WK in the hippocampus (HP) and the cerebral cortex (CX), causing a correlated replay of activity patterns across two-neuron or many-neuron ensembles. To date, experience-dependent brain reactivation during sleep has been observed in rodents, nonhuman primates, humans, and even songbirds, pointing to a very general biological phenomenon. Importantly, postacquisition brain reactivation during sleep has been shown to be proportional to memory acquisition in rats and humans.

2In spite of the positive evidence, the brain reactivation hypothesis for memory consolidation during sleep faces several objections. First, the neocortical reactivation detected to date is extremely subtle and decays rapidly within less than 1 h of memory trace formation. Such transient reactivation falls short of explaining the disruption of memory traces by sleep deprivation several hours and even days after initial acquisition. Second, strictu sensu neuronal reactivation during sleep in mammals has only been investigated in the hippocampocortical loop, making it difficult to ascertain whether the phenomenon is particular to this neural circuit or whether it represents global experience-dependent changes in the brain. Third, brain reactivation has mostly been observed in highly trained animal subjects, raising skepticism about its relevance for the acquisition and consolidation of novel information. Finally, experience-dependent brain reactivation has been reported to occur in all behavioral states, including WK. Although the first finding in this regard has hinted at a possible predominance of reactivation during slow-wave (SW) sleep, a comprehensive comparison of the relative contributions of WK, SW, and rapid eye-movement (REM) sleep for brain reactivation is still missing. To further complicate the issue, recent studies have raised the possibility that neuronal processing may occur at either slower or faster speed than normal physiological rates during REM and SW sleep, respectively. Thus, it is uncertain at the moment how brain reactivation relates to different behavioral states.

3In order to address these objections, we set out to investigate the effects of a transient novel tactile experience on the long-term evolution of ongoing brain activity across the major behavioral states of the rat. In each of the five animals studied, extracellular activity of 59–159 neurons per animal and local field potentials (LFPs) representing larger-scale neural rhythms were simultaneously recorded from four different brain regions: HP, primary somatosensory “barrel field” CX, ventral posteromedial thalamic nucleus (TH), and putamen (PU) (Figure 1A and 1B; see also Figures S6 and S7). These brain regions were chosen because they comprise three major forebrain circuit loops essential for rodent species-specific behaviors. Rats are nocturnal gatherers that exhibit a variety of exploratory behaviors during the night, sleeping intermittently and mostly during the day (upper panel in Figure 1C). In the wild, rats rely on spatial navigation and superb whisker-based tactile discrimination to explore new territories in search of food (Nowak 1999). The corticothalamic, corticohippocampal, and corticostriatal loops probed in this study have been implicated in tactile information processing, spatial navigation and memory formation, and the execution of complex motor sequences.

4In our studies, neural signals were continuously recorded across the natural sleep–wake cycle for 48–96 h, with a single 1-h exposure to four complex objects placed in the four corners of the recording box (Figure 1C). All objects were strictly novel to the subjects and were designed to maximize shape, texture, and behavioral value differences (Figure S1). Objects were presented half-way through the recording time (01:16 a.m. ± 00:51, mean ± SD) when lights were off and WK peaked (see Figure 1C), so as to maximize the drive for whisker-based tactile exploration of the environment. The experiment, therefore, consisted of a naturalistic behavioral paradigm involving multiple novel sensory and spatial cues; it was designed to maximize changes induced by exposure to novel objects, as opposed to changes related to repeated behavioral training. As expected, this paradigm increased WK relative to sleep during the exposure time (Figure S2), leading to novel complex sensory stimulation (CSS). Other than novel stimulation and the periodic removal of waste and introduction of food pellets and water, animals were kept undisturbed in the same environment throughout the recordings. Our paradigm produced marked and acute exploratory behavior (Figure S2) without disrupting the large-scale sleep–wake structure across the many hours of recording (see upper panel in Figure 1C).

5In order to investigate the long-term effects of novel stimulation on the spatiotemporal evolution of ongoing neuronal activity, we took advantage of a neuronal ensemble correlation method previously shown to detect experience-dependent reactivation of rodent hippocampal ensembles during SW and REM sleep. This method generalizes the concept of pairwise neuronal correlations to an arbitrarily large number of neurons, quantifying the degree of similarity between spatiotemporal patterns of neuronal activity by way of a firing rate-normalized template-matching algorithm (see Figure 1D). Templates of alert WK neuronal ensemble activity were selected from moments when animals made whisker contact with the novel objects (n= five templates per animal). Control templates were selected from epochs of alert WK 24 h or 48 h before novel stimulation (24 h for rats 1–3, 48 h for rats 4 and 5; n = five templates per animal), during which familiar tactile stimulation was produced by the contact of whiskers with the smooth walls of the recording box, to which animals had been habituated. Templates were matched against the entire record of neuronal activity using the neuronal ensemble correlation method (see Figure 1E). The resulting correlation temporal profiles were averaged for each template set, aligned with reference to the light–darkness cycle to control for possible circadian effects, and compared.

6First, we tested whether the neuronal ensemble correlation method could detect any trace of neuronal reverberation lasting at least more than 1 h after exposure to novel stimulation. For this, we examined correlation profiles obtained for all recorded neurons (three to four brain areas pooled together) in each animal. As shown in Figure 2A, postnovelty average correlation distributions were significantly right-shifted relative to prenovelty distributions (ANOVA of mean pre- and postnovelty correlations over 24 h or 48 h; n = five animals, F = 9.5, d.f. 1, p = 0.016). This indicates that the neuronal firing patterns concomitant with novel stimulation persisted significantly more during the ensuing time than patterns sampled 24 h or 48 h before novel stimulation, when animals were in the same behavioral state (alert WK), but without novel objects to explore. The effect was independently observed, to a variable degree, in all the five animals studied (Bonferroni comparison, p < 0.01). Figure 2B shows the temporal evolution of neuronal ensemble correlations for 24 h (upper three panels in Figure 2B) and 48 h (lower two panels in Figure 2B).

7Despite the marked interanimal variability in the shapes and magnitudes of these profiles, a significant increase of neuronal correlations after exposure to novel stimulation was observed in most recording sites. Importantly, these increases lasted well above 1 h, as revealed by the temporal evolution of the p values generated by Bonferroni comparisons between post- and prenovelty correlation distributions (Figure 2C). These results indicate that significant experience-dependent neuronal reverberation could be detected in the forebrain up to 48 h after exposure to novel stimulation.

8In order to assess the contributions of different neurons to total ensemble correlations, we ran the correlation analysis on a neuron-by-neuron basis. We found that no one subset of neurons was particularly responsible for the reverberation effect, since the contribution of individual neurons was highly variable in time (data not shown). This indicates that the neuronal changes associated with exposure to novel stimuli were highly distributed through the neuronal populations sampled. Furthermore, judging by the maximum neuronal ensemble correlations observed (rat 3, peak at 0.35; Figure 2B), one would conclude that novel stimulation-specific neuronal activity was not perfectly repeated, but was rather loosely reverberated for several hours.

9In order to assess the anatomical distribution of experience-dependent neuronal reverberation, we performed the neuronal ensemble correlation analysis for each area separately (Figure 3A). At first glance, differences between pre- and postnovelty correlation profiles were noticeable in all animals, with predominant effects in a different subset of areas for each animal. For example, rat 4 showed marked reverberation in the HP, but small changes in the CX, while rat 5 showed just the opposite. In the majority of the recorded sites, postnovelty traces (red in Figure 3A) run above prenovelty traces (black in Figure 3A), but the reverse also occurs, suggesting some sort of antireverberation. The most widespread reverberation was seen in rat 1, which showed sustained reverberation in the CX and decaying reverberation in the HP and TH. A somewhat similar pattern was seen in rat 5, while rat 3 showed strong reverberation only in the PU, and rat 4 in the HP and TH. Rat 2 showed the least reverberation of all, with somewhat stronger effects in the PU.

10Despite the interanimal and interarea differences in the magnitude and shape of correlation profiles, significant changes between pre- and postnovelty correlations were observed in all areas studied (CX, five of five rats; HP, three of four rats; PU, four of four rats; and TH, four of five rats; Bonferroni comparison, p < 0.05). Indeed, experience-dependent changes were not statistically different across different forebrain areas (ANOVA, F = 0.24, d.f. 3, p = 0.86). The temporal evolution of p values (Bonferroni comparison) associated with single-area correlation profiles shows that significant reverberation was present in 16 of 18 recording sites for several hours after exposure to novel stimulation (Figure 3B). It also confirms that neuronal ensemble reverberation (post-/prenovelty correlations) is not the only kind of experience-dependent change possible. Some animals showed significant long-lasting antireverberation (pre-/postnovelty correlations), i.e., patterns of activity that were statistically more dissimilar from novel stimulation templates than expected by chance. Antireverberation (indicated by blue hues in Figure 3B) occurred in the HP (one of four rats), PU (two of four rats), and TH (one of five rats), but not in the CX. Single-area postnovelty average correlations showed peaks of the order of 0.4 (rat 5, PU; Figure 3A), but typically ranged from 0.1 to 0.2. Therefore, high-fidelity replay of neuronal firing patterns was not observed even when single areas were considered.

11An intriguing observation came from the scrutiny of no-contact templates of neuronal activity, sampled within the novel stimulation 1-h period during alert WK, but excluding moments of contact between whiskers and objects. Surprisingly, no-contact templates yielded correlation profiles that were almost indistinguishable from those obtained when animals had tactile contact with the novel objects (Figure 3C). Therefore, the exploration of the novel environment enhanced the reverberation of all the neuronal activity patterns concomitant with the experience and not just of those corresponding to moments in which animals received sensory inputs from the objects. This rules out the possibility that stimulus complexity, rather than novelty, is the underlying cause of the enhanced neuronal reverberation observed after exploration of the objects.

12Single-area results indicate that neuronal ensemble correlations often peak during discrete epochs that last a few hours. We also noticed marked oscillations of the correlation trace in several recorded sites (e.g., rat 5, CX). These observations suggest that some underlying biological process, with slow evolution but with sharp phase transitions, governs the long-term reverberation of neuronal firing patterns. To test whether transitions in the wake–sleep cycle could amount for these effects, we investigated how experience-dependent changes in neuronal correlations varied across the three major rat behavioral states: WK, SW sleep, and REM sleep. A comparison across states of post-/prenovelty correlation ratios calculated from averages of entire recordings indicated a significant state-specific effect (ANOVA, F = 9.289, d.f. 2, p= 0.0004), with SW ratios being significantly higher than those of both WK (Bonferroni comparison, p < 0.05) and REM (Bonferroni comparison, p < 0.003). Indeed, significant state-specific differences in post-/prenovelty correlation ratios were individually detected in four of five animals (ANOVAs, d.f. 2: rat 2, PU, F = 4.13, p = 0.039; rat 3, CX, F = 6.45, p = 0.026; rat 4, HP, F = 3.99, p = 0.029; rat 5, CX, F = 13.81, p < 0.0001).

13The mean correlation values found in those recording sites for the three behavioral states reveal that SW correlations were systematically larger than WK correlations (Figure 4A). Several other recorded sites displayed similar but nonsignificant trends. Meanwhile, the REM correlations mea-sured were variable and could not be consistently ranked in relation to WK and SW sleep. Comparable neuronal reverberation between SW and REM sleep was observed in only one animal (rat 5, CX). A major effect of SW sleep on neuronal reverberation was corroborated by the temporal evolution of successive state-specific Bonferroni comparison p values calculated for pre- and postnovelty 4-h average correlations across all animals and brain areas studied (Figure 4B). The strongest contrast between pre- and postnovelty neuronal correlations was clearly seen during SW sleep, with less effect seen in WK and even less in REM sleep. Figure 4C depicts the state-sorted pre- and postnovelty correlations for rat 5, CX, illustrating both the general SW effect and the much less-prevalent REM sleep changes.

14Next, we tested the possibility reported in hippocampal place cells that experience-dependent replay of neuronal firing patterns during sleep can be slower (REM) or faster (SW) than during WK. Template-to-target matches at different speed factors were obtained by comparing 250 ms-binned templates with targets binned within a range of different bin sizes (from 12.5 ms to 500 ms). By temporally compressing and expanding “target” spike records before matching them to templates, we determined the magnitude of neuronal ensemble correlations for speed factors ranging from 0.5 to 20 times the physiological WK processing speed, which covers the reported optimum speed ranges for SW and REM sleep. A predominance of neuronal reverberation during SW sleep was seen for all speed factors, as indicated by Bonferroni comparisons (Figure 4D). However, no significant differences were seen when post- and prenovelty correlation ratios (calculated from averages of entire recordings) were compared across different speed factors (ANOVA, F = 1.496, d.f. 5, p = 0.19). Figure 4D shows that within any given state or area, neuronal reverberation did not vary systematically with speed factor, and the temporal distribution of correlation hot-and-cold spots was largely insensitive to speed factor.

15Thus, we found no evidence that forebrain neuronal reverberation can be optimized assuming replay speeds different from the WK normal rate. Indeed, a subtle but consistent decrease of p values can be observed for speed factors 10 times and 20 times faster than normal WK rates, while speed factors near the physiological range (2×–0.5×) show stronger and similar effects. This was the case even in the HP, in contrast with previous findings in hippocampal place cells recorded in highly trained animals performing a spatial navigation task. At present, it is unclear whether this discrepancy reflects differences in stimulus familiarity (novel versus habitual), stimulation modality (tactile exploration versus spatial navigation), the very low representation of place cells in our hippocampal samples (less than 5%), or possible differences in the analyses used in previous studies, based on the statistical boot-strapping of relatively small datasets.

16All together, our results indicate that neuronal reverberation was consistently stronger during SW sleep, decreasing during WK. This is remarkably well-illustrated by a superimposition of behavioral state classification and neuronal ensemble correlations (middle panel in Figure 5A), which reveals an exquisite long-term temporal match between epochs of increased neuronal correlations and SW episodes (red in Figure 5A). Likewise, neuronal correlation troughs show a tight correspondence with WK episodes (blue in Figure 5A). This characteristic state-dependency persisted throughout the 45 h of postnovelty recording (see Figure 4B). Notice that REM sleep only showed SW-like results in one out of five animals (rat 5, CX; depicted in Figures 4 and 5). In the remaining animals, REM correlations were either closer to WK levels than to SW levels or in between (see Figure 4A). Given this marked variability and the very short duration of total REM sleep in comparison with total SW sleep (WK 52%, SW 40%, and REM 8% of total recording time for five animals), this indicates that REM sleep plays a minor role in neuronal reverberation.

17Interestingly, a comparison of the correlation temporal profile with the concurrent neuronal firing record (upper panel in Figure 5A) reveals that SW correlation peaks correspond to periods of decreased firing rate, while WK correlation troughs match epochs of increased neuronal activity. This is better shown in Figure 5C, which depicts data segments approximately 2-h long, comprising the three major behavioral states studied. The first segment (shown by a single asterisk in Figure 5A) corresponds to 60 min of novel stimulation and the immediately ensuing sleep–wake cycles, while the second segment (shown by double asterisks in Figure 5A) illustrates sleep–wake episodes occurring approximately 15 h after the original experience. Thus, although novel stimulation templates of neuronal activity were selected from WK episodes characterized by high firing rates, ensuing reverberation of these neuronal firing patterns was most pronounced during SW sleep, under lower firing rates.

18In order to assess several objections to the replay hypothesis for memory consolidation during sleep, we conducted long-term continuous neuronal recordings on animals subjected to a naturalistic behavioral paradigm, which involved multiple novel sensory and spatial cues. Our results indicate that large-scale neuronal firing patterns generated during the exploration of novel objects can recur for several hours after the reference experience throughout most of the forebrain, while firing patterns associated with familiar stimulation (i.e., the walls of the recording box) are substantially less detectable over time.

19Our results fend off three major objections to the notion that neuronal reverberation during sleep may underlie memory consolidation. First, significant experience-depen-dent changes in neuronal ensemble correlations can be tracked as late as 48 h after the reference novel experience, being therefore compatible with memory impairment effects of sleep deprivation applied hours or days after training. Second, these effects were observed in rats completely naïve with respect to the reference stimuli, ruling out the possibility that only the performance of highly trained behaviors would be followed by neuronal reverberation. Third, neuronal ensemble correlations were significantly enhanced during SW sleep and decreased during WK, while REM sleep produced variable results. The data indicate that novel experience caused sustained neuronal reverberation rather than discrete reactivation, in the sense that reverberation decreased, but did not disappear, during WK (see Figure 4C). The consistent increase in neuronal reverberation during SW sleep, the high interanimal variability of neuronal reverberation during REM sleep, and the small contribution of REM sleep to total sleep time indicate that the cognitive effects of experience-dependent neuronal reverberation must be largely attributed to SW sleep. Therefore, our results suggest a major role for SW sleep in the reverberation of new memory traces.

20An important result of the present study is the inverse correlation between neuronal correlations and concurrent firing rates. Although all neuronal activity templates were taken from epochs of high arousal WK when firing rates were generally high, their neuronal reverberation during subsequent WK was not very prominent (see Figures 4 and 5). In contrast, reverberation of the same activity templates peaked during SW sleep, when the firing rates of forebrain neurons are generally low (see Figures 4 and 5). This suggests that reverberating patterns of neuronal activity associated with past novel experience are largely—but not completely—masked during WK by incoming sensory inputs unrelated to the reference experience. By the same token, peak neuronal correlations arise during SW sleep, when sensory interference ceases. Taken together, these observations corroborate the notion that the importance of sleep for memory consolidation stems from the offline processing of memory traces, i.e., from the absence of sensory interference. Were there differences in average firing rate before and after exposure to the novel environment, and could such differences account for the effects seen in neuronal correlations? The neuronal ensemble correlation method involves a normalization of firing rates after binning, and therefore it is insensitive to moderate changes in the mean firing rate. In our experiment, firing rates for individual neurons varied substantially during exposure to novelty, with some neurons firing more and other neurons firing less than before exposure. This variability caused a moderate but nonsignificant increase in the average activity of the cells so that the firing rates within novelty templates were on average approximately 10% higher than the neuronal firing rates of preexposure templates. Neuronal firing rates increased during exposure to novel objects and persisted elevated for up to 1 h after removal of the objects, returning to baseline afterwards. This contrasts with the timecourse of neuronal correlations changes, which were increased for up to 48 h. Finally, as explained above, neuronal reverberation was inversely correlated with firing rates. Thus, mean firing-rate differences were not responsible for the reverberation effect, which should rather be attributed to specific firing-rate relationships across multiple neurons.

21Our results impose some clear constraints on future sleep and learning theories.

22First, no sign of neuroanatomical specificity was found in the correlations measured, and in particular no significant differences between hippocampal and extrahippocampal areas could be detected. Despite considerable interanimal variability in the magnitude of the correlations observed in the different brain structures, statistically significant neuronal reverberation produced by novel stimulation was observed in 16 out of 18 recorded brain sites, comprising the CX, HP, PU, and TH. This broad forebrain reverberation was related to the free exploration of four novel and complex objects, placed in four well-separated places and including the presence of novel food. Thus, novel experience involving tactile, gustatory, olfactory, spatial, and motor components is able to engage multiple forebrain structures, all similarly capable of reverberating neuronal patterns of activity after novel stimulation.

23Second, neuronal ensemble correlations measured across the forebrain were typically small (on the order of 0.1–0.3), agreeing with values previously reported for pairwise or many-neuron correlations. Qualitatively similar results were observed for bins ranging from 5 ms to 1,000 ms, with higher correlation values for larger bin sizes. This suggests that neurons of multiple forebrain areas, once exposed to novel experience, do not accurately replay prior WK activity patterns longer than 5 ms. Instead, they show a mild but long-lasting bias towards (or against) the reference activity patterns. Indeed, not a single template-to-target match (out of 979,200 matches sampled) yielded correlation values higher than 0.45, indicating that novelty-induced neuronal reverberation occurs at low fidelity. It has been proposed that a high-fidelity replay of neuronal firing patterns during sleep may be achieved assuming that replayed patterns can undergo time compression and expansion. We assessed this possibility thoroughly, but found no evidence of such effects in any of the forebrain sites recorded. Thus, in the face of consistently low neuronal correlation values, the “high-fidelity replay hypothesis” for timeperiods larger than 5 ms should be rejected, at least in mammals. It remains to be seen whether more precise patterns of spike-to-spike correlations may reverberate in intervals smaller than 5 ms.

24A third important point regards the observation that neural activity sampled when animals were aroused, but not touching the objects, yielded neuronal reverberation that was nearly identical to that obtained when animals made sensory contact with the objects. This indicates that the kind of experience-dependent neuronal reverberation detected by the neuronal ensemble correlation method does not reflect the specific features of the stimuli, but is related to the overall behavioral salience of the novel stimulation period, i.e., is context- rather than stimulus-specific. In principle, these results are compatible with a slow and progressive process of memory consolidation, proportional to the novelty of the experience, and able to bind together a multitude of contextual cues related to its core sensory elements.

25It has been suggested that the neuronal reverberation of newly acquired synaptic changes during SW sleep may lead to the recall and storage of new memories by way of “calcium-mediated intracellular cascades” capable of opening the “molecular gates to plasticity”. This hypothesis is partially contradicted by evidence that calcium-dependent gene expression related to synaptic plasticity is upregulated during REM sleep, but not during SW sleep. The present findings and the current literature suggest instead that SW and REM sleep play separate roles on memory consolidation, with memory recall occurring during SW sleep and memory storage taking place during REM sleep. According to this view, the deleterious effects of sleep deprivation on memory consolidation would be a consequence of the disruption of the underlying neuronal reverberation and gene expression during SW and REM sleep, respectively.

26The fact that neuronal reverberation is sustained for long epochs during SW sleep suggests that unconsolidated synaptic changes may not only be recalled, but also amplified over time during SW sleep. Indeed, a progressive increase of neuronal correlations across single SW sleep episodes was often observed (upper panel in Figure 5B). A model of how such amplification may arise is presented in Figure 6. We have recently proposed that the cyclical reiteration of trace amplification during SW sleep and trace storage during REM sleep promotes the postsynaptic propagation of memory traces, as suggested by the hippocampofugal pattern of gene expression during REM sleep. Potentially, this propagation could cause memory traces to progressively reach farther and farther away from the original synaptic trajectory activated at initial encoding. Over time, this sleep-dependent propagation could lead to deeper encoding within the CX, as well as hippocampal disengagement. The notion that the two major phases of sleep play distinct and complementary roles in memory consolidation is in line with evidence that SW and REM sleep have synergistic effects on human procedural learning. Put in historical perspective, our model argues that sleep separately harbors both mechanisms postulated in the past to be necessary for memory consolidation: postacquisition neuronal reverberation and structural synaptic plasticity. In conclusion, sustained neuronal reverberation during SW sleep, immediately followed by plasticity-related gene expression during REM sleep, may be sufficient to explain the beneficial role of sleep on the consolidation of new memories.

27Multielectrode arrays (Figure S3) were surgically implanted according to National Institutes of Health (NIH) guidelines in five adult male Long–Evans rats (250–300 g). The following coordinates in millimeters relative to Bregma were used to center the arrays: HP (+2.8 anteroposterior [AP], +1.5 mediolateral [ML], −3.3 dorsoventral [DV]), CX (+3.0 AP, +5.5 ML, −1.5 DV), PU (−1.0 AP, +2.5 ML, −5.0 DV), and TH (+3.0 AP, +3.0 ML, −5.0 DV). Hippocampal data pool together neurons recorded with staggered electrodes from the CA1 field and the dentate gyrus. Locations of implants were histologically verified by comparing cresyl-stained frontal brain sections with reference anatomical planes (Figure S4). A Multineuron Acquisition Processor (128 channels; Plexon Inc., Dallas, Texas, United States) was used to perform recordings of neuronal spikes and LFPs, as previously described (Figures S5 and S6). A waveform-tracking technique involving periodic template adjustment was employed for the continuous recording of individual units (see Figure S5). Units that showed nonstationary waveforms, unstable firing-rate profiles, or both were discarded. The stability of firing rates within each behavioral state can be appreciated in the upper panel in Figure 5.

28Before the beginning of the experiment, animals were individually habituated to an empty recording box for 5–7 entire days (12 h:12 h light:dark schedule, lights on at 06:00 a.m.) so as to reach steady-state behavioral activity and baseline wake–sleep cycles. Behaviors were continuously recorded by way of two infrared-sensitive CCD video-cameras; infrared illumination was used to monitor behavior when visible lights were off. Behavioral states were coded as WK, SW sleep, or REM sleep based on a spectral analysis of LFPs and visual inspection of videotaped behaviors, according to previously described criteria (Figure S7).

29Data were processed and analyzed by custom-made MATLABTM (MathWorks, Natick, Massachusetts, United States) code running in a computer cluster comprising 32 CPU (EvolocityTM, LNXI, Sandy, Utah, United States). Neuronal ensemble correlations were calculated following the method in ). In brief, a stretch of data (target) is scanned for similarity to certain multineuron temporal patterns of activity (templates). Templates consist of CxN matrices corresponding to simultaneously recorded spike trains of C neurons binned into N intervals of a certain length. Our results were obtained using 9 s-long templates binned with 250 ms-long bins with as many neurons as were available for the brain areas in question. Hence, each template pattern yielded a Cx36 matrix. Target data comprising 24 h or 48 h were sparsely sampled (every 30 s) and binned into matrices of the same dimensions. For a C×N template data matrix x and a target data matrix y sampled at time t, the ensemble correlation index Ct is obtained by:

30where

31(the same applies to Yc, y¯, and σy).

32Templates were selected by careful scrutiny of behavior recorded on videotapes and of the corresponding spectral characteristics of hippocampal LFPs, so as to assure sampling during alert WK (Figure S7). Particular care was used to select template sets with a comparable prevalence of θ (5–8 Hz) over δ (2–4 Hz) frequencies (θ/δ hippocampal spectral ratios, ≥10). All templates and targets were individually sampled without overlap. Because binning may destroy higher-frequency phenomena, control analyses were run to assess the consequences of binning the data with bins of different sizes. Results obtained with bin sizes in the interval of 5–1,000 ms are qualitatively equivalent to the ones obtained using 250 ms bins (Figure S8). In order to detect temporally expanded or contracted repetitions of template patterns, a range of larger- or smaller-than-template bin sizes, respectively, was used when binning each target sample (12.5 ms, 25 ms, 125 ms, 250 ms, 375 ms, and 500 ms). This allowed us to compare targets spanning different temporal scales with templates of neuronal activity sampled at WK speed. The bin size range used allowed the detection of patterns temporally compressed by factors of 20, 10, and 2 (bin sizes of 12.5 ms, 25 ms, and 125 ms), temporally expanded by factors of 1.5 and 2 (bin sizes of 375 ms and 500 ms) as well as replayed at the same speed (250 ms). StatViewTM (SAS, Cary, North Carolina, United States) and MATLABTM software were used for descriptive statistics and hypothesis testing.