Cortical Ripples During NREM Sleep And Waking in Humans Part 2
Nov 02, 2023
Ripple detection. The median center frequency of hippocampal ripples in nine studies was;85 Hz in humans (Bragin et al., 1999; Staba et al., 2004; Clemens et al., 2007; Axmacher et al., 2008; Le Van Quyen et al., 2008; Staresina et al., 2015; Jiang et al., 2019b; Norman et al., 2019; Vaz et al., 2019), whereas in rodents, sharp-wave–ripple frequency is;120-140 Hz (Buzsáki, 2015).
Ripple detection is a technology that measures human brain activity. It captures the electrical activity of neurons in the brain by placing electrodes on the scalp and converts the signal into an electroencephalogram recorded on a computer. In recent years, with the in-depth study of the human brain, ripple detection has become an artifact for studying cognition, learning, emotion, etc., and has gradually entered our daily lives.
Regarding memory, research shows that the ripple activity of the human brain plays a key role in the formation and maintenance of memories. When the human brain performs learning and memory tasks, it can produce ripples of specific frequencies between the cerebral cortex, such as theta waves and gamma waves. These waves can enhance the encoding and storage of memory, and also help with subsequent retrieval and recall of the past. stored information. Therefore, ripple detection can be used to study the memory mechanism of the human brain and has the potential to treat cognitive disorders and memory disorders clinically.
In addition, ripple detection also helps us broaden our learning methods. When students participate in classroom teaching, they can learn about individual learning characteristics and subject preferences by recording the ripple activities of the brain during the learning process, to learn and improve in a more targeted manner. In addition, when studying, appropriate music and sounds also have a great impact on the learning effect. Through ripple detection, the reactions of different groups of people to different tones and volumes can be determined, so that teachers or students themselves can choose the most appropriate music and sounds to improve learning effects.
To sum up, ripple detection plays a great role in our life and study. By measuring ripple activity at specific frequencies of the human brain, we can better understand our cognition, learning, and memory characteristics, and help us improve our learning effects and quality of life. As the level of technology improves, ripple detection will continue to deepen its research on cognitive neuroscience and neurological diseases, bringing us more prospects and possibilities. It can be seen that we need to improve memory, and Cistanche deserticola can significantly improve memory, because Cistanche deserticola can also regulate the balance of neurotransmitters, such as increasing the levels of acetylcholine and growth factors. These substances are very important for memory and learning. In addition, Meat can also improve blood flow and promote oxygen delivery, which can ensure that the brain receives sufficient nutrients and energy, thereby improving brain vitality and endurance.
Click know 10 ways to improve memory
In humans, putative 90 Hz hippocampal ripples have the same characteristic relation to sharp waves, intrahippocampal localization, modulation by sleep stage, and relation to cortical sleep waves as in rodents, and occur in hippocampi that have no signs of epileptic involvement. Furthermore, in the rodent hippocampus, the distinction between higher frequency g bursts and ripples is not always sharp, leading to the suggestion that for simplicity they both be referred to as “ripples” (Stark et al., 2014), which we follow here.
Ripple detection was performed in the same way for all structures and states and was based on a previously described hippocampal ripple detection method (Jiang et al., 2019a,b). Requirements for inclusion and criteria for rejection were determined using an iterative process across patients, structures, and states. Data were first bandpassed at 60-120 Hz (sixth-order Butterworth filter with zero-phase shift), and the top 20% of 20 ms moving root-mean-squared peaks were detected.
It was further required that the maximum z score of the analytic amplitude of the 70- 100 Hz bandpass was .3 and that there were at least three distinct oscillation cycles in the 120 Hz lowpass signal, determined by shifting a 40 ms window in increments of 5 ms across 650 ms relative to the ripple midpoint and requiring that at least one window have at least three peaks. Adjacent ripples within 25 ms were merged. Ripple centers were determined as the time of the maximum positive peak in the 70-100 Hz bandpass.
Ripple onsets and offsets were identified on each side of the center peak when the 70-100 Hz analytic amplitude fell below 0.75 z scores. To reject epileptiform activities or artifacts, ripples were excluded if the absolute value of the z score of the 100 Hz highpass exceeded 7 or if they occurred within 2 s of a 3 mV/ms change in the broadband LFP. Ripples were also excluded if they fell within 6500 ms of putative IISs, detected as described below.
To exclude events that could be coupled across channels because of epileptiform activity, we excluded events that coincided with a putative IIS on any cortical or hippocampal SEEG channel included in the analyses. Events in SEEG recordings that had only one prominent cycle or deflection in the broadband LFP that manifested as multiple cycles above the detection threshold in the 70-100 Hz bandpass were excluded if the largest valley-to-peak or peak-to-valley absolute amplitude in the broadband LFP was 2.5 times greater than the third largest.
For each channel, the mean ripple-locked LFP and mean ripple band were visually examined to confirm that there were multiple prominent cycles at ripple frequency (70-100 Hz) and the mean time-frequency plot was examined to confirm there was a distinct increase in power within the 70-100 Hz band. In addition, multiple individual ripples in the broadband LFP and 70-100 Hz bandpass from each channel were visually examined to confirm that there were multiple cycles at ripple frequency without contamination by artifacts or epileptiform activity. Channels that did not contain ripples that met these criteria were excluded from the study.
Of note, a recent study has identified ripples based on these criteria in a patient without epilepsy (Rubin et al., 2022). The oscillation frequency of each ripple was computed by first counting the number of 70-100 Hz bandpass zero crossings (each representing half a cycle) during the ripple event and adding any remaining partial half cycle (i.e., the remaining phase angle over p). The number of half cycles in the ripple was divided by 2 to calculate the total number of cycles during the ripple. Finally, the oscillation frequency in Hertz was calculated as the number of cycles divided by the ripple duration in seconds.
Subtraction of unit spikes from LFPs. One challenge in detecting high-frequency oscillations, such as ripples in the LFP recorded by a microelectrode, is that unit spikes may “bleed through” into the microLFP (Ray, 2015), thus resulting in the detection of spurious relationships between ripples and unit spikes. Although unit spikes are fast events, simply downsampling (which requires first low-passing to prevent aliasing) may not completely remove the influence of the action potential on the micro-LFP.
To address this, we used a modified unit spike waveform subtraction technique (Pesaran et al., 2002). Specifically, the average spike waveform (500 to 1600 ms around the trough) of each unit was subtracted from the unfiltered 30 kHz Utah Array micro-LFP of the same channel centered on each spike. The data were then downsampled to 1 kHz and ripple detection was performed as described above. We confirmed that this method led to the detection of true oscillations through extensive visual confirmation of events in the 30 kHz micro-LFPs.
IIS detection and rejection. Ripples and other sleep waves were excluded if they were within 6500 ms from putative IIS detected as follows: A high-frequency score was computed by smoothing the 70- 190 Hz analytic amplitude with a 20 ms boxcar function, and a spike template score was generated by computing the cross-covariance with a template IIS. The high-frequency score was weighted by 13, the spike score was weighted by 25, and an IIS was detected when these weighted sums exceeded 130. In each patient, detected IIS and intervening epochs were visually examined from hippocampal and cortical channels (when present) to confirm high detection sensitivity and specificity.
Detection of downstates, upstates, and sleep spindles. Downstates and upstates were detected as previously described (Jiang et al., 2019a,b), where the broadband LFP from each channel was bandpassed from 0.1- 4 Hz (sixth-order Butterworth filter with zero-phase shift), and consecutive zero crossings separated by 0.25-3 s were selected. The top 10% amplitude peaks were selected, and the polarity of each signal was inverted if needed so that downstates were negative and upstates were positive, by ensuring that the average analytic amplitude of the 70-190 Hz bandpass within 6100 ms around the peaks was greater for upstates than downstates. A total of 2,649,563 downstates were detected, with an average SD (across channels) density (occurrence rate) of 12.9 6 4.7 min1 and amplitude of 237.3 6 169.8 mV. A total of 2,922,211 upstates were detected with a density of 14.6 6 4.7 min1 and amplitude of 163.8 6 84.9 mV.
Spindles were detected as previously described (Hagler et al., 2018), where data were bandpassed at 10-16 Hz, then the absolute values were smoothed via convolution with a tapered 300 ms Tukey window, and median values were subtracted from each channel. Data were normalized by the median absolute deviation, and spindles were detected when peaks exceeded 1 for at least 400 ms. Onsets and offsets were marked when these amplitudes fell below 1. Putative spindles that coincided with large increases in lower (4-8 Hz) or higher (18-25 Hz) band power were rejected to exclude broadband events as well as theta bursts, which may extend into the lower end of the spindle range (Gonzalez et al., 2018). A total of 694,168 spindles were detected with an average and SD (across channels) density of 3.0 6 3.0 min1, amplitude of 29.1 6 13.5 mV, oscillation frequency of 12.4 6 0.7 Hz, and duration of 633.5 6 67.2 ms.
Ripple temporal relationships. Peri-cortical ripple time histograms of cortical sleep waves on the same channel were computed. Event counts were found in 50 ms bins for sleep waves within 61500 ms around cortical ripple centers at t = 0. A null distribution was generated by shuffling the event times relative to the ripples at t = 0 within this 3 s window 200 times each. Pre-FDR p values were computed by comparing the observed and null distributions within each bin over 61000 ms for cortical sleep waves. These p values were then FDR-corrected for the number of channels across patients multiplied by the number of bins per channel (Benjamini and Hochberg, 1995). A channel was considered to have a significant modulation if there were three or more consecutive bins with FDR-corrected p values less than a = 0.05. Whether events were leading or lagging cortical ripples at t = 0 was computed for each channel with a two-sided binomial test with an expected value of 0.5, using event counts in the 1000 ms before versus 1000 ms after t = 0 for sleep waves. Plots had 50 ms Gaussian smoothed event counts with 50 ms bins.
Conditional probabilities of a ripple given the following sleep waves or their sequences were computed: downstate, spindle, upstate, downstate-spindle, and spindle-upstate. A ripple given a spindle (R | SS) was determined if the ripple center occurred during the spindle (average spindle duration was 634 ms). A ripple was considered to precede an upstate (R | US) or follow a downstate (R | DS) if the ripple center occurred within 634 ms before or after the peak of the upstate or downstate, respectively.
Unit detection, classification, quality, and isolation. Unit detection and classification were performed according to our published procedures (Peyrache et al., 2012; Chan et al., 2014; Dehghani et al., 2016; Le Van Quyen et al., 2016; Telenczuk et al., 2017; Eichenlaub et al., 2020; Dickey et al., 2021). Data were bandpassed at 300-3000 Hz and putative unit spikes were detected when the filtered signal exceeded 5 times the estimated SD of the background noise. Units were k-means clustered using the first three principal components of each spike. Overlaid spikes were examined visually and those with abnormal waveforms were excluded. Based on their waveforms, firing rates, and autocorrelograms, action potentials were clustered as arising from putative PYs or INs. PYs had spike rates of;0.1-0.8 Hz, long valley-to-peak and half-width intervals, sharp autocorrelations, and bimodal interspike interval (ISI) distributions, reflecting a propensity to fire in bursts.
By contrast, INs had spike rates of;1-5 Hz, short valley-to-peak and half-width intervals, broad autocorrelations, and a predominantly unimodal ISI distribution. The Utah Array patient included in the study had 69 PY with a total of 231,922 spikes, as well as 23 IN with a total of 462,246 spikes. The average and SD PY valley-to-peak amplitude was 44.8 6 12.9 mV, spike rate was 0.28 6 0.17 Hz, valley-to-peak width was 0.49 6 0.5 ms, half-peak width was 0.62 6 0.04 ms, and bursting index was 0.03 6 0.02. The average and SD IN valley-to-peak amplitude was 28.9 6 11.9 mV, spike rate was 1.67 6 1.63 Hz, valley-to-peak width was 0.29 6 0.05 ms, half-peak width was 0.34 6 0.05 ms, and bursting index was 0.01 6 0.01. Valley-to peak amplitude, spike rate, valley-to-peak width, half-peak width, and bursting index were all significantly different between PY and IN (p, 0.0001, two-sided two-sample t-test).
Single-unit quality and isolation were confirmed according to previously established guidelines (Kaminski et al., 2020 ). Unit spikes were verified to well exceed the noise floor based on large peak signal-to-noise ratios (PY: 10.1 6 3.4; IN: 5.9 6 3.1). Since the neuronal spiking refractory period is ;3 ms, the percent of ISIs,3 ms estimates the degree of single-unit contamination by spikes from different units, which was very low among the units included in this study (PY: 0.12 6 0.15%; IN: 0.31 6 0.50%). Furthermore, single units detected on the same contact were highly separable according to their projection distances (Pouzat et al., 2002) (PY: 49.2 6 25.8 SD; IN: 50.9 6 28.6 SD). Last, temporal stability of unit spikes over time was confirmed based on the consistency of the mean waveform shape and amplitude of each unit across recording quartiles.
Analyses of unit spiking during ripples. Unit spiking was analyzed concerning local ripples detected on the same contact. Ripple phases of unit spikes were determined by finding the angle of the Hilbert transform of the 70-100 Hz bandpass signal (zero-phase shift) at the times of the spikes. The circular mean ripple phase was determined for each unit according to Berens (2009). Phase analyses were only performed on units that had at least 30 spikes during local ripples. The Rayleigh test was used to assess for unimodal deviation from the uniformity of the circular mean phases across units. PY–PY and PY–IN unit pair cofiring within 5 ms was assessed based on Dickey et al. (2021). To evaluate cofiring when there was a ripple at either or both sites versus baseline when no ripples were occurring, we randomly selected non-ripple epochs matched in number and duration to the ripples but during which there were no ripples detected on either channel. Cofiring during ripples with observed versus random spike times was computed by shuffling the spike times of each unit within the intervals of the ripples.
Experimental design and statistical analyses. All statistical tests were evaluated with a = 0.05. All p values involving multiple comparisons were FDR-corrected according to Benjamini and Hochberg (1995). FDR corrections across channel pairs were done across all channel pairs from all patients included in the analysis. Box-and-whisker plots show median, mean, and interquartile range, with whiskers indicating a 1.5 interquartile range with outliers omitted. Kernel density plots were produced using methods from Bechtold (2016). Significance and shuffling statistics of peri-ripple time histograms were computed as described above. Unit spiking statistics were computed as described above.
Statistical comparisons between ripple characteristics were performed using two-sided paired or two-sided two-sample t-tests between channel means. Differences in ripple characteristics between states (NREM or waking) or regions (cortex or hippocampus) were determined using linear mixed-effects models with state and region as fixed effects and patient as a random effect, according to the following model:
Data availability. The deidentified raw data that support the findings of this study are available from the corresponding authors on reasonable request provided that the data sharing agreement and patient consent permit that sharing.
Code availability. The code that supports the findings of this study is available from the corresponding authors upon reasonable request.
Results
Ripples are ubiquitous across states and structures with a characteristic and focal frequency (;90 Hz) and duration (;70 ms)
Ripples were detected in intracranial cortical and hippocampal SEEG recordings from 17 patients undergoing monitoring for seizure focus localization (Table 1). Bipolar transcortical derivations ensured the measurement of locally generated LFPs. Ripples were detected exclusively from nonlesional, non epileptogenic regions and were required to have at least three cycles of increased 70-100 Hz amplitude without contamination by epileptiform activity or artifacts (Fig. 1A). Recording epochs and channels with possible contamination by epileptiform activity were rigorously rejected. Ripples were found during both waking and NREM in all cortical areas sampled (Fig. 2A, B, E–J; Table 3) as well as the hippocampus (Fig. 2C, D, K, L; Table 3).
Across states and cortical regions, ripple frequency was remarkably consistent at;90 Hz. Specifically, the mean cortical ripple frequency within the region ranged from 88.8-90.0 Hz during NREM and 89.2-89.8 Hz during waking (Table 3; Fig. 3). The mean 6 SD ripple oscillation frequency across all cortical channel means during NREM was 89.1 6 0.8 Hz and during waking was 89.5 6 0.7 Hz. These data are also presented as histograms of individual ripple characteristics (amplitude, frequency, duration, associated changes in .200 Hz amplitude) in Figure 4 as well as individual patients in Figure 5. The basic characteristics of cortical ripples were very similar in a supplemental analysis which only included cortical channels free of IISs (Fig. 6A).
Oscillation frequency was not only highly consistent at;90 Hz across ripples, but increased amplitude was highly concentrated during ripples at;90 Hz in broadband recordings, with a steep drop-off in amplitude at higher and lower frequencies. This focal power increase, which is especially prominent during NREM, can be seen in example channels (Fig. 2A, E–H) as well as grand average time-frequency plots that are averages across all channel averages of all ripples (Fig. 2I–L).
We conducted supplementary analyses to examine whether this regularity and focality of ripple frequency could be because of our method of detecting them. First, we conducted a complete reanalysis of all SEEG channels and epochs using two different frequency ranges for ripple detection and selection. One used a detection bandpass from 70 to 100 Hz as in our primary analysis, and the other expanded the detection bandpass to 65-120 Hz. Mean 6 SD NREM ripple oscillation frequency across channels (N = 273) from all SEEG patients (S1-S17) is highly similar when using either a 70- 100 Hz bandpass (mean 6 SD = 89.1 6 0.8 Hz) or a 65-120 Hz bandpass (92.7 6 2.2 Hz) (Fig. 1B). Furthermore, we verified that the 60 and 120 Hz notch filters (to remove line noise) as well as the 70-100 Hz bandpass did not artificially produce a peak in power at;90 Hz (Fig. 1C). Thus, the ripple oscillation frequency of;90 Hz appears to be physiologic rather than being driven by specific detection methods.
The mean ripple duration was also remarkably consistent (;70 ms) across states and cortical regions (Table 3). The duration and consistency are not explained by the detection requirement of having at least 3 cycles (which at 90 Hz is 33.3 ms), suggesting that this duration is also a physiological characteristic. Other characteristics, including density (NREM: 8.36 6 2.69 min1, waking: 5.77 6 3.42 min1 ), peak amplitude (NREM: 5.10 6 2.25 mV, waking: 7.38 6 4.17 mV), and change in .200 Hz amplitude (NREM: 3.096 2.75%, waking: 33.026 36.90%) were also highly similar across channels (Figs. 2, 3). Ripple characteristics were also consistent across patients (Fig. 5). However, small but significant differences were noted between regions, cortex versus hippocampus, and NREM versus waking, which are described below.
Ripples exhibit small but significant differences between the cortex and hippocampus
Human hippocampal ripples have previously been selected using a variety of methods (for review, see Jiang et al., 2019c). In our previous studies, we required that ripples be superimposed on sharp waves (characteristic of the anterior hippocampus) (Jiang et al., 2019a) or spindles (characteristic of the posterior hippocampus) (Jiang et al., 2019b). In the current study, we reanalyzed these data using the same criteria and procedures as we used for the detection and selection of cortical ripples, to avoid any methodological ambiguities; that is, we did not require the presence or absence of any associated lower-frequency LFP signature.
Using the same detection criteria, we found that cortical and hippocampal ripples share the same basic characteristics with relatively small differences. The basic hippocampal ripple characteristics were very similar in a supplemental analysis, which only included hippocampal channels free of IISs (Fig. 6B). During NREM, hippocampal ripples had average time-frequency plots (Fig. 2C, K) that resembled the corresponding cortical plots (Fig. 2A, E–I) in having concentrated oscillatory activity at;90 Hz, but differed in being superimposed on the peak of a local sharpwave–ripple rather than occurring just before the local upstate as is seen in the cortex. During waking, average time-frequency plots of hippocampal (Fig. 2D, L) and cortical (Fig. 2B, J) ripples again show concentrated oscillatory activity at;90 Hz, but also with activity that stretches into higher frequencies as noted above.
Statistical analyses (Fig. 3) revealed several differences which, although small, were nonetheless significant, attributable to the large numbers of channels. Specifically, compared with cortical ripples, hippocampal ripples were on average: slightly less dense during NREM, and more dense during waking; slightly lower frequency, especially during NREM (but still,3 Hz difference); longer duration, especially during NREM (but,10 ms difference); larger amplitude (by;2.5); and accompanied by a smaller but still significant increase in .200 Hz amplitude during waking compared with NREM. In addition, cortical ripple characteristics were not significantly correlated with the hippocampal connectivity density to their local parcel (Fig. 7). Overall, cortical and hippocampal ripples appear to be very similar, except notably in their amplitudes and associated slower waves during NREM (Fig. 2; Fig. 8A–F).
Ripples can be distinguished from high-frequency limbic oscillations
Oscillatory activity at .60 Hz in humans has been studied previously, mainly in the hippocampus of epilepsy. Le Van Quyen et al. (2010) reported likely nonpathologic activity, mainly from parahippocampal gyrus sites. They imposed a minimum duration of 100 ms, which would have eliminated the vast majority of the events that we studied (Figs. 3D, 4), and is longer than those previously reported for human hippocampal ripples (Jiang et al., 2019c), human cortical ripples (Vaz et al., 2019), and rodent hippocampal ripples (Buzsáki, 2015).
In addition, being on average;7 longer than the events we describe here, those described by Le Van Quyen et al. (2010) also differ in that they contain several oscillatory frequencies (whereas those described here are strongly centered at;90 Hz) and have larger amplitudes. Their events are similar, however, in being related to upstates, and in modulating unit firing (see below). We were therefore interested in determining the relationship between the mainly parahippocampal events described by Le Van Quyen et al. (2010) and the widespread cortical events described here.
We subselected our parahippocampal channels (N = 5 channels from 5 patients), and implemented the selection criteria described by Le Van Quyen et al. (2010): data were bandpass at 40-120 Hz, minimum event duration was 100 ms, and otherwise, events were detected as described in Materials and Methods. Using these selection criteria, we found very low densities of events. The average ripple density of parahippocampal channels during NREM using our criteria was 9.08 min1 (range: 7.23-10.77 min1 ), whereas that using the Le Van Quyen et al. (2010) detection criteria was only 0.40 min1 (range: 0-1.53 min1 ). Thus, the events described here do not correspond to those previously reported by Le Van Quyen et al. (2010).
Ripples exhibit small but significant differences between primary sensory-motor and association cortices
Previous work in sleeping rats (Khodagholy et al., 2017) and waking humans (Vaz et al., 2019) found cortical ripples to be more common in association areas than in early sensory and motor regions where they were absent or infrequent. In contrast, we found that ripple density, amplitude, and accompanying .200 Hz amplitude (a proxy for unit firing) (Mukamel et al., 2005) in primary cortex were all significantly higher than in association areas, as indicated by a positive correlation with the myelination index (Fig. 8G–I) (Rosen and Halgren, 2021). Oscillation frequency and duration were not correlated with myelination index (Fig. 9). In some cases, this effect was small, for example, explaining only 1.7% of the variance in density during waking and 4.4% during NREM. In other cases, the effect was substantial, for example, explaining 19.4% of the variance in amplitude during NREM. Comparing NREM and waking, in NREM there was a significantly stronger correlation between myelination index and density (p = 0.02, t = 2.1), amplitude (p = 8 104, t = 3.2), and .200 Hz modulation (p = 2 106, t = 4.7; one-sided paired r test; MATLAB: r_test_paired) (Steiger, 1980). In sum, human cortical ripples appear to be modestly more numerous and robust in sensory and motor than association areas.
Ripples exhibit small but significant differences between NREM and waking
Most cortical and hippocampal ripple characteristics were similar between NREM and waking (Fig. 3; Table 3), but because of the large numbers of events, even small differences were significant. Considering only differences of .10%, cortical ripple density was 31% higher and amplitude was 43% lower during NREM than waking. Ripple duration was 13% higher in the cortex and 20% higher in the hippocampus during NREM than waking. The most striking differences were the percent change in mean .200 Hz amplitude during ripples compared with a 2 to 1 s baseline (Fig. 3E), which increased 969% from NREM to waking in the cortex, and 304% in the hippocampus.
This difference is seen to be broadband in the waking ripple triggered time-frequency plots from the cortex (Fig. 2Bii, J) and hippocampus (Fig. 2Dii, L), and thus would typically be interpreted not as an oscillation, but as an indication of increased multiunit activity. Thus, compared with waking, cortical ripples are somewhat denser, longer, and smaller in NREM; hippocampal ripples are also longer. The only major difference is that both cortical and hippocampal ripples appear to be associated with a much larger increase in putative multiunit activity during waking than NREM.
Cortical ripples lock to sleep waves crucial for memory consolidation
Oscillation couplings are important for memory consolidation (Latchoumane et al., 2017). Previous studies have shown that cortical ripples occur on upstate and spindle peaks in cats (Grenier et al., 2001) and rats (Khodagholy et al., 2017), but these relationships have not been evaluated in humans. We detected downstates and upstates, with polarities determined based on associated high-frequency activity (for details, see Materials and Methods), as well as spindles, and found that cortical ripples were precisely coupled to the sequence of sleep waves described above (Fig. 10; Table 4). Specifically, in 95% of cortical channels, ripples were significantly associated with downstates and upstates (Fig. 10B, D, E), usually on the down-to-upstate transition, as seen in individual trials (Figs. 2Aiii, 10A), and in ripple-triggered averages of the broadband LFP (Fig. 2Ai, E–H).
Peri-ripple histograms show that, on average, cortical ripple centers occurred 450 ms after downstate maxima and 100 ms before upstate maxima (Fig. 10B, D–G). Less frequently, ripples occurred during spindles (significant association in 29% of channels; Fig. 10C, E), as seen in individual trials (Fig. 10A), and in peri-ripple time-frequency plots (Fig. 2Aii, E). Peri-ripple histograms show that spindles began on average 225 ms before the cortical ripple center, indicating that ripples tend to occur during spindles (Fig. 10C, E–G). The probability of ripples occurring during spindles preceding upstates was greater than that for ripples occurring during spindles, or before upstates (Fig. 10H). These results suggest that the timing of cortical ripples during NREM is appropriate for facilitating consolidation, guided by sequential activation of sleep waves.
For more information:1950477648nn@gmail.com
