Cortical Ripples during NREM Sleep and Waking in Humans

Abstract

Hippocampal ripples index the reconstruction of spatiotemporal neuronal firing patterns essential for the consolidation of memories in the cortex during non-rapid eye movement sleep (NREM). Recently, cortical ripples in humans have been shown to enfold the replay of neuron firing patterns during cued recall. Here, using intracranial recordings from 18 patients (12 female), we show that cortical ripples also occur during NREM in humans, with similar density, oscillation frequency (∼90 Hz), duration, and amplitude to waking. Ripples occurred in all cortical regions with similar characteristics, unrelated to putative hippocampal connectivity, and were less dense and robust in higher association areas. Putative pyramidal and interneuron spiking phase-locked to cortical ripples during NREM, with phase delays consistent with ripple generation through pyramidal-interneuron feedback. Cortical ripples were smaller in amplitude than hippocampal ripples but were similar in density, frequency, and duration. Cortical ripples during NREM typically occurred just before the upstate peak, often during spindles. Upstates and spindles have previously been associated with memory consolidation, and we found that cortical ripples grouped cofiring between units within the window of spike timing-dependent plasticity. Thus, human NREM cortical ripples are as follows: ubiquitous and stereotyped with a tightly focused oscillation frequency; similar to hippocampal ripples; associated with upstates and spindles; and associated with unit cofiring. These properties are consistent with cortical ripples possibly contributing to memory consolidation and other functions during NREM in humans.SIGNIFICANCE STATEMENT In rodents, hippocampal ripples organize replay during sleep to promote memory consolidation in the cortex, where ripples also occur. However, evidence for cortical ripples in human sleep is limited, and their anatomic distribution and physiological properties are unexplored. Here, using human intracranial recordings, we demonstrate that ripples occur throughout the cortex during waking and sleep with highly stereotyped characteristics. During sleep, cortical ripples tend to occur during spindles on the down-to-upstate transition, and thus participate in a sequence of sleep waves that is important for consolidation. Furthermore, cortical ripples organize single-unit spiking with timing optimal to facilitate plasticity. Therefore, cortical ripples in humans possess essential physiological properties to support memory and other cognitive functions.

Keywords: cortex; hippocampus; humans; ripples; sleep; waking.

Figure 1.

Ripple detection and event rejection. A, Broadband LFP single sweeps show events that exceeded amplitude thresholds and were accepted for or rejected from the analyses. Traces represent an example accepted ripple, with black arrows indicating multiple 70-100 Hz oscillation cycles, a rejected artifact, a rejected event with a single dominant cycle, and a rejected possible IIS. Putative ripples within ±500 ms of possible IIS detected on the same channel, as well as putative ripples coinciding with the sharp component of possible IIS on any cortical or hippocampal channel were rejected. B, Mean ± SD NREM ripple oscillation frequency (red line) across channels (N = 273; black lines) from all SEEG patients (S1-S17) is highly similar when using either a 70-100 Hz bandpass (mean ± SD = 89.1 ± 0.8 Hz) or a 65-120 Hz bandpass (92.7 ± 2.2 Hz). C, Average power spectral density of unfiltered, notched (60 Hz and harmonics), notched and 70-100 Hz bandpassed, as well as notched and 65-120 Hz bandpassed (N = 1000 randomly selected 27-s-long epochs over 450 min of recording from Patient S2). Dashed black lines indicate 70-100 Hz bandpass used for the main analyses in this study. Dashed red line indicates 90 Hz, the approximate average ripple frequency across states, structures, and filter settings.

Figure 2.

Cortical and hippocampal ripples are generated during NREM and waking. A, Orbitofrontal ripples from one channel during NREM. Ai, Average broadband LFP. Aii, Average time-frequency plot. Aiii, Example broadband unfiltered 3 s sweep (black) with 70-100 Hz bandpass analytic amplitude (blue) and below the same example, 400 ms sweep (black) with 70-100 Hz bandpass (blue). B, Same as in A, but during waking. C, Same as in A, but with hippocampal ripples (note ripple on sharp-wave peak). D, Same as in C, but during waking. E–H, Same as in A, but from other cortical regions. I–L, Grand average time-frequency plots across all channels in neocortex (N = 273) during NREM (I) and waking (J), as well as in hippocampus (N = 28) during NREM (K) and waking (L). Note the highly consistent and focal concentration of power centered at ∼90 Hz, the occurrence of cortical ripples on the down-to-upstate transition during NREM, and that some channels show increased power in the 10-16 Hz spindle band coinciding with the ripples. All plots represent ripples detected on bipolar SEEG channels. ERSP, Event-related spectral power.

Figure 3.

Neocortical and hippocampal ripple characteristics in NREM and waking. A-E, NREM and waking cortical ripple density (A), peak 70-100 Hz analytic amplitude (B), oscillation frequency (C), duration (D), and percent change in mean >200 Hz analytic amplitude during ripples compared with a −2 to −1 s baseline (E). Distributions are comprised of channel means (SEEG Patients S1-S17; N = 273 neocortical channels, N = 28 hippocampal channels). Circles represent medians. Horizontal lines indicate means. Vertical lines indicate interquartile ranges. FDR-corrected p values, linear mixed-effects models with post hoc analyses, patient as random effect. ns, Nonsignificant factor precluding post hoc analysis. For distributions across individual ripples, see Figure 4. For distributions across individual patients, see Figure 5. Figure 1B, C shows that the ∼90 Hz ripple frequency we measured is not because of filtering including the detection bandpass.

Figure 4.

Characteristics of individual cortical ripples. Histograms of ripple characteristics across individual events (N_NREM = 1,906,502, N_waking = 3,415,232) from all channels (N = 273 cortical, N = 28 hippocampal) from all SEEG patients (S1-S17). Values are mean ± SD. Circles represent medians. Horizontal lines indicate means. Vertical lines indicate interquartile ranges.

Figure 5.

Cortical ripple characteristics by patient. Cortical ripple characteristics are consistent across patients (S1-S17). Horizontal lines indicate medians. Boxes represent interquartile ranges. Whiskers represent 1.5 × interquartile range.

Figure 6.

Characteristics of ripples detected on interictal-free channels. A, NREM and waking cortical ripple density, peak 70-100 Hz analytic amplitude, oscillation frequency, duration, and percent change in mean >200 Hz analytic amplitude in interictal-free cortical channels (N = 232 channels; Patients S1-S17). B, Same as in A, but with interictal-free hippocampal channels (N = 5 channels; Patients S4, S6, S9, and S17).

Figure 7.

Cortical ripple characteristics versus estimated hippocampo-cortical connectivity density. Cortical ripple density, peak 70-100 Hz analytic amplitude, oscillation frequency, duration, and >200 Hz amplitude modulation (N = 273 channels from Patients S1-S17) as a function of the connectivity density (Rosen and Halgren, 2021) between the hippocampus and each cortical parcel. The absence of significant correlations suggests that cortical ripples are related to corticocortical integration rather than being driven primarily by the hippocampus.

Figure 8.

Distributions of ripple characteristics across the cortex and hippocampus in NREM and waking. A-F, Cortical maps with hippocampal map insets of channel coverage (A) as well as NREM and waking mean ripple densities (B), peak 70-100 Hz analytic amplitudes (C), oscillation frequencies (D), durations (E), and changes in mean >200 Hz analytic amplitude during ripples compared with baseline (−2 to −1 s) (F) (SEEG Patients S1-S17; N = 273 channels). Left and right hemisphere channels were mapped onto a left hemisphere template. The parcellation scheme (specified in Table 2) is low resolution, and more extensive sampling may detect more spatially differentiated response characteristics. Values for each cortical region are reported in Table 3. G, H, Average cortical ripple density (G) and amplitude (H) by channel were significantly correlated with cortical parcel myelination index (i.e., lower densities and amplitudes in association compared with primary areas) (Rosen and Halgren, 2021) during NREM and waking (linear mixed-effects models with patient as random effect, FDR-corrected p values). Colors represent individual patients. I, Same as in G, H, but with change in mean >200 Hz amplitude z scores, for which there was a significant correlation during NREM but not waking.

Figure 9.

Cortical ripple oscillation frequency and duration versus myelination index. Neither cortical ripple oscillation frequency nor duration was significantly correlated with myelination index (Rosen and Halgren, 2021) during NREM or waking (linear mixed-effects models with patient as random effect, FDR-corrected p values, N = 273 channels from Patients S1-S17). Colors represent individual patients. Notably, higher myelination indices correspond to primary, whereas lower myelination indices correspond to associative cortical regions.

Figure 10.

Cortical ripples occur during sleep spindles on the down-to-upstate transition. A, Example cortical ripple occurring during a spindle on a down-to-upstate transition during NREM. B, Times of downstate peaks plotted relative to local cortical ripple centers at t = 0 across significant channels (N = 258/273, Patients S1-S17) during NREM. Downstate maxima occurred on average 450 ms before the cortical ripple center. Dashed red lines indicate 99% CI of the null distribution (200 shuffles/channel). C, Same as in B, but with spindle onsets (N = 80/273). Spindles began on average 225 ms before the cortical ripple center, indicating that ripples occurred during spindles (yellow shaded area represents average spindle interval of 634 ms). D, Same as in B, but with upstate peaks (N = 260/273). Upstate maxima occurred on average 100 ms after the cortical ripple center. E, Percent of channels with significant peri-ripple modulations of sleep waves detected on the same channels within ±1000 ms (e.g., US | R represents upstate peaks relative to cortical ripples at t = 0; one-sided randomization test, 200 shuffles, 50 ms nonoverlapping bins, 2 consecutive bins with post-FDR p < 0.05 required for significance). DS and US were significantly associated with ripples in ∼95% of channels, sleep spindles less frequently. F, Percent of channels with significant modulations that had significant sidedness preference around t = 0 (post-FDR p < 0.05, one-sided binomial test, −1000 to −1 ms vs 1-1000 ms, expected = 0.5). G, Percent of channels with significant sidedness preference that had cortical ripples leading the other sleep waves (according to counts in −1000 to −1 ms vs 1-1000 ms). Downstate peaks and spindle onsets typically preceded ripples, and upstate peaks followed. H, Probabilities of ripple centers preceding upstates, following downstates, occurring during spindles in isolation or during spindles following downstates (DS-SS), or during spindles preceding upstates (SS-US). The time window used following a downstate or preceding an upstate was 634 ms, which was the average spindle duration. The probability of a ripple occurring was greatest during spindles preceding upstates (post-FDR p < 0.0001, two-sided paired t test, channel-wise). Table 4 contains results from E-G in tabular format. DS, Downstate; SS, sleep spindle; US, upstate.

Figure 11.

Pyramidal leads interneuronal cell firing at cortical ripple peaks. A, Superior temporal gyrus granular/supragranular layer ripples detected during NREM in a Utah Array recording. Top, Average broadband LFP. Middle, Average time-frequency. Bottom, Single-trial example trace in broadband LFP (black) and 70-100 Hz bandpass (blue, left represents analytic amplitude; right represents sweep). B, Mean broadband LFP locked to cortical ripple centers (black) and associated local PY (blue; N = 69) and IN (red; N = 23) spike rates during NREM. C, Circular mean 70-100 Hz phase of spikes of each PY (N = 47, mean = 2.97 rad, p = 5 × 10⁻²¹, Rayleigh test) and IN (N = 22, mean = 3.25 rad, p = 3 × 10⁻¹³) during local cortical ripples (minimum 30 spikes per unit). PY spiking preceded IN spiking by 0.28 rad (p = 0.02, Watson–Williams test). The 0 rad corresponds to the trough, and π rad corresponds to the peak of the ripple. D, Circular mean ripple phase-lags of spikes from each PY (N = 47) to each IN (N = 22) (N = 1034 unit pairs, mean = 0.31 ± 0.63 rad, p = 4 × 10⁻⁵¹, one-sided one-sample t test). E, Pyramidal Interneuron Network Gamma ripple generation mechanism consistent with single-unit recordings, animal studies (Stark et al., 2014), and modeling (Buzsáki and Wang, 2012). Abrupt depolarization causes synchronous PY and IN firing, which then spike rhythmically, separated by fixed intervals because of recurrent inhibition. F, Pairs of PYs (PY–PY) cofire ∼7 times more often within 5 ms of each other during ripples compared with randomly selected epochs in between ripples matched in number and duration to the ripples. Similarly, PY–IN pairs cofire ∼10 times more during ripples. IN, Putative interneuron unit; PY, putative pyramidal unit.