Sleep stages and sleep rhythms in 107 subjects.

a, Sleep rhythms and task schematic. Subjects slept first half of the night with simultaneous EEG-fMRI recordings. Since detecting hippocampal ripples directly from scalp EEG is challenging, our focus was on capturing SOs, spindles, and their couplings. Regions of interest (ROIs) are colour-coded: green for the thalamus (spindle), purple for the mPFC (SOs), and orange for the hippocampus (ripples). b, Sleep staging and EEG spectrogram. N2/3 sleep stages (red line) were initially identified using an offline automatic sleep staging algorithm (Vallat & Walker, 2021) and then manually validated. (c) Schematic of EEG data across different sleep stages, using preprocessed data from the C3 electrode. (d) Proportion of each sleep stage in the dataset. (e) Amplitudes (μV) of SOs (left) and spindles (right) across sleep stages. The SO amplitudes used here are from 0.16-1.25 Hz band EEG data, and the spindle amplitudes used here are from 12-16 Hz band EEG data. Each dot represents an individual participant. Error bars indicate SEM. *** p < 0.001.

Sleep rhythms and SO-spindle coupling.

a, The SO-spindle coupling in the temporal frequency domain. The upper two rows illustrates the spindle (12-16 Hz) phase-locked in the transition to UP-state of SO (0.16-1.25 Hz). The bottom row shows the averaged temporal frequency pattern across all instances of SO-spindle coupling and over all subjects. b, SO-spindle coupling density across different sleep stages. c, Differences between coupled and uncoupled sleep rhythms. The left panel shows the difference in amplitude between spindles coupled with SOs (Coupling) and spindles not coupled with SOs (Other). The right panel displays the difference in amplitude between SOs coupled with spindles (Coupling) and SOs not coupled with spindles (Other). d, Phase modulation of SO-spindle coupling. Spindle peaks cluster slightly before the UP-state peak of SO (i.e., 0°), where –π/2 reflects the transition from DOWN to UP-state. The histogram represents the distribution of coupling directions across all subjects, with the red line showing the mean. Coupling phases for each subject are plotted on the circle, with coupling strength color-coded. e, Distribution of spindle peaks on the SO phase during all SO-spindle coupling events across participants. The distribution is represented by a probability density function, and the density is evaluated at 100 equally spaced points covering the data range. Each dot represents data from an individual subject. Error bars indicate the SEM. * p < 0.05, ** p < 0.01, *** p < 0.001.

Brain-wide activation associated with sleep rhythms.

a, Simultaneous EEG-fMRI analysis framework for detecting brain-wide activation during sleep rhythms. Detected SOs, spindles, and their coupling were convolved with the hemodynamic response function (HRF) and downsampled to match fMRI temporal resolution. These events formed the design matrix for the general linear model (GLM) analysis of fMRI activity during sleep, linking the EEG-derived timing of sleep rhythms to the corresponding brain responses in fMRI. b, Brain-wide activation associated with SOs. The upper row illustrates SOs, and the lower row shows the fMRI activation pattern during SO events, whole-brain family-wise error (FWE) corrected at the cluster level (p < 0.05) with a cluster-forming voxel threshold of p < 0.001. c, Brain-wide activation associated with spindles. Same as panel b, but for spindle events. d, Brain-wide activation associated with SO-spindle coupling (compared to non-coupling events).

Functional connectivity changes during SO-spindle coupling.

a, The PPI analysis framework for detecting brain-wide connectivity changes during SO-spindle coupling. This starts by setting a specific ROI (e.g., the hippocampus) as the seed to extract the BOLD signal (physiological condition) and using identified SO-spindle coupling events as the psychological condition to compute the interaction term. The design matrix includes the main effects of the physiological and psychological conditions, along with their interaction. This analysis examines whether whole-brain communication with the hippocampus changes as a function of SO-spindle coupling. b, Hippocampus-based functional connectivity with the whole brain (main effect of hippocampus BOLD signal in PPI analysis). The hippocampus ROI is bilateral, anatomically defined (bottom, orange colour). Brain-wide connectivity is shown with whole-brain FWE correction at the cluster level (p < 1e-7) with a cluster-forming voxel threshold of punc. < 0.001 for visualization purpose. c, Same with panel b, but based on thalamus (bilateral anatomically defined ROI). d, Same with panel b, but based on the mPFC (bilateral functionally defined ROI, detailed in Methods). e, Functional connectivity changes during SO-spindle coupling for hippocampus-based (left bottom, orange colour), thalamus-based (middle, green colour), and mPFC-based (right bottom, purple colour) connectivity. The results of ROI analysis for each direction are shown on the arrows. * p < 0.05, ns., not significant. Abbreviations: FC - functional connectivity, PPI - psychophysiological interaction.

Removal of MRI gradient noise from simultaneous collected EEG data.

a, Time series of both raw and preprocessed EEG data. The top row depicts the raw EEG data, which contains noise primarily from the MRI gradient magnetic field and electrocardiographic artifacts. The bottom row showcases the preprocessed EEG data (detailed in Methods). b, Power spectral density of the raw and preprocessed EEG data estimated by the fast Fourier transform. The raw EEG data is shown in the top row, while the preprocessed EEG data is in the bottom row. c, Time-frequency spectrogram of the raw and preprocessed EEG data, by the short-time Fourier transform. The top row represents the raw EEG data, and the bottom row displays the preprocessed EEG data.

ERPs of SOs and spindles coupling during different sleep stages across all 107 subjects.

a. ERP of SOs in different sleep stages using the broadband (0.1–30 Hz) EEG data. We align the trough of the DOWN-state of each SO at time zero (see Methods for details). The orange line represents the SO ERP in the N1 stage, the black line represents the SO ERP in the N2&N3 stage, and the green line represents the SO ERP in the REM stage. b. ERP of spindles in different sleep stages using the broadband (0.1–30 Hz) EEG data. We align the peak of each spindle at time zero (see Methods for details). The color scheme is the same as in panel a.

ERP and time-frequency patterns of SO-spindle coupling in the N1 stage.

The averaged temporal frequency pattern and ERP across all instances of SO-spindle coupling, computed over all subjects, following the same procedure as in Fig. 2a, but for N1 stage.

ERP and time-frequency patterns of SO-spindle coupling in the REM stage.

The averaged temporal frequency pattern and ERP across all instances of SO-spindle coupling, computed over all subjects, again following the same procedure as in Fig. 2a, but for REM stage.

Brain-wide activity between SO UP-state (peak) and DOWN-state (trough).

a, Brain activity with SO DOWN-state (trough) modelled as event onset, whole-brain FWE corrected at the cluster level (p < 0.05) with a cluster-forming voxel threshold of punc. < 0.001. b, Brain activity with SO UP-state (peak) modelled as event onset. c, Differences in brain activity corresponding to SO UP-state and SO DOWN-state. The whole-brain results were displayed at an uncorrected threshold of p < 0.01 for visualization purpose only. No brain region was found significant in this contrast.

Influence of the percentile threshold for SO detection on hippocampal activation (ROI) during SO-spindle coupling.

We changed the percentile threshold for SO event detection in the EEG data analysis and then reconstructed the GLM design matrix based on the SO events detected at each threshold. The brain-wide activation pattern of SO-spindle couplings in the N2/3 stage was extracted using the same method as shown in Fig. 3. The gray horizontal line represents the significant range (71%–80%). * p < 0.05.

Functional decoding using the ROI association method in Neurosynth.

a, Decoding results using positive activation. b, Decoding results using negative activation. Each row corresponds to the brain-wide activation patterns for sleep rhythms shown in Fig. 3b-d, while each column corresponds to topics in the Neurosynth database (detailed in Methods). Only topics with a decoded significance level of p < 0.05 are displayed.

Descriptive results of demographic information and sleep characteristics.

Note: The total recorded time is equal to the awake time plus the total sleep time. The sleep onset latency is the time taken to reach the first sleep epoch. The Sleep Efficiency is the ratio of actual sleep time to total recording time.

Statistics of sleep duration, SO, spindle and coupling event numbers and densities for all 107 subjects during N1 stage.

Statistics of sleep duration, SO, spindle and coupling event numbers and densities for all 107 subjects during N2&N3 stage.

Statistics of sleep duration, SO, spindle and coupling event numbers and densities for all 107 subjects during REM stage.

Peak and significant cluster of fMRI activity during SO main effect.

We used the SO main effect whole-brain activation patterns in Fig. 3b. ROIs were defined anatomically (see Methods). Cluster sizes are reported punc. < 0.001.

Peak and significant cluster of fMRI activity during spindle main effect.

We used the Spindle main effect whole-brain activation patterns in Fig. 3c. ROIs were defined anatomically (see Methods). Cluster sizes are reported punc. < 0.001.

Peak and significant cluster of fMRI activity during SO-spindle interaction.

We used the SO-spindle interaction effect whole-brain activation patterns in Fig. 3d. ROIs were defined anatomically (see Methods). Cluster sizes are reported punc. < 0.001.