Sleep spindles in humans: insights from intracranial EEG and unit recordings

Abstract

Sleep spindles are an electroencephalographic (EEG) hallmark of non-rapid eye movement (NREM) sleep and are believed to mediate many sleep-related functions, from memory consolidation to cortical development. Spindles differ in location, frequency, and association with slow waves, but whether this heterogeneity may reflect different physiological processes and potentially serve different functional roles remains unclear. Here we used a unique opportunity to record intracranial depth EEG and single-unit activity in multiple brain regions of neurosurgical patients to better characterize spindle activity in human sleep. We find that spindles occur across multiple neocortical regions, and less frequently also in the parahippocampal gyrus and hippocampus. Most spindles are spatially restricted to specific brain regions. In addition, spindle frequency is topographically organized with a sharp transition around the supplementary motor area between fast (13-15 Hz) centroparietal spindles often occurring with slow-wave up-states, and slow (9-12 Hz) frontal spindles occurring 200 ms later on average. Spindle variability across regions may reflect the underlying thalamocortical projections. We also find that during individual spindles, frequency decreases within and between regions. In addition, deeper NREM sleep is associated with a reduction in spindle occurrence and spindle frequency. Frequency changes between regions, during individual spindles, and across sleep may reflect the same phenomenon, the underlying level of thalamocortical hyperpolarization. Finally, during spindles neuronal firing rates are not consistently modulated, although some neurons exhibit phase-locked discharges. Overall, anatomical considerations can account well for regional spindle characteristics, while variable hyperpolarization levels can explain differences in spindle frequency.

Figure 1.

Data overview and spindle detection. A, Illustration of flexible probes used for simultaneous recording of depth EEG (platinum contacts, blue) and unit activity (microwires, green). B, Overview of 129 depth electrode locations in 13 individuals spanning multiple brain regions as seen from medial view. SM, Supplementary motor; PC/P, posterior cingulate/parietal cortex; PH, parahippocampal gyrus; HC, hippocampus; E, entorhinal cortex; Am, amygdala; LH, left hemisphere; RH, right hemisphere. C, Example power spectra of scalp EEG in one representative individual in stage N2 sleep (blue), N3 sleep (red), and REM sleep (green). Note high power in slow-wave (<4 Hz) and spindle (9–16 Hz) range in NREM sleep. D, Hypnogram: time course of sleep–wake stages in the same individual. W, Wake; R, REM sleep; N1–N3, NREM sleep, stages 1–3. E, Spindle detection step 1: channels with spindle activity in NREM sleep are chosen for further analysis. Spectral profiles of a typical selected channel (parietal lobe, top) and a typical discarded channel (amygdala, bottom). For each channel, power in NREM sleep (red) was compared with a 1/f^α model (blue), and channels with a significant peak (asterisks) in the range of 9–16 Hz are considered for further analysis. Yellow dots denote frequency with maximal power difference and error bars denote SEM across 10 s epochs of NREM sleep. F, Spindle detection step 2: individual spindles were selected based on their power and duration. Raw EEG (top row) is filtered in the spindle range (9–16 Hz, second row). The instantaneous amplitude is extracted via the Hilbert transform (red trace, third row). A detection threshold is set at mean + 3 SD of spindle power across NREM sleep (horizontal black line, fourth row) and peaks exceeding this threshold (red dot) are considered putative spindles. A start/end threshold is set at mean + 1 SD of spindle power across NREM sleep (horizontal green line, bottom row), thereby defining start and end times (cyan and green dots, respectively) and determining spindle duration (between 0.5 and 2 s). G, Spindle detection step 3: channels in which power increases of detected events were specific to the spindle range rather than broadband were chosen for final analysis. Spectral profiles of detected events in a typical selected channel (parietal lobe, top) and a typical discarded channel (entorhinal cortex, bottom). For each channel, power of detected events (red) is compared with power of random 1 s segments (blue), and channels with a significant peak which is specific to the spindle range (9–16 Hz) are chosen for final analysis. Note that detected events in entorhinal cortex had a diffuse broadband power increase. Yellow dots and error bars as above.

Figure 2.

Fast centroparietal spindles differ from slow frontal spindles. A, Average frequency of spindles across all depth electrodes (n = 50 electrodes in 13 individuals). The color of each circle denotes the mean spindle frequency in an individual electrode according to its precise anatomical location. Green outlines mark electrodes placed more laterally than the midline. Note the contrast between slow (9–12 Hz) frontal spindles and fast (13–16 Hz) centroparietal spindles. The two outliers in the medial prefrontal cortex (red circles) were the only electrode placements in one atypical individual in whom parietal spindles may be even faster. Numbers as in legend of B. B, Distribution of spindle frequencies grouped by region. Slow spindles are found in prefrontal regions (1–3) while fast spindles are found in centroparietal regions (4–5). Note the difference in spindle frequency between SMA and adjacent pSMA. C, In each brain region, spectral frequency decreases during individual spindles. Spectrogram shows mean frequency dynamics in a representative centroparietal electrode in the SMA (left, n = 159 spindles) and in a frontal electrode in the OF (right, n = 237 spindles). Black crosses mark the mean instantaneous frequency around the beginning and end of spindles (13.8 and 12.5 Hz for fast spindles, 11.5 and 10.3 Hz for slow spindles). Note that frequency decreases during fast spindles and to a lesser degree during slow spindles.

Figure 3.

Fast centroparietal spindles precede slow frontal spindles. A, Four examples of individual spindles in three individuals showing depth EEG along with corresponding spectrograms in the spindle range (9–16 Hz) during 4 s of NREM sleep. When spindles occur in multiple brain regions, fast centroparietal spindles precede slow frontal spindles. B, Quantitative analysis of time offsets in spindle occurrence. A graph showing the order in which spindles are detected across multiple regions (node color) and the mean temporal delays between each pair of regions (edge color). Mean order and timing across spindles for all individuals (n = 12) indicate that centroparietal spindles precede frontal spindles. Abbreviations as in Figure 1. C, An example of differences in timing and frequency between PC and AC in one individual. Spectrograms are aligned to the mean of peak power in AC and PC spindles. Note that AC shows a hint of an early fast component aligned with peak power in PC while its peak is at lower frequencies and delayed by >500 ms. D, Frequency decreases (y-axis) and temporal delays (x-axis) are closely related in individual spindles (r = −0.90, n = 9741 spindles in 50 electrode pairs).

Figure 4.

Association of spindles with slow-wave up-states. A, Example of a single spindle associated with a slow-wave up-state as recorded in the depth EEG of the anterior cingulate cortex. B, Number of detected spindles around slow-wave down-states (left, positive peaks in depth EEG) and up-states (right, negative peaks in depth EEG). Blue traces show average slow waves (n = 13 individuals). Red shading, SEM across channels; green horizontal lines, confidence interval (α = 0.05). Thicker green bars, significant deviations from chance. Note that around slow waves, spindles occur more often after transition to the up-state. C, Regional differences in association of spindles with slow-wave up-states. Note that fast centroparietal spindles (regions 4–5, red-brown colors) are more tightly associated with slow-wave up-states. Asterisks mark the average time to up-states for each region.

Figure 5.

Local sleep spindles. A, Two examples in two different individuals of depth EEG along with corresponding spectrograms in the spindle frequency range (9–16 Hz) during 15 s of slow-wave sleep. Note that regardless of slow waves, local spindles robustly occur without spindle activity in other regions, including homotopic regions across hemispheres and regions with equivalent signal-to-noise ratio showing the same global slow waves. B, Quantitative analysis of spindle occurrence across pairs of channels. Top row (concordant spindles, 45% of cases) shows spectrograms for cases in which a spindle was detected in the seed channel as well as in the target channel (n = 27,338 in 170 pairs of regions in 12 individuals). Bottom row (non-concordant spindles, 55% of cases) shows spectrograms for cases in which a spindle was detected in the seed channel but not in the target channel (n = 32,797 in 170 pairs of regions in 12 individuals). Note that spindle power in non-concordant target channels is at near chance levels, indicating that our detection can reliably separate between presence and absence of spindle activity. C, Most sleep spindles are local. Distribution of involvement (percentage of monitored brain structures expressing each spindle) for all spindles (blue bars, n = 21,240 spindles in 49 electrodes of 12 individuals) and for isolated spindles (red bars, spindles without a slow wave within ± 1.5 s, 37% of events). Note that 73% of spindles are observed in <50% of electrodes indicating that most spindles are local. D, Scatter plot of spindle amplitudes as a function of involvement shows that global spindles have some tendency to be of higher amplitude (r = 0.52, p < 0.0001, n = 170).

Figure 6.

Deep sleep and high SWA are associated with lower spindle frequencies. A, Frequency of slow spindle changes between early and late NREM sleep (n = 5 individuals, see Materials and Methods). Left, Frequency changes in centroparietal channels (black bars) and frontal channels (red bars). Right, Frequency changes for all channels (blue), fast spindles in centroparietal channels (black), slow frontal spindles (red) and occasional slow spindles in centroparietal channels (green). B, Example of time course of SWA and spindle frequency dynamics throughout sleep in the anterior cingulate of one individual. Note that within NREM cycles, spindle frequency is lowest when SWA is highest (vertical purple bars) and increases toward transitions to REM sleep. C, Quantitative analysis across the entire dataset (n = 8 individuals) comparing sleep depth (as measured by SWA, blue) with spindle frequency (red) and density (green). Note that deep sleep (high SWA on the right) is accompanied by fewer spindles (r = −0.73, p = 0.02) with lower frequencies (r = −0.81, p = 0.005). Error bars denote SEM across NREM sleep intervals (n = 16, see Materials and Methods).

Figure 7.

Unit discharges during sleep spindles. A, Spindle-triggered averaging of unit spiking activity (n = 207 units), time locked to the middle point of the spindles in depth EEG (n = 40 channels). Left, Spindles occurring within 500 ms following up-states (8.5% of events). Right, All spindles. Red shades denote SEM across neurons. Black traces below bar graphs show the average depth EEG at those times. In contrast to robust firing rate modulations around transitions to UP states, firing rates modulations are largely absent during spindles. B, Example of phase-locking analysis for one neuron and one spindle. For each detected spindle, raw depth EEG (blue trace) is filtered in the spindle range (9–16 Hz, cyan trace). The instantaneous phase is computed (green trace) and phase values are extracted for each action potential (red bars). C, Phase-locking of spikes for the same unit displayed in A across all spindles (real distribution) and for spikes randomly shuffled within each spindle (shuffled spikes). A unit is considered phase-locked only if the Rayleigh p-value for its real distribution is smaller than the critical p-value defined as the 95th percentile across shuffled iterations. D, Cumulative histogram of preferred firing phase for all phase-locked neurons (n = 41/212, 20% in 12 individuals). Note the tendency of phase-locked units to fire between the negative and positive peaks of spindles in the depth EEG.