The Degree of Nesting between Spindles and Slow Oscillations Modulates Neural Synchrony

Abstract

Spindles and slow oscillations (SOs) both appear to play an important role in memory consolidation. Spindle and SO "nesting," or the temporal overlap between the two events, is believed to modulate consolidation. However, the neurophysiological processes modified by nesting remain poorly understood. We thus recorded activity from the primary motor cortex of 4 male sleeping rats to investigate how SO and spindles interact to modulate the correlation structure of neural firing. During spindles, primary motor cortex neurons fired at a preferred phase, with neural pairs demonstrating greater neural synchrony, or correlated firing, during spindle peaks. We found a direct relationship between the temporal proximity between SO and spindles, and changes to the distribution of neural correlations; nesting was associated with narrowing of the distribution, with a reduction in low- and high-correlation pairs. Such narrowing may be consistent with greater exploration of neural states. Interestingly, after animals practiced a novel motor task, pairwise correlations increased during nested spindles, consistent with targeted strengthening of functional interactions. These findings may be key mechanisms through which spindle nesting supports memory consolidation.SIGNIFICANCE STATEMENT Our analysis revealed changes in cortical spiking structure that followed the waxing and waning of spindles; firing rates increased, spikes were more phase-locked to spindle-band local field potential, and synchrony across units peaked during spindles. Moreover, we showed that the degree of nesting between spindles and slow oscillations modified the correlation structure across units by narrowing the distribution of pairwise correlations. Finally, we demonstrated that engaging in a novel motor task increased pairwise correlations during nested spindles. These phenomena suggest key mechanisms through which the interaction of spindles and slow oscillations may support sensorimotor learning. More broadly, this work helps link large-scale measures of population activity to changes in spiking structure, a critical step in understanding neuroplasticity across multiple scales.

Keywords: correlation; sleep; slow oscillations; slow waves; spiking; spindle.

Figure 1.

Recording setup and sleep oscillation detection. A, Sleeping rat along with anatomic location of multielectrode arrays and example LFP and spiking data from one recording channel. B, Example of sleep classification in both power spectral density (PSD; top) and temporal spaces. Each dot represents the PSD in the δ (0.1-4 Hz) and γ (30-60 Hz) frequency bands during 6 s windows. A k-means classifier was used to cluster epochs into two clusters: Awake/REM (blue) and NREM (orange). The average LFP trace is plotted across a 2 h sleep session (blue). Identified sleep epochs are highlighted with orange boxes. C, Examples of detected sleep oscillations, highlighting the automatic methods used for detection. The broadband LFP (top) is decomposed into a lower-frequency band (middle) and spindle band (bottom) components. SOs must have had sufficient positive and negative amplitudes (black lines; middle) and sufficiently long durations (highlighted blue; middle). The spindle band envelope (purple line; bottom) must have exceeded an upper threshold (solid black line; bottom) for one sample and a lower threshold (dashed black line; bottom) for at least 500 ms. Purple represents the detected spindle duration. D, Average spindle-triggered waveform (black line; right) and spectrogram (heat map; right). Average spindle-triggered power spectrum (black line; left) and baseline power spectrum (dashed line; left). Solid and dashed boxes on the heat map represent the timing of the spindle and baseline periods used to calculate the power spectra.

Figure 2.

Spindle modulation of spiking. A, An example of the average waveform for a single unit recorded across many channels using a custom polytrode probe. Inset, Spike waveforms (gray) on one channel of many spiking events along with the average waveform (red). This example unit is used for all panels in this figure. B, Spindle-triggered spiking for the example unit. The average spindle waveform (blue) is plotted with the average normalized firing rate (orange). A raster of spike times is displayed below. C, Phase extraction methods. Spindle band (10-16 Hz) LFP is plotted during a detected spindle (top) and is replotted immediately below (second from the top) with a finer time resolution. The Hilbert phase of the spindle activity is plotted below (second from the bottom) and fluctuates between –π and π. The spiking activity is displayed on the bottom (bars) and as dots in the spindle band and phase subplots. The highlighted portion of these plots extends from –2π to 2π and shows spikes that are used to compute the phase-locking value (C,D). D, The spike phase distribution for the example unit, plotted as a circular histogram (blue). The average phase vector is overlaid in red and copied below for clarity. The magnitude and direction of this vector are defined as the phase-locking value and preferred spindle phase, which are collected for all units. E, Summary of the phase-locking value (bottom) and preferred spindle direction (top) for all units. Gray bars (bottom) represent the phase-locking to the spindle band during control epochs. *Significant difference between the distributions.

Figure 3.

Spindle cycle analysis of phase-locking. A, Generation of the spike phase distribution across spindle cycles. Spike-triggered phases are extracted from single-unit spiking during specific spindle cycles. Phases are aggregated across actual spindle epochs (right) or control epochs (left). Blue represents analyses of actual spindles. Black represents analyses of control spindle. B, Summary of all neurons' spike phase distribution statistics across spindle cycles. The average spindle band waveform for each spindle cycle (top) is plotted along with the average spike rate (second from top), preferred spindle direction (second from bottom), and phase-locking value (bottom). Blue lines indicate averages during actual spindle epochs. Black/gray lines indicate averages during control epochs. Error bars indicate SEM. C, Summary of spiking dynamics. Spike rates, preferred spindle directions, and phase-locking values are combined into three categories: (1) CTRL, the two cycles at the center of the control epochs; (2) TAIL, the two cycles farthest from the spindle peaks; and (3) PEAK, the two cycles nearest the spindle peaks. Bar plots represent the newly categorized data. *Significant differences between the categories. D, Relationship between phase-locking and spike count across neurons. Each neuron's phase-locking value and spiking rate in the PEAK are shown as a scatter plot (left) along with each neuron's change in phase-locking value and spiking rate (PEAK – TAIL; scatter plot, right). Regression lines are superimposed on the scatter plots along with Pearson correlation coefficients and associated p values.

Figure 4.

Spindle cycle analysis of synchrony. A, Generation of the CCH. For a pair of units, relative spike times are extracted during each spindle cycle. Spike times are then aggregated across spindles relative to one another. Additionally, a shuffled CCH is constructed using a similar procedure. Relative spike times are extracted during specific spindle cycles, but one neuron's spikes are shuffled across spindle epochs. B, CCH correction. Within each spindle cycle, the shuffled CCH is subtracted from the raw CCH to generate a corrected CCH. The peak, and time of peak (red arrows) are collected for each pair of neurons in each spindle cycle. C, Summary of all neurons' corrected CCH statistics across spindle cycles. The average spindle band waveform for each spindle cycle (top) is plotted along with the average corrected CCH peak (middle) and time of peak (bottom). Blue lines indicate averages during actual spindle epochs. Black/gray lines indicate averages during control epochs. Error bars indicate SEM. D, The distribution of peak cofiring probability for controls (black) and spindles (purple) plotted as a CDF (top). Gray lines divide the distributions into sextiles (top). The difference in sextile dividers is plotted with 95% CIs as error bars (bottom). E, The peak and time of peak are reproduced for the same categories as in Figure 3D. *Significant difference between the distributions.

Figure 5.

Effect of nesting. A, Timing of spindles relative to SOs. B, Example of binning spindles based on timing relative to SO in 1 animal. Bar graph represents the distribution of time delays. Windows representing the spindle bin closest to (dark blue rectangle), and farthest from (light blue rectangle) SOs, are overlaid. Blue dotes represent intermediate window starts. The color gets lighter as the window start gets farther from the SOs. C, The distribution of spindle-induced pairwise correlations for each bin. Colors match the window colors in B. The average pairwise correlation across bins is plotted with a linear fit (left inset), and the SD of pairwise correlations across bins in plotted with a linear fit (right inset). D, The distributions of spindle-induced pairwise correlations are reproduced for the bins closest to SOs (0, 1.5) and farthest from SOs (4.5, 6) (top). Gray lines divide the distributions into sextiles (top). The difference in sextile dividers is plotted with 95% CIs as error bars (bottom). *Significant difference between the distributions.

Figure 6.

Effect of novel motor task engagement on nested spindles. A, Description of timeline. Each rat's sleep was monitored before (Sleep₁, green) and after (Sleep₂, orange) engaging in a novel motor task. B, Nested spindle rates for each rat plotted as a function of Sleep₁ and Sleep₂. Each dot indicates spindles detected on a different electrode. Dots above the diagonal indicate an increase in the nested spindle rate. C, The distribution of nested spindle-induced pairwise correlations for Sleep₁ and Sleep₂ (top). Gray lines divide the distributions into sextiles (top). The difference in sextile dividers is plotted with 95% CIs as error bars (bottom). *Significant difference between the distributions.

Figure 7.

Summary of spindle-induced correlations and interactions with SO and novel motor engagement. A, Idealized distributions of pairwise correlations during control epochs (black) and spindle epochs (blue). Gray lines indicate sextile dividers. Blue arrows indicate a rightward shift in the distribution of correlations across all sextiles during spindles. B, Idealized distributions of spindle-induced pairwise correlations near SO (dark blue) and far from SO (light blue). Gray lines indicate sextile dividers. Dark blue arrows indicate a tightening of the distribution of correlations for spindles nearer to SO. C, Idealized distributions of spindle-induced pairwise correlations pre-engagement (green) and post-engagement (orange) in a novel motor task. Gray lines indicate sextile dividers. Orange arrows indicate a rightward shift in the middle of the distribution of correlations.