Abstract
Temporal nesting of cortical slow oscillations, thalamic spindles, and hippocampal ripples indicates multiregional neuronal interactions required for memory consolidation. However, how the thalamic activity during spindles organizes hippocampal dynamics remains largely undetermined. We analyzed simultaneous recordings of anterodorsal thalamus and CA1 in male mice to determine the contribution of thalamic spindles in cross-regional synchronization. Our results indicated that temporal hippocampo-thalamocortical coupling was more enhanced during slower and longer thalamic spindles. Additionally, spindles occurring closer to slow oscillation trough were more strongly coupled to ripples. We found that the temporal association between CA1 spiking/ripples and thalamic spindles was stronger following spatial exploration compared with baseline sleep. We further developed a hippocampal-thalamocortical model to explain the mechanism underlying the duration and frequency-dependent coupling of thalamic spindles to hippocampal activity. Our findings shed light on our understanding of the functional role of thalamic activity during spindles on multiregional information transfer.
SIGNIFICANCE STATEMENT The contribution of thalamic spindles with differential properties to cross-regional synchronization and information transfer still remains poorly understood. Using simultaneous anterodorsal thalamic and hippocampal recordings from naturally sleeping mice before and after exploration, we found strong coupling of CA1 units to anterodorsal thalamic spindles and increase of this coupling following spatial experience. We further showed that the temporal coupling of CA1 units and hippocampal ripples with thalamic spindles and the spindle-associated modulation of CA1 units with ripples were stronger for spindles with slower frequency of oscillations. Our experimental as well as computational findings using a hippocampal-thalamocortical model provide the first demonstration that spindle frequency and duration can provide valuable information about the underlying multiregional interactions essential for memory consolidation computations.
- anterior thalamic spindles
- hippocampal ripples
- memory consolidation
- neural mass model
- phase amplitude coupling
Introduction
The precise temporal relationship of cortical slow oscillations (SOs), thalamic spindles, and hippocampal ripples, which are three main NREM rhythms, are believed to be important for memory consolidation during sleep (Niknazar et al., 2015; Maingret et al., 2016; Latchoumane et al., 2017; Helfrich et al., 2018). Especially the mediating role of thalamic induced spindles in the hippocampal-neocortical dialogue essential for memory processing has been proposed by previous studies (Diekelmann and Born, 2010; Fogel and Smith, 2011; Latchoumane et al., 2017; Ngo et al., 2020). However, how spindles with differential properties contribute to the multiregional communication still remains poorly understood. Although the spindles are generated in the thalamus and transferred to cortex via the thalamocortical projections (Timofeev and Bazhenov, 2005), the existing literature provides evidence for spindle-ripple coupling focused on the spindle activity detected from the cortex or hippocampus (Siapas and Wilson, 1998; Sirota et al., 2003; Mölle et al., 2006; Clemens et al., 2011; Staresina et al., 2015; Latchoumane et al., 2017; Helfrich et al., 2019; Jiang et al., 2019; Ngo et al., 2020; Varela and Wilson, 2020). Recent studies revealed modulation of the thalamic units to the hippocampal ripples (Yang et al., 2019; Varela and Wilson, 2020; Viejo and Peyrache, 2020), but the precise temporal association of CA1 units and thalamic spindles is still unknown. A recent study reported that spike-field coupling during spindles was stronger within anterodorsal (AD) thalamus compared with the one computed within barrel cortex (Bandarabadi et al., 2020). In addition, slow wave-spindle coupling was stronger when slow waves and spindles were detected from thalamic than cortical LFPs. Because of strong anatomic projections from the thalamus to the hippocampus, it is possible that thalamic spindles also drive the hippocampal activity directly. Hence, given the latency between the thalamic and cortical LFPs during the spindles (Mak-McCully et al., 2017) and their differential coupling to the unit activity (Bandarabadi et al., 2020), it is important to investigate the precise temporal coordination of hippocampal unit activity/ripples and the spindles detected from the thalamus. Here, we addressed this gap by characterizing the fine temporal coordination between thalamic spindles and hippocampal activity using simultaneous recordings of AD thalamus and CA1 (single units and local field potentials [LFPs]) in freely moving mice before and after spatial exploration. We specifically tested the hypothesis that the hippocampal-thalamocortical temporal coordination depends on the oscillating frequency and duration of the thalamic spindles as well as their locking phase to SOs. A wide frequency range of 7-15 Hz is typically considered for spindles in rodents. Within this frequency band in human sleep, slow and fast spindles with differential properties and functional relevance for memory consolidation are discriminated (Ayoub et al., 2013; Klinzing et al., 2016; Hashemi et al., 2019; Fernandez and Lüthi, 2020; Dehnavi et al., 2021). We hypothesize that the precise temporal modulation of CA1 units/ripples with spindles is frequency-dependent so that spindles with lower frequencies provide longer windows of opportunity for cross-regional synchronization and information transfer. We further developed a simplified hippocampal-thalamocortical neural mass model that can spontaneously generate SOs, slow/fast spindles, and hippocampal ripples to investigate the mechanisms underlying multiregional temporal interactions. Our findings provide evidence for the functional role of slow long-duration thalamic spindles in the hippocampal-thalamocortical interaction required for successful memory consolidation.
Materials and Methods
Mice
We used a publicly available dataset (Peyrache and Buzsáki, 2015; Peyrache et al., 2015) recorded in Professor Gyorgy Buzsáki's laboratory and taken from the public data sharing repository (http://crcns.org). This dataset consists of 18 recording sessions (4 male mice labeled Mouse 17, Mouse 12, Mouse 20, and Mouse 32, weighing ∼30 g, 3-6 months) each with 6 recording sites in CA1 (tungsten wires) and 64 recording site (8 shanks, separated by 200 μm) in anterior thalamus. Each session consists of successive epochs of baseline sleep (∼1.5 h), spatial exploration for ∼30 min, and postexploration-sleep (∼1.5 h). During the exploration, the animals foraged for food (sweetened cereals or regular food pellets) in an open environment. The environment was a circular arena surrounded by 21-cm-high, black-painted walls on which were displayed two salient visual cues.
Electrophysiological recordings
Electrophysiological signals were acquired at 20 kHz on a 256-channel Amplipex system (16-bit resolution) and were downsampled to 1.25 kHz for LFP analyses. Slow wave sleep (SWS) was detected using CA1 LFP spectrogram (Viejo and Peyrache, 2020) as periods with high 1-4 Hz and 10-15 Hz activity during sleep (a long period of immobility tracked with the LEDs). All the analyses were conducted during SWS. For classification of CA1 units as interneurons or pyramidal cells, the average waveform of each unit was first computed. Units with peak-to-trough time<0.4 ms and mean firing rate >5 Hz were classified as interneurons and units with peak-to-trough time >0.4 ms were classified as pyramidal cells (Ravassard et al., 2013).
Detection of sleep spindles, ripples, and SOs
The CA1 and thalamic LFPs were used to detect ripples and SO/spindles, respectively. Since in this study we sought to investigate the effect of spindle frequency, we detected spindles separately for 7 overlapping frequency ranges of 7-9, 8-10, 9-11, 10-12, 11-13, 12-14, and 13-15 Hz. The algorithm for detecting spindles was similar to the one used in a previous study (Klinzing et al., 2016). Thalamic LFP signal was first band pass filtered in the corresponding spindle frequency range using finite-impulse response (FIR) filters from the EEGLAB toolbox (Delorme and Makeig, 2004) (FIR bandpass filter, filter order corresponds to 3 cycles of the low-frequency cutoff). Hilbert transform was used to compute instantaneous amplitude, which was then smoothened using a 300 ms Gaussian window. Periods with amplitude >3 SDs for a duration >0.5 and<3 s were selected as spindle events. Two nearby events were merged if they were closer than 0.5 s. There was no significant difference between the density of the spindles with different frequencies (p > 0.1, Friedman test, median ± median absolute deviation [MAD], density = 0.26 ± 1.11 × 10–3 1/s for 7-9 Hz, 0.25 ± 1.01 × 10–3 1/s for 8-10 Hz, 0.25 ± 1.2 × 10–3 1/s for 9-11 Hz, 0.24 ± 1.03 × 10–3 1/s for 10-12 Hz, 0.24 ± 1.3 × 10–3 1/s for 11-13 Hz, 0.23 ± 1.05 × 10–3 1/s for 12-14 Hz, and 0.23 ± 1.5 × 10–3 1/s for 13-15 Hz, n = 18 sessions). On average, there was an overlap of 19.53 ± 1.25% (mean ± SD) between spindles at different frequency ranges. The time of the spindle event was defined as the time of the largest spindle peak. The algorithm for detecting ripples was similar to the one used in a previous paper (Levenstein et al., 2019). The CA1 LFP was first band pass filtered in the ripple frequency range (150-200 Hz, fourth-order zero-phase delay Butterworth). The RMS signal was calculated and smoothed using a Gaussian window (50 ms). A ripple event was identified when the smoothed RMS signal was higher than median + 4 SD for a minimum duration of 30 ms. The time of the ripple event was defined as the time of the largest ripple peak. The FMA toolbox (Zugaro et al., 2018) was used to detect SOs. First, the signal was filtered between 0.5 and 4 Hz (FIR bandpass filter, filter order corresponds to 3 cycles of the low-frequency cutoff). SOs were identified in the LFP when (1) the distance between consecutive positive-to-negative zero crossings was between 0.5 and 2 s (corresponding to 0.5-2 Hz), (2) positive peaks >2 SDs were identified as SO positive peaks, and (3) difference between the positive and negative peaks was >3.5 SDs.
Phase amplitude coupling (PAC)
The method to calculate PAC was similar to previous papers (Staresina et al., 2015; Dehnavi et al., 2021). First, time–frequency representations (TFRs) were calculated for every spindle event by mtmconvol function of the FieldTrip toolbox (Oostenveld et al., 2011) with frequency steps of 0.25 Hz in the frequency range of 5-300 Hz. Sliding (10 ms steps) Hanning tapered windows with a variable length, including 5 cycles were used to ensure reliable power estimates. TFRs of all epochs around an event were then normalized as percentage change from the pre-event baseline (−2.5 to −1.5 s). To quantify the modulation of hippocampal ripple amplitude with the phase of thalamic spindle, first TFR bins around each spindle were averaged across the ripple frequency ranges to obtain the power of ripple time series (−3 to 3 s around the spindle peak). To ensure proper phase estimation, both LFP and power of the ripple time series (obtained from TFR) were filtered in the spindle frequency range (two-pass FIR band pass filter, order = 3 cycles of the low-frequency cutoff). Next, we extracted the phase values of these time series using the Hilbert transform and defined a synchronization index (SI) for each spindle event as the vector mean of unit vectors each showing the phase difference between ripple amplitude and spindle event at different time points around the spindle peak as follows:
Phase-locking analyses
To find the phase of the spindles at the time of the unit spikes/ripples, first the instantaneous phase of thalamic LFP filtered in the spindle frequency ranges was estimated by Hilbert transform. Then the circular mean of the phases of the spikes occurring ± 0.25 s around the spindle peaks and phases of the ripples occurring between the spindle onset and offset were computed. To obtain the phase-locking of the ripples/spindles to SO, the circular mean of the instantaneous phases of thalamic LFP filtered in the SO frequency range at the time of the ripple/spindle was computed.
Modulation index (MI)
To examine neural spiking activity around spindles/ripples, peri-event spike histograms (PETHs) were generated ±2 s around each event and smoothed with a Gaussian window of 20 ms. Similar to a previous paper (Yang et al., 2019), the PETHs around the true events were z score normalized to the PETHs around surrogate events obtained by randomly distributing (repeated 100 times) the same number of events detected every 4 s in a given session. The time series of surrogate PETHs were subtracted from the corresponding values of the true PETHs, and the resulting PETHs were then z score normalized to the firing rate during the entire 4 s time window of the surrogate PETHs. To quantify modulation of neural spiking to the spindle/ripple events, an MI was defined as the difference between peak to trough firing rate in a 0.25 s window centered at 0 on the PETHs.
Event correlation histogram
To analyze the temporal relationships between spindles and ripples, event (ripple) correlation histograms were calculated within ±1.5 s window around the reference event (spindle onset) with a bin size of 100 ms. The event correlation histograms were then normalized to a 1-s pre-event interval (form −2.0 to −1.0 s before the reference event at 0 s). For statistical analysis, a randomization procedure similar to a previous paper (Mölle et al., 2011) was applied by randomizing the time points of all detected ripples (repeated 100 times). Next, for each session, the event correlation histograms for the randomized data were recalculated; and finally, each bin in the true event correlation histogram was compared with the corresponding bin in the random condition using two-tailed paired-samples t test.
Hippocampal-thalamocortical model
The neural mass hippocampal-thalamocortical model consists of two hippocampal networks representing CA1 and CA3 networks: one thalamic and two cortical networks. The networks were described in our previous papers (Ghorbani et al., 2012; Hashemi et al., 2019; Azimi et al., 2021). In short, each of cortical and hippocampal networks consists of one group of identical inhibitory (I) neurons (number of neurons:
Here,
Table 10-1
The values of the parameters of single neurons (for details, see Materials and Methods). t and r show TC and RE neurons. e (1) represents cortical (CX) or hippocampal (CA1/CA3) excitatory (inhibitory) neurons. Download Table 10-1, DOCX file.
The thalamic network consists of one group of identical excitatory neurons representing the thalamocortical neurons (TC) and one group of identical inhibitory neurons representing the thalamic reticular neurons (RE, number of neurons:
If
The within- and between-network connections are considered based on anatomic connections (Extended Data Table 10-2). The probability and strength of connections from neuron m of network k to the neuron n of network h are given by
Here
Statistical analysis
Unless otherwise stated, MATLAB (version 2017b) was used for all statistical analyses. Particularly, MATLAB CircStat toolbox (Berens, 2009) was used for all statistical analyses on circular data. The circular mean of m angels (
Table 10-2
The values of the probability and strength of connections from neuron m of network k to the neuron n of network h given by
Results
AD neurons were more strongly coupled to slower spindles
We used simultaneous recordings of AD thalamus and CA1 (single units and LFPs) in freely moving mice to investigate the mediating role of thalamic spindles in multiregional interactions during SWS. Thalamic and CA1 LFPs were used to detect spindles and ripples (150-200 Hz), respectively (Fig. 1A). During the peak of SOs detected from thalamic LFP, AD units decreased their activity (Fig. 1B). We hypothesized that spindles with lower frequencies provide a longer temporal window for hippocampal synchronized activity enhancing multiregional communication. To explore the covariation of the thalamic/hippocampal activity with the spindle frequency, we partitioned the broad spindle frequency range (7-15 Hz) into 7 overlapping frequency ranges of 7-9, 8-10, 9-11, 10-12, 11-13, 12-14, and 13-15 Hz (for details, see Materials and Methods). Figure 1C, D displays the average of spindle peak-locked time frequency representation for low-frequency (slow, 7-9 Hz) and high-frequency (fast, 13-15 Hz) spindles.
Thalamic spindles/SOs and hippocampal ripples during SWS. A, Sample unit spikes and LFP traces simultaneously recorded in AD and CA1. AD LFPs were used to detect SOs (marked above the 0.5-2 Hz filtered AD LFP by a red circle) and spindles (slow spindles: marked above the 7-9 Hz filtered AD LFP by blue bracket, fast spindles: marked above the 13-15 Hz filtered AD LFP by red bracket). Sharp-wave ripple (marked above the 150-200 Hz filtered CA1 LFP by magenta stars) were detected using CA1 LFP. B, Top, Representative average AD LFP traces aligned to the SO peaks for one recording session. Bottom, Peri-SO-peak histograms of AD spikes from one unit (n = 3161 SOs). C, Top, Grand average (±SEM) of slow (7-9 Hz) spindles time-locked to slow spindle peaks, respectively, for the same session (n = 1241 spindles) (B). Bottom, Average of slow spindle peak-locked TFR (percentage change from pre-event baseline). D, Same as in C, but for the fast (13-15 Hz) spindle (n = 1216 spindles).
To investigate the modulation of AD neurons with thalamic spindles, we computed the z score normalized AD unit activity during spindles (Fig. 2A,B; for details, see Materials and Methods). AD unit activity increased around the peak of spindles, which is consistent with a recent study (Bandarabadi et al., 2020). We next calculated the MI as the difference between the peak to trough of the curve of each peri-spindle-peak spike histograms ± 0.25 s around time 0. To show the covariation of the thalamic activity with the spindle frequency, we used Friedman test to compare the effect of spindle frequency factor on the MI. There was a significant effect of the spindle frequency indicating a stronger modulations of AD neurons to slower spindles (Fig. 2C, p = 2.47
Modulation of AD neurons to thalamic spindles. A, Top, Representative average AD LFP traces filtered in the slow spindle frequency range (7-9 Hz) aligned to the spindle peaks. Spike rasters (middle) and peri-slow spindle-peak histograms (bottom) of AD spikes from one unit (z score normalized to the peri-event spike histograms around surrogate events; for details, see Materials and Methods, mean firing rate = 0.7 Hz, MI = 6.15 z score, the correlation between peri-spindle-peak spike histogram and the mean LFP filtered in the spindle frequency range, r = 0.65). B, Same as in A, but for fast spindles (13-15 Hz; MI = 3.12 z score, r = 0.57 z score). C, Comparison of the MI of AD units defined as the difference between peak to trough of the curve of each peri-spindle-peak spike histograms ± 0.25 s around time 0 among the 7 spindle frequency ranges of 7-9 Hz (slow spindle, blue), 8-10, 9-11, 10-12, 11-13, 12-14, and 13-15 Hz (fast spindle, red, p = 2.47
Table 2-1
The number of sessions, recording sites, units, sleep/wake duration and SO/spindle/ripple measures for 4 mice separately. Download Table 2-1, DOCX file.
We further studied the phase-locking of AD units to thalamic spindles by computing the phase of Hilbert transform of filtered thalamic LFP in the spindle frequency range at the time of the spikes of AD units occurring ± 0.25 s around the spindle peaks; 58.51% and 47.28% of the AD units were phase-locked to slow and fast spindles, respectively (p < 0.05, Rayleigh test, n = 464). Across all phase-locked units, the preferred phases were also nonuniformly distributed for the slow and fast spindles so that, for most of the cells, the preferred phase of spike occurrence was between −90° and 90° (i.e., around the spindle peak 0°) (Fig. 2E, slow spindle: p = 7.66
Temporal correlation of CA1 units with AD spindles
We next examined the modulation of CA1 units with AD thalamic spindles by computing peri-spindle-peak spike histograms of CA1 units. Figure 3A, B displays the strong firing modulations of CA1 units with thalamic slow and fast spindles for one representative unit. Friedman test for the MI revealed a significant effect of the spindle frequency so that CA1 units were more strongly modulated by slower spindles (Fig. 3C, p = 4.94
Modulation of CA1 units to thalamic spindles. A, Top, Representative average AD LFP traces filtered in the slow spindle range (7-9 Hz) aligned to the spindle peaks. Spike rasters (middle) and peri-slow spindle-peak histograms (bottom) of CA1 spikes from one unit (z score normalized to the peri-event spike histograms around surrogate events, mean firing rate = 0.71 Hz, MI = 4.51 z score, the correlation between peri-spindle-peak spike histogram and the mean LFP filtered in the spindle frequency range, r = 0.64). B, Same as in A, but for fast spindles (13-15 Hz; MI = 1.52 z score, r = 0.46). C, Comparison of the MI of CA1 units defined as the difference between peak to trough of the curve of each peri-spindle-peak spike histograms ± 0.25 s around time 0 for the 7 spindle frequency ranges of 7-9 Hz (slow spindle, blue), 8-10, 9-11, 10-12, 11-13, 12-14, and 13-15 Hz (fast spindle, red). Each circle represents one CA1 unit (p = 4.94
We next investigated the spindle-related modulation of CA1 units to the hippocampal ripples; 30.04% and 28.17% of the ripples co-occurred with slow and fast spindles, respectively (ripples occurred between onset and offset of the spindles, median ± MAD, density = 0.35 ± 2.1
The modulation of CA1 units to the ripples was spindle-dependent. A, Histogram of the duration of ripples co-occurring with slow spindles (blue, ripples occurred between the onset and offset of the spindles), ripples co-occurring with fast spindles (red), and isolated ripples (magenta) for one session (left, P refers to probability). Vertical lines indicate the median of the distributions (median ± MAD, Ripple + slow spindle = 0.041 ± 2.1
In sum, we found that the modulation of CA1 units with spindles and the spindle-associated modulation of CA1 units with ripples were stronger for slower spindles. These results suggest that spindles with lower frequencies (i.e., longer cycles within a spindle) provide a wider depolarizing window for the temporal coordination of the hippocampal and thalamocortical networks and enhancement of the multiregional interactions.
Stronger modulation of hippocampal ripples to slower and longer thalamic spindles
We next explored the temporal association between hippocampal ripples and thalamic spindles. First the event correlation histograms of ripple events time-locked to the spindle onset were computed, which revealed increase of ripple occurrence after the spindle onset with a peak of ∼200 ms for both slow and fast spindles (Fig. 5A). To further explore the coupling between spindles and ripples, we next quantified the modulation of hippocampal ripple amplitude with the phase of spindles detected from the thalamic LFP using PAC analyses (Fig. 5B–D; for details, see Materials and Methods). The ripples were mostly coupled to the phases around the peak of thalamic spindles with the SI angles dominantly after the spindle peak for the slow spindles and before the spindle peak for the fast spindles (slow spindle: p = 2.21 × 10–60, z = 671.61, Rayleigh test, mean ± SD, SI angle = 56.51 ± 40.11°, fast spindle: p = 2.13 × 10–70, z = 780.21, Rayleigh test, SI angle = –63.91 ± 30.13°, n = 18 sessions). In addition, the ripples were more strongly coupled to the spindles with slower frequency (Fig. 5E,F, p = 3.11
Temporal coordination of spindles with ripples. A, The event correlation histograms of ripple events time-locked to the spindle onset for slow (left) and fast (right) spindles. Thick lines indicate the significant (p < 0.05) increase (red) or decrease (blue) in ripple rates (pairwise comparison with the randomized data; for details, see Materials and Methods). B, Raw (top) and filtered (bottom) traces of AD and CA1 (black) LFPs during one slow (left, blue) and fast (right, red) spindle co-occurring with ripple events. The AD and CA1 LFPs were filtered in the spindle (slow spindle: 7-9 Hz, fast spindle: 13-15 Hz) and ripple (150-200 Hz) frequency ranges, respectively. C, Left, Average of slow spindle peak-locked TFR of the CA1 LFP (percentage change from pre-event baseline, n = 1500 slow spindles) for one session. Black curves represent grand average filtered thalamic LFP in the slow spindle frequency range (7-9 Hz) aligned to the spindle peak (time 0). Right, Circular histogram of SI angles of the ripples relative to the slow spindles. SI angles were nonuniformly distributed (p = 5.18 × 10–54, z = 221.12, Rayleigh test, n = 1500 slow spindles). D, Same as in C, but for fast spindles (13-15 Hz, p = 3.18
We next compared the duration and amplitude of spindles co-occurring with ripples (spindles with at least one ripple between their onset and offset) with isolated spindles; 27.11% and 26.17% of the slow and fast spindles were coupled to the ripples. The duration and amplitude of both slow and fast spindles co-occurring with ripples were larger than the isolated spindles (Fig. 6A, slow spindle: p = 1.42
Stronger coupling of ripples to longer spindles. A, Left, Histograms of the duration (top) and amplitude (bottom) of slow spindles co-occurring with ripples (blue) and isolated slow spindles (black) for one session. Vertical lines indicate the median of the distributions (p = 1.41
Spindles occurring closer to the SO trough were more strongly coupled to the ripples
We next investigated the temporal interaction of SOs and ripples/spindles; 52.19% of the ripples co-occurred with SOs (SO peak occurred within ± 0.5 s interval around ripple peak, median ± MAD, density = 0.39 ± 1.11 ×
73.18% and 72.12% of the slow and fast spindles co-occurred with SOs, respectively (SO peak occurred within ± 1.5 s interval around spindle maximum peak, density = 0.18 ± 2.11 ×
We next computed the phase of thalamic LFP in the SO frequency range at the time of the spindle event (spindle peak) using Hilbert transform. Figure 7E displays the circular histogram of the SO phases at the time of the slow and fast spindle events for one session. For 60.11% and 59.03% of the sessions (n = 18), the SO phases at the time of slow and fast spindles, respectively, were nonuniformly distributed (p < 0.05, Rayleigh test). Across all the sessions also, the mean phases were nonuniformly distributed with mean phases between 90° and 180° corresponding to down to up transition phase of SO (slow spindle: p = 2.21 ×
Stronger coupling of CA1 units to thalamic spindles after exploration
We next examined whether the temporal association of CA1 units/ripple and thalamic spindles was task-relevant. To this end, we computed the spindle/ripple properties, spindle-associated modulation of CA1 units and spindle-ripple coupling during sleep periods (∼1.5 h each) that preceded and followed spatial exploration. During the exploration, the animals foraged for food for ∼30 min. The spindle/ripple/SO density did not change significantly between sleep after and before the exploration (p > 0.13, Wilcoxon signed rank test, n = 18 sessions). The durations of both slow and fast spindles as well as the ripples were longer for sleep after than before the exploration (Fig. 8A, slow spindle: p = 1.01 ×
SO-spindle coupling phase predicts spindle-ripple coupling strength. A, Histogram of the duration (left) and amplitude (right) of the ripples co-occurring with SOs (magenta, SO peak occurred within ± 0.5 s interval around ripple maximum peak) and isolated ripples (black) for one session. Vertical dotted lines indicate the median values (p = 1.72 × 10–25 and p = 3.21 × 10–14 for duration and amplitude, respectively, Mann–Whitney U test, median ± MD, duration = 0.042 ± 9.21 × 10–4 s and 0.038 ± 1.1 × 10–3 s, amplitude = 4.02 ± 1.08 SD and 4.89 ± 1.04 SD, n = 342 and 218 for ripples co-occurring with SOs and isolated ripples, respectively). Inset, The distributions of the ripple duration (left) and amplitude (right) for ripples co-occurring with SOs (magenta) and isolated ripples (black). Each circle represents one session (n = 18 sessions). B, The circular histogram of the SO phases at the time of the ripples for the same session as in A revealing a nonuniform distribution (p = 1.82
The duration and amplitude of the events before and after exploration. A, Histogram of the duration of slow spindles (left), fast spindles (middle), and ripples (right) before (black) and after (blue, red, and magenta represent slow spindles, fast spindles, and ripples, respectively) exploration for one session. Vertical lines indicate the median of the distributions (slow spindle: p = 2.74 ×
We explored whether the modulation of the CA1 cells with spindles in sleep after exploration was associated with the activity of the cell during the exploration. We computed the correlation coefficient between the CA1 MI and mean firing rate of the cells during exploration. We found that the MI for slow spindles in sleep after exploration was positively correlated with the mean firing rate during exploration for both pyramidal cells and interneurons so that more active cells during exploration were more strongly modulated with the slow spindles in the sleep after exploration (Fig. 9D–F, pyramidal cells: r = 0.63, p = 3.06 ×
In addition, PAC analyses revealed that the spindle-ripple coupling strength was larger during sleep after than before exploration for both slow and fast spindles (Fig. 9G,H). Two-way ANOVA further revealed significant main effect of both spindle frequency factor and after/before exploration factor and their interaction on the spindle-ripple coupling strength (Fig. 9I, p = 2.12 ×
To check that the difference we observed between before and after exploration is not simply because of changes in the sleep measures over time of sleep, we further partitioned the duration of sleep before and after exploration into tertiles and compared the spindle/ripple density, MI of CA1 units with spindles and spindle-ripple coupling strength between the first and third tertiles for sleep before and after exploration separately. We did not find any significant difference in these measures between the first and third tertiles of sleep (p > 0.05, Wilcoxon signed-rank test).
Together, these results reveal task-specific interactions between the hippocampal and thalamic activity during spindles/ripples, suggesting a functional role of thalamic spindles on cross regional information transfer.
Hippocampal-thalamocortical model for multiregional interactions during spindles
We next developed a simplified hippocampal-thalamocortical neural mass model to investigate the mechanisms underlying the hippocampal-thalamocortical temporal interactions during thalamic spindles. The model consists of thalamocortical and CA1-CA3 neural networks with long-range bidirectional hippocampal-thalamocortical and cortico-thalamic projections. As we previously described (Ghorbani et al., 2012; Hashemi et al., 2019; Azimi et al., 2021), in this model the up- and down-states with high-frequency activity during the up-states are generated in recurrently connected cortical networks because of dendritic spike frequency adaptation. Similar to previous studies (Destexhe et al., 1996; Destexhe and Sejnowski, 2003; Cona et al., 2014; Hashemi et al., 2019), spindles are generated in a bursting recurrent thalamic network, which in our model simply consists of one identical group of inhibitory neurons representing the thalamic reticular neurons and one identical group of thalamocortical neurons. The ripples are produced because of interactions between CA1 excitatory and inhibitory neurons. While the model is deterministic (with no external or intrinsic noise), because of nonlinear interaction of multiple thalamocortical and hippocampal oscillators, it can generate spindles and ripples with variable frequency, duration, and amplitude by chaotic dynamics (Fig. 10A) (Ghorbani et al., 2012; Hashemi et al., 2019).
Modulation of CA1 units and ripples to thalamic spindles enhances after exploration. A, Top, Representative average AD LFP traces filtered in the slow spindle range (7-9 Hz) aligned to the slow spindle peaks. Spike rasters (middle) and peri-slow spindle-peak histograms (bottom) of CA1 spikes from one unit (z score normalized) before (left) and after (right) exploration. B, Same as in A, but for fast spindles (13-15 Hz). C, Comparison of the MI of CA1 units with spindles before and after exploration for the 7 spindle frequency ranges of 7-9 Hz (slow spindle, blue), 8-10, 9-11, 10-12, 11-13, 12-14, and 13-15 Hz (fast spindle, red). Each circle represents one CA1 unit (two-way ANOVA, n = 58 CA1 units). D, Top, Representative average AD LFP traces filtered in the slow spindle frequency range (7-9 Hz) aligned to the slow spindle peaks. Spike rasters (middle) and peri-slow spindle-peak histograms (bottom) of CA1 spikes (z score normalized) in sleep after exploration from one unit with high activity during exploration (mean firing rate during exploration = 5.9 Hz, MI = 4.57 z score). E, Same as in D, but for one unit with low activity during exploration (mean firing rate during exploration = 1.46 Hz, MI = 4.35 z score). F, The MI of CA1 units with slow spindles for sleep after exploration versus the mean firing rate of the cells during exploration for pyramidal cells (left, n = 39) and interneurons (right, n = 19). G, Average of spindle peak-locked TFR of the CA1 LFP (percentage change from pre-event baseline) before (left) and after (right) exploration for slow (7-9 Hz, top) and fast (13-15 Hz, bottom) spindles for one session. Black curve represents grand average filtered thalamic LFP in spindle frequency range aligned to the spindle peak (time 0). H, Histograms of the spindle-ripple coupling strength before (black) and after (blue and red for slow and fast spindles, respectively) exploration for slow (left) and fast (right) spindles for the same session as in G. Vertical lines indicate the median of the distributions (slow spindle: p = 1.54 ×
Hippocampal-thalamocortical temporal interactions during spindles in the model. A, From top to bottom, respectively: Broadband trace of the membrane potential of excitatory neurons of the cortical network, broadband trace of the membrane potential of thalamocortical neurons, filtered (8-12 Hz) trace of the membrane potential of thalamocortical neurons with detected slow spindles (blue bracket), filtered (12-16 Hz) trace of the membrane potential of thalamocortical neurons with detected fast spindles (red bracket), broadband trace of the membrane potential of excitatory neurons of CA1 network and filtered (150-200 Hz) traces of the membrane potential of CA1 excitatory neurons with detected ripples (magenta stars). B, Average membrane potential of the thalamocortical neurons filtered in the slow (8-12 Hz, left) and fast (12-16 Hz, right) spindle frequency range aligned to the spindle peaks (top). Average of slow spindle peak-locked firing rate of thalamocortical (middle) and CA1 excitatory (bottom) neurons. C, Left, Average of slow spindle peak-locked TFR of the CA1 excitatory neuron membrane potential (n = 385). Black curve indicates grand average filtered thalamocortical neuron membrane potential in the slow spindle frequency range (8-12 Hz) aligned to the spindle peak (time 0). Right, Circular histogram of SI angles of the ripples relative to the slow spindles. SI angles were nonuniformly distributed (p = 2.50 × 10–30, z = 65.23, Rayleigh test). D, Same as in C, but for fast spindles (12-16 Hz, p = 2.75 × 10–16, z = 35.13, Rayleigh test). E, Histogram of the spindle-ripple coupling strength for slow (blue) and fast (red) spindles. Vertical lines indicate the median of the distributions (slow spindle: median ± MAD, SI strength = 0.76 ± 0.16, fast spindle: SI strength = 0.67 ± 0.16). F, The effect of applying input currents (bottom) oscillating in the slow spindle (left, 10 Hz) and fast spindle (right, 14 Hz) frequency ranges on the trace of CA1 excitatory neurons when the thalamocortical network was removed. The stimulation was applied during ripple refractory time. The control and stimulated raw (top) and filtered (150-200 Hz) traces (middle) are shown in black, blue (slow spindle)/red (fast spindle), respectively. G, Spindle-ripple coupling strength versus the stimulation frequency. Line indicates the linear fit to the data (p = 1.7
To investigate the modulation of the thalamocortical and CA1 neurons with thalamic spindles in the model, we first computed the average firing rate of thalamocortical and CA1 neurons around the thalamic spindle peak separately for slow (8-12 Hz) and fast (12-16 Hz) spindles (Fig. 10B). Similar to the experimental findings, the firing rate of both thalamocortical and CA1 neurons was more strongly modulated by slow than fast spindles. We next quantified the spindle-ripple coupling strength and phase using PAC analyses in the model data (Fig. 10C,D). Consistent with the experimental results, the hippocampal ripples were mostly coupled to the phases around the peak of thalamic spindles with the SI angles dominantly after the spindle peak for the slow spindles and before the spindle peak for fast spindles (slow spindle: p =
In agreement with the experimental results (Fig. 6), in the model data, the duration and amplitude of both slow and fast spindles co-occurring with ripples were larger than the isolated spindles (Fig. 10H, slow spindle: p = 6.70
To assess whether the spindle-ripple coupling depended on the duration and amplitude of the spindles in the model data, we next partitioned the distributions of the spindle durations and amplitudes into tertiles and compared the coupling strength of ripples to long (third tertile) versus short (first tertile) spindles for spindles with low and high amplitudes. Consistent with our experimental findings, the ripples were more strongly coupled to long versus short spindles for both low- and high-amplitude spindles (Fig. 10J, slow spindle: p = 1.9
Discussion
Together, our results provide the first demonstration that spindle frequency and duration can provide valuable information about the hippocampo-cortical interactions essential for memory consolidation computations. Our findings indicate enhanced thalamic spindle duration and spindle-associated activity of CA1 units during postexploratory sleep. The other key finding revealed by the current study is that the modulation of the CA1 units to the hippocampal ripples is thalamic activity-dependent so that the depolarizing window during the thalamic spindle peaks provides the fine-tuned window for CA1 activity coordination. We further described how the low-frequency long-duration spindles can increase the temporal multiregional coordination by developing a simplified hippocampal-thalamocortical model.
Several studies have inferred the phase-locking of hippocampal ripples to neocortical (Siapas and Wilson, 1998; Sirota et al., 2003; Mölle et al., 2006; Peyrache et al., 2011; Ngo et al., 2020; Varela and Wilson, 2020) or hippocampal/parahippocampal (Clemens et al., 2011; Staresina et al., 2015; Helfrich et al., 2019; Jiang et al., 2019; Ngo et al., 2020) spindles. Studies involving manipulation of sleep rhythms indicated the role of spindle-ripple coupling in successful memory consolidation (Maingret et al., 2016; Latchoumane et al., 2017; Xia et al., 2017; Binder et al., 2019). The modulation of the CA1 unit activity with cortical spindles has been shown previously (Sirota et al., 2003; Wierzynski et al., 2009; Sullivan et al., 2014; Varela and Wilson, 2020). However, none of these previous studies used thalamic spindles (i.e., spindles detected in the thalamus) to explore the precise temporal coupling of hippocampal activity to spindles that are originated in the thalamus and can reach the hippocampus by direct projections from the thalamus to CA1 (Vertes, 2015). In addition, how the hippocampal-thalamocortical interactions during spindles depend on spindle properties was not investigated in the previous studies. We sought to address this gap by focusing on thalamic spindles to investigate the differential contribution of spindles with different duration and oscillating frequency to multiregional coordination.
Previous reports have shown association between an increase in spindle frequency during NREM and memory impairments in old subjects (Taillard et al., 2019) and negative correlations between spindle frequency and cognitive abilities in children (Geiger et al., 2011; Chatburn et al., 2013). However, the significance of spindle frequency in rodents remains to be elucidated. A recent study suggested that memory impairment in a hippocampus-dependent object place recognition task in young sleep-deprived mice could be associated with an increase in spindle frequency (Yuan et al., 2021).
The functional relevance of long-duration ripples for successful memory consolidation has been underscored recently (Fernández-Ruiz et al., 2019; Ngo et al., 2020). Consistently, we found that the ripple duration increased after exploration. We further found an increase in the spindle duration after behavior and provided evidence for the importance of long-duration spindles for hippocampal-thalamocortical interactions. Increase in spindle duration in the sleep following learning was reported in humans (Fogel et al., 2007; Morin et al., 2008). Several neuropsychiatric disorders are also associated with reduction of spindle duration (Ferrarelli et al., 2007; Fernandez and Lüthi, 2020), and spindle duration also decreases with age (Nicolas et al., 2001). In accordance with these studies, our findings highlight the significance of spindle duration for successful memory consolidation, which needs to be explored more directly in the future manipulating studies.
Consistent with studies reported the coupling of cortical SOs and spindles in rodents (Mölle et al., 2006; Peyrache et al., 2011; Niethard et al., 2018), we found that thalamic spindles were coupled to SOs occurring mainly during the down to up-state transition. The phase of spindles relative to SOs was not found to be different between high-frequency (7-9 Hz) and low-frequency (13-15 Hz) spindles. Studies in humans revealed the existence of two distinct types of cortical sleep spindles: frontal slow spindles (8-12 Hz) and more posterior fast spindles (12-16 Hz). Whereas cortical fast spindles occur during the SO down to up-state transition (Andrillon et al., 2011; Clemens et al., 2011; Mölle et al., 2011; Klinzing et al., 2016; Mak-McCully et al., 2017; Oyanedel et al., 2020), EEG studies in humans reported the main occurrence of cortical slow spindles during the cortical up to down-state transition (Clemens et al., 2011; Mölle et al., 2011; Klinzing et al., 2016; Dehnavi et al., 2021). Unlike the human EEG/LFP spectrum in which both slow and fast spindle peaks are distinguishable (Andrillon et al., 2011; Dehnavi et al., 2021), the rodent spectrum shows no distinct peak at the spindle frequency range (7-15 Hz) (Mölle et al., 2009; Fernandez and Lüthi, 2020), so that distinguishing fast and slow spindles from the spectrum is not possible in rodents. In addition, the differential topographical distribution of slow and fast spindles in humans has not been observed in rodents (D. Kim et al., 2015). Most of the studies in humans did not distinguish between slow and fast spindles. The studies that analyzed the slow and fast spindles separately mainly suggest that slow and fast spindles involve differently in memory consolidation during sleep (Barakat et al., 2011; Mölle et al., 2011). Previous studies also reported the association between slow spindles and memory performance in humans (Schmidt et al., 2006; Schabus et al., 2008; Mednick et al., 2013; Lustenberger et al., 2015; Fernandez and Lüthi, 2020). Our results provide further support for the functional relevance of spindle frequency for memory consolidation.
Computational models showed that the interaction of thalamocortical and reticular neurons in bursting recurrent thalamic networks can generate fast spindles (Destexhe et al., 1996; Destexhe and Sejnowski, 2003; Cona et al., 2014; Hashemi et al., 2019). In our model, cortical activity initiates interregional interaction, which impacts spindle frequency (Hashemi et al., 2019) and duration (Azimi et al., 2021). Consistent with previous reports (Bonjean et al., 2011), the beginning and termination of spindles in our model are controlled by the long-range cortico-thalamic input (Hashemi et al., 2019). We also recently reported larger SO up-state associated excitatory input from the cortex to the thalamus during longer spindles (Azimi et al., 2021). Progressive changes in the activity of thalamocortical (Bal and McCormick, 1996; Lüthi and McCormick, 1998; Lüthi et al., 1998) or reticular neurons (Bal et al., 1995; U. Kim and McCormick, 1998; Barthó et al., 2014) during the spindles was also proposed as other mechanisms for spindle termination controlling the spindle duration. Despite growing evidence for the significance of spindle-ripple coupling in memory consolidation (Maingret et al., 2016; Latchoumane et al., 2017; Xia et al., 2017; Binder et al., 2019), the mechanism underlying spindle-ripple coupling remains unclear. Selective enhancement of the incidence of sharp wave ripple-delta-spindle by closed-loop electrical stimulation resulted in successful memory consolidation (Maingret et al., 2016). It has also been shown that the inhibition of CA1 parvalbumin-positive cells disrupted the spindle-ripple coupling as well as contextual fear memory consolidation (Xia et al., 2017). In addition, silencing the projections from hippocampus to mPFC impaired the modulation of ripples by spindles and spatial memory consolidation (Binder et al., 2019). Thalamic spindles can reach the hippocampus either directly by projections from the thalamus to CA1 (Jankowski et al., 2013; Vertes, 2015; Fernandez and Lüthi, 2020) or indirectly by the prefrontal (Rajasethupathy et al., 2015) or entorhinal cortical inputs to the hippocampus (Preston and Eichenbaum, 2013; Witter et al., 2017). Recently, it has been reported that AD receives inhibitory projections from CA3 (Vetere et al., 2021). The activity of CA3 inhibitory neurons during sharp waves and the resulting inhibitory input to AD neurons in addition to the inhibitory input from TRN (Shibata, 1992) could also control the spindle-ripple coupling. In addition, AD and CA1 can communicate with each other via input from anterior cingulate cortex (Shibata and Naito, 2008; Rajasethupathy et al., 2015). The excitation provided by the depolarizing phase of the spindle can generate the sharp wave ripples in the hippocampal network phase-locked to the spindles (Azimi et al., 2021). Because of large hippocampo-cortical projection, cortex receives a large depolarizing input at the time of the SWRs. When this depolarizing input to the cortex arrives at the time of the spindle occurrence, it can change the phase of the spindle toward more depolarizing phases which results in the increase of the spindle-ripple coupling strength. Here, we further showed that larger temporal depolarizing window provided by the slower spindles resulted in stronger spindle-ripple coupling.
In conclusion, using simultaneous thalamic and hippocampal recordings from naturally sleeping mice before and after exploration, we provide the first evidence for the strong coupling of CA1 units to thalamic spindles and increase of this coupling following spatial experience. Our experimental and computational findings suggest that low-frequency (long cycles within a spindle) and long-duration thalamic spindles enhance multiregional temporal coordination by providing large enough window of opportunity for information processing during sleep. These results shed light on our understanding of the mechanisms underlying the hippocampal-thalamocortical dialogue required for memory consolidation.
Footnotes
We thank Prof. Adrien Peyrache and Fereshteh Dehnavi for helpful discussions; and Prof. Adrien Peyrache and Prof. Gyorgy Buzsáki for sharing their datasets.
The authors declare no competing financial interests.
- Correspondence should be addressed to Maryam Ghorbani at maryamgh{at}um.ac.ir