Interregional phase-amplitude coupling between theta rhythm in the nucleus tractus solitarius and high-frequency oscillations in the hippocampus during REM sleep in rats

Abstract

Cross-frequency coupling (CFC) between theta and high-frequency oscillations (HFOs) is predominant during active wakefulness, REM sleep and behavioral and learning tasks in rodent hippocampus. Evidence suggests that these state-dependent CFCs are linked to spatial navigation and memory consolidation processes. CFC studies currently include only the cortical and subcortical structures. To our knowledge, the study of nucleus tractus solitarius (NTS)-cortical structure CFC is still lacking. Here we investigate CFC in simultaneous local field potential recordings from hippocampal CA1 and the NTS during behavioral states in freely moving rats. We found a significant increase in theta (6–8 Hz)-HFO (120–160 Hz) coupling both within the hippocampus and between NTS theta and hippocampal HFOs during REM sleep. Also, the hippocampal HFOs were modulated by different but consistent phases of hippocampal and NTS theta oscillations. These findings support the idea that phase-amplitude coupling is both state- and frequency-specific and CFC analysis may serve as a tool to help understand the selective functions of neuronal network interactions in state-dependent information processing. Importantly, the increased NTS theta-hippocampal HFO coupling during REM sleep may represent the functional connectivity between these two structures which reflects the function of the hippocampus in visceral learning with the sensory information provided by the NTS. This gives a possible insight into an association between the sensory activity and REM-sleep dependent memory consolidation.

Graphical Abstract

Graphical Abstract.

Statement of Significance.

Nucleus tractus solitarius (NTS), the primary integrative center for autonomic functions in the dorsal brainstem, exhibited theta oscillations which are highly coupled to hippocampal high frequency oscillations during REM. This state-dependent interregional cross-frequency coupling provides a new insight into a functional characteristic of REM sleep and neural mechanisms that establish the long-range connectivity between NTS and hippocampal networks.

Introduction

Brain states are partly defined by interactions between frequency bands within the same neuronal network [1]. Unique frequency bands originating from remote brain structures can also coexist and interact with one another [2]. The power of LFPs follows the 1/f law; i.e. the power of specific rhythms is generally inversely correlated to their frequency [3]. This relationship results from neural architecture. A larger neuronal network is recruited during low-frequency oscillations whereas high-frequency oscillations are restricted to the local site [4]. These characteristics imply that widespread low-frequency oscillations could modulate fast-frequencies locally [5]. This interaction among different frequencies or frequency bands is called CFC.

Phase to amplitude coupling (PAC), in which the phase of low-frequency oscillations modulates the amplitude of high-frequency oscillations, is the most-studied type of CFC observed in rodents [6–9] and in humans [10–14] and is relevant to many physiological [11–13, 15] and pathological states [16–19]. PAC analysis reveals that certain frequency pairs are coupled during specific brain states. Theta (4–12 Hz) phase-gamma (30–100 Hz) amplitude modulation is the best-defined PAC in rodent hippocampus [8, 9, 20, 21], and cortex [22] during memory and learning tasks, active wakefulness and REM sleep. In humans, theta-gamma coupling occurs across cortical sites [10, 11, 23] and gamma amplitude can be modulated by theta or alpha in a task-dependent manner [14]. High-frequency oscillations (HFOs) modulated by low-frequency oscillations extend beyond the gamma range. Theta-associated higher frequencies (>100 Hz) also occur [8, 22, 24]. Here, we denote HFOs of 100–200 Hz as HFOs, which is also called “fast gamma” [25] or ripples [26, 27] in other studies. In contrast to intermittent ripple oscillations (100–250 Hz) [28, 29] associated with sharp waves in the hippocampus during slow wave sleep and waking immobility (WI) states [28, 30, 31], HFOs are sustained and coupled to the theta oscillations during active exploration (AE) and REM sleep in rodent hippocampal structures [22, 24] and are increased by behavioral and learning tasks [8]. The analysis of LFPs and CFC in the hippocampus have provided a possible neural mechanism for memory consolidation in awake rodents [8, 9, 32], however, the underlying mechanisms of memory consolidation during REM sleep are still unclear [33, 34].

The NTS is an important hub for processing visceral afferent information and transmitting to several brainstem and forebrain sites [35–38]. Although no direct pathway links the NTS and the hippocampus, multisynaptic pathways connect between them [39]. The NTS → nucleus paragigantocellularis → locus coeruleus → CA1 is one of the pathways that has been shown to contribute to memory consolidation of object recognition in rats [40, 41]. Further, vagal nerve stimulation elicits changes in hippocampal EEG [42, 43] and increases memory in awake rats [44, 45]. In awake humans, the electrical stimulation of the hippocampus evokes unpleasant gastric feelings [46]. These findings suggest the connection between visceral influences and higher neurobehavioral processes but the neural mechanisms that establish the long-range functional connectivity between hippocampus and NTS has not yet been elucidated.

In this study, we investigated the association between hippocampal HFOs (120–160 Hz) and theta (6–8 Hz) oscillations in the CA1 of the hippocampus and the NTS of freely moving rats. During active exploration (AE) and REM sleep states, HFOs were modulated by hippocampal theta with the greatest coupling during REM sleep. We also found that the NTS exhibited peak theta activity during REM sleep and NTS theta-hippocampal HFO coupling is more pronounced during REM when compared to AE. Finally, the preferred theta phases that modulated hippocampal HFOs were different among brain regions. These findings emphasize the differential expression of neuronal connectivity during behavioral states and help explain how neural oscillations contribute to intra-hippocampal and hippocampal-NTS networks.

Methods

All experimental procedures were reviewed and approved by the Animal Research Committee at the University Health Network (Toronto, Ontario, Canada), in accordance with the guidelines of the Canadian Council on Animal Care.

Implantation of Recording Electrodes

Eighteen adult male Wistar rats (300–350 g), Charles River, Canada, were used in this study. Preoperatively, rats were given subcutaneous (s.c.) injection of Ketoprofen (5 mg/kg), Buprenorphine (0.01–0.05 mg/kg, s.c., every 12 h) and 2–3 ml of 0.9% sterile saline (s.c.), and repeated every 12 h, if necessary. Rats were anesthetized with 1.5%–2% isoflurane. After exposure to the skull bone, four stainless steel screws were fixed to the skull as anchor screws. Two of them that were fixed on the right frontal bone and left occipital bone were connected and used as a ground. Two twisted bipolar electrodes (MS303/3-B, polyimide-insulated stainless steel, 150 µm, Plastic-1 Inc., Roanoke, VA) were stereotaxically implanted into the CA1 region of the right hippocampus (AP, −3.8 mm; L, −2 mm; V, −2.4 mm), and the right NTS (AP, −13.3 mm; L, −1.0 mm; V, 8.0 mm). The two tips of the twisted bipolar electrodes were separated by 500 µm. Another pair of stainless-steel wires (150 µm) was inserted into the neck muscle for muscle activity recording. In addition to CA1 and NTS coordinates, a twisted bipolar electrode was implanted into the left parietal cortex (AP, −4.3 mm; L, +5.2 mm; V, 2.5 mm) of additional six rats (three electrode locations in each rat) in a different set of experiment for phase coherence analysis. The animals were allowed to recover for 1 week after the surgery.

LFP Measurements

To prevent stress-induced LFP changes, rats were conditioned to a recording apparatus for 3 consecutive days, 20 min each day, before the experimental day. On the experimental day, rats were placed into a Plexiglas bowl filled with fresh bedding and connected to the recording dual-channel AC microelectrode amplifier with extended head stages (model 1800, AM Systems, Carlsborg, WA). The input frequency band of the amplifiers was set at 1–1000 Hz with a sampling rate of 5000 Hz (Spike 2 software version 7, and Micro-1401, Cambridge Electronic Design). All data were amplified 1000 times. Rats were also video-monitored using 2 webcams (Logitech) positioned in front of and above the recording cage. LFPs were recorded for 4–6 h during the afternoon (12:00–16:00 h) in the quiet recording room and without any interruption from the experimenters.

Histological Assessment

Histological assessments of brain tissues were conducted as previously described by Lertwittayanon et al. [47]. Briefly, rats were deeply anesthetized with an intraperitoneal injection of sodium pentobarbital (70 mg/kg, i.p.). Electrode locations were marked by passing a direct current of 1 mA for 5 s through the electrodes. Next, rats were transcardially perfused with heparinized phosphate buffered saline (10 units/ml) followed by 10% neutral buffered formalin solution. Brains were removed from the skull and placed in formalin solution at 4°C overnight for further fixation. Brains were then transferred to a 30% sucrose solution for 3–4 days before being frozen at −80°C. A series of coronal sections were obtained at 50 µm thickness and stained with cresyl violet. In 18 rats that were subjected to simultaneous LFP recordings in HC and NTS, only data from 10 rats with confirmed electrode locations by histological assessment were analyzed. In eight rats that were excluded from the study, five rats had the brainstem electrodes in the cuneate nucleus (Cu) while the other three rats had the electrodes in other brainstem areas. To eliminate the data misinterpretation due to the observation of a possible volume conduction effect between the hippocampus and the NTS, we also analyzed the LFPs in five rats that the brainstem electrodes were incorrectly inserted into the Cu, which is 0.5 mm dorsal to the NTS. Finally, the histological assessment confirmed the correct locations of electrodes in all six rats that were used in another set of experiments for phase coherence analysis.

Data Analysis

Behavioral analysis

A trained technician visually annotated the behavioral states by visual inspection of behaviors, LFP features, and off-line spectral analyses. Four behavioral states were defined as follows: (1) AE, the animal exhibited exploratory behavior (walking, turning, rearing), with large theta (7–8 Hz) power peak in the hippocampal LFPs. Neck muscle activity was high. (2) WI, the animal was motionless but remaining awake (sitting/lying down quietly with a slight lateral or vertical movement of the head or limbs), with low theta (6–7 Hz) power peak in the hippocampus. Neck muscle activity was high but lower than during AE. (3) Slow wave sleep (SWS), the animal was lying down with eyes closed. The hippocampal LFPs showed predominant delta waves (1–4 Hz) with low neck muscle activity. (4) Rapid eye movement sleep (REM), the animal was lying down with eyes closed and atonic. High theta (7–8 Hz) power was observed in the hippocampus [48].

LFPs segment selection

In WI, SWS, and REM, LFP recordings were divided into 10-s epochs, with 9–11 epochs per rat. In AE, a variable amount of time was chosen, between 4 and 10 s, avoiding movement artifact. Thus, the number of epochs analyzed also varied between rats depending on the duration of each epoch. However, the total duration of AE recordings analyzed was approximately 100 s/rat.

Power analysis

LFP recordings were analyzed using MATLAB 2016a (The Mathworks Inc., Narick, MA). Data were digital bandpass filtered at 1–1000 Hz. 60 Hz power and corresponding harmonics were notched filtered using Matlab's FIR filter with a ±0.5 Hz cutoff. Continuous wavelet transform (CWT) (Matlab 2016a’s cmor6-0.8125 basis function) was used to investigate time-frequency power profiles. The frequency range of interest was 1–200 Hz with 0.5 Hz frequency resolution. To create a frequency-power plot, the absolute power of each frequency was averaged over time. Next, the absolute power of an individual frequency or frequency band was divided by the total power and represented as a percent total power. For state comparisons, the percent total power of all epochs in one rat were averaged and represented as one data sample. The percent total powers of all rats (8–10 rats) were then pooled together. The mean, and standard error of the mean (SEM) were performed across rats (Figure 1).

Figure 1.

Power spectrum during behavioral states. (A) left, raw local field potentials (LFPs) recorded in the HC and NTS during AE and REM states, right, schematic representation of LFP recording electrode locations. (B) Mean power spectra of 10 s-segments of HC and NTS LFP recordings during four behavioral states: active exploration (AE), waking immobility (WI), slow wave sleep (SWS), and rapid eye movement sleep (REM) (n = 8–10 rats: AE = 9 rats, WI = 10 rats, SWS = 8 rats, REM = 8 rats). Note the different power scales (dB) in the 20–60 Hz and 60–200 Hz frequency ranges compared to 1–20 Hz frequency range (%total power). Shadings indicate SEM. Arrows point to the power peaks for theta in both brain areas and HFOs activity in the hippocampus during REM. (C) Bar diagrams show mean band power for theta (6–8 Hz), HG (60–100 Hz) and HFOs (120–160 Hz) during AE and REM sleep in HC and NTS. Asterisks indicate significant differences (Student’s t-test, p < .05). HC: hippocampus, NTS: nucleus tractus solitarius, HG: high gamma, HFOs: high-frequency oscillations.

Analysis of phase-amplitude coupling

The modulation index (MI) as described by Tort et al. [8, 49] was used to evaluate phase-amplitude coupling strength in this study. The MI assesses whether the phase of slower-oscillation frequency (f_p) modulates the amplitude of faster oscillation frequency (f_A). Briefly, the MI was calculated as follows: (1) the phases of f_p were binned into 20 degree intervals and the mean amplitude of f_A in each phase bin was calculated; (2) the mean of f_A in each phase bin was normalized by the sum of the means over all bins, giving rise to a phase-amplitude “distribution,” P. (3) MI is a measure that quantifies the deviation of P from the uniform distribution, i.e. if there is no phase-amplitude coupling between a frequency pair (f_p, f_A) being investigated, the amplitude distribution P over the phase bins is uniform. The higher the coupling, the more the amplitude distribution deviates from the uniform distribution. Finally, the normalization was applied to achieve the distribution distance values between 0 and 1 (Figure 2).

Figure 2.

Steps in the computation of the modulation index (MI) modified from Tort et al. CWT is used to filter the raw field potentials recorded from the hippocampus during REM (A) at the phase frequency (6–8 Hz) range (B) and at the amplitude frequency (120–160 Hz) range (D, thin line), then the phase time series (C) and the amplitude time series (D, thick line) are calculated from the filtered signals. A composite phase-amplitude time series is then constructed and used to obtain the mean amplitude distribution over phase bins. The MI is obtained by measuring the divergence of the observed amplitude distribution from the uniform distribution. HFOs: High-frequency oscillations.

These analyses were performed for all pairs of target frequencies. The phase frequencies of interest were 1–20 Hz with 1-Hz step and the amplitude frequencies of interest were 22–200 Hz with 2-Hz step. The comodulogram is obtained by plotting the MI of all frequency pairs in a two-dimensional pseudocolor plot where the x-axis represents the phase-modulating and y-axis represents the amplitude-modulated frequencies. The single difference from Tort et al.'s method was that instead of applying Hilbert transform, a complex CWT was applied to extract the amplitude and phase information. For each epoch, an MI value was calculated for each frequency pair. The MI values of all epochs in one rat were averaged and represented as one data sample. The MI values of all rats were then pooled together. The mean and SEM were performed across rats (Figures 3A, Supplementary Figure S1).

Figure 3.

Cross-frequency couplings (CFCs) in different behavioral states. (A) Mean phase-to-amplitude comodulograms obtained from 10 s-segments within the HC (HC _phase-HC _Amp) and between NTS and HC (NTS _Phase-HC _Amp) during behavioral states (n = 8–10 rats: AE = 9 rats, WI = 10 rats, SWS = 8 rats, REM = 8 rats). The x-axis denotes the phase-modulating frequencies and y-axis denotes the amplitude-modulated frequencies. (B) Mean theta (6–8 Hz)-HFOs (120–160 Hz) CFCs are significantly higher in REM when compared to AE in both brain location pairs (HC _theta-HC _HFOs and NTS _theta-HC _HFOs) (B, Student’s t-test, p < .05). AE: active exploration, WI: waking immobility, SWS: slow wave sleep, REM sleep: rapid eye movement sleep, HC: hippocampus, NTS: nucleus tractus solitarius, HFOs: high-frequency oscillations.

Analysis of the relationship between CFC and theta power

The relationship between theta power and CFC as shown in Figure 4 was achieved in the following steps: (1) the total power between 1 and 200 Hz was calculated for each 0.5-s segment; (2) theta power (6–8 Hz) was divided by the total power and represented as a percent total power; (3) The segments in each animal were pooled and divided into bins of 0.1 values of percent total power and the mean theta-HFO MI was calculated. (4) The data from all animals were pooled. The mean and SEM of theta-HFO MI were calculated across animals.

Figure 4.

Theta-HFO coupling strength depends on theta power, both in active exploration (AE) and rapid eye movement (REM) sleep. (A) hippocampal theta (6–8 Hz) power is significantly correlated with HC_theta-HC_HFOs coupling (B) NTS theta (6–8 Hz) power is significantly correlated with NTS_theta-HC_HFOs coupling both in AE and REM sleep. Best linear fit (blue line for AE and red line for REM), indicating their positive correlation (p < 0.05). Data were from the average across rats (AE = 9 rats, REM = 8 rats). HC: hippocampus, NTS: nucleus tractus solitarius, HFOs: high-frequency oscillations .

Averaged theta peak time-locked plot and histogram of the theta phases at which the HFO peaks occurred

To create Figures 5A and 6A, Supplementary Figure S2A and C we first identified 0.5 s segments centered on the peak of the theta oscillation. We then applied a CWT to calculate the power of each LFP segment during REM and then normalized by dividing the power calculated at each frequency with maximum power and averaged across segments to obtain a mean normalized time-frequency plot of power distribution (Figures 5A and 6A, Supplementary Figure S2A and C, upper). Next, mean normalized power at 130 Hz was obtained by calculating the normalized power at 130 Hz of each segment and taking the average across segments. The raw signal of each segment was also filtered at theta frequency range (6–8 Hz) and averaged across segments. Finally, the average of both normalized power at 130 Hz and theta-filtered time series were plotted on the same axis as shown in Figures 5A and 6A, Supplementary Figure S2A and C, lower.

Figure 5.

The amplitude of hippocampal HFOs (120–160 Hz) is maximal at the falling phase of co-occurring hippocampal theta oscillations (6–8 Hz) during REM sleep. (A) (Upper) Mean normalized time-frequency plot of power distribution obtained from a continuous wavelet transform (CWT) applied to 0.5 s of hippocampal LFP recording during REM. (A) (Lower) Plot showing the mean normalized power at 130 Hz (blue dash-dot line, left axis). The falling phase of the averaged hippocampal theta filtered signal (red solid line, right axis) is time-locked to the HFO peaks. Averaged hippocampal theta filtered signal is obtained by aligning the LFP traces at the peaks of the theta oscillations (n = 8 rats). (B) Plot showing the peaks of HFOs (red arrow) co-occur with the falling phase of hippocampal theta oscillations (blue solid line) obtained from one epoch of the same representative animal as in Figure 2. (C) Corresponding histogram of the theta (6–8 Hz) phases at which the HFO peaks occurred (n = 8 rats) (D) angular histograms of the hippocampal theta phases where the HFO peaks occurred. The red arrow indicates the direction (avgAng ± 95% CI) and magnitude (r) of the mean resultant vector. Rayleigh test was used to detect a unimodal deviation from uniformity (p < .05). REM sleep: rapid eye movement sleep, HC: hippocampus, HFOs: high-frequency oscillations.

Figure 6.

The amplitude of hippocampal HFOs (120–160 Hz) is maximal near the peak of co-occurring NTS theta oscillations (6–8 Hz) during REM sleep. (A) (Upper) Mean normalized time-frequency plot of power distribution obtained from a continuous wavelet transform (CWT) applied to 0.5 s of hippocampal LFP recording during REM. (A) (Lower) Plot showing the mean normalized power at 130 Hz (blue dash-dot line, left axis). The phase around the peak of the averaged NTS theta filtered signal (red solid line, right axis) is time-locked to the HFO peaks. Averaged NTS theta filtered signal is obtained by aligning the LFP traces at the peaks of the theta oscillations (n = 8 rats). (B) Plot showing the peaks of HFOs (red arrow) co-occur near the peak of NTS theta oscillations (blue solid line) obtained from one epoch of the same representative animal as in Figure 2. (C) Corresponding histogram of the theta phases at which the HFO peaks occurred (n = 8 rats). (D) angular histograms of the NTS theta phases where the HFO peaks occurred. The red arrow indicates the direction (avgAng ± 95% CI) and magnitude (r) of the mean resultant vector. Rayleigh test was used to detect a unimodal deviation from uniformity (p < .05). (E) Box plot displaying the distribution of the theta phase differences between HC and the NTS (n = 8 rats). The red line indicates the median and the magenta diamond marker indicates the mean of the data. REM sleep: rapid eye movement sleep, HC: hippocampus, NTS: nucleus tractus solitarius, HFOs: high-frequency oscillations.

The results shown in Figures 5B and 6B were obtained as described by Tort et al. [8]. First the signal was filtered at the HFO range under study (120–160 Hz). Then, a time series indicating the times of the peaks of the filtered signal was created with the requirement that the peak times are separated by at least 100 ms from each other. A time series of theta filtered signals was also shown on the same plot. The time series of the theta phases were also extracted from the theta filtered signal. Then the phases where the HFO peaks occurred were assessed and used to create a histogram. A corresponding histogram (Figures 5C and 6C, Supplementary Figure S2B and D-upper) showing the number of hippocampal HFO peaks in each theta phase. The histograms were constructed from 0.5 s LFP segments from eight rats.

Circular histogram

The theta (6–8 Hz) phases where the HFO peaks occurred were assessed as mentioned above. Then, CircHist toolbox was used to calculate mean resultant vector (avgAng), 95% confidence interval (95% CI), resultant vector length (r) and to create circular histograms (Figures 5D and 6D, Supplementary Figure S2B and D-lower) [50]. Rayleigh test of uniformity was calculated using the CircStat toolbox [51].

Mean theta phase difference

The phase difference is a measure of connectivity between two LFP signals from different locations. The phase time series of theta (6–8 Hz) in the hippocampus and NTS in each 0.5 s-epoch were extracted using CWT. Then the phase difference between the signals from hippocampus and NTS is computed by subtracting their individual phases. The absolute phase difference was computed by squaring and then taking the square root of the squared difference. The absolute phase difference was calculated in radians and converted to degrees. The mean theta phase difference between the hippocampus and the NTS was calculated across animals (Figure 6E).

Wavelet phase coherence analysis

Phase coherence measures the consistency in phase differences between two signals from different brain regions. In this study, phase coherence was calculated based on the method previously described by Lachaux et al. [52, 53]. Briefly, the phase time series of a frequency between 1 and 200 Hz (1-Hz step) was extracted from signal x $\left(x\right)$ and signal y $\left(y\right)$ using CWT (Matlab 2016a's cmor6-0.8125 basis function). Next, a time-frequency distribution of the phase difference was obtained. The phasor with the phase difference (P_xy) was then calculated using: $P_{xy}\left(t\right)=e^{i(x\left(t\right)-y\left(t\right))}$ for each time point. Next, the phase coherence was calculated by taking the absolute value of the average of P_xy over time. The phase coherence is a real number which varies between 0 (independent signals) and 1 (constant phase-lag between two signals). For group comparisons, the phase coherence of all epochs in one rat were averaged and represented as one data sample. The phase coherence values of all rats were then pooled together. The mean and SEM were performed across rats (Figure 7).

Figure 7.

Cross-regional coherence spectrum during REM sleep shows a peak at 7 Hz and low coherence values in the high-frequency oscillation (120–160 Hz). HC-CT: hippocampus and cortex. HC-NTS: hippocampus and NTS. Data were obtained from six rats in another set of experiments. The parietal cortex (AP, −4.3 mm; L, +5.2 mm; V, 2.5 mm) was implanted with the same electrode type and configuration as used in the HC and NTS. Each coherence spectrum (red line) is compared to the coherence computed from 100 phase-scrambled surrogates (blue line), demonstrating a significant coherence across spatial regions. The dashed blue lines represent a 95% confidence interval. REM sleep: rapid eye movement sleep, NTS: nucleus tractus solitarius.

Surrogate data analysis

To validate that the interregional coherence measures were significant, a set of phase-scrambled surrogate data was generated from the electrophysiological recordings (10-s epoch). 100 surrogate recordings were generated for each physiological recording by using a randomized phase-scrambling of the physiological data. For each surrogate, the amplitude remained identical to the physiological trace, however, the phase information was identical only until a randomly determined point within the first 0.1 s of the physiological trace. After this point, the remainder of the trace was subdivided into 0.1 s blocks of phase information, and each block was randomly shuffled with other 0.1 s blocks from within the same trace. This ensured each surrogate possessed physiological phase information what would be uncoupled from the amplitude signal, providing a baseline coherence value from which to assess the physiological coherence of each state [54]. A 95% confidence interval was generated based on all surrogate data of the same location pair.

The statistical significance of the MI values was assessed by evaluating the recorded modulation against a 95% confidence interval generated from a surrogate data set. For each recording, 200 surrogate recordings were artificially manufactured through randomized block shuffling of the phase information of the recorded sample. This was done to ensure the modulation was normalized to a physiological signal. The 95% confidence interval was determined as 2 standard deviations away from the mean of the surrogate data set, and MI values existing beyond the confidence interval (Mthresh) were deemed a statistically significant finding (i.e. p < .05).

Statistical Analyses

Results were expressed as mean ± SEM. The data from all animals in each state were pooled together and statistics were performed across that data. Student’s t-test were used for the reported statistics. Statistical tests were conducted via Sigmaplot software (12th version; Systat Software Inc, San Jose, CA). Significance was p < .05, unless stated otherwise.

Data Availability

The data used in this study are not publicly available for download but may be retrieved from the corresponding author upon reasonable request.

Results

LFPs recorded from hippocampus (HC), and NTS in 10 freely moving rats were used in this study. LFP spectral feature of the hippocampus, neck muscle activity and overt behaviors were used to classify the animal’s behavioral states in time (Figure 1A). The mean power spectrum clearly showed a distinct peak of theta activity (6–8 Hz) during AE and REM sleep in the hippocampus and NTS. A peak of HFOs (HFOs, 120–160 Hz) power was also seen in the hippocampus during REM (Figure 1B). Theta band (6–8 Hz) in both brain areas had higher power during REM when compared to AE (HC_theta (AE) = 7.23 ± 0.38 VS HC_theta (REM) = 8.66 ± 0.72; NTS_theta (AE) = 1.65 ± 0.26 VS NTS_theta (REM) = 2.83 ± 0.16) (p < .05, n = 8–10 rats). The higher frequencies in the range of high gamma (HG, 60–100 Hz) and HFOs (120–160 Hz) were not different between states in the hippocampus. In contrast, their powers in the NTS during REM were significantly lower than that of AE (p < .05, n = 8–10 rats) (Figure 1C).

We therefore focused our analysis on the CFC between the phase of theta (6–8 Hz) activity in the hippocampus or NTS and the amplitude of HFOs (120–160 Hz) in the hippocampus during REM. Visual inspection of the unfiltered signal during REM clearly showed that theta waves were superimposed by high-frequency oscillations (Figure 2A). The intensity of phase-amplitude coupling was assessed by calculating the MI as described previously [49] (Figure 2A–D).

Next, mean phase-to-amplitude comodulograms were created to visualize the coupling intensity between hippocampal or NTS theta phases and the hippocampal HFO amplitude during behavioral states. The results showed that theta (6–8 Hz)-HFO (120–160 Hz) coupling during REM was significantly higher than during AE between the following pairs; (1) theta in the hippocampus and HFOs in the hippocampus (HC_theta-HC_HFOs) (Figure 3A top row) (MI_AE = 0.0028 ± 0.0005 (n = 9) vs. MI_REM = 0.0054 ± 0.001 (n = 8); p < .005; Figure 3B, left), and (2) theta in the NTS and HFOs in the hippocampus (NTS_Theta-HC_HFOs) (Figure 3A bottom row) (MI_AE = 0.0009 ± 0.0001 (n = 9) vs. MI_REM = 0.0016 ± 0.0001 (n = 8); p < .005; Figure 3B, right). Despite high gamma (60–100 Hz) peak power was seen in both brain regions (Figure 1B), the gamma amplitude was not modulated by any low-frequency phases, regardless of origin, and behavioral states (Figure 3, Supplementary Figure S1). These data showed that CFC is highly selective for different high-frequency bands and behavioral states.

The strength of coupling between theta phase and HFOs could be positively dependent on theta power [11, 22, 55]. Thus, stronger theta-HFO coupling observed during REM could possibly be explained by higher mean theta power during REM when compared to AE. To test this hypothesis, we compared theta-HFO coupling strength as a function of theta power. In Figure 4, we showed that theta-HFO coupling was dependent on theta power for both AE and REM, irrespective of the location where the theta oscillations were detected. Linear regression analysis showed significant correlations between hippocampal theta power and HC_theta-HC_HFOs MI (r² = 0.0458, p < .05 for AE, n = 9 rats; r² = 0.124, p < .001 for REM, n = 8 rats, Figure 4A). The same was true for NTS theta power and NTS_theta-HC_HFOs MI (r² = 0.0857, p < .05 for AE, n = 9 rats; r² = 0.118, p < .05 for REM, n = 8 rats, Figure 4B). However, modulation indices were higher during REM sleep when compared to AE (Figure 4A and B, red line compared to blue line), particularly at higher theta power levels. Taken together, our results suggest that the strength in the couplings of HC_theta-HC_HFOs and NTS_theta-HC_HFOs highly depend on behavioral state but the dependence is not solely explained by state-dependent changes in theta power.

The theta frequency ranges (6–8 Hz) in both brain structures that modulate hippocampal HFOs during REM sleep were similar, but the couplings exhibited different phase preferences (Figures 5 and 6). Hippocampal HFO peaks were phase-locked to the falling phase (circular Reyleigh test, p < .05, n = 8 rats) of hippocampal theta wave (Figure 5), consistent with the hippocampal LFP recording during T-maze tasks in rats [8]. The preferred NTS theta phase for the hippocampal HFO power was around the peak (circular Reyleigh test, p < .05, n = 8 rats) (Figure 6). The results shown in Figures 5 and 6 also suggest that the phases of HFO appearance when comparing the hippocampal theta cycle with NTS theta cycle may be related to the inherent theta phase difference between these regions. To investigate this possibility, the mean theta phase difference between HC and NTS was calculated. The result showed that the mean theta phase difference between two regions was 121.88 ± 5.32 degree (Figure 6E) which is not comparable to the difference of the mean theta phases of HFO appearance between HC (Figure 5D) and NTS (Figure 6D). This result suggests the independence between HC and NTS theta. Finally, we analyzed the mean theta phase that modulates HFOs during AE. The result showed a similar pattern of preferred hippocampal and NTS theta phases for the hippocampal HFO power to that of REM sleep (Supplementary Figure S2). These findings indicate that the preferred theta phase of HFOs is consistent across behavioral states. As suggested by Tort et al. [56], this characteristic is believed to be one of the criteria that used to distinguish HFOs from spike-leaked high-frequency oscillations, caused by extracellular spiking activity, that often contaminate HFO recordings from CA1 area of the hippocampus in some studies.

The origin of theta in the NTS is currently not known. The main argument is that the theta activity detected in the NTS may be volume-conducted from the hippocampus in the same way as theta activity detected in the cortex [2, 28, 57–59]. To eliminate the confounding factor due to volume conduction, we performed another set of experiment in which the LFPs in the hippocampus, cortex (CT) and NTS were simultaneously recorded using the same electrode configuration and recording paradigm as used in this study. The results showed that phase coherences of both location pairs (HC-CT and HC-NTS) were higher for theta oscillations (6–8 Hz) compared to surrounding frequencies but were negligible for HFOs (Figure 7). These results support the idea that HFOs are locally generated and thus independent, while low-frequency oscillations are more associated with the larger neuronal network and can passively spread over a large area from their origins [22]. Interestingly, unlike the high value of HC-CT theta phase coherence (coherence value = 0.55), which is known to be due to volume conduction, HC-NTS theta phase coherence value was low (coherence value = 0.2) and comparable to the coherence value for HFOs (coherence value = 0.15). This result combined with high NTS_theta-HC_HFOs coupling may suggest that both HC and NTS can independently generate theta rhythms which directly modulate the HC_HFOs.

To further disentangle local contributions from volume conduction effects, we also analyzed the LFPs recorded from the neighboring brainstem region, the Cuneate nucleus (Cu), located 0.5 mm dorsal to the NTS, during REM. Five recordings were identified histologically to have the implanted electrode misplacement into the Cu instead of the NTS and therefore be excluded from the NTS analysis. Figure 8A showed representative histological images of the electrode traces in HC, NTS, and Cu corresponding to the schematic coronal sections of the rat brain. The raw HC and NTS LFPs showed clear theta activity when compared to that of the Cu (Figure 8C). The power spectral analysis of the LFPs recorded from the Cu showed no theta (4–10 Hz) peak (Figure 8B) and significantly lower power of theta frequency range (6–8 Hz) compared to that of the NTS (NTS_theta = 2.83 ± 0.16 VS Cu_theta = 1.43 ± 0.22) (p < .05) (Figure 8D). Importantly, CFC was exclusively present between NTS_theta-HC_HFOs (MI, NTS_theta-HC_HFOs = 0.0016 ± 0.0001 vs. CU_theta-HC_HFOs = 0.0004 ± 0.00006, p < .05) (Figure 8E and F)

Figure 8.

Clarification of the origin of theta in the NTS during REM sleep. (A) (Left) Schematic representation of the coronal sections of the rat brain showing identified local field potential (LFP) recording sites in the HC, NTS, and Cu (adapted from the Rat Brain Atlas http://labs.gaidi.ca/rat-brain-atlas/). (A) (Right) representative images of histological verification of electrode traces (arrows) corresponded to the left schematic atlas. (B) Simultaneous recordings were performed in the HC and NTS (n = 8 rats) or HC and Cu (n = 5 rats) during REM. Higher power in the theta range (4–10 Hz) (rectangle) was found in the HC and NTS coordinates but not Cu coordinates. (C) Raw LFPs showing pronounced theta activity in the HC and NTS but irregular activity in the Cu. (D) Corresponding theta (6–8 Hz) power calculated from the LFPs recorded in the NTS and Cu coordinates as shown in B. (E) Mean phase-to-amplitude comodulograms obtained from 10 s-segments between NTS and HC (NTS _Phase-HC _Amp) and between Cu and HC (Cu _Phase-HC _Amp). (F) Mean theta (6–8 Hz)-HFO (120–160 Hz) CFCs are significantly higher in NTS_theta-HC_HFOs when compared to Cu_theta-HC_HFOs (Student’s t-test, p < .05). REM sleep: rapid eye movement sleep, HC: hippocampus, NTS: nucleus tractus solitarius, Cu: cuneate nucleus, HFOs: high-frequency oscillations.

Discussion

Hippocampal theta-associated HFOs in memory functions and REM sleep

HFOs in the range of 110–160 Hz have been recently described and shown to be coupled with theta activity in many studies [2, 8, 22, 24, 25]. Some studies reported the co-occurrence of theta-gamma and theta-HFO couplings [8, 24]. In our study, we observed exclusive theta (6–8 Hz)-HFO (120–160 Hz) coupling without theta-gamma coupling in the hippocampus. This is because the patterns of theta-phase coupling are layer-specific. Theta-HFO (140 Hz) coupling is strongest in the stratum oriens-alveus (superficial CA1 layer) which is the recording site in our study while theta-gamma (80 Hz) coupling is highest in the stratum lacunosum moleculare (deeper CA1 layer) [24].

The functional role of hippocampal theta-associated HFOs is currently not understood. Exaggerated, or impaired CFC in the hippocampus has been reported in many neurodegenerative diseases [16–19]. Recent evidence showed that theta-HFO coupling is behavioral state specific [8, 22, 60]. Here, in agreement with previous reports [22, 55], we found that REM sleep has been shown to elicit stronger theta-HFO coupling when compared to the AE state. REM sleep is the state that has the highest level of the plasticity-related neuromodulator, acetylcholine, compared to active waking, quiet waking, and non-rapid eye movement sleep [61–64]. Also, theta-associated REM sleep was related to memory consolidation in the hippocampus [65]. Thus, theta-HFO coupling augmented during REM sleep would represent neural integration that is needed for local communication within the hippocampus and possibly associated with memory functions. Our work also showed that theta-HFO coupling strength is positively correlated with theta power which is congruent with many studies [11, 22, 55]. Therefore, theta power should be considered as a confounding factor when studying the role of theta-HFO coupling. Altogether, our data showed that not only brain state but also theta power that have an influence on theta-HFO coupling strength.

NTS theta-hippocampal HFO coupling during REM sleep

The NTS receives inputs from peripheral receptors and other parts of the brain to regulate autonomic function [66, 67]. NTS projection targets such as medial prefrontal cortex [68], nucleus accumbens [93], thalamus, septal area, bed nucleus of the stria terminalis, dorsomedial, and paraventricular hypothalamus [37] suggest a role in modulating cortical and subcortical activities associated with sleep-wake cycle and learning processes. Although there is no direct anatomical projection from the NTS to the hippocampal formation, NTS has been proposed to play a role in hippocampal theta generation [69] probably through the medial septal nucleus and vertical limb of the nucleus of the diagonal band of Broca (MS/vDBB). The direct chemical (carbachol) [70] or electrical stimulation [71] of the NTS or the electrical vagal nerve stimulation [72] increased the production of hippocampal theta rhythm. Studies have found that vagal nerve stimulation can cause changes in hippocampal EEG in waking animals [42, 43] and enhance memory in both rats [44, 45] and humans [73]. These results suggest that anatomical connections between NTS and hippocampal structure do exist through multisynaptic pathways [39]. One of the proposed pathways is NTS → Paragigantocelluaris → locus coeruleus → CA1 region of dorsal hippocampus [74] which contributes to consolidation of object recognition memory [39–41]. This pathway demonstrates a connection between visceral influences and higher neurobehavioral processes.

Most of the CFC studies include only the cortical and subcortical structures and focus principally on the coupling within a single brain region. To our knowledge, the study of NTS-cortical structure CFC is still lacking, in part, due to the difficulty of recording from vital brainstem areas in awake animals. In this study, we showed for the first time that the NTS exhibits theta oscillations (6–8 Hz) which are coupled to the HFOs (120–160 Hz) in the hippocampus during AE and REM. This coupling showed that distant neural structures, like the NTS, can establish and express functional connectivity with a cerebral structure. Given that the hippocampus plays an important role in modulating visceral learning mediated by the NTS [40], one can speculate that NTS theta-hippocampal HFO coupling reflects the acquisition and consolidation of memory, the processes which are highly dependent on the sensory information provided by the NTS. However, the lack of hippocampal theta phase modulation of NTS HFOs raises the possibility that this coupling is asymmetric.

Although most sensory inputs do not reach conscious perception during sleep, the sensory information is still being processed [75, 76]. In this study, we found that both NTS theta power and NTS theta-hippocampal HFO coupling were more pronounced during REM compared to AE which suggest that the NTS essentially modulates the hippocampal local network during active sleep. The NTS, a sensory hub of the brainstem, may still continuously report to the higher brain center about the internal and external environment. Given that REM sleep has been shown to support memory processing [77–79] and sensory information plays an important role in sensory-modulated memory formation [80], augmented NTS theta-hippocampal HFO coupling during REM may represent the ongoing sensory processing and changes in the central nervous system regarding learning, memory, and other body functions during sleep. Furthermore, recent data suggest the positive correlation between high-frequency (HF: 0.15–0.4 Hz) component of heart rate variability, a variance between consecutive heartbeats, which is related to parasympathetic activity, and the cognitive performance during REM sleep in humans [68]. This raises the possibility that autonomic inputs integrated by the NTS reach the hippocampus and NTS theta-associated hippocampal HFOs may be the neural mechanism that maintains this long-range connectivity.

The patterns of theta phase-hippocampal HFO amplitude coupling during REM sleep

The modulation of hippocampal HFOs by theta phase exhibited consistent patterns. During REM, HFOs peaked at the falling phase of hippocampal theta wave. This result is similar to HFO power that was maximal on the falling phase or near the peak of the theta wave in LFPs that were recorded from the superficial layers in the CA1 region of the hippocampus of the rats undergoing T-maze tasks [8]. On the other hand, the preferred NTS theta phase for hippocampal HFO powers is around the peak. Like the phase coding in hippocampal place cells [81] and phase-coding model of working memory [82], our studies suggest that the patterns of cross-frequency coupling as well as the preferred phase of coupling may help define the state of vigilance and characterize the brain areas that are involved in cognitive states.

The origin of NTS theta

Theta rhythm is very well known to be generated within the hippocampus and volume-conducted to other brain regions such as the cortex. This study showed for the first time that the theta rhythms are also detected in the NTS and NTS theta modulates the hippocampal HFOs. Given the fact that the NTS theta power detected in this study is quite low and cellular mechanisms of NTS theta generation are still unknown, one may argue that theta rhythms are not generated in the NTS itself. Rather, they may emerge from the network formed by the hippocampus and medial septum-diagonal band of Broca [59]. To assess whether NTS theta oscillations were locally generated rather than volumed-conducted from other structures, like hippocampus, we conducted several analyses and the results suggested that NTS theta rhythm is largely generated within the NTS. First, the NTS theta oscillations were detected by twisted bipolar electrodes which allow the removal of common average activity that may be due to the external sources. This type of electrode has been proved to record field potentials generated in a particular circuit even during a surge of excessive electrical activity such as epileptic seizures [47, 83]. Second, the HC_HFOs were highly modulated by both HC_theta and NTS_theta at different but consistent phase preferences. Also, the mean theta phase difference between HC and NTS was not related to the different phases of HFO appearance when comparing the hippocampal theta cycle with NTS theta. Both results suggested the independence of theta oscillations between these two regions. Third, the phase coherence in the theta range between HC and NTS was low while both HC_theta-HC_HFO coupling and NTS_theta-HC_HFO coupling were high. The low phase coherence values between these two regions reduces the possibility of passive field spread. This would imply that both HC and NTS theta frequencies are not only independent, but both contribute to modulating the HFO rhythms in HC. Fourth, LFPs recorded from Cu, which is located near the NTS, did not show a peak of theta activity. Also, Cu_theta-HC_HFOs coupling during REM was not present. If NTS theta rhythms were not locally generated but volume-conducted from the distant brain area such as the hippocampus, the NTS and Cu field potentials should equally exhibit theta oscillations. Altogether, our data support the idea that theta oscillations observed in NTS LFPs are not contaminated by volume-conducted signals.

The next interesting question is how theta oscillations were generated in the NTS. It was proposed that a given brain structure capable of producing local field theta oscillations should possess rhythmically firing theta-on phasic neurons. Unfortunately, this information is currently unknown for the NTS. Therefore, detailed studies of the physiology and pharmacology of theta rhythm observed locally in the NTS seem to be particularly important. The simultaneous recording of spiking activity and LFPs in the NTS during REM should also be conducted to assess if spikes favorably occurred at a certain phase of theta oscillations and to identify theta-related neurons within the neural circuits of the NTS. The NTS is composed of quite heterogeneous neurons [84, 85]. A significant portion of the cells are glutamate decarboxylase 2-positive GABAergic neurons, forming an inhibitory network within the NTS [86, 87]. The presence of GABAergic networks in the NTS suggests the capability of theta rhythm production through similar mechanisms as occur in the hippocampal network. Apart from the intrinsic properties of the neurons, theta rhythm generation also depends on the adequate inputs that can induce or modulate theta oscillations. The lack of inputs from hippocampus or medial septum, the regions involved in theta generation, and low synchronization between NTS and HC theta during REM emphasized that the theta rhythms in these two regions may not share a common neural generator. The NTS receives inputs from the pedunculo pontine tegmentum (PPT) [88], the cholinergic nucleus of the brainstem that controls theta rhythm generation in the hippocampus [89] and is known to be involved in the initiation of REM sleep [90]. Here, we proposed that the NTS theta generation may arise through the cholinergic inputs from the PPT.

Altogether our work provides a support for the idea that the NTS theta is generated locally. However, the in-depth study into the electrical/chemical stimulation, chemical lesions, firing properties of the NTS neurons with the induction of theta waves in the hippocampus or other structures complicated in theta oscillations and the correlation of NTS neuronal activity and behavioral states are needed to uncover the cellular mechanisms of theta generation within the NTS.

In summary, apart from the well-known CFC within the hippocampus, we reported that NTS also exhibits theta oscillations and NTS theta-hippocampal HFO CFC is prominent during REM sleep. These results suggested that phase-amplitude coupling may reflect the recruitment of different neuronal circuits under different vigilance states. The increased NTS theta-associated hippocampal HFO during REM may indicate a possible role for the NTS as a sensory input to the hippocampal network. If so, this input may help support the basic mechanisms for memory encoding and retrieval [91, 92] of the hippocampus during REM. In addition, the peak amplitude of hippocampal HFO is modulated by a consistent theta phase recorded from hippocampus and NTS. This differential modulation of HFO peaks by theta phases from different brain regions may extend the understanding of the function of neuronal network in health and disease. Further experiments are needed to test whether exposure to a novelty such as performing the object recognition task would change the state-dependence and the pattern of NTS-hippocampal coupling. Electrophysiological experiments both in vitro and in vivo are also needed to uncover a possible role of theta-nested HFO oscillations as well as a cellular mechanisms underlying the generation of those interactions.

Funding

This work was financially supported by the Canadian Institutes of Health Research (CIHR) and Finding A Cure for Epilepsy and Seizures (FACES) foundation, USA.

Disclosure statement

The authors declare no competing financial interests and no conflicts of interest in the writing of this article.