Aberrant thalamocortical coherence in an animal model of tinnitus

Abstract

Electrophysiological and imaging studies from humans suggest that the phantom sound of tinnitus is associated with abnormal thalamocortical neural oscillations (dysrhythmia) and enhanced gamma band activity in the auditory cortex. However, these models have seldom been tested in animal models where it is possible to simultaneously assess the neural oscillatory activity within and between the thalamus and auditory cortex. To explore this issue, we used multichannel electrodes to examine the oscillatory behavior of local field potentials recorded in the rat medial geniculate body (MBG) and primary auditory cortex (A1) before and after administering a dose of sodium salicylate (SS) that reliably induces tinnitus. In the MGB, SS reduced theta, alpha, and beta oscillations and decreased coherence (synchrony) between electrode pairs in theta, alpha, and beta bands but increased coherence in the gamma band. Within A1, SS significantly increased gamma oscillations, decreased theta power, and decreased coherence between electrode pairs in theta and alpha bands but increased coherence in the gamma band. When coherence was measured between one electrode in the MGB and another in A1, SS decreased coherence in beta, alpha, and theta bands but increased coherence in the gamma band. SS also increased cross-frequency coupling between the phase of theta oscillations in the MGB and amplitude of gamma oscillations in A1. Altogether, our results suggest that SS treatment fundamentally alters the manner in which thalamocortical circuits communicate, leading to excessive cortical gamma power and synchronization, neurophysiological changes implicated in tinnitus. Our data provide support for elements of both the thalamocortical dysrhythmia (TD) and synchronization by loss of inhibition (SLIM) models of tinnitus, demonstrating that increased cortical gamma band activity is associated with both enhanced theta-gamma coupling as well as decreases alpha power/coherence between the MGB and A1.

NEW & NOTEWORTHY There are no effective drugs to alleviate the phantom sound of tinnitus because the physiological mechanisms leading to its generation are poorly understood. Neural models of tinnitus suggest that it arises from abnormal thalamocortical oscillations, but these models have not been extensively tested. This article identifies abnormal thalamocortical oscillations in a drug-induced tinnitus model. Our findings open up new avenues of research to investigate whether cellular mechanisms underlying thalamocortical oscillations are causally linked to tinnitus.

INTRODUCTION

Subjective tinnitus is a phantom auditory sensation often described as a ringing, buzzing, or sizzling sound (Eggermont 2007; Henry et al. 2014; Mühlnickel et al. 1998). Approximately 12–15% of adults have experienced tinnitus, but 1% suffer from severe or disabling subjective tinnitus for which they seek medical treatment (Heller 2003; Hoffman and Reed 2004; Shargorodsky et al. 2010). Despite the severity of its symptoms and widespread occurrence, there are no drugs approved by the Food and Drug Administration (FDA) to treat this disabling brain disorder (Salvi et al. 2009). Most cases of subjective tinnitus are associated with sensorineural hearing loss, which in many cases goes undetected because of incomplete assessment of the audiogram or because the underlying cochlear pathologies are not detected by conventional clinical audiometric testing (Salvi et al. 2017; Weisz et al. 2006). There is growing recognition that cochlear damage gives rise to tinnitus by inducing aberrant activity in neural networks within the central nervous system (Chen et al. 2015; De Ridder et al. 2011, 2014; Eggermont and Roberts 2004; Henry et al. 2014; Leaver et al. 2011; Rauschecker et al. 2010; Roberts et al. 2010; Salvi et al. 2010; Seydell-Greenwald et al. 2012). While the central origins of tinnitus are now widely accepted, the exact location and nature of these central auditory changes are still debated and many models of subjective tinnitus have been proposed (Henry et al. 2014). Several lines of evidence suggest that hearing loss triggers abnormal synchronization of neural activity within the classical auditory pathway as well as nonauditory brain regions (De Ridder et al. 2014; Rauschecker et al. 2010; Seydell-Greenwald et al. 2012). Thus, aberrant interregional synchrony is postulated to be a key contributor to tinnitus perception and/or distress (De Ridder et al. 2015; Llinás et al. 1999, 2005).

The coordinated activation of large groups of neurons distributed across and within specialized brain regions is thought to be essential for cognition and perception (Melloni et al. 2007). Large-scale synchronization of neural activity is evident in the rhythmic oscillatory behavior of voltage fluctuations observed with extracellular electrophysiological recordings, and there is an immense body of research linking the power and phase-coherence of different frequencies of brain oscillations to cognitive processes (Sedley and Cunningham 2013). Electroencephalography (EEG) and magnetoencephalography (MEG) are noninvasive recordings techniques that have allowed for the characterization of cortical oscillatory behavior from tinnitus patients. Numerous studies have revealed abnormal rhythms in delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta (15–25 Hz), and gamma (30–90 Hz) frequency bands. One of the most consistent findings in these studies is increased gamma band activity in tinnitus patients (De Ridder et al. 2011, 2014; Hartmann et al. 2014; Lorenz et al. 2009; Moazami-Goudarzi et al. 2010; Ortmann et al. 2011; Sedley et al. 2012; van der Loo et al. 2009; Vanneste et al. 2010, 2011a; Weisz et al. 2005a, 2006, 2007a, 2011). Increased gamma band activity is also associated with changes in slow wave cortical oscillations, such as a reduction in alpha and/or an increase in delta/theta band activity (Adjamian et al. 2012; Ashton et al. 2007; Llinás et al. 1999; Lorenz et al. 2009; Sedley et al. 2012; Weisz et al. 2007a). These findings and others have led to models of tinnitus generation focused on abnormal neural synchrony within and between the cortex and thalamus, in particular the thalamocortical dysrhythmia (TD) and the synchronization by loss of inhibition (SLIM) models (Jeanmonod et al. 1996; Llinás et al. 2005; Schlee et al. 2008; Weisz et al. 2007a).

According to the TD and SLIM models, the loss of feedforward input to the central auditory system (deafferentation) due to cochlear hearing loss causes thalamocortical auditory networks to reverberate in an abnormal rhythmic pattern, resulting in the perception of sound in the absence of stimuli (i.e., tinnitus). Thalamocortical neural networks are essential for acoustic perception (Lee 2013) and imaging studies indicate that the auditory cortex and thalamus are overactivated during tinnitus (Gu et al. 2010; Lanting et al. 2014; Lockwood et al. 2001; Rauschecker et al. 2010; Reyes et al. 2002). Consistent with these observations, morphological changes have been observed in the medial geniculate body (MGB) of tinnitus patients (Mühlau et al. 2006) and partial resection of the suprageniculate-limitans area of the MGB can relieve tinnitus symptoms (Jeanmonod et al. 1996). The TD model proposes that loss of auditory input leads to increased rhythmic bursting of thalamic cells, reflected by increased low-frequency thalamic oscillations in the delta/theta range, which in turn activates the auditory cortex, resulting in increased activity in the delta and gamma ranges (Llinás et al. 1999). The SLIM model suggests deafferentation-induced hearing loss leads to a downregulation of inhibitory inputs to the auditory cortex. This results in decreased alpha band activity, which in turn leads to increased synchronization of activity between auditory cortical neurons, reflected in increased power and coherence of gamma oscillations (Weisz et al. 2007a). These models suggest not only a neural signature for tinnitus but also a plausible anatomical locus and molecular mechanisms for tinnitus generation, which could enable the development of new pharmacotherapies.

While EEG and MEG studies have provided support for the TD and SLIM models, these surface recordings lack the precision needed to pinpoint the exact sources of the neural generators in the thalamus and cortex. Thus, the precise role of thalamocortical dysrhythmias in tinnitus remains poorly understood. Electrophysiological recordings from electrodes implanted in both the A1 and MGB of animals could test some predictions put forward by the TD and SLIM models, provided that tinnitus can be reliably induced. High doses of sodium salicylate (SS), the active ingredient in aspirin, have long been known to reliably induce sensorineural hearing loss and tinnitus in humans (Dequeker and Mardjuardi 1981; Halla and Hardin 1988; Mongan et al. 1973). SS treatment also consistently induces behavioral evidence of hearing loss and tinnitus in diverse animal models as well (Chen et al. 2013; Jastreboff et al. 1988; Myers and Bernstein 1965; Yu et al. 2016). The consistent and reliable induction of tinnitus by salicylate has made it an indispensable tool for understanding the neural correlates of tinnitus (Stolzberg et al. 2012). Therefore, we carried out experiments in which we treated rats with a high dose of SS known to induce tinnitus while simultaneously making extracellular multisite recordings in the MGB and A1 cortex to determine whether SS disrupts thalamocortical oscillations as predicted by the TD and SLIM models.

METHODS

Subjects.

Ten Sprague-Dawley rats (male, ~2 mo old, ~230 g) were used in this study (Charles River Laboratories, Wilmington, MA). The animals were housed in the laboratory animal facility at the University at Buffalo (22°C, 12:12-h light-dark cycle) and given free access to food and water. All procedures regarding the use and handling of animals were reviewed and approved by the Institutional Animal Care and Use Committee at the University at Buffalo.

Salicylate treatment.

There is a strong linear relationship between plasma levels of unbound salicylate and decreased auditory sensitivity (Day et al. 1989). Tinnitus and hearing loss start within minutes after salicylate ingestion and tinnitus subsides within 24–48 h posttreatment (Guitton et al. 2003; Mongan et al. 1973). In animal models, behavioral evidence of tinnitus and hearing loss is evident within 1 h and disappears 48–72 h posttreatment (Lobarinas et al. 2004; Radziwon et al. 2015). The lowest dose that induces tinnitus sensation is around 150 mg/kg (Lobarinas et al. 2004; Stolzberg et al. 2012). SS treatment reduces the neuronal output of the cochlea within 2 h posttreatment while it paradoxically increases sound-evoked neuronal responses in the MGB and A1 (Chen et al. 2013; Jiang et al. 2017; Sheppard et al. 2014). Generally, spontaneous activity is not changed or decreases slightly postsalicylate (Stolzberg et al. 2012). Based on these results we used a 200 mg/kg dose of SS (Sigma; St. Louis, MO) and a time course of 2 h to investigate whether SS treatment modulates MGB-A1 neuronal coherence as proposed by SLIM and TD models of tinnitus.

Surgical procedures.

Details of the experimental procedures can be found in our earlier publications (Chen et al. 2013; Stolzberg et al. 2011). Rats were anesthetized with a cocktail of ketamine (60 mg/kg ip) and xylazine (6 mg/kg ip) supplemented with half-doses of the mixture approximately every hour or as needed to maintain a suitable plane of anesthesia. Body temperature was maintained at 37°C using a homeothermic heating pad (Harvard Apparatus, Holliston, MA). Anesthetized rats were placed in a stereotaxic apparatus with blunt ear bars. Using appropriate stereotaxic coordinates, an opening was made on the dorsal and lateral skull to gain access to the thalamus and left A1, respectively. After the dorsal and lateral surfaces of the skull were exposed, a head bar was firmly attached using screws and dental cement. The head bar assembly was used to firmly hold the rat’s head in the stereotaxic frame after removing the right ear bar to allow for acoustic stimulation using a free-field loudspeaker (FT28D Dome Tweeter, Fostex, Tokyo, Japan). The dura mater was carefully removed from the surface of the cortex. Initially, a tungsten microelectrode (~1 MΩ; FHC, Bowdoin, ME) mounted on a microdrive and guided by stereotaxic coordinates was directed toward the MGB while presenting noise bursts (50 mg, 5 ms rise/fall time, 10 Hz); these initial recordings were used to confirm the location of the MGB. A similar procedure was used to identify the boundaries of A1. After the locations of A1 and the MGB were verified, 16-channel silicon microelectrode arrays (NeuroNexus Technologies; Ann Arbor, MI) were inserted into A1 (A4x4-3mm-100-177) and the MGB (A1x16-10mm-100-177).

Acoustic stimuli.

Broadband noise (1–42 kHz) and tone bursts (50 ms; 5 ms rise/fall cos² gating) were generated using a TDT RX6-2 processor (~100 kHz sampling rate). Tones were presented in 20 logarithmic frequency steps between 1 and 42 kHz with intensities from 0 to 100 dB SPL in 10-dB steps (10 repetitions per frequency/level; pseudorandom order).

Electrophysiological recordings.

Spontaneous and tone-evoked multiunit spike discharges and local field potential (LFP) signals were acquired with a Tucker-Davis System 3 processor (TDT; Alachua, FL). Signals were preamplified by a RA16PA (bandpass: 2 Hz to 7.5 kHz, 3 dB down) and sampled at 25 kHz by a RX5 base station. Custom-written data acquisition and analysis software (MATLAB R2007b, MathWorks) were used to acquire the data as previously described (Stolzberg et al. 2011; Chen et al. 2013). Spike detection was performed online and maintained throughout the experiment by using a manually set voltage threshold. Spike discharges were digitally filtered between 300 and 3,500 Hz. LFPs were continuously acquired and digitally resampled at 610 Hz and bandpass filtered from 2 to 300 Hz. Recordings were obtained before and after an intraperitoneal injection of SS (Sigma; 200 mg/kg). The stimuli were delivered through a loudspeaker (FT28D, Fostex) located 10 cm directly in front of the right ear. Stimuli were calibrated using ¼-in. microphone, microphone preamplifier (Larson Davis, model 2221), and sound level meter (Larson Davis, model 2520).

Multiunit analysis.

We computed the instantaneous firing rates from each multiunit cluster (MUC) for each stimulus presentation, which were used to construct peristimulus time histograms (PSTHs) and raster plots from −50 to 250 ms relative to tone burst onset. This was accomplished using 1 ms bins for each stimulus trial; each 1 ms bin contained 1 or 0 timestamps representing the discharges from the MUC as a series of zeros and ones across the −50 to 250 ms analysis window that was displayed as a raster plot (Fig. 1). Then, each column of this matrix was smoothed with a 3-ms Gaussian sliding window, which transformed each binned vector of zeros and ones into a continuous waveform that represents the instantaneous firing rate of each trial for each MUC. This procedure allowed us to compute the mean and standard deviation of the instantaneous firing rate across 50 trials; thereby, allowing us to statistically compare the effect of SS treatment in the firing rate of each MUC. The matrix of PSTHs was used to define the neuron’s frequency receptive field (FRF), which was plotted as a color-coded heat map (Fig. 1, A and B). The map was constructed by using mean evoked firing rates to tones of different frequency and intensities. In addition to electrode position, the response latencies, shape and sharpness of the FRFs and features of the LFP waveform were consistent with recordings from MGB and the A1 (Fig. 1). FRFs were only generated before SS treatment to ensure that the recordings were A1 and MGB, but significant changes in shape and sharpness of FRF in MGB and A1 have been previously described by our group and others (Chen et al. 2015; Ochi and Eggermont 1996).

Fig. 1.

Tuning and hyperactivity in medial geniculate body (MGB) and primary auditory cortex (A1). Representative frequency receptive field (FRF) of a multiunit cluster (MUC) in A1 (A) and MGB (F). Heat map on right shows firing rate associated with each color in the intensity vs. frequency FRF. Raster plots from one representative MUC in A1 (B and C) and one MUC in MGB (G and H) pre- and 2 h post-SS. Representative peristimulus time histograms (PSTHs) from one MUC in A1 (D) and one MUC in MGB (I) in response to 50 trials of a 90 dB SPL, 50 ms noise burst pre- (red line) and 2 h post-sodium salicylate (SS) (blue line). Symbols on PSTH identify time points when the firing rate post-SS were significantly greater than pretreatment (Wilcoxon signed-rank test with P-values adjusted for multiple comparisons using the false discovery rate approach; values with P < 0.01 considered significant). Boxplot (red line is mean, blue box is 95% CI of the mean and circle dots are baseline firing in each MUC) of the baseline firing rate in A1 (E) and MGB (J) displaying statistical comparison of the firing rate before and after SS treatment (t-test, P < 0.05).

Spectral analysis.

We recorded 10 min of spontaneous LFP baseline activity for each experimental condition. The baseline LFP activity was subdivided in 30 trials of 10 s long. These trials were used to compute the power spectral density (PSD), spectrograms, and coherence analysis for each individual LFP site. Prior to spectral analysis, the LFP signal was visually inspected and noisy sections of the signal were discarded. To remove 60 Hz noise that contaminated a small fraction of the recording channels, the LFP was filtered around 60 Hz using a linear finite impulse response (FIR) filter provided by the “eegfilt” function from the EEGLAB Toolbox (Delorme and Makeig 2004) . This filter uses the two-way least-squares FIR filtering method with the following parameters: low-pass of 55 Hz, high-pass of 65 Hz, 3-point filter order with a roll-off of 18 dB/octave. Pre-SS baseline measures were compared with LFP activity acquired 2 h post-SS and analyzed in the same manner.

Oscillations in the MGB and A1 were estimated by computing the spectrogram and power spectrum of the LFP. These spectral analyses were performed as previously described (Vianney-Rodrigues et al. 2011) by using the multitaper method provided by the MATLAB Chronux toolbox (chronux.org). The multitaper method has been used extensively to compute spectrograms, PSD, and coherence between two signals in several brain areas; details of the methodology are reported elsewhere (Jarvis and Mitra 2001; Pesaran et al. 2002; Prechtl et al. 1997). This method utilizes an optimal set of orthogonal tapers known as prolate spheroidal functions (Slepian functions) that have optimal variance and bias properties; the tapers are applied before performing the fast Fourier transform (FFT) algorithm. Slepian functions are concentrated in a specific time duration (T) and frequency bandwidth (W); for each choice of T and W; a maximum of K = 2TW−1 tapers can be used for spectrum estimation. Each data epoch is multiplied for each of these orthogonal tapers and then Fourier transformed, which gives the windowed Fourier transform: ${\tilde{X}}_k\left(f\right)={\displaystyle \sum }_t^N{W}_k\left(t\right)X_t{e}^{-2\pi ift}$ in which X_t is the time series of the signal being considered (LFP) and W_k(t) are taper functions.

Coherence is the magnitude squared of the correlation between two Fourier transform signals (the cross-spectrum of two signals) normalized by the power spectrum of each signal. The cross-spectrum is the sum of the product of the two-signal spectrum correlation. The spectrum and cross-spectrum are averaged over trials/channels to compute the coherence. Coherence measures the strength of the correlation between two signals as a function of frequency and provides a normalized linear association of phase and amplitude between two signals (Halliday and Rosenberg 1999). Coherence is a normalized value between 0 and 1 where 0 indicates no coherence between two signals and 1 indicates that the phase and amplitude of the two signals are perfectly coherent.

Cross-frequency coupling analysis.

To assess cross-frequency coupling (CFC) between MGB and A1 electrode pairs, we computed the modulation index (MI) between the phase of theta (4–8 Hz) oscillations in the MGB and the amplitude of oscillations in A1 across frequencies (5–90 Hz). Phase-amplitude coupling between MGB and A1 electrode pairs was computed for 40 electrode pairs from each animal (n = 10), for a total of 400 combinations, before and 2 h post-SS treatment. The 10-min recordings of spontaneous LFP activity were subdivided into thirty 20-s trials for each experimental treatment. MI and CFC analyses were performed as described previously (Scheffer-Teixeira et al. 2012; Tort et al. 2010). Briefly, the MI measures the CFC strength between a phase-modulating signal (in our case, 4–8 Hz which is the theta frequencies in MGB) and an amplitude-modulated signal (in our case, 5- to 90-Hz frequencies in A1). First, the raw LFP in A1 and MGB are filtered for the frequencies of interest by using the eegfilt filter as described above. Then, a standard Hilbert transform was applied to the filtered data to extract the time series of the amplitude envelope in the A1 signal and the phase in the MGB signal. Next, the phase values of the MGB signal was binned into eighteen 20° intervals (0° to 360°), the mean amplitude of the A1 signal over each phase bin is computed, and the mean amplitude is normalized by dividing each bin value by the sum of all bin values. This procedure generates a phase-amplitude distribution, in which a uniform phase-amplitude distribution means that there is no phase-amplitude coupling. The MI measures how divergent our empirical phase-amplitude distribution P is from a uniform distribution U. This can be computed by the Kullback-Leibler divergence: KL(P,U) = log(N) + Σpjlog(pj), where N is the number of phase bins (18) and pj is the amplitude across phase bins. MI is a normalized measure between 0 and 1 and it is given by MI = KL(P,U)/log(N). A comodulation heat-map was obtained by computing the MI across the phase-modulated (x-axes in Fig. 7) and amplitude-modulated signal (y-axes in Fig. 7) band pairs in a bidimensional map. Warm colors represent the strength to which theta phase in the MGB modulates the amplitude of frequencies in A1.

Fig. 7.

Sodium salicylate (SS) alters cross-frequency coupling (CFC) between medial geniculate body (MGB) and primary auditory cortex (A1). Representative trials showing CFC heat maps between theta phase in MGB and gamma amplitude in A1 before (Ai) and after (Aii) SS. B: mean CFC plots (n = 30 trials, 10 s each) of a representative MGB/A1 electrode pair pre-SS (Bi), 2 h post-SS (Bii), and statistical comparison (Biii) showing P-value heat maps (Wilcoxon signed-rank test with P-values adjusted for multiple comparisons using a false discovery rate method; P < 0.05 in red). C: barplot of the CFC modulation index between theta phase in MGB and gamma amplitude in A1 before and after SS treatment (permutation t-test, 5,000 permutations, P-values shown on histogram, P < 0.01 considered significant).

Statistics.

The Wilcoxon sign-rank test was used to evaluate differences in MUC firing rates, LFP spectrograms, coherograms, and comodulograms before and after SS treatment. The Wilcoxon sign-rank test comparisons between the pre- and post-SS spectrograms, coherograms, and comodulograms were performed on a pixel by pixel basis, which yielded ~4,000 tests. To correct the P-values for multiple comparisons, we used a false discovery rate (FDR) method to reduce Type I errors due to multiple comparisons (Benjamini and Hochberg 1995; Groppe et al. 2011). P-values ≤ 0.05 were used for assigning statistical significance in the spectrogram, coherogram, and comodulogram plots. Differences in mean PSD values across all animals (n = 10) and trials (n = 30) were determined using a permutation t-test with 5,000 permutations to construct null distributions. All values displayed in the bar plots were expressed as mean and P ≤ 0.05 was used for assigning statistical significance. All statistical analyses were performed in MATLAB (MATLAB R2014b, MathWorks).

RESULTS

LFP tuning.

The tone-evoked LFPs and MUCs recorded from most electrodes were tuned to a narrow frequency range. This is illustrated by the FRF heat maps obtained from one electrode site in A1 (Fig. 1A) and from another electrode in the MGB (Fig. 1B). Most FRFs from MUCs in A1 and the MGB had a low-threshold, narrowly tuned tip and a high-threshold, broadly tuned low-frequency tail. In this example, the MUC in A1 had a characteristic frequency of ~30 kHz while the one in the MGB had a characteristic frequency of ~15 kHz. These results confirm that our recordings were made from primary auditory thalamus and cortex.

Salicylate effects on spontaneous and sound-evoked spiking activity.

While SS administration significantly reduces the neural output of the cochlea, it paradoxically enhances suprathreshold sound-evoked activity in A1 and MGB (Chen et al. 2013; Stolzberg et al. 2011; Sun et al. 2009; Wei et al. 2010). To confirm that our dose of SS induced hyperactivity, we obtained MUC raster plots and PSTHs in response to 50-ms noise bursts presented at 90 dB SPL pre- and post-SS. Figure 1 shows representative raster plots and PSTHs of MUC in A1 (Fig. 1, A–D) and MGB (Fig. 1, F–I) before and after SS treatment. The PSTHs show the mean instantaneous firing rate (IFR) as function of time pre- and 2 h post-SS. There was a significant enhancement (symbols on PSTH) of the IFR in A1 (Fig. 1D) at ~15 ms and a significant increase in IFR in the MGB (Fig. 1I) around 10 ms (Wilcoxon test, P < 0.01, FDR-corrected for multiple comparisons). Approximately 42% of neurons in A1 and 49% of MGB neurons showed a significant increase in firing rate post-SS consistent with our previous report (Chen et al. 2013). We found a small but significant reduction in baseline firing rate in the thalamus (Fig. 1J; Wilcoxon test, P = 0.03, n = 79 units) but no difference in the baseline firing rate of A1 (Fig. 1E; Wilcoxon test, P = 0.06, n = 127 units), consistent with previous results (Ochi and Eggermont 1996).

A1 oscillations.

Using two multichannel electrodes, we recorded spontaneous LFPs from 155 electrode sites in A1 and 100 sites in the MGB pre- and post-SS. Figure 2 shows representative spectrograms of spontaneous LFPs from A1 pre-SS (Ai and Bi) and 2-h post-SS (Aii and Bii); the low-frequency (Fig. 2A) and high-frequency (Fig. 2B) data were plotted separately for ease of comparison. The heat map corresponds to the power (mV²/Hz) at each frequency over time. Figure 2, Aiii and Biii are statistical maps comparing the pre-SS and post-SS spectrograms; the heat map shows P-values corresponding to the frequency time point on the spectrogram. Prior to SS treatment, A1 displayed strong oscillatory activity in the 8–15 Hz range (Fig. 2Ai); the activity in this band decreased, but the reduction was not statistically significantly 2 h post-SS (Fig. 2Aiii; Wilcoxon signed-rank test P > 0.01, FDR-corrected for multiple comparisons). However, there was a statistically significant decrease in power in the theta (4–8 Hz) band (Fig. 2Aiii, Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons). Similar statistical comparisons were carried out for each of the 155 LFP recording sites in A1. Theta and beta band activity were significantly altered in 61 and 56% of the recording sites, respectively. Figure 2Ci shows the low-frequency mean (± SE) PSD plots for all 155 recording sites pre- and 2 h post-SS (note: because the low-frequency cutoff (−3 dB) of our data acquisition system is 2 Hz, there is little power below 4 Hz in our data). The histograms in Fig. 2, Di, Dii, and Diii show the mean power for theta, alpha, and beta bands. SS significantly decreased power in the theta (4–8 Hz) and beta (15–25 Hz) bands (P < 0.05, pairwise permutation t-test, 5,000 permutations) but did not have a major effect on the alpha band (8–12 Hz) (P = 0.49, pairwise permutation t-test, 5,000 permutations).

Fig. 2.

Sodium salicylate (SS) alters spontaneous oscillations in primary auditory cortex (A1). Representative low-frequency (A) and high-frequency (B) mean spectrograms (Hz vs. time) pre-SS (Ai and Bi) and 2 h post-SS (Aii and Bii). Mean calculated using the multitaper method (1-s window at 0.1-s steps; n = 30 trials of 10 s each). Heat map on right shows power (mV²/Hz). Statistical comparison between low-frequency (Aiii) and high-frequency (Biii) spectrograms pre- vs. post-SS (P-values heat map; red hue shows P < 0.01; Wilcoxon signed-rank test, corrected by false discovery rate multiple comparisons). C: mean (± SE, shaded area) power spectral densities (PSDs) averaged across all A1 electrodes sites (n = 155 channels, 10 rats, 30 trials each) for low frequencies (Ci) and high frequencies (Cii) pre- (red) and 2 h post-SS (blue). D: pre- and post-SS mean PSD for theta (Di; 4–8 Hz), alpha (Dii; 8–12 Hz), beta (Diii; 15–25 Hz), and gamma (Div; 30–90 Hz) bands (n = 155, 10 rats). Statistical comparison performed by pairwise permutation t-test (5,000 permutations; P-values shown on histograms; P < 0.05 considered statistically significant).

Figure 2, Bi and Bii show representative high-frequency mean spectrograms for a representative A1 spontaneous LFP pre- and post-SS. In this example, pronounced oscillatory activity occurred around 45–55 Hz pre-SS; the oscillation power around 35–50 Hz increased at 2 h post-SS. A statistical comparison of pre- and post-SS spectrograms revealed significant increases mainly around 35–45 Hz plus occasional increases at higher frequencies (Fig. 2Biii, Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons). Similar analyses from all 155 recording sites revealed significant differences in gamma activity in 80% of the cases. Figure 2Cii shows the mean PSD (±SE) across all 155 recording sites pre- and 2 h post-SS. The mean PSD shows a clear increase in gamma band activity 2 h post-SS. The histogram in Fig. 2Div shows the mean gamma band (30–90 Hz) power pre- and 2 h post-SS; there was a significant increase in gamma band activity 2 h post-SS (P < 0.01, pairwise permutation t-test, 5,000 permutations). These results indicate that SS mainly decreased power in the low-frequency bands but increased power in the high-frequency gamma band.

A1 coherence.

Coherence occurs when the phase difference in a given frequency band is consistent over time. To determine whether SS altered the coherence of oscillatory activity within A1, we computed the spectral coherence from 789 electrode pairs pre- and post-SS. Figure 3A shows an example of a mean low-frequency coherogram from one LFP pair (30 trials, 10 s per combination) pre- (Ai) and 2 h post-SS (Aii). The pre-SS coherogram shows a quasi-random pattern; however, 2 h post-SS, coherence had decreased in the 4–8 Hz theta band, but increased in the beta band. Figure 3Aiii shows a statistical comparison between the pre- and post-SS coherograms. Red hues on the P-value heat bar identify regions with significant difference pre- vs. post-SS (Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons). There was a significant decrease in coherence for the theta band post-SS for this electrode pair. When the same analysis was performed on all 789 LFP electrode pairs, SS induced a significant change in coherence in 51% of the electrode pairs in the theta, alpha or beta frequency ranges (Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons). Figure 3B shows the mean high-frequency coherogram pre- (Bi) and 2 h post-SS (Bii) from the same electrode pair in Fig. 2A. The pre-SS coherogram displayed modest coherence between 50 and 60 Hz; a slight increase was observed from 35 to 60 Hz 2 h post-SS. Statistical analysis of the pre- and post-SS data (Fig. 2Biii) revealed significant differences in patches along the frequency-time statistical map (Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons). When the same analysis was performed on all 789 LFP pairs, 57% of the pairs showed significant differences post-SS (Wilcoxon signed-rank test P < 0.01, FDR-corrected for multiple comparisons).

Fig. 3.

Sodium salicylate (SS) alters coherence within primary auditory cortex (A1). A: low-frequency mean coherograms from a representative spontaneous local field potential electrode pair (n = 30 trials) pre- (Ai) and 2 h post-SS (Aii). Coherence values shown on heat map. Aiii: pixel by pixel statistical comparison of pre- and post-SS coherograms with P-values shown on heat map; red hues shows P < 0.01 (Wilcoxon signed-rank test, corrected by false discovery rate (FDR) multiple comparisons). B: high-frequency coherograms pre- (Bi) and 2 h post-SS (Bii). Coherence values shown on heat map. Biii: pixel by pixel statistical comparison of high-frequency coherograms pre- and post-SS (P-values heat map; red hue shows P < 0.01; Wilcoxon signed-rank test, corrected by FDR multiple comparisons). C: mean (±SE, shaded areas) coherence values as function of frequency pre- (red) and 2-h post-SS (blue) for low frequencies (Ci) and high frequencies (Cii). Values computed across all channel-pairs. D: pre- and post-SS mean coherence values for theta (Di; 4–8 Hz), alpha (Dii; 8–12 Hz), beta (Diii; 15–25 Hz), and gamma (Div; 30–90 Hz) bands (30–90 Hz). SS caused a significant decrease in theta and alpha band coherence and a significant increase in gamma band coherence (permutation t-test, 5,000 permutations, P-values shown on histograms; P < 0.05 considered statistically significant).

To provide an overview of the SS-induced changes, the mean (±SE) coherence functions for all 789 LFP electrode pairs were computed for the low-frequency (Fig. 3Ci) and high-frequency bands (Fig. 3Cii) pre- and 2 h post-SS. Visual inspection of the mean coherence functions suggests a decrease in the low frequencies and an increase in the high frequencies. Figure 3D shows the mean theta (Di), alpha (Dii), beta (Diii), and gamma band (Dvi) coherence values pre- and 2 h post-SS. There were small, but significant decreases in theta and alpha band coherence and a significant increase in gamma band coherence (P < 0.05, pairwise permutation t-test, 5,000 permutations).

MGB oscillations.

It is unclear whether spontaneous oscillatory activity in the MGB is disrupted by high doses of SS. To explore this issue, we recorded spontaneous LPFs from electrodes in the MGB pre- and 2 h post-SS. Figure 4Ai shows a representative mean spectrogram of the spontaneous LFP from the MGB pre-SS. Strong oscillations were present around 8 Hz; however, this activity was reduced 2 h post-SS (Fig. 4Aii). A pre- vs. post-SS comparison (Fig. 4Aiii) revealed a statistically significant decrease between 5 and 10 Hz (P < 0.01, pairwise permutation t-test, 5,000 permutations). The spontaneous LFP from this recoding site in the MGB also displayed prominent neuronal oscillations between 40 and 50 Hz pre-SS (Fig. 4Bi). After SS treatment, the high-amplitude gamma band oscillations spread to 40 to 65 Hz (Fig. 4Bii). A comparison of the pre- and post-SS spectrograms (Fig. 4Biii) revealed significant increases mainly from 55 to 85 Hz (Wilcoxon signed-rank test P < 0.01, corrected by FDR multiple comparisons). The same analysis was performed on all 100 recording sites in the MGB. Statistically significant differences (increase or decreases) were observed in theta, alpha, beta, and gamma bands in 73, 51, 41, and 79% of the individual cases, respectively.

Figure 4, Ci and Cii displays the mean PSD (±SE) averaged across all 100 MGB electrode sites pre-SS (red) and 2 h post-SS (blue). Prominent decreases in power occurred in the theta, alpha, and beta bands, but the gamma band showed only a small decrease. Figure 4, Di–Div show the mean PSDs for the theta, alpha, beta, and gamma bands pre- and 2 h-post SS. The alpha, theta, and beta bands were significantly reduced 2 h post-SS (P < 0.01, pairwise permutation t-test, 5,000 permutations). Although there was a decrease in the gamma band, the decrease did not reach statistical significance (P = 0.07). Thus, SS significantly decreased low-frequency power in MGB theta, alpha, and beta bands.

Fig. 4.

Sodium salicylate (SS) modulates spontaneous oscillations in the medial geniculate body (MGB). Representative low-frequency (A) and high-frequency (B) mean spectrograms (Hz vs. time) pre-SS (Ai and Bi) and 2 h post-SS (Aii and Bii). Spectrogram calculated using the multitaper method (1 s window at 0.1 s steps; n = 30 trials of 10 s each). Heat map on right shows power (mV²/Hz). Statistical comparison between low-frequency (Aiii) and high-frequency (Biii) spectrograms pre- vs. post-SS (P-values heat map; red hue shows P < 0.01; Wilcoxon signed-rank test, corrected by false discovery rate multiple comparisons). C: mean (±SE, shaded areas) power spectral density averaged across all MGB electrodes sites (n = 100 channels, 10 rats, 30 trials each) for low-frequency (Ci) and high-frequency (Cii) pre-SS (red) and 2-h post-SS (blue). D: pre-SS and post-SS mean pwer spectral density for theta (Di; 4–8 Hz), alpha (Dii; 8–12 Hz), beta (Diii; 15–25 Hz), and gamma (Div; 30–90 Hz) bands. SS significantly decreased the power in theta, alpha, and beta bands, but not gamma (pairwise permutation t-test, 5,000 permutations, P-values shown on histograms, P < 0.01 considered significant).

MGB coherence.

To determine whether SS altered the coherence between electrode pairs in the MGB, spectral coherence was measured for 480 MGB electrode pairs pre- and post-SS. Figure 5 shows the mean coherograms (n = 30 trials, 10 s each) of a representative MGB electrode pair before (Fig. 5Ai) and 2 h after SS treatment (Fig. 5Aii) along with the pre-post statistical comparison (Fig. 5Aiii). There was a high degree of coherence in theta and alpha bands pre-SS (Fig. 5Ai) that decreased 2 h post-SS (Fig. 5Aii); these decreases were statistically significant (Fig. 5Aiii; Wilcoxon signed-rank test P < 0.01, corrected by FDR multiple comparisons). There was low coherence in the gamma band pre-SS (Fig. 5Bi), but a large increase 2 h post-SS particularly around 65 Hz (Fig. 5Bii); these increases were statistically significant (Fig. 5Biii; Wilcoxon signed-rank test P < 0.01, corrected by FDR multiple comparisons) for this MGB electrode pair. To gain a global perspective, the mean (±SE) coherence was computed for all 480 electrode pairs (Fig. 5C) before and 2 h post-SS. Figure 5, Di–Div display bar plots of the mean coherence for theta, alpha, beta, and gamma bands. SS significantly decreased coherence in theta, alpha, and beta bands (P < 0.05, permutation t-tests, 5,000 permutations), and significantly increased coherence in the gamma band (P < 0.05).

Fig. 5.

Sodium salicylate (SS) alters coherence within medial geniculate body (MGB). Mean low-frequency (A) and high-frequency (B) coherograms of a representative MGB electrode pair pre- (Ai and Bi) and post-SS (Aii and Bii). Coherence values shown on heat map. Pre- and post-SS statistical comparison of low-frequency (Aiii) and high-frequency (Biii) coherograms; P-value heat map on right (n = 30 trials, Wilcoxon signed-rank test, red hue shows P < 0.01, corrected by false discovery rate multiple comparisons). C: mean (±SE, shaded area) coherence values for low-frequency (Ci) and high-frequency (Cii) coherograms of all MGB electrodes pairs (n = 100 channels, 480 combinations, 30 trials each) pre-SS (red) and 2-h post-SS (blue). D: mean coherence pre-SS and post-SS in theta (Di), alpha (Dii), beta (Diii), and gamma (Dvi) bands pre-SS and 2 h post-SS. SS caused a significant decrease in theta, alpha, and beta bands and a significant increase in gamma band (permutation t-test, 5,000 permutations, P-values shown on histograms, P < 0.05 considered significant).

Coherence between A1 and MGB.

Because the MGB and A1 are interconnected, SS-induced changes in one region likely influence the other. To examine this possibility, we computed the spectral coherence between 1568 MGB/A1 electrode combinations before and 2 h post-SS (30 trials, 10 s per combination). Figure 6, Ai, Aii, Bi, and Bii show representative mean spectral coherograms pre- and post-SS for a single A1-MGB electrode pair. In this example, SS decreased coherence in the low-frequency band and increased coherence in the high-frequency, gamma band. Figure 6, Aiii–Biii show statistical comparisons between the low-frequency and high-frequency coherograms pre- and 2 h post-SS. SS caused a significant decrease in coherence mainly from 10 to 20 Hz and a significant increase in coherence from 60 to 70 Hz (Wilcoxon signed-rank test P < 0.01, corrected by FDR multiple comparisons).

Fig. 6.

Sodium salicylate (SS) alters coherence between medial geniculate body (MGB) and primary auditory cortex (A1). Mean low-frequency and high-frequency coherograms (n = 30 trials, 10 s each) of a representative MGB/A1 electrode pair pre-SS (Ai and Bi), 2 h post-SS (Aii and Bii); coherence heat map shown on right. Statistical comparison of low-frequency (Aiii) and high-frequency (Biii) coherograms; P-value heat map shown on right (Wilcoxon signed-rank test with P-values adjusted for multiple comparisons using a false discovery rate method; P < 0.01 in red). C: mean (±SE shaded area) low-frequency (Ci) and high-frequency (Cii) coherence from all 1568 MGB/A1 electrode pairs pre-SS (red) and 2 h post-SS (blue). D: mean coherence in theta (Di), alpha (Dii), beta (Diii), and gamma (Dvi) bands pre- and 2-h post-SS. SS caused a significant decreases in theta, alpha, and beta bands and a significant increase in the gamma band (permutation t-test, 5,000 permutations, P-values shown on histogram, P < 0.01 considered significant).

To obtain a global perspective on the interaction between the MGB and A1, the mean coherence was computed across the 1,568 MGB/A1 electrode pair combinations. Figure 6, Ci and Cii display the mean (± SE) coherence in the low-frequency and high-frequency bands before and after SS treatment. There was a large reduction of coherence in the low-frequency band 2 h post-SS and modest increase in the high-frequency gamma band around 60–70 Hz. The bar plots in Figure 6, Di–Div show the mean coherence pre- and 2 h post-SS in theta, alpha, beta, and gamma frequency bands. SS treatment decreased the coherence in theta, alpha, and beta frequency bands (P < 0.01, pairwise permutation t-test, 5,000 permutations) and increased coherence in the gamma band (P < 0.01, pairwise permutation t-test, 5,000 permutations).

Cross-frequency coupling between A1 and MGB.

In addition to coherence within isolated frequency bands, the coordination of neural activity across oscillatory frequencies is likely important for integrating activity across brain regions (Lisman and Jensen 2013). Pathological CFC of theta and gamma oscillatory activity has been suggested to contribute to tinnitus and this is a core component of the TD model (Llinás et al. 1999). It has been proposed that the phase of theta oscillations in MGB modulates the amplitude of gamma oscillations in A1, and that excessive theta-gamma coupling occurs during tinnitus. To examine this possibility, we computed the modulation index (MI; see methods) between 400 MGB/A1 electrode combinations before and 2 h post-SS (30 trials of 20 s per combination). Figure 7, Ai and Aii show representative comodulograms before and after SS treatment for a single trial of a MGB-A1 electrode pair. These examples show that the phase of theta oscillations in MGB modulates the amplitude of oscillations in A1 in the gamma frequency bands before (Fig. 7Ai) and after SS (Fig. 7Aii). On average, the phase of theta oscillation in MGB significantly influenced the amplitude of high-frequency gamma oscillations in 64% of trials in A1 before SS treatment. After SS treatment, theta phase modulated gamma amplitude in 74% of the trials in A1, significantly more than before SS treatment (P = 0.002, pairwise permutation t-test, 5,000 permutations). Therefore, theta phase in MGB transiently modulates gamma amplitudes in A1 and this phase-amplitude coupling is enhanced following SS treatment.

Figure 7B shows the mean comodulogram from one MGB-A1 electrode pair (n = 30 trials) pre-SS (Bi) and post-SS (Bii) and their statistical significance (Biii). SS significantly increased the CFC of theta oscillations in MGB and gamma amplitude in A1 from this representative electrode pair (Wilcoxon signed-rank test P < 0.05, corrected by FDR multiple comparisons). In total, 38% of the MGB-A1 electrode pairs showed significant differences in theta-gamma CFC before and after SS treatment. To obtain a global perspective on phase-amplitude coupling between the MGB and A1, the mean MI of the CFC between theta (4–8 Hz) in MGB and gamma (30–80 Hz) in A1 was computed across the 400 MGB/A1 electrode pair combinations. Figure 7C displays the mean (± SE) CFC MI before and after SS treatment. The MI was significantly increased after SS treatment (P < 0.01, pairwise permutation t-test, 5,000 permutations), thereby showing that SS treatment induces an increase in CFC between theta phase in MGB and gamma amplitude in A1. These results suggest that theta oscillations in the MGB influence cortical gamma power, which may contribute to the observed increased in cortical gamma band activity following salicylate treatment.

DISCUSSION

Neural oscillations, which occur at many different frequencies, provide a mechanism for synchronizing responses from large ensembles of neurons and binding neural activity across brain areas (Gray et al. 1989; Singer 1999). Aberrant neural oscillations in the central auditory system are believed to contribute to the perception of tinnitus, but the exact nature of these changes is still poorly understood (Adjamian et al. 2012; Dohrmann et al. 2007; Llinás et al. 1999, 2005; Sedley et al. 2015; Weisz et al. 2007b). In the current study, we attempted to identify how thalamocortical oscillations were altered in an ototoxic drug model of tinnitus. To accomplish this, we recorded spontaneous neural activity from multiple electrode sites within the MGB and A1 of rats before and after administering a dose of SS that induces tinnitus (Brozoski et al. 2007; Cazals 2000; Chen et al. 2015, 2017). To verify that we were recording from auditory areas and confirm that SS was functioning as expected, we measured sound-evoked responses in A1 and MGB and found that SS enhanced suprathreshold neural activity consistent with our previous reports (Chen et al. 2013, 2015). SS treatment also significantly altered neural oscillatory activity in the MGB and A1 as described in the next paragraph.

SS alters neural oscillations.

Prior to drug treatment, we observed stable oscillatory activity in theta, alpha, beta, and gamma bands in the MGB and A1. Our PSDs in A1 were similar to those reported by others in anesthetized subjects (Bojak et al. 2013; Feshchenko et al. 2004; Hayashi et al. 2007; Kochs et al. 1996; Tsuda et al. 2007) except for reduced power <5 Hz that was likely due to the low-frequency cutoff of our data acquisition system. SS reduced low-frequency power in theta, alpha, and beta bands in the MGB (Fig. 4), whereas in A1 it reduced power in the theta and beta bands but increased power in the gamma band (Fig. 2). Thus, the one consistent effect of SS was to reduce theta and beta band power in both the MGB and A1. To provide a global measure of neural synchrony within a region, we computed the mean coherence (i.e., phase synchrony) of neural oscillatory activity between electrode pairs located within MGB or pairs within in A1 (Halliday and Rosenberg 1999). Within the MGB, SS reduced the overall coherence in the low-frequency theta, alpha, and beta bands but increased coherence in the high-frequency gamma band (Fig. 5). Within A1, SS reduced coherence only in the theta and alpha bands but increased coherence in the gamma band (Fig. 3). When coherence was computed between electrode pairs across structures, one located in the MGB and the other in A1, SS decreased low-frequency coherence in theta, alpha, and beta bands but increased coherence in the gamma band (Fig. 6).

Gamma band activity (>30 Hz) is important for sensory processing and may directly influence perception (Aru et al. 2012; Singer 2001). Many EEG and MEG studies have demonstrated that tinnitus patients exhibit increased cortical gamma oscillations (De Ridder et al. 2011, 2014; Hartmann et al. 2014; Lorenz et al. 2009; Moazami-Goudarzi et al. 2010; Ortmann et al. 2011; van der Loo et al. 2009; Vanneste et al. 2010, 2011a; Weisz et al. 2005a, 2006, 2007a, 2011). Consistent with these studies, we found that SS increased gamma band power in A1 (Fig. 2) and increased gamma band coherence between electrode pairs within A1 (Fig. 3) and between A1 and the MGB (Fig. 6). Previous studies from our laboratory and others have demonstrated this same SS treatment reliably induces transient tinnitus-like behavior (Chen et al. 2014, Chen et al. 2017; Guitton et al. 2003; Jastreboff et al. 1988). Analogous to our SS-induced tinnitus model, MEG studies performed on subjects with acute noise-induced tinnitus revealed a rapid rise in gamma power in the auditory cortex (Ortmann et al. 2011). Moreover, a decrease in tinnitus loudness is associated with lower gamma power (Adamchic et al. 2012, 2014; Vanneste et al. 2011b) whereas an increase in tinnitus loudness is associated with increased gamma power (van der Loo et al. 2009). Taken together, these results suggest that increased cortical gamma activity is a neurophysiological correlate of tinnitus.

Theta and gamma.

Oscillations in different frequency bands can interact with one another and this CFC is thought to influence neural processing (Hyafil et al. 2015). In some models, the phase of low-frequency theta oscillations biases the amplitude of gamma waves, helping neuron ensembles to fire together (Belluscio et al. 2012; Lisman and Jensen 2013). Aberrant theta-gamma coupling is at the core of the TD model of tinnitus (Llinás et al. 1999). Abnormal rhythmic bursts of spikes with a frequency around 4 Hz occur in the thalamus of some patients with tinnitus and other deafferenting neurological disorders such as neuropathic pain (Jeanmonod et al. 1996). According to the TD model, this increase in low-frequency thalamic oscillations reportedly enhances high-frequency gamma band activity in the auditory cortex, thereby contributing to the phantom sound of tinnitus (Fox and Armstrong-James 1986; Llinás et al. 1999; McLelland and VanRullen 2016). EEG and MEG studies have provided mixed results, with some seeing increased theta-gamma frequency coupling in the cortex of tinnitus patients as compared with healthy humans (Adamchic et al. 2014; De Ridder et al. 2011), whereas other reports suggest that there is no such effect (Ahn et al. 2017; Zobay and Adjamian 2015; Zobay et al. 2015). One issue with the noninvasive recording techniques used in these studies is that they are restricted to examination of oscillatory activity across broad cortical regions, whereas the TD model proposes that the coupling between theta and gamma should be specifically observed between the thalamus and cortex. To our knowledge, this is the first study to simultaneously assess thalamic and cortical oscillatory activity as well as thalamocortical oscillatory coupling in an animal model of tinnitus.

A key component of the TD model is that thalamic theta oscillations modulate cortical gamma oscillations (Llinás et al. 1999; Walton and Llinás 2010). As such, one would expect strong frequency coupling between theta and gamma oscillations in the MGB and A1, respectively. Consistent with the TD model, we observe significant theta-gamma CFC between the MGB and A1 (Fig. 7). SS also increased gamma band coherence across electrode pairs, as predicted by the TD model (Fig. 3). Contrary to predictions of TD models, however, SS reduced rather than enhanced theta power in MGB (Fig. 4) and A1 (Fig. 3). SS also decreased theta, alpha, and beta band coherence between the MGB and A1 (Fig. 6). However, despite the reduction in power and coherence of low-frequency oscillations in MGB and A1, we did observe significant enhancement in theta-gamma coupling between the MGB and A1 following SS treatment, reflected by the increased in phase-amplitude CFC (Fig. 7). Overall, our data generally support the TD model with the notable exception of decreased rather than increased theta power in the MGB.

Alpha and gamma.

A growing body of evidence suggests that alpha band oscillations can exert critical influence on neuronal function and perception (Dugué et al. 2011; Jensen and Mazaheri 2010; Weisz et al. 2011). Alpha band desynchronization is linked to sound perception and the magnitude of alpha power in the human auditory cortex is inversely correlated with loudness of the phantom Zwicker tone, akin to tinnitus (Leske et al. 2014; Weisz et al. 2011). In patients with chronic tinnitus, the magnitude of alpha power is reduced (Llinás et al. 1999; Moazami-Goudarzi et al. 2010; Weisz et al. 2005b), but alpha activity increases when patients are given neurofeedback that reduces their tinnitus symptoms (Adamchic et al. 2012, 2014; Hartmann et al. 2014; Ortmann et al. 2011; van der Loo et al. 2009; Vanneste et al. 2011b). Interestingly, some studies of patients with chronic tinnitus have found that increased gamma power in the auditory cortex is correlated with decreased alpha activity (Lorenz et al. 2009). In the SLIM model of tinnitus, alpha band activity normally suppresses gamma; however, decreased sensory input caused by hearing loss reduces alpha-mediated inhibition resulting in enhanced gamma and greater neural synchrony that presumably gives rise to tinnitus (Weisz et al. 2007b).

The thalamus is generally considered a major generator of alpha activity, although the cortex may also play a role (Andersen et al. 1968; Bazhenov and Kleshchevnikov 1999; Bollimunta et al. 2008; Csercsa et al. 2010; Schreckenberger et al. 2004). Our analyses revealed strong spontaneous oscillation in the alpha band in both A1 and the MGB, consistent with previous recording from awake, behaving rats (Stolzberg et al. 2013). In our SS-induced tinnitus model, the increase in gamma power in A1 was associated with a large decrease in alpha power in the MGB and decreased alpha band coherence between electrodes in the MGB and A1. The changes in alpha power could occur because SS reduces synaptic transmission and alters the intrinsic membrane properties of neurons in the MGB (Su et al. 2012). Thus, our data are compatible with the SLIM model of tinnitus in which the increase in cortical gamma is associated with a decrease in alpha power in the thalamus and decreased alpha coherence between the thalamus and auditory cortex.

Limitations.

To optimize data acquisition from high-density recording electrodes from multiple sites, our experiments were carried out on anesthetized animals. In addition to potential confounds of anesthesia, this precluded behavioral measurement of tinnitus from the same animals in which the recordings were performed. However, the dose of SS used in this study reliability induces tinnitus sensation in awake animals (Jastreboff et al. 1988; Lobarinas et al. 2004; Myers and Bernstein 1965), with some studies even demonstrating tinnitus-like behavior in 100% of subjects (Rüttiger et al. 2003). Therefore, it is likely that the neurophysiological measures collected in a sound-attenuating booth in the absence of controlled sound stimulation are likely related to tinnitus generation and/or perception. That being said, we cannot rule out the possibility that the observed neurophysiological changes are related to SS-induced hearing loss or hyperacusis, but this seems less likely given these perceptual phenomena require sound stimulation (Radziwon et al. 2015, 2017). It is also possible that the observed neurophysiological changes could relate to both tinnitus and hyperacusis, as these two disorders are highly comorbid (Baguley 2003). For instance, increased gamma band activity could conceivably give rise to both tinnitus and hyperacusis sensation, as gamma-band coherence modulates the gain of sensory-evoked responses (Ni et al. 2016) and increased neuronal gain has been strongly linked to hyperacusis (Auerbach et al. 2014, 2018).

Anesthetics typically lead to an overall depression in neuronal activity. While some anesthetics, such as urethane, substantially change the oscillatory dynamics of the cortex (Aylwin et al. 2009; Liu et al. 2012), the dissociative anesthetic ketamine plus xylazine used in this study exert substantially fewer effects on the oscillatory behavior of neural networks. For instance, neural oscillations in the olfactory bulb of mice anesthetized with ketamine/xylazine displayed similar properties to those in awake animals, except for a decrease in overall power (Chery et al. 2014). Thus, the anesthetic regimen used in this study is likely to preserve many of the fundamental aspects of thalamocortical oscillatory behavior and therefore may not have grossly affected the SS-induced changes in oscillatory activity that we observed in the MGB and A1.

Delineating a causal relationship between the observed neurophysiological changes and tinnitus perception will require chronic electrophysiological recordings from animals trained on validated behavioral paradigms (Stolzberg et al. 2013). Nevertheless, experiments in anesthetized animals have been indispensable for uncovering basic physiological processes underlying perception. Indeed, the results from this study confirm some of the most common MEG and EEG findings from tinnitus patients, such as increased cortical gamma band activity. Our results extend those studies by directly assessing changes to neural oscillatory activity in the MGB and demonstrating that altered thalamic oscillatory activity can directly influence cortical synchronization/activity. Therefore, the results of this study provide novel insight into tinnitus-related changes in thalamocortical circuitry and provide a set of testable hypothesis for future studies in awake, behaving animals.

While SS is a powerful experimental tool for reliably inducing tinnitus, it may not accurately recapitulate chronic tinnitus associated with sensorineural hearing loss. Because SS-induced tinnitus is transient, it may differ mechanistically from long-term changes in the central nervous system that give rise to chronic tinnitus (Grapp et al. 2013; Mulders and Robertson 2011; Wallhäusser-Franke et al. 2017). While SS does induce cochlear hearing loss (Cazals 2000), there is growing evidence that it also exerts a wide range of effects on the central nervous system (Bauer et al. 2000; Chen et al. 2015, 2017; Gong et al. 2008; Su et al. 2012; Wei et al. 2010). Therefore, it is possible that SS directly disrupts cortical and subcortical oscillations. Whole-cell recordings in MGB slices treated with SS show that it disrupts rebound depolarization following membrane hyperpolarization in MGB neurons (Su et al. 2012). The disruption of rebound depolarization following hyperpolarization could provide a potential synaptic mechanism for the reduction of theta rhythms observed in our experiments. Such effects are unlikely to be present in noise-induced tinnitus.

Synopsis.

Our results show that theta, alpha, beta, and gamma oscillations are present in auditory thalamocortical circuits and SS modulates these thalamocortical oscillations in ways similar to those seen in tinnitus patients with acute and chronic tinnitus. Importantly, we observed a significant increase in gamma band activity in auditory cortex, which may relate to tinnitus sensation. Cortical gamma increases were associated with a significant decrease in alpha band activity in the MGB consistent with the SLIM model of tinnitus (Gray et al. 1989; Singer 1999; Weisz et al. 2007b) ; these changes were associated with decreased alpha and increased gamma coherence between MGB and auditory cortex. We also found that SS increased the coupling of theta phase in MGB and gamma amplitude in A1, consistent with the TD model of tinnitus. Thus, our data support aspects of both the SLIM and TD models. It should be noted that these models are not mutually exclusive but are concordant in several ways. At their core, both models are predicated on auditory deafferentation leading to increased cortical synchronization, potentially via decreased inhibition. Interestingly, while both the TD and SLIM models suggest that tinnitus is associated with increased activity in the theta band, we found that SS resulted in generally decreased activity in low-frequency bands in both the cortex and thalamus. This discrepancy may be due to certain aspects of our experimental design, such as the use of anesthesia or the salicylate toxicity model. It is also conceivable that the observed changes to cortical gamma power may be independent of changes in theta band or other low-frequency oscillations, for instance as a result of increased gamma coherence between the MGB and A1. Further studies are needed to identify the cellular mechanisms that lead to these electrophysiological changes and pharmacologic manipulations that can reverse these effects and suppress tinnitus.

GRANTS

This research was supported in part by grants from NIH (R01DC014452 and F32DC015160).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

P.V.R., B.D.A., and R.S. conceived and designed research; P.V.R. performed experiments; P.V.R. analyzed data; P.V.R. drafted manuscript; R.S. and B.D.A. edited and revised manuscript; P.V.R., B.A., and R.J.S. approved final version of manuscript.