The major repetitive transcranial magnetic stimulation (rTMS) paradigm applied to the treatment of tinnitus has been the 1-Hz variant due to its alleged inhibitory effects. Clinical effects have, however, been hampered by great interindividual variability as well as the fact that TMS includes no explicit mechanism to modulate excitability in circumscribed regions of tonotopically organised auditory fields. Following studies showing that the effect of TMS depends on the activational state preceding the stimulation, participants were exposed to 10 min of either notch- or bandpass-filtered noise prior to 1-Hz rTMS applied to the left auditory cortex. A control group was additionally assessed using bandpass noise – albeit with subsequent sham stimulation – to assess whether effects were due to the differential sounds alone or to a genuine interaction between sound and rTMS. Electroencephalogram was recorded from 128 electrodes before and after the experimental treatment while participants performed an auditory intensity discrimination task. While state-dependency effects from the behavioural data are not conclusive, several condition × (sound) frequency effects (some specific to the stimulated side) could be observed. Importantly, many of these could not be explained by the use of rTMS or the filtered noise alone. The resulting patterns are, however, complex and temporally variable, which currently prohibits recommendations on how to design a clinically effective approach to treat tinnitus. Nevertheless, our study gives the first evidence that state-dependency principles can induce sound frequency-specific effects in the auditory cortex, providing a crucial proof-of-principle upon which future studies can build.
Transcranial magnetic stimulation (TMS) is a non-invasive method for depolarising cortical neurons. Its repeated application to a single cortical target is referred to as repetitive TMS (rTMS), a method that induces changes of cortical excitability that outlast the stimulation period (Fitzgerald et al., 2006). Most studies have examined TMS effects in the motor system using the motor-evoked potential (MEP) as a measure of cortical excitability – in general, low-frequency rTMS (≤ 1 Hz; ‘inhibitory’) has been shown to reduce MEPs, and vice versa for high-frequency rTMS (≥ 5 Hz; ‘excitatory’). Even though the exact mechanisms have not yet been completely understood [potentially involving long-term potentiation (LTP)/long-term depression (LTD)-like changes in synaptic efficacy], owing to its potential for modulating excitability over an extended period of time, high expectations have been placed upon rTMS as a therapeutic tool for diverse disorders (Ridding & Rothwell, 2007).
One of these disorders is tinnitus, the conscious perception of sound(s) without an objectively identifiable sound source. Empirical data and theoretical notions (see Weisz et al., 2007 for a review) emphasise the role of increased ‘spontaneous activity and/or synchronisation’ in circumscribed regions of the auditory cortex. Thus, reducing aberrant activity using particularly ‘inhibitory’ rTMS should lead to marked reductions of tinnitus symptoms. However, clinical success can only be described as moderate, with an extremely high interindividual variability (Kleinjung et al., 2007). Apart from uncertain practical issues (e.g. ideal target location, coil orientation, stimulation intensity and frequency, etc.), an important factor hampering progress in this field is that stimulation protocols have simply been taken over from the motor system without precise knowledge of the effects that rTMS exerts on auditory cortical activity.
In a previous study (Lorenz et al., 2010), we were able to show that different variants of rTMS had an overall suppressive effect on evoked auditory cortical activity. However, these effects were not specific with regards to sound frequency, thus implying that rTMS could impact neurons similarly over tonotopically organised auditory regions defined as target. Current models of tinnitus (Llinás et al., 1999; Weisz et al., 2007; Rauschecker et al., 2010) differ with regards to the extent and frequency representations that may be overexcited. However, these do not generally assume an overexcitation of the ‘entire’ auditory cortex. From this perspective, a specific targeting of the actual affected regions would be desirable.
In this study, we present an approach that aims to overcome this limitation of rTMS. The background of our approach rests upon recent studies (e.g. ‘adaptation paradigm’; Silvanto et al., 2008; ‘preconditioning paradigm’; Siebner et al., 2004) that call the simplified division of rTMS into ‘excitatory’ vs. ‘inhibitory’ as well as the notion of single-pulse TMS as ‘virtual lesion’ tool into question, but rather imply that the current ‘activational state’ of the population of neurons prior to stimulation determines the outcome (state-dependency). While the works of Siebner et al., 2004 nicely show that rTMS can have enhancing or suppressive effects depending on the kind of prior transcranial Direct-Current Stimulation (tDCS) demonstrates that the spatial resolution of TMS can be selectively enhanced by features of pre-TMS stimuli. The common outcome of both approaches is an increase in the excitability of neurons in a low activation state and vice versa for neurons in a high activation state. The present study combines features of both approaches, i.e. employing preconditioning sensory stimulation in order to enhance the spatial resolution of rTMS when applied to the auditory cortex. Using differently filtered sounds prior to the application of 1-Hz rTMS to the left auditory cortex, we are able to show differential effects on auditory-evoked activity at specific sound frequencies. This is the first study to demonstrate that state-dependency principles can be successfully exploited in order to modulate auditory cortical activity in a frequency-specific manner.
Materials and methods
Thirty-one healthy young volunteers (range – 19–34 years; 16 females equally distributed across groups) with no history of neurological, psychiatric or audiological disorders took part in the study. With one exception, all participants were right-handed as assessed using the Edinburgh inventory (Oldfield, 1971). All participants gave written informed consent concerning the experiment procedures, which conformed to the standards set by the Declaration of Helsinki and approved by the Ethics Committee of the University of Konstanz. The first 21 participants were assigned to one of two groups (‘Notch’ or ‘Bandpass’ in the following; 10 and 11 participants, respectively) in a pseudo-random manner (ABAB design), which received sound stimulation combined with 1-Hz rTMS. Based on the outcome, we decided to recruit an additional 10 participants, who then received sound stimulation alongside sham rTMS in order to estimate the impact of sound stimulation alone (‘Sham’ group in the following). The detailed experimental procedures are described in the next section.
Stimuli and experimental procedure
Participants were told that they were taking part in a study investigating the influence of brain stimulation on auditory intensity discrimination capabilities. The basic experimental setup is depicted in Fig. 1. At the beginning and end of the experiment [electroencephalogram (EEG) Block 1 and Block 2, respectively], participants listened to 40-Hz amplitude-modulated tones and were instructed to respond via the press of a button to rare target sounds that were of lower intensity (−12 dB relative to standard intensity; see below; presented on 10% of the trials). During sound stimulation, participants fixated their attention on a white cross at the centre of a black screen. Prior to the experiment, the intensity of the standard sounds was adjusted to 40 dB above individual threshold using in-house software (http://sourceforge.net/apps/mediawiki/pytunesounds/ ). Sounds lasted for 800 ms (10 ms linear rise/fall time) and the interstimulus interval randomly varied between 1000 and 2000 ms. Critically, sounds could be of one of three different carrier frequencies that were chosen in such a manner that they logarithmically formed the centre frequencies of the noise bands used in the next block (see more below). The carrier frequencies were at 495, 990 and 1980 Hz. Over all three blocks, each carrier frequency was presented 450 times, thus yielding 135 trials per sound frequency at standard intensity (+ 15 target sounds). Targets were interspersed into standard sounds in a pseudo-random manner so that each target had to be separated by at least two standard sounds and the first 10 trials of a block could not contain a target sound. EEG activity was recorded while participants performed the intensity discrimination task.
Following the first block of the intensity discrimination task, participants underwent an experimental manipulation consisting of a passive noise stimulation followed by an rTMS session (see below for details). The acoustic stimulus consisted of a narrow-band noise (350–2800 Hz) that was filtered in such a way that either a middle part (700–1400 Hz; notched noise) was filtered out or selectively passed, permitting only energy through the middle part (bandpass noise). Sounds lasted for 10 min and were created using fast Fourier transform (FFT) filters implemented in Matlab. As with the probe tones, individual thresholds were determined for the noise and set to 50 dB above sensation level. During noise presentation, participants passively watched a movie showing an aquarium. The rTMS session (see below for details) then followed immediately after the 10 min of noise. Based on noise type (notched vs. bandpass) and rTMS type (1 Hz vs. Sham), participants were assigned to one of three possible groups, as already mentioned above (Fig. 1) –‘Notch’, ‘Bandpass’ or ‘Sham’. The most important contrast in this study was between the ‘Notch’ and ‘Bandpass’ groups, who both received the identical and commonly held ‘inhibitory’ form of 1-Hz rTMS stimulation, differing solely according to which noise type they were presented with prior to brain stimulation. Because differential effects were found with regards to brain activity (see Results), we recruited a control group (‘Sham’) to verify that the prior observed effects were indeed due to a ‘combination’ of sound stimulation and rTMS. The ‘Sham’ group listened to the bandpass noise, which was, however, followed by a sham rTMS stimulation (see below).
For a precise localisation of the TMS coil, we employed a neuronavigation system (ANT; asa software 22.214.171.124) together with individually acquired magnetic resonance images (MRIs). The target for the neuronavigation was individually defined for each subject. The focus of the magnetic field was directed at the main generator of the N1 (Fig. 2A), obtained from a previous magnetoencephalography (MEG) study (Lorenz et al., 2010). In this study, we stimulated with 40-Hz amplitude-modulated tones (800 ms duration) at three different frequencies (250, 1000, 4000 Hz) on five different days. Using an lcmv beamformer (van Veen et al., 1997), source analysis was performed for the N1 period of each measurement. Source data for each individual were interpolated into MNI space using SPM2 and averaged over tones and measurements sessions in order to sharpen the focus of activity. The maximum point of activation for the N1 was localised to the auditory cortex in all participants. The respective MNI coordinates for the left auditory cortical source were averaged across participants and defined as left auditory cortex target in the present study. The coordinates of this target were then transformed into coordinates of the individual head coordinate system and used as a target for neuronavigation for each participant (Fig. 2A). We chose to stimulate the left auditory cortex, as this is the most commonly used target of 1-Hz rTMS in the treatment of tinnitus (Kleinjung et al., 2007).
We used a biphasic MAGSTIM system (Rapid2, MAGSTIM, Whitland, Dyfed, UK) and an air-cooled figure-of-eight coil (Magstim Air Film Coil, 70 mm) for magnetic stimulation. For the ‘Notch’ and ‘Bandpass’ groups, the rTMS session consisted of 1000 pulses administered to the left auditory cortex at a frequency of 1 Hz with an intensity of 50% of the maximum output of the stimulator. For the ‘Sham’ group, the same parameters as above were used, however, the coil was angulated by 45°. Earplugs were provided to the participants in order to prevent hearing damage due to the clicking sound of the TMS. The subjects were blinded to the stimulation conditions.
EEG recordings and data analysis
During Block 1 and Block 2 (Fig. 1), EEG (Advanced Neuro Technology, Enschede, the Netherlands) was continuously recorded from 128 electrodes sampled at 2048 Hz. We extracted 4-s time windows from the continuous data stream (2 s pre- and post-stimulus onset) around standard sounds (i.e. non-target), discarding deviant sounds due to their low amount. All trials in which a participant erroneously responded to the standard sound were excluded. The epochs were then downsampled and detrended in order to remove the DC offset. The subsequent artefact rejection was performed in several steps: first, we chose 50 trials at random from all three sound types and blocks (i.e. 300 trials overall) for further processing in an independent component analysis (ICA). In order to optimise the performance of the ICA, we conducted an initial ‘coarse’ artefact rejection excluding large artefacts of unphysiological origins (e.g. channel jumps, dead channels, etc.). Following this step, 100 randomly sampled trials were entered into an ICA using the logistic infomax ICA (Makeig et al., 1996) implemented in the ‘runica’ function of the EEGLAB toolbox (Delorme & Makeig, 2004). This analysis yields 128 maximally temporally independent components, each with a distinct time course of activity and topography. Using visual inspection, we identified those components that captured eye movements and large muscular activity. Afterwards, the ICA weights were applied to each of the six data sets (three sounds × two blocks), artefact components were rejected and the raw data were finally reconstructed without artefacts. A final visual inspection was then performed for each respective data set and residual artefacts were removed if they occurred in a time window from 400 pre- to 800 ms post-stimulus onset. In order to warrant balanced data across all six data sets, we determined the minimal trial number within a data set and randomly chose that amount of trials from the other data sets. For the later source analysis (see below), we used the data set with the 300 randomly picked trials described above. We performed a stricter visual inspection of artefact trials for this data set, excluding all trials with eye movements, strong muscle activity, etc. The manually cleaned data set was then used for the calculation of spatial filters of the beamformer analysis, avoiding rank deficiency issues during the inversion of the cross-spectral density matrix.
Because the laterality of the combined noise + rTMS effects was of fundamental interest, we decided to pursue all following signal processing steps on regions of interest (ROIs); that is, the left and right auditory cortex, which were determined by group average source localisation of the N1 – a component that could be reliably detected in all participants (Fig. 2A and B). For this purpose, all artefact-free trials were concatenated and filtered between 2 and 25 Hz prior to averaging. Based on the filtered single trial data, the covariance matrix was calculated between −250 and 700 ms, thus encompassing pre- and post-stimulus periods, for later calculation of a common spatial filter. An estimation of source activation was achieved via a lcmv-beamformer (van Veen et al., 1997) by separately applying the common spatial filter to the averaged pre- (−250 to −50 ms) and post-stimulus (50–250 ms) data, thereby yielding a dipole moment value for each of these respective time periods. For the calculation of the forward model we utilised standard electrode positions and a standard Boundary Element Model supplied by the EEG system vendor. Subsequently, a single activation value was calculated by (post − pre)/pre, thus yielding relative changes with respect to the pre-stimulus period. These values were then interpolated onto a standard T1 MRI image, also supplied by the EEG system vendor. Positions of maximum increase within the left and right auditory cortex were then defined as ROIs, upon which all subsequent analyses were performed (also known as ‘virtual electrode’ analysis).
Overall, two measures of interest were analysed at the ROI level – the evoked response; and time–frequency dynamics. For the transient evoked response, a common spatial filter was constructed for both ROIs from 2 to 25 Hz filtered data (concatenated from all conditions), and the covariance matrix was estimated between −300 and 300 ms. Because the AM also elicited an auditory steady-state response (aSSR), a spatial filter was also constructed, albeit with a sharp filtering of data around the AM frequency (38–43 Hz) and the covariance window set to −300 to 800 ms. These filters were then applied to the analogously filtered raw data of the individual conditions in order to yield single trial data at the ROI level. These source-level data were then averaged over trials and the transient Event-Related Potential (ERP) was finally baseline corrected by subtracting the average activation in the time period from −100 to 0 ms. For the aSSR, time–frequency analysis (hanning tapered FFT on fixed time windows of 200 ms shifted in steps of 4 ms) was applied to the evoked data, baseline corrected for a −300 to −100-ms period, and the average temporal profile extracted between 38 and 43 Hz. Similar to the analysis of the aSSR, the time window for the estimation of the common spatial filter was set to −300 to 800 ms for the calculation of the time–frequency (induced + evoked) dynamics, although a broad filter was utilised between 2 and 90 Hz. At an ROI level, single trials were then submitted to a time–frequency analysis (hanning tapered FFT), calculating power between 5 and 25 Hz in steps of 1 Hz. The time window for the FFT (shifting in steps of 4 ms) was set to five cycles of the respective frequency. Time–frequency data of the single trials were then averaged and baseline corrected using a time period between −400 and −100 ms.
All offline steps of EEG analysis were performed using the Matlab-based open-source Fieldtrip Toolbox (Robert Oostenveld et al., 2011).
We assessed the amount of false alarms and misses for each participant. Because we were interested in changes, we subtracted the behavioural measures of Block 1 from Block 2 (i.e. Block 2 − Block 1). Behavioural data were assessed for statistically significant effects using 3 × 3 repeated-measure anovas with the within-subject factor sound frequency (‘Frequency’) and the between-subject factor stimulation condition (‘Condition’). Please note that the dependent variable already constitutes a difference measure as we were specifically interested in frequency and/or condition-specific changes across the experiment. Effects were considered to be statistically significant when P < 0.05, and were subsequently followed up by t-tests.
For the early transient evoked potential (< 300 ms; ERP) as well as the aSSR, we investigated time periods that contiguously exhibited a Side × Condition × Frequency interaction or a Condition × Frequency interaction for a minimum of 10 ms. By these means, we could identify three time windows of interest for the ERP (two-way interaction – 116–126 and 286–298 ms; three-way interaction – 126–138 ms) and two for the aSSR (two-way interaction – 236–268 and 640–748 ms), which are described in further detail below. Prior to the calculation of the anova, individual values of Block 1 were subtracted from Block 2 (analogous to the behavioural data). Effects were considered to be statistically significant when P < 0.05, and were subsequently followed up by t-tests. For the post-stimulus-induced (time–frequency) responses, we extracted the average profile for the alpha dynamics (8–12 Hz) prior to statistical testing, as this is the most sensitive auditory cortical feature following auditory input (Weisz et al., 2011). This was then submitted to an analogous anova. Even though desynchronisations could be observed in both hemispheres (see Fig. 6A), the exploratory anova did not yield any contiguous time periods in which the relevant interactions were P < 0.05. We will therefore not describe these results in further detail.
Finally, it is possible that the experimental treatments already have a differential effect on ongoing brain activity. Due to the absence of a true resting measurement (discarded in order to rapidly assess the sound-induced changes), we investigated such effects on the pre-stimulus periods using the non-baseline-corrected time–frequency data described above. Because participants could not predict the sequence of sound frequencies, we averaged over all three sound frequencies within a time window of −500 to −100 ms prior to stimulus onset, thereby obtaining a spectrum for each block. The spectrum of Block 1 was subtracted from Block 2 in order to focus on the changes following the experimental treatments. Because no a priori hypothesis existed of how rTMS + noise might affect ongoing brain activity, we first computed a non-parametric cluster permutation test described in detail by Maris & Oostenveld (2007). As a statistic, we used an independent F-statistic (implemented in the statfun_indepsamplesF.m function of the Fieldtrip toolbox), which identifies frequencies in which the arithmetic mean of the power changes differs between the conditions. The permutation statistic functions first by searching for contiguous clusters (here – neighbouring frequencies) exhibiting a significant effect (P < 0.05), and then by summing up the F-values within this cluster. This procedure is repeated 1000 times on shuffled data, upon which each permutation of the maximum summed F-value is stored, thus allowing the comparison of the empirically observed clusters with those gained following permutation. Using this procedure, a significant cluster was observed in the left auditory cortex (P = 0.02; 9–25 Hz) and another at a trend level in the right auditory cortex (P = 0.07; 7–14 Hz), both approximately in the alpha to beta range (Fig. 6C). Based on these results, paired t-tests were calculated on the average power change in a 9–15-Hz frequency band in order to elucidate the exact origin of the differences.
Intensity discrimination task
In order to assure that participants closely listened to the sound and to test potential behavioural impacts of our experimental manipulation, sounds were presented within the context of an intensity discrimination task in which participants had to respond to rare deviant sounds of lower intensity by pressing a button. The behavioural results indicate that this was a task that participants could perform very accurately (misses – mean 1.4, range 0–9; false alarms – mean 1.63, range 0–64). A comparison of behavioural data across the two blocks (i.e. pre- vs. post-noise/rTMS treatment) does not indicate any frequently selective impact specific to the experimental conditions, as indicated by absent condition × frequency interactions (all F4,56 < 1, P > 0.5). Tellingly, the ‘Notch’ group had greater reaction time improvements than the other two groups (−9.78%, ± 2.13; ‘Bandpass’–−0.31%, ± 2.06; ‘Sham’–−1.24%, ± 2.81), although this effect was not statistically significant (F2,28 = 2.24, P = 0.12). For the false alarms, a significant main effect could be observed for Frequency (F2,56 = 3.24, P = 0.04), which was mainly driven by a significant improvement in particular for the highest test frequency (t30 = −2.55, P = 0.02). For the lowest frequency, no improvements could be observed (t30 = −1.22, P = 0.23), whereas during the highest frequency test sounds the effect was marginally significant despite being absolutely the greatest (t30 = −1.97, P = 0.06; Fig. 3A). The latter result was mainly due to two outliers that exhibited a high amount of false alarms prior to rTMS as well as an associated strong decrease in false alarms (−67 and −59), leading to a strong variability. The exclusion of these two participants, however, lead to a similar (and here statistically significant) improvement (t28 = −2.19, P = 0.03). Overall, false alarms were reduced at middle and high frequencies in the second block, although in a manner unrelated to the experimental manipulation. With regards to misses, there was only a trend for the main effect Condition (F2,28 = 2.84, P = 0.07), which could be attributed to a deviant pattern for the ‘Bandpass’ condition (Fig. 3B) – whereas ‘Sham’ and ‘Notch’ went along with a reduction in misses, this could not be observed for the ‘Bandpass’ condition (‘Bandpass’ vs. ‘Sham’–t61 = 2.89, P = 0.005; ‘Bandpass’ vs. ‘Notch’–t61 = 3.02, P = 0.004).
ROI transient evoked activity
For the transient evoked response (< 300 ms), the anova over the entire time span yielded three non-overlapping time windows exhibiting either a Condition × Frequency or a Side × Condition × Frequency interaction lasting for at least 10 ms (see also Materials and methods). They are described here averaged over the relevant time periods. A Condition × Frequency interaction was observed 126–138 ms post-stimulus onset (F4,56 = 3.06, P = 0.02; main effect Condition also significant, F2,28 = 5.23, P = 0.01; Fig. 4A), indicating an effect independent of the stimulation side. Therefore, results were collapsed over both ROIs and subjected to separate anovas for each condition in order to separately assess the frequency-specific influences of our experimental treatments on each subcondition. This analysis revealed significant effects for Frequency in the ‘Bandpass’ (F2,20 = 3.06, P = 0.05) and ‘Notched’ conditions (F2,18 = 4.13, P = 0.03); however, no statistically significant effect for the ‘Sham’ condition (F2,20 = 0.07, P = 0.93). Post hoc testing revealed in ‘Bandpass’ that, compared with the middle frequency, evoked activity significantly increased for the high frequency (t10 = 2.20, P = 0.04). A similar pattern was also observed for the low frequency when compared with the middle frequency, although the effect was not statistically significant (t10 = 1.75, P = 0.11). No difference in amplitude changes could be found between the low and high frequency in the ‘Bandpass’ condition (t10 = 0.58, P = 0.57). In contrast to the ‘Bandpass’ condition, in which relative ‘suppressive’ effects following rTMS were specifically observed at the middle frequency, such an effect was most pronounced for the high frequency of the ‘Notched’ condition. The evoked activity significantly increased for the low (t9 = 3.37, P = 0.008) and middle (t9 = 2.51, P = 0.03) frequency compared with the high frequency. The changes in evoked activity did not differ between the low and middle frequency. Compared with the ‘Bandpass’ condition, post-rTMS activity was enhanced for the ‘Notched’ condition at low and middle frequencies, though this difference failed to reach statistical significance (low –t19 = 1.53, P = 0.14; middle –t19 = 1.81, P = 0.08). Despite this, ‘Bandpass’ exhibited a significant increase at the highest frequency compared with ‘Notched’ (t19 = 2.23, P = 0.04). The frequency-selective effects of the experimental conditions are also confirmed when comparing with ‘Sham’– significantly increased activity was found for ‘Bandpass’ vs. ‘Sham’ at the lowest (t19 = 2.46, P = 0.02) and highest frequencies (t19 = 2.07, P = 0.05), whereas this pattern was present at the lowest (t18 = 3.09, P = 0.006) and middle (t18 = 2.51, P = 0.02) frequencies when contrasting ‘Notched’ with ‘Sham’.
In an earlier non-overlapping time window to the two-way interaction (between 116 and 128 ms), we also observed a significant three-way interaction (i.e. Side × Condition × Frequency; F4,56 = 3.22, P = 0.02). These are of particular interest in this study as they may reflect effects specific to the stimulation site. For this time window, a follow-up anova that separately tested each hemisphere indeed confirmed a significant Condition × Frequency interaction for the left (stimulated) auditory cortex (F4,56 = 3.46, P = 0.01; Fig. 4B, left panel), which was absent in the right auditory cortex (F4,56 = 1.42, P = 0.24; Fig. 4B, right panel). The interaction in the left auditory cortex was to a great extent driven by the ‘Notched’ condition, in which a significant post-rTMS increase could be observed at the lowest frequency relative to the middle (t9 = 2.70, P = 0.02) and highest (t9 = 3.27, P = 0.009) frequencies. At the lowest frequency, activity significantly increased [also relative to ‘Sham’ (t18 = 2.87, P = 0.01)] and marginally failed trend level when compared with ‘Bandpass’ (t19 = 1.67, P = 0.11). However, changes in ‘Bandpass’ also appear to have contributed to this interaction effect pronounced at the highest frequency; these were admittedly only revealed when comparing between the groups – contrasting with the ‘Notched’ condition particularly reveals significantly increased activity at the highest frequency for ‘Bandpass’ (t19 = 2.24, P = 0.03), which was also present on a trend level when the comparison was made with ‘Sham’ (t19 = 1.72, P = 0.10).
Finally, in a later time window following the effects in the N1 latency window, a three way interaction was also present between 286 and 298 ms (F4,56 = 3.79, P = 0.009). Separate anovas for the left and right auditory cortex again confirmed that a Condition × Frequency effect was evident only for the stimulated side (F4,56 = 3.21, P = 0.02; Fig. 4C, left panel), but not on the unstimulated side (F4,56 = 0.71, P = 0.59; Fig. 4C, right panel). In contrast to the previously reported evoked effects, modulations in the ‘Notched’ condition did not appear to contribute to this relatively late effect (all P-values involving ‘Notched’ > 0.12). However, marked frequency-specific modulations could be observed in the left auditory cortex for ‘Bandpass’ and ‘Sham’. Despite receiving the same sound stimulation prior to the rTMS block, the modulations for these two groups appeared to follow slightly opposing patterns. Whereas for ‘Bandpass’ the lowest frequency exhibited a reduction of evoked activity following rTMS relative to the middle (t10 = −2.43, P = 0.04) and highest frequencies (t10 = −3.84, P = 0.003), evoked activity in the ‘Sham’ condition was significantly enhanced for the highest frequency relative to the middle frequency (t9 = 2.75, P = 0.02) and tendentiously also compared with the lowest frequency (t9 = 1.94, P = 0.09). Furthermore, a direct comparison between ‘Sham’ and ‘Bandpass’ revealed a significant difference at the middle frequency, with a relative increased evoked activity for ‘Bandpass’ following rTMS.
In summary, this section demonstrates that rTMS has frequency-specific impacts on transient evoked responses that emerge from a pre-treatment with a filtered sound and differ from the pattern observed by processing the sound ‘without’ real rTMS. The effects are mainly observed on the stimulated side, but are also partially observed in the auditory cortex contralateral to stimulation. The generators underlying the present effect are mainly located in second auditory regions, corresponding to the stimulation site as determined using the neuronavigation system. In the following section, we address the issue of to what extent similar effects on brain responses putatively reflecting primary auditory cortical activity may be observed.
ROI steady-state evoked activity
Reliable aSSR activity was observed in all participants, and descriptively (Fig. 2B, right panel) the typical hemispheric asymmetry pattern could be observed with greater aSSR responses in the right auditory cortex (see Ross et al., 2005). In this figure, it can also be seen that the aSSR takes ∼200 ms to evolve, and rapidly wears off immediately following sound offset (i.e. ∼800 ms post-sound onset). Within this 200–800 ms time window, no significant Side × Condition × Frequency interaction could be identified, indicating that our treatment did not have frequency-specific effects that were differential for the left and right auditory cortex (e.g. specific effects restricted to stimulation side). However, in two time windows (236–268 and 640–748 ms post-stimulus onset) a Condition × Frequency effect could be found. Data within these respective time windows were averaged, and are described separately in the following discussion.
For the earlier (236–268 ms) time window, the Condition × Frequency interaction was statistically significant (F4,56= 2.82, P = 0.03), whereas the full three-way interaction was far from statistical significance (F4,56= 0.64, P = 0.63). Separate anovas for each condition showed that solely the ‘Notched’ condition lead to frequency-specific modulations of aSSR activity (F2,18= 6.77, P = 0.006), whereas no such effect could be observed in ‘Bandpass’ (F2,18= 0.33, P = 0.72) and ‘Sham’ (F2,18= 1.57, P = 0.24). In Fig. 5 (left panel), it can be seen that the ‘Notched’ condition is particularly marked by increased aSSR responses at the lowest and middle frequency, which cannot be observed at the highest frequency. This descriptive impression was confirmed by post hoc testing when contrasting lowest vs. highest frequencies (t9= 2.93, P = 0.02), as well as middle vs. highest frequencies (t9= 2.63, P = 0.03). For the lowest frequency, the modulation for ‘Notched’ was also significantly stronger than that observed during ‘Sham’ (t18= 2.07, P = 0.05), with the difference being present on a trend level for the middle frequency (t18= 1.76, P = 0.10). From Fig. 5 (left panel), the impression could be gained that ‘Sham’ lead to frequency-specific modulations, in particular a selective decrease in aSSR following the experimental treatment. However, as mentioned above, the Frequency’s main effect was not significant following separate calculation of an anova for this condition. Post hoc testing also confirmed the absence of such an effect (all P > 0.22).
As for the earlier time window, the later (640–748 ms) time window was marked by a highly significant Condition × Frequency effect (F4,56 = 4.24, P = 0.005) in the absence of an interaction with Side (F4,56 = 0.47, P = 0.76). Analogously to the former effect, a pronounced Frequency effect was observed in the ‘Notched’ (F2,18 = 6.83, P = 0.006) but absent in the ‘Bandpass’ condition (F2,20 = 1.36, P = 0.28). However, in contrast to the previous time window, a significant Frequency effect was also obtained for ‘Sham’ (F2,18 = 4.60, P = 0.02). In Fig. 5 (right panel), one can notice that the effect for the ‘Notched’ condition is again driven by a relative increase in aSSR modulation for the lowest (vs. highest –t9 = 3.27, P = 0.01) and middle frequencies (vs. highest –t9 = 3.08, P = 0.01). For ‘Sham’, the reduction in aSSR post-experimental treatment, also present in the earlier time window, was significant when compared with the lowest frequency (t9 = −2.52, P = 0.03), and at a trend level also against the highest frequency (t9 = −1.90, P = 0.09). At the middle frequency, ‘Notched’ and ‘Sham’ also differed significantly (t18 = 2.89, P = 0.01), with relatively increased aSSRs in the former. Finally, when contrasting ‘Bandpass’ with ‘Notched’, a marginally significant difference could be obtained at the highest frequency (t19 = 2.02, P = 0.06), with relatively increased aSSRs for the former.
ROI pre-stimulus oscillatory activity
Prior to analysing modulations of induced activity following sound presentation, we first describe potential changes of pre-stimulus power following our experimental treatment. Figure 6B separately depicts the modulations for each ROI and condition. At a descriptive level, it can be observed that, in contrast to the two experimental conditions that included a real rTMS session, ‘Sham’ lead to widespread decreases in pre-stimulus power below 25 Hz in both auditory cortices (no significant effects were observed for high frequencies; data not shown). A non-parametric cluster permutation test revealed a significant (P = 0.02) cluster in the left auditory cortex ranging from 9 to 25 Hz, with a maximum effect at ∼11 Hz and a near-to-significant cluster (P = 0.07) in the right auditory cortex, which was spectrally more restricted between 7 and 14 Hz, with a peak at ∼10 Hz (Fig. 6C, left panel). The test indicates that a difference between the conditions is present for these alpha to beta frequencies. In order to follow up this effect in more detail, individual effects were separately averaged between 9 and 15 Hz for both hemispheres. An anova with Condition and Side as independent variables revealed a highly significant main effect for Condition (F2,28 = 7.12, P = 0.003), whereas the Side and Condition × Side effects were not significant (F < 2.36, P > 0.14). Because the statistic showed the absence of any impact of side, data were averaged over both ROIs (Fig. 6C, right panel), and post hoc tests were performed. This analysis confirmed the descriptive impression that ‘Sham’ lead to a reduction in pre-stimulus alpha power that was significantly stronger than ‘Notched’ (t18 = −2.66, P = 0.02) or ‘Bandpass’ (t19 = −3.55, P = 0.002). Even though Fig. 6C (right panel) suggests a stronger reduction for ‘Notched’ relative to ‘Bandpass’, this effect was statistically not significant (t19 = −1.30, P = 0.21).
The central motivation of the present study was the question of whether excitation within the auditory cortex can be modulated in a selective manner for specific sound frequencies by using distinctly filtered noise stimuli prior to 1-Hz rTMS. While the behavioural task was not sufficiently sensitive to capture any sound frequency-specific effect of noise pre-exposure on subsequent rTMS, several ROI reconstructed evoked response results unequivocally underline their presence. Due to the design, in order to demonstrate that our intervention indeed elicited topographically (or tonotopically) focal alterations of ‘states’, requires: (i) a (Side ×) Condition × Frequency interaction; and (ii) the demonstration that patterns of differences are different between Bandpass vs. Notched (thus excluding rTMS alone as source of the effect), and additionally a difference between Bandpass vs. Sham (thus excluding noise stimulation alone as source of the effect). Altogether the criteria (when applicable) argue for true interaction effects between noise type and rTMS. While a Condition × Frequency effect could be observed for the aSSR, they appear to be mostly driven by the notched noise, while Bandpass and Sham largely did not differ. Therefore for the aSSR, the influence of noise type as driving the effect can not be excluded. The case, however, is different for the early transient responses (∼100–300 ms), in which several patterns of effects could be observed in which neither rTMS nor noise stimulation alone can explain the results. Partly these effects were also restricted to the stimulation site, while the non-stimulated auditory cortex did not show such modulations (Fig. 4B). Given the criteria outlined above, the most pronounced effects implying genuine interaction between noise type and rTMS are an increase of the evoked response in an early time window (∼116–138 ms) at the high frequency for the Bandpass condition, and a reversed (mirror image) effect at the middle frequency between Sham (evoked response reduction) and Bandpass (evoked response enhancement) in a late time window (∼286–298 ms). While the late evoked response effect was strictly restricted to the stimulated auditory cortex, the early effect is stimulation side specific in an early interval (∼116–128 ms), however involves both sides at later intervals (∼126–128 ms). Before some of the results are discussed in further detail, it is important to emphasise that these evoked response results ride on top of a relative alpha enhancement of the rTMS conditions compared with Sham, thus implying a part of global rTMS effects the presence of a fine-structure in dependence of the prior noise stimulation.
The basic idea of the project was inspired by the growing amount of literature demonstrating state-dependency effects of TMS generally in the sense that the excitability of activated neuronal populations is decreased and vice versa for suppressed neuronal populations (see Silvanto et al., 2008 for review) using exactly the same TMS protocol. Evidence for this comes from multiple angles, such as basic motor responses (Siebner et al., 2004), cognitive tasks (e.g. Cattaneo et al., 2010), as well clinical phenomena. With regards to the repetitive form of TMS (i.e. rTMS), state-dependency has been shown, for example, in the motor system using ‘preconditioning’, by which for instance anodal or cathodal tDCS is used to alter baseline excitability prior to the application of rTMS. By this means, Siebner et al., 2004 could show that the identical 5-Hz rTMS treatment either increases or decreases MEPs depending on the form of preconditioning. Clinical data from Brighina et al. (2002) indirectly demonstrate state-dependency effects of 1-Hz rTMS by showing that when applied to the visual cortex in migraine patients, excitability (inferred from phosphene thresholds) increases, whereas in healthy controls it decreases. As pointed out by Siebner (2004), the underlying physiological mechanism for state-dependency effects may be related homeostatic principles adapting thresholds for LTP or LTD according to the level of neural activity, thus maintaining time-averaged postsynaptic activity within distinct limits. Disregarding the exact physiological mechanisms, applying a variant of the adaptation paradigm (Silvanto et al., 2008) to the auditory system with the potential capability of (sound) frequency-specific modulation of auditory cortical activity may have an important impact on the treatment of tinnitus. This condition is putatively marked by hyperexcitability of circumscribed portions of the tonotopic map related to the perceptual characteristics of the phantom sound (Weisz et al., 2007). Current rTMS protocols have no intrinsic property that would allow such regionally specific changes in neuronal activity (Lorenz et al., 2010); therefore, pre-exposure to specifically filtered sounds might aid in targeting specific auditory cortical regions that are affected by hyperexcitability. Furthermore, auditory cortical activity changes following rTMS in tinnitus are marked by a considerable interindividual variability (Lorenz et al., 2010). In the present study we used filtered noise in order to modulate pre-rTMS states as a previous MEG study by Okamoto et al. (2007) was able to show short-term (< 1 s) reductions of the N1m following brief presentations (3 s) of notch-filtered noise. Also a perceptual illusion known as the Zwicker tone (e.g. Norena et al., 2000) argues for at least short term (yet excitatory) impacts of filtered noise on auditory cortical activity. Effects on a longer time range (from seconds to minutes) can be inferred from a phenomenon in the tinnitus literature known as residual inhibition (Roberts, 2007), in which tinnitus becomes suppressed following the offset of a masking noise. The several effects (evoked responses as well as non-evoked pre-stimulus alpha activity) observed for the Sham group argues that prolonged application of bandpass-filtered noise alone can alter auditory cortical excitability over a period of at least several minutes. More importantly, however, we were indeed successful in showing that the identical 1-Hz rTMS stimulation over the left auditory cortex (a common rTMS approach to treat tinnitus), commonly assumed to be an ‘inhibitory’ stimulation form, differs depending on the pre-treatment with filtered noise. We could furthermore elucidate differences between the combined exposure to bandpass noise and rTMS compared with noise stimulation alone (together with sham stimulation). Altogether, these are the first results that unequivocally demonstrate state-dependency effects of rTMS in the auditory system.
Nevertheless, in contrast to the previously published data from other modalities (Siebner, 2004) or in cognitive tasks (Silvanto et al., 2008), the pattern of results is more complex in the case of our auditory system data. At a behavioural level, some effects not specific to the experimental treatment could be seen (fewer false alarms for the high-frequency tone), probably reflecting practice effects. Furthermore, a trend could be observed for misses, caused by the absence of a behavioural improvement in the Bandpass condition as compared with the Notched and Sham conditions. This effect is interesting as it provides behavioural evidence for state-dependency effects, namely in a sound frequency-unspecific manner. The lack of such a Condition × Frequency interaction at a behavioural level makes the effect very difficult to interpret in a functional sense, as depending on the previously applied noise type rTMS could lead to behavioural consequences that either differ from or mimic those observed following Sham treatment. It is possible that the discrimination task could have already been too functionally complex to reveal sound frequency-specific changes and that a simple detection task, for example, may have been more appropriate. However, in the present study we focussed on supra-threshold stimulation as one ambition was to elicit reliable evoked responses. Furthermore, it must be emphasised that with the low number of mistakes, participants were already performing at a ceiling level, presumably with insufficient space for large performance changes. Indeed, in contrast to the behavioural results, several Condition × Frequency interactions emerged, some of which were also specific to the stimulated auditory cortex. An interesting finding for the aSSR, a measure of primary auditory cortical excitability (Ross et al., 2005; Bidet-Caulet et al., 2007), is that a decrease at the middle frequency (i.e. the noise centre) could be observed in the Sham condition (i.e. following bandpass noise stimulation without active rTMS), particularly relative to the sideband frequencies, as well as to the Notched condition and especially in later phases of the aSSR (∼640–750 ms). The addition of active 1-Hz rTMS (i.e. the Bandpass condition) abolishes this effect. The statistical test indicates that this effect is not restricted to the side of stimulation, indicating a spread of 1-Hz rTMS effects beyond the stimulated site to homologue auditory cortical regions. It must, however, be clearly stated that the pattern of aSSR effects does not fully fit into the adaptation logic, as suggested by Silvanto et al. (2008), from which for instance significant increases might have been expected at the middle frequency in the Bandpass condition and/or at sideband frequencies for the Notched condition. However, based on the definition of the neuronavigated target derived from the N1 source of a previous study, it is likely that this is not optimal for inducing aSSR effects. Indeed, the transient evoked responses that putatively rather reflect activity from secondary auditory cortical regions reveal (among others) stimulation side-specific Condition × Frequency interactions, whereby the pattern resembling most notions that conform to the adaptation paradigm logic evolve relatively late, at about 286–298 ms. For this time window, it can be clearly seen that at the middle frequency, the Bandpass and the Sham conditions (differing only by the use of active rTMS) produce opposing patterns – that is, enhancements in the case of Bandpass and reductions in the case of Sham. The Notched condition did not lead to marked changes in evoked responses in this time window, excluding the possibility that the use of active rTMS alone was responsible for this effect. However, effects in earlier time windows (∼116–138 ms) dismiss the conclusion that this could be a general pattern derived by the combined use of sound and 1-Hz rTMS. Together with the aSSR responses, the results from the transient evoked responses point to complex effects of the pre-exposure of auditory cortex-filtered noise to 1-Hz rTMS in terms of their spatial location as well as temporal dynamics. Future studies are needed to enhance our understanding of the influence of diverse parameters, for example the filter characteristics, sound intensity, stimulation site and frequency, on the observed effects, making these more predictable and yielding a greater potential for exploiting state-dependency principles in order to sharpen the impact of rTMS on auditory cortical activity. Among the so far unknown factors, the inevitable noise accompanying each TMS pulse also needs to be further considered in future. For the present study it has to be kept in mind that each TMS pulse not only directly activates auditory areas beneath the coil, but also in the bottom-up manner by the auditory stimulus. This parallel sensory stimulation may add to the complex patterns of results in a so far unknown manner, even though the abundant differences between the Bandpass and Notched group exclude this factor as a trivial cause for the results reported in this study. Related to this issue, it has to be admitted that in the present study we can not clearly infer the effects of noise stimulation alone, as even in the Sham group the subsequent sound stimulation during rTMS may interact with the preceding noise stimulation. A future study should therefore also focus on the immediate effects following noise stimulation.
A word of caution should be mentioned in that the evoked responses presented in this study do not directly represent ongoing (spontaneous) brain activity and that the relationship between these two aspects of brain activity is complicated. On a global level, however, the present study could show that ongoing brain activity in the auditory cortex is modulated via TMS – whereas the Sham condition went alongside an overall decrease in spectral power, this was prevented in the active rTMS conditions, with the effect being most pronounced in the alpha frequency band. Accepting the notion that brain oscillation in this frequency band represents an ongoing inhibitory activity (Weisz et al., 2011), this is the first evidence for 1-Hz rTMS acting in an inhibitory manner on auditory cortical activity. With regards to the application to tinnitus, from various theoretical perspectives it would be desirable to reduce ongoing brain activity in circumscribed regions of the auditory cortex (i.e. those characterized by a hyperactivity). At the moment, it is impossible to state how the observed changes in the evoked responses for particular sound frequencies relate to changes of ongoing brain activity in specific auditory cortical regions, a project that would require invasive studies in animals as the spatial resolution of non-invasive methods are insufficient. Another aspect that will need to be considered prior to the application of combined sound/rTMS treatments to tinnitus is that baseline activities in tinnitus are most likely already changed, which may unpredictably influence the outcome – for example, whereas Brighina et al. (2002) could show a likely decrease in visual cortical excitability in patients with migraine, invasive measurements by Pasley et al. (2009) showed enhanced activity following rTMS if pre-rTMS neuronal activity was elevated. On the other hand, pre-exposure with filtered noises prior to rTMS may help to reduce the impact of the interindividual variability of rTMS effects (Lorenz et al., 2010), which is most likely an important source for the thus far rather moderate clinical impacts of this brain stimulation technique.
To conclude, the main message that can be taken from this study is that 1-Hz rTMS applied to the auditory cortex can not be regarded as generally inhibitory, but that depending on the preceding stimulation context a lot of fine-structure (differential sound frequency-specific evoked potential effects) can be observed following an identical 1-Hz rTMS. This is the first study to demonstrate such state-dependency effects of rTMS applied to the auditory cortex. Another relevant take-home message is that with a deeper understanding of the interaction effect between sound stimulation characteristics and rTMS, this knowledge may be successfully employed to improve the spatial accuracy as well as directionality of rTMS effects when applied to treat chronic tinnitus. The resulting patterns of the present study are, however, too complex to directly derive a clinical application to tinnitus, which will require further studies. At this stage, the present study serves as a proof-of-principle and encouragement that state-dependency principles may be employed to optimise current rTMS stimulation forms.
This study was supported by grants from the Deutsche Forschungsgemeinschaft (DFG) and the Tinnitus Research Initiative (TRI). We thank Katharina von Bardeleben and Barbara Awiszus for their assistance in collecting the data.