Abstract

The processing of valence is known to recruit the amygdala, orbitofrontal cortex, and relevant sensory areas. However, how these regions interact remains unclear. We recorded cortical electrical activity from 7 epileptic patients implanted with depth electrodes for presurgical evaluation while they listened to positively and negatively valenced musical chords. Time–frequency analysis suggested a specific role of the orbitofrontal cortex in the processing of positively valenced stimuli while, most importantly, Granger causality analysis revealed that the amygdala tends to drive both the orbitofrontal cortex and the auditory cortex in theta and alpha frequency bands, during the processing of valenced stimuli. Results from the current study show the amygdala to be a critical hub in the emotion processing network: specifically one that influences not only the higher order areas involved in the evaluation of a stimulus's emotional value but also the sensory cortical areas involved in the processing of its low-level acoustic features.

Introduction

Emotional valence refers to the intrinsic attractiveness (positive emotional valence) or aversiveness (negative emotional valence) of a given object (or event) and applies across sensory domains such that, for instance, unpleasant smells and sad facial expressions are negatively valenced while pleasant smells and happy facial expressions are positively valenced. However, while functional neuroimaging work has shown that processing of valenced stimuli recruits medial temporal and frontal lobe structures as well as relevant sensory areas, much less is known regarding how these regions interact.

In the auditory domain, the effects on a listener of so called consonant and dissonant musical chords are fairly consistent: in contrast to the former, which tend to be considered pleasant, the latter generally evoke a negatively valenced emotion—a feeling of unpleasantness (Plomp and Levelt 1965; Wedin 1972; Costa et al. 2000). Music can convey a range of complex and rich emotions, from wonder to nostalgia, and various attempts to examine the neural correlates of such specific aesthetic emotions are emerging (Trost et al. 2011). However, research into the processing of the simpler pleasant versus unpleasant dichotomy that consonant and dissonant chords evoke is well underway, and critically, the role of regions such as the amygdala and orbitofrontal cortex in such processing has been confirmed (Blood et al. 1999; Gallagher and Chiba 1996; Koelsch et al. 2006).

The fact that processing of consonant and dissonant musical chords recruits both the amygdala and orbitofrontal cortex makes such stimuli relevant in addressing outstanding issues regarding the emotion cerebral network. A number of studies have reported on the latency of processing in the amygdala and orbitofrontal cortex (Oya et al. 2002; Eimer and Holmes 2002; Eger et al. 2003; Streit et al. 2003; Pourtois et al. 2004; Naccache et al. 2005; Pourtois et al. 2005; Gothard et al. 2007; van Heijnsbergen et al. 2007; James et al. 2008; Mormann et al. 2008). Indeed, some have suggested a dependence of processing latency on factors such as the extent to which the stimulus's valence is attended to (Krolak-Salmon et al. 2004) and the degree of expertise the participant has with the stimulus (James et al. 2008). However, while for instance, Streit et al. (2003) suggested a similar onset latency of amygdala and orbitofrontal cortex activity (between 100 and 160 ms) during the processing of visual emotional stimuli, Krolak-Salmon et al. (2004) suggested an earlier and more transient response in the amygdala (at ∼200 ms) relative to that in the orbitofrontal cortex (from ∼300 ms). Further support for a potentially modulatory effect of the amygdala on the orbitofrontal cortex comes from functional neuroimaging results showing that increased amygdala activation led to increased activation in the ventral prefrontal cortex areas during the processing of emotional stimuli (Lerner et al. 2009). Thus, one key aim of the current study was to examine the directed communication that exists between the amygdala in the medial temporal lobe and the orbitofrontal cortex in the frontal lobe.

Another aim of the current study was to elucidate the nature of the interaction between the amygdala and sensory cortices during the processing of emotional stimuli. Previous studies in the auditory domain have shown the amygdala to respond to aversive sounds (Mirz et al. 2000; Zald and Pardo 2002) and it has been argued that this area plays a modulatory role on activity in early sensory areas (Morris et al. 1998; Vuilleumier et al. 2004). However, the extent to which sensory areas are necessarily implicated in emotional processing is still under discussion with some studies failing to show a modulation of auditory cortical activity by valence (Blood et al. 1999) and at least one neuropsychological study suggesting that the integrity of the auditory cortex is not necessary for emotion recognition (Peretz et al. 2001).

In a single case study reported by our lab (Dellacherie et al. 2009), we took advantage of the high temporal and spatial resolution of intracranial electroencephalography (EEG) recordings to examine responses to consonant and dissonant sounds in the amygdala, orbitofrontal cortex, and auditory cortex. However, while that study provided evidence of valence processing in all 3 regions, it did not examine how they might interact. Indeed, in this regard, it has been suggested that oscillations in the theta band bring about integration and synchronization of long-range activity within the emotion network (Lewis 2005). Support for the notion that theta oscillations are involved in the emotion network, and potentially therefore in the communication between structures therein, comes from findings in human and animal models showing that theta band activity is modulated by emotional valence (e.g., Aftanas et al. 2001; Knyazev 2007; Maratos et al. 2009; Pourtois et al. 2010; Rutishauser et al. 2010; Meletti et al. 2012) and may increase during emotional arousal (Paré et al. 2002). One prediction that may be made, therefore, is that such low frequency activity will be implicated during musical emotion processing.

The current study sought to provide a better understanding of the way the amygdala, the orbitofrontal cortex, and the auditory cortex respond and interact during the processing of emotionally valenced stimuli. Initial analyses, examining oscillatory activity in the amygdala, orbitofrontal cortex, and auditory cortex using data from all participants, regardless of whether they had contacts in 1, 2, or all 3 areas of interest, allowed the identification of any patterns of oscillatory activity that may be specific to the processing of positively and negatively valenced stimuli. However, the main analysis focused on the data from a smaller subset of 3 participants, all of whom had at least one contact in each of the 3 areas. Here we investigated the patterns of communication that take place across the 3 regions, by specifically carrying out analysis of functional connectivity. Spectral Granger causality, a variant of the statistical tool known as Granger causality (Granger 1969), seeks to predict oscillations in one area from the past history of oscillations in another and in doing so allows the determination of the direction in which oscillations are being propagated. Critically, this analysis technique has proven useful in inferring causality effects within functional networks in nonhuman and human primates (e.g., Brovelli et al. 2004; Bollimunta et al. 2008; Anderson et al. 2010; Cohen et al. 2012; Cohen and van Gaal 2013). In the current study, it allowed us to examine the transmission of information between structures in the emotion processing network.

We predicted that the current study would confirm processing of musical emotional valence in the amygdala, orbitofrontal and auditory cortex. Specifically, we predicted that time–frequency analysis would implicate low frequency rhythms, specifically the theta rhythm, in the processing of these valenced musical stimuli, and most importantly, that the examination of the propagation of low frequency oscillations would reveal a modulatory effect of the amygdala both on the orbitofrontal and the auditory cortex.

Materials and Methods

Participants and Intracranial Recordings

Data were collected from 11 participants who suffered from severe, pharmacoresistant partial epilepsy and had therefore been implanted with depth electrodes for presurgical evaluation. One participant had to be excluded as a result of unusable data and 3 others were not analyzed as they were not implanted with electrodes in relevant structures, resulting in a final sample of 7 participants (3 males; mean age = 34.85 ± 8.9 years). Experiments were approved by the Ethical Committee for Biomedical Research of Pitié-Salpêtrière Hospital in Paris (agreement 19-07 issued on 26 March 2007). All participants gave informed consent and all clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki.

The invasive exploration had been planned according to the brain locations expected to be at the origin of epileptic seizures. These localization hypotheses resulted from the analysis of previously gathered data (clinical examination and history, prolonged continuous surface EEG-video, brain magnetic resonance imaging (MRI) and sometimes Fluorodeoxyglucose Positron Emission Tomography and ictal Single Photon Emission Computed Tomography). Electrodes were composed of 4–10 contacts, 2.3 mm long, 5 to 10 mm apart, mounted on a 1 mm wide flexible plastic probe (Ad-Tech Medical Instruments, Racine, WI) and were stereotactically inserted using a Leksell frame (Adam et al. 1996). Cartesian coordinates were calculated for each recording site after normalizing the anatomical cerebral MRI into Tailarach space. Local field potential data were acquired with a Nicolet 6000 (Nicolet-Viasys, Madison, WI, US) at a sampling rate of 400 Hz (band-pass: 0.05–150 Hz). Data were analyzed with bipolar montage in order to minimize the influence of distant sources. Bipolar derivations were also preferred to monopolar signals since they prevent potentially nonsilent references from contaminating the results of Granger causality analysis (Bollimunta et al. 2009). Analysis with bipolar montage involved subtracting the signals recorded from adjacent sites. For each bipolar montage, the Cartesian coordinates (x, y, z) were calculated as the location at mid-distance between the 2 adjacent recording sites. A skilled anatomist (D.H.) examined the location of each of the contacts individually in the MRI images in order to verify and confirm their location in the different brain structures. In Figure 1, the distribution of contacts is shown for each participant whose data were used. Table 1 shows the number of contacts in each of the areas of interest in all 7 subjects.

Table 1

Total number of bipolar contacts in each area for each subject and total number of pairwise combinations of contact signals in the 3 subject subsamplea

 Number of bipolar contacts per area
 
   
Subject Amyg Orb Aud    
    
   
Number of connections per area pair 
Amyg–Orb Aud–Orb Aud–Amyg 
Three-subject subsample 
6(4) 
3(2) 
 Number of bipolar contacts per area
 
   
Subject Amyg Orb Aud    
    
   
Number of connections per area pair 
Amyg–Orb Aud–Orb Aud–Amyg 
Three-subject subsample 
6(4) 
3(2) 

Note: In parentheses are presented the number of contacts in the dominant hemisphere for the subsample of 3 subjects. Granger causality analysis was carried out only on the contacts in this hemisphere.

aAmyg, Orb, and Aud to indicate amygdala, orbitofrontal cortex, and auditory cortex, respectively.

Figure 1

The distribution of all the contacts in each of the 7 patients whose data are considered. Amygdala (pink), orbitofrontal cortex (green), and auditory cortex (black) contacts used in the analysis are indicated in color. The 3 subjects with contacts in the amygdala, orbitofrontal, and auditory cortex, whose data are therefore most extensively reported, are indicated with asterisks.

Figure 1

The distribution of all the contacts in each of the 7 patients whose data are considered. Amygdala (pink), orbitofrontal cortex (green), and auditory cortex (black) contacts used in the analysis are indicated in color. The 3 subjects with contacts in the amygdala, orbitofrontal, and auditory cortex, whose data are therefore most extensively reported, are indicated with asterisks.

Stimuli and Task

Stimuli and task were the same as in Dellacherie et al. (2009). Three-tone consonant and dissonant chords were created using the music computation and notation software, Sibelius. 50% of consonant chords were made of perfect major chords while the other 50% were made of minor chords. Dissonant chords were composed of 2 minor seconds or a combination of a minor and a major second. The duration of each chord was 1800 ms while the interstimulus interval (time period between the presentation of individual chords) varied between 300 and 700 ms. Participants were presented with the consonant and dissonant chords in 12 blocks of 40 chords each. Fifty percent of each block was consonant chords and the other 50% was dissonant chords. Eighty percent of each block was comprised of chords of one timbre (piano or organ) while the remaining 20% was presented in the other timbre and constituted “target chords”. Participants were required to count the number of target chords in each block and report this number at the end of the block. For all analyses, only trials corresponding to non-target chords in each block were considered.

Data Preprocessing

Data were epoched from −1000 ms to 3000 ms relative to the onset of the chords. An automatic artifact removal of epileptic activity was carried out. This excluded 1) trials whose maximum amplitude exceeded the mean amplitude of the trial by at least 5 standard deviations (SD) as well as 2) contacts in which >5% of trials were excluded (following the previous criterion). Trials were also checked for spikes and abnormal rhythmic activity.

Time–Frequency Analyses

Estimates of event-related (log) spectral perturbations (ERSPs) were computed using the newtimef function from the EEGLAB toolbox (Delorme and Makeig 2004). A fast Fourier transform with Hanning window tapering (EEGLAB's default algorithm) and a padding ratio of 4 was used. ERSPs were computed over a frequency range of 4–30 Hz. A prestimulus baseline of 300 ms relative to the onset of the chords was used to compute the differences in log spectral power following the onset of the chord. Trials were collapsed across contacts and across subjects to obtain a set of consonant and dissonant trials for each region. The built-in permutation statistical test with a Bonferroni-corrected alpha value of 0.0167 (0.05/3, where 3 is the number of areas being considered) was then used to test for significant differences between consonant and dissonant chord responses within each region of interest.

Granger Causality Analyses

Granger causality analysis was performed on the data from 3 participants who had electrodes implanted in all 3 regions of interest (the orbitofrontal cortex, the auditory cortex, and the amygdala). In contrast to measures such as synchrony (Engel and Singer 2001) and phase coherence (Nunez et al. 2001), which provide estimates of undirected functional connectivity between signals, Granger causality analysis allows the assessment of directed functional connectivity and therefore the examination of the putative direction of information flow within a system. According to the theory and practice of Granger causality, a variable x causes y if information in the past of x can help to predict the future of y (in a linear regression model of x and y), with more accuracy than is possible when using only the past of y itself (Seth 2010). The analyses used here, MultiVariate AutoRegressive (MVAR) modeling, involves the testing of the consistency between data and model by comparing covariance structures estimated from both (Ding et al. 2000).

In the current study, Granger causality analysis based on MVAR modeling was carried out using adapted functions from the Matlab toolbox developed by Anil K. Seth (Seth 2010). Preprocessing steps included the removal of line noise (50 Hz) by multitaper filtering (whereby a 50 Hz sinusoidal oscillation was fit to the noise and then subtracted), the removal of deterministic linear trends (by subtracting the best straight line fit from each trial), and the removal of the temporal mean (by subtracting the value of the mean of the entire trial from the value at each sample of the trial). Finally, the ensemble mean, the mean value for each variable across trials at each time point, was removed as recommended for meeting the covariance stationarity assumption that is required for autoregressive modeling (Brovelli et al. 2004; Seth 2010). Note, however, that while this step removes intracranial evoked response potentials, these may vary on a trial-to-trial basis both in terms of amplitudes and latencies and indeed it has been demonstrated that time–frequency and Granger causality results may, for instance, demonstrate artefactual temporal modulations simply due to the trial-to-trial variability of event-related potentials (Truccolo et al. 2002; Wang and Ding 2011; Wang and Ding 2011).

We implemented spectral Granger causality, the frequency domain interpretation of Granger causality obtained by performing a Fourier transform on the time domain MVAR model, as it allowed the examination of causality effects within distinct frequency bands, in this case the theta, alpha, and beta bands. As spectral Granger causality measures lack a known statistical distribution, bootstrap confidence intervals were constructed to identify Granger causality values that were significantly greater than zero. The model order parameter required for the MVAR model is sometimes estimable using the “Akaike information criterion” or “Bayesian Information criterion”. However, in the event that these methods are not able to provide an optimal model order, a range of model orders may be tested (Seth 2010). Thus, we carried out analysis using model orders of 20 and 10, downsampling the data for the theta band analysis to 200 Hz and keeping the sampling rate at 400 Hz for the alpha and beta band. For both types of analysis, we report results using a model order of 20 (at a sampling rate of 200 Hz for the theta band and 400 Hz for the alpha and beta band, this is the equivalent of a 100 and 50 ms lag, respectively) since other models produced similar results. These time lags of 100 and 50 ms for theta and alpha–beta, respectively, were also favored as they are in line with the previous literature (Brovelli et al. 2004; Bressler et al. 2007; Bollimunta et al. 2008; Zhang et al. 2008; Anderson et al. 2010; Hoerzer et al. 2010; Bollimunta et al. 2011).

To avoid the complication of cross hemisphere effects that would vary across participants by virtue of varying implantation distributions, the contacts from only one hemisphere were considered for each participant: this was the right hemisphere in Subject 3 and the left hemispheres in Subjects 5 and 6. Importantly, this resulted in data from only 3 out of 21 contacts being discarded (2 orbitofrontal cortex contacts from Patient 5, and 1 amygdala contact from Patient 6) due to the general tendency for patients to have greater coverage in one hemisphere than the other. Table 1 shows the resulting number of contacts in each of the 3 areas for each of the 3 subjects of interest. For pairwise computation of Granger causality values, 3 area pairs are possible (amygdala and orbitofrontal cortex, auditory cortex and orbitofrontal cortex, and auditory cortex and amygdala). Thus for each subject, each contact from each area was pairwise combined with each of the contacts from a given other area (Anderson et al. 2010). Table 1 shows the number of such contact pairs for each area pair for each subject. In the results, we report the mean Granger causality values across contact pairs to obtain a single causality profile per area pair (in each direction) per subject.

In addition to seeking to observe causality effects as they change over time, the choice of the size of the analysis time window on which to compute Granger causality values was governed by methodological considerations. The Granger causality analysis approach assumes that all variables concerned have an unchanging mean and variance over time; that is, the variables are covariance stationary. With the current data, it was necessary to carry out analysis on windows as small as 125 ms (0–125, 125–250 ms, etc.) to achieve perfect stationarity (100% performance on the Kwiatkowski–Phillips–Schmidt–Shin test of nonstationarity which tests the null hypothesis of no unit roots). However, although allowing the MVAR model to meet the covariance stationarity criterion to somewhat lesser extent (∼90% covariance stationarity), time windows of 250 ms (0–250, 250–500, and 500–750 ms) were necessary to obtain meaningful estimates of causality in the theta range: 250 ms corresponds to one full cycle oscillation of the lowest frequency of interest in the theta band, 4 Hz. Given these considerations, we used the 250 ms windows for the examination of Granger causality in the theta band (4–7 Hz) and the more reliable and greater temporally resolved 125 ms analysis windows for the examination of Granger causality in the alpha (8–12 Hz) and beta bands (13–30 Hz). The power spectra of the data used in these analyses are shown in Supplementary Figure 1 and the peaks in power for the 250 and 125 ms analysis windows at ∼7 (yellow and green lines) and 10 Hz (blue and red lines), respectively, demonstrate the effect of the analysis window size parameter on the amount of theta power that could be estimated. Importantly, the peaks observed in the power spectrum can be seen to correspond with the peaks of the Granger causality values. Finally, inferences using the Granger causality method may be considered valid only if the estimated MVAR model adequately captures the correlation structure in the data and it was therefore ensured that all models considered had an appropriately high model consistency (a measure of the variance accounted for by an MVAR model) of not <90%.

Results

Time–frequency

To examine, for each region, whether either consonant or dissonant chords were associated with relatively disproportionate changes in power, trials from all contacts across relevant participants in each region were aggregated to obtain 2 sets of trials (one “consonant” and the other “dissonant”), which were then compared. Figure 2 plots, for each region, time-frequency representations of the output of ERSP analysis carried out on all trials from all subjects (upper 9 plots) as well as on all trials from contacts in the 3-subject subsample (bottom 9 plots).

Figure 2

Grand average time–frequency diagrams showing power in response to consonant and dissonant chords and the significant difference between these responses. Color in the columns showing responses to consonant and dissonant chords (first 2 columns) indicate the log increase or decrease in power relative to a baseline of the 300 ms preceding the sound onset. Color in the third column indicates the presence and strength of significant differences between spectral perturbations in response to the 2 types of sounds as evaluated using permutation statistical testing with a Bonferroni-corrected alpha value of 0.0167 (0.05/3).

Figure 2

Grand average time–frequency diagrams showing power in response to consonant and dissonant chords and the significant difference between these responses. Color in the columns showing responses to consonant and dissonant chords (first 2 columns) indicate the log increase or decrease in power relative to a baseline of the 300 ms preceding the sound onset. Color in the third column indicates the presence and strength of significant differences between spectral perturbations in response to the 2 types of sounds as evaluated using permutation statistical testing with a Bonferroni-corrected alpha value of 0.0167 (0.05/3).

The figure shows a large-band transient increase in power in auditory cortex peaking at a latency of ∼100 ms after sound onset and a more gradual increase in theta and alpha power over time, reaching a peak between 200 and 600 ms, in the amygdala and orbitofrontal cortex. In the orbitofrontal cortex, results from data of the 3 subjects alone indicated greater theta, alpha, and low beta power (between 13 and 20 Hz) for consonant relative to dissonant sounds that were centered at ∼250, 400 and 150 ms, respectively. Also observed in this region was a drop in high beta power (between 21 and 30 Hz) for consonant relative to dissonant chords between 100 and 400 ms. Results of analysis considering data from all subjects indicated a similar pattern of results although the lower beta power difference previously observed was absent. In the auditory cortex, both types of analysis (all subjects, and the 3 subjects alone) revealed a drop in low beta power from consonance to dissonance centered at ∼800 ms and another late drop in high beta power from consonance to dissonance centered at ∼1300 ms. In contrast to the orbitofrontal cortex and auditory cortex, no consistent significant power differences between the consonant and dissonant conditions were observed in the amygdala.

Granger Causality

This analysis sought to determine whether there was any evidence of directed functional connectivity among the 3 regions of interest with a focus on the theta, alpha, and beta frequency bands, given that frequencies in this range are held to be involved in long-range neural communication during emotional processing (Lewis 2005). Figure 3 shows, for each subject, how the mean significant Granger causality values computed for each of the 6 different directions (amygdala to orbitofrontal cortex, orbitofrontal cortex to amygdala, auditory cortex to orbitofrontal cortex, orbitofrontal cortex to auditory cortex, auditory cortex to amygdala and amygdala to auditory cortex) evolved over time from sound onset in windows of 250 ms for the theta band (0–250; 250–500 … 1500–1750 ms) and 125 ms for the alpha and beta bands (0–125; 125–250 … 1625–1750 ms). The value at time 0 indicates the significant causality observable for the given directed connection in the 250 ms preceding the onset of the sound.

Figure 3

Plots showing the evolution over time of significant causal flow for consonant (blue) and dissonant (red) chords in the theta, alpha, and beta bands for the 3-subject subsample. For all curves, solid lines indicate the forward direction of Granger causality within the pair (thus Amygdala → Orbitofrontal cortex, Auditory Cortex → Orbitofrontal cortex, Auditory Cortex → Amygdala) while broken lines indicate causality in the reverse direction (Orbitofrontal cortex → Amygdala, Orbitofrontal cortex → Auditory Cortex, Amygdala → Auditory Cortex). The value at time 0 indicates the significant causality observable in the 250 ms preceding the onset of the sound.

Figure 3

Plots showing the evolution over time of significant causal flow for consonant (blue) and dissonant (red) chords in the theta, alpha, and beta bands for the 3-subject subsample. For all curves, solid lines indicate the forward direction of Granger causality within the pair (thus Amygdala → Orbitofrontal cortex, Auditory Cortex → Orbitofrontal cortex, Auditory Cortex → Amygdala) while broken lines indicate causality in the reverse direction (Orbitofrontal cortex → Amygdala, Orbitofrontal cortex → Auditory Cortex, Amygdala → Auditory Cortex). The value at time 0 indicates the significant causality observable in the 250 ms preceding the onset of the sound.

Examination of the patterns of causal flow in the 6 possible directions, for each frequency band separately, revealed a number of interesting patterns. With regard to the theta band, when considering the 6 directed connections for which Granger causality was calculated, significant flow was observed from the amygdala to the orbito-frontal cortex (Amyg → Orb). While one subject did not show any flow either in this direction or the opposite (Orb → Amyg, Subject 3), another participant (Subject 5) showed much larger flow from the amygdala to orbitofrontal cortex relative to the opposite direction (Orb → Amyg), while the third also showed an asymmetry, especially for consonant sounds, despite a higher baseline activity in the direction from amygdala to orbitofrontal cortex. Comparable to the flow between amygdala and orbitofrontal cortex, with regard to magnitude, was the flow from the amygdala to the auditory cortex (Amyg → Aud). Causal flow between the amygdala and auditory cortex showed even higher consistency here with all 3 subjects showing greater flow in this direction than in the reverse (Aud → Amyg). A final observation was that there was only little flow between the auditory cortex and the orbitofrontal cortex and this flow tended to be from the latter to the former (Orb → Aud) with 3 participants showing flow in this direction and 2 showing flow in the reverse direction.

With regard to the alpha band, once again the greatest causal flow observable was from the amygdala to the orbitofrontal cortex (Amyg → Orb). All 3 participants showed greater flow in this direction than in the reverse (Orb → Amyg). Further, the peak of causal flow was fairly consistent at ∼500 ms in all subjects. As for the flow to and from the auditory cortex, there was bidirectional flow between the auditory cortex and amygdala: with all 3 subjects showing significant flow from auditory cortex to amygdala (Aud → Amyg) and 2 out of the 3 participants showing flow from amygdala to auditory cortex (Amyg → Aud). Bidirectional flow was also observed between auditory cortex and orbitofrontal cortex, with 3 subjects showing significant flow from Orb to Aud (Orb → Aud) and 2 showing flow in the reverse (Aud → Orb). A potential effect of valence could be observed with amount of causal flow for all 3 directions being generally greater for consonant than dissonant sounds. Finally, with regard to the beta band, the greatest causal flow was once again observed from amygdala to orbitofrontal cortex, although this was not as consistently large or as time locked across participants as in the alpha band. Significant causal flows in the 5 other directions were generally lower than that which was observable in the theta and alpha bands.

To summarize the patterns of causal flow between the 3 different structures making up the distributed network considered here, Figure 4 shows the total causality over the duration of the stimuli in each of the 6 possible directions, averaged across consonant and dissonant sounds and across the 3 subjects. This highlights 2 noteworthy findings: First, a tendency for the causal flow from the amygdala to the orbitofrontal cortex to be greater than in the reverse direction in all 3 bands (although with greater consistency across subjects in the alpha band) and second, a considerable and reliable causal flow from amygdala to auditory cortex, specifically in the theta band. Supplementary Figure 2, showing the results when the causal flow in the 250 ms preceding sound onset are subtracted from significant Granger causality values in each time window, confirm these main findings to be robust to any variation in baseline levels of connectivity.

Figure 4

Barplots showing mean significant causal flow in each of the 6 possible directions averaged across consonant and dissonant sounds and across the 3 subjects. The number of participants showing causal flow in each given direction is indicated.

Figure 4

Barplots showing mean significant causal flow in each of the 6 possible directions averaged across consonant and dissonant sounds and across the 3 subjects. The number of participants showing causal flow in each given direction is indicated.

Discussion

By examining intracranial depth electrode recordings collected while participants were presented with consonant and dissonant chords, the current study was able to elaborate on the neural dynamics of emotional valence processing in 3 areas of interest: the amygdala, the orbitofrontal cortex, and the auditory cortex. We initially analyzed the data of all patients from whom data were collected in these regions (regardless of whether they had contacts in one, 2 or all 3 areas of interest), before focusing on the data from a subset of 3 participants, all of whom had at least one contact in each of the 3 areas. Our approach allowed us to fully exploit the data available while addressing the dual objectives of characterizing the patterns of neural responses within each region separately and the patterns of communication that take place across the 3 regions.

Initial time–frequency analysis suggested a special role of the orbitofrontal cortex in the processing of positively valenced stimuli. Specifically, greater power in the theta and alpha frequency bands was observed during the processing of consonant stimuli relative to dissonant stimuli, while no such trends (i.e., greater oscillatory activity in response to either positive or negatively valenced stimuli) were seen in the amygdala. This finding replicates results in a previous noninvasive EEG study (Sammler et al. 2007) showing, in response to consonant sounds, increased frontal midline theta, held to arise from the medial prefrontal cortex (Asada et al. 1999). Thus, perhaps the most novel contribution of the study was its demonstration of directed communication within the emotional processing network: firstly, from the amygdala to the orbitofrontal cortex and secondly, from the amygdala to the auditory cortex, suggesting that orbitofrontal and auditory cortex responses are modulated by the amygdala.

Indeed, our main finding was that in contrast to causal flow in the reverse directions, the causal flow from the amygdala to the orbitofrontal cortex and from the amygdala to the auditory cortex was considerable and reliable across subjects. Various previous functional magnetic resonance imaging (fMRI) and positron emission tomography studies have been able to identify the amygdala and orbitofrontal cortex as key players in the processing of emotionally valenced sounds (Blood et al. 1999; Blood and Zatorre 2001; Koelsch et al. 2006). However, these haemodynamic studies had not been able to elaborate either on the time course of processing in these areas or the degree of influence that they have on each other. The current study, which along with our previous case study is the first to examine intracranial EEG data in these areas during the processing of emotional musical stimuli, was well placed to address such questions. For one, while the ability to record from intracranial electrodes provided the necessary spatial resolution for describing the activity in the specific areas, the method's high temporal resolution allowed the inference of functional connectivity using Granger causality analysis. This analysis is dependent on there being an invariable (if any) response delay across the different structures being considered, and is therefore more reliable for electrophysiological than, for instance, fMRI data, where a difference in haemodynamics in different areas can present a severe confound.

While the classical view of the amygdala associates it chiefly with the processing of negative stimuli (LeDoux et al. 1990; Morris et al. 1998), we demonstrate here that the amygdala is a critical hub in the musical emotion processing network and may have a rather more valence-general role (Lane et al. 1999; Taylor et al. 2000; Davis and Whalen 2001; Garavan et al. 2001; Zald 2003). One part of this role, as indexed by the flow from this region to the orbitofrontal cortex, might be to influence higher order areas, which are involved in the evaluation of a stimulus' value. Another might be to influence the sensory cortices, which are involved in the processing of stimuli's low-level features. Indeed, adaptive behavior necessitates that emotionally salient information is processed particularly efficiently, and our results here suggesting a modulation of the auditory cortex by the amygdala alludes to this potential tendency for sensory level processing of a stimulus to be influenced by classical emotion processing areas.

The notion of a modulatory role of the amygdala on sensory cortical processing is in agreement with the extensive anatomical connections between the amygdala and auditory cortex (Amaral and Price 1984; Iwai and Yukie 1987) as well as previous neuroimaging studies. Several authors have suggested that the amygdala may be responsible for the modulation of activity observable in the visual and auditory cortices when participants are presented with emotional stimuli (Lang et al. 1998; Morris et al. 1998; Plichta et al. 2011; Viinikainen et al. 2011). In the auditory domain, the idea of a modulatory role of the amygdala has been supported by an fMRI study (Kumar et al. 2012), which examined the specific effect of acoustic features and emotional valence on the coupling between amygdala and the auditory cortex. Using dynamic causal modeling, a tool for inferring effective connectivity, the authors showed that while acoustic features of aversive sounds modulated the connectivity from the auditory cortex to the amygdala, it was the sounds' valence that modulated connectivity from the amygdala to the auditory cortex. Our findings extend those from Kumar and colleagues in demonstrating connectivity from the amygdala to the auditory cortex during processing of positively valenced stimuli and in doing so support the notion of the amygdala being involved in the processing of positive as well as negative stimuli (Davis and Whalen 2001; Garavan et al. 2001; Zald 2003; Lane et al. 1999; Taylor et al. 2000). Our results also extend the notion of a modulatory role of the amygdala on sensory cortices to the musical domain and calls for further investigation into the previously suggested dissociation between perceptual and emotional processing in the auditory cortex (Peretz et al. 2001).

Strong anatomical connections are known to exist between the amygdala, orbitofrontal cortex, and auditory cortex (Amaral and Price 1984; Iwai and Yukie 1987; Ebeling and von Cramon 1992; Petrides and Pandya 2007; Thiebaut de Schotten et al. 2012) and the current study contributes to the literature by showing that communication between these areas is underlied by low-frequency oscillations. The notion that low frequency oscillations may allow long-range synchronization of distributed brain regions (Varela et al. 2001; Fries 2005; Buzsaki 2006) has been supported by studies showing theta coherence between frontal and posterior areas during cognitive tasks (Sarnthein et al. 1998; Sauseng et al. 2005; Summerfield and Mangels 2005). It had also been suggested that theta might play a specific role in the synchronization of medial temporal and prefrontal lobe structures during the processing of emotional stimuli (Lewis 2005). The current work supports such a notion further in showing that in addition to the theta band, such synchronized activity during emotion processing may also occur in the alpha frequency band in line with an increasing number of studies associating this frequency band with the processing of emotional stimuli (Knyazev et al. 2008; Parvaz et al. 2012).

In summary, our work confirms previous studies, both in the visual and auditory domain, that demonstrate processing of emotional valence in the amygdala, orbitofrontal cortex, and auditory cortex. It demonstrates that activity in both the theta and alpha frequency bands play an important role in the processing of emotional stimuli. Moreover, it makes a particularly significant contribution in its clarification of the nature of the directed communications that exists in the emotion network during the processing of valenced stimuli. However, there are some limitations in the current study, notably regarding the consistency of the Granger causality analysis results across patients. Indeed while we emphasize certain trends, namely for amygdala to drive orbitofrontal cortex and auditory cortex, the timing and size of causal flows vary considerably across participants. These variations may have cognitive and/or behavioral bases. The stimuli used in the experiments were fairly long at 1800 ms and indeed it is possible that participants might have varied in how they considered the stimuli over time. For example, some perhaps might have focused on the sounds for the whole duration while others may have focused during just the first few hundred milliseconds. Also while some participants might have tended to make evaluative judgments (even if not asked to), others may have been less inclined to do so, resulting in differential modulation of the network. Another source of inconsistencies across subjects may be in the variability in terms of where contacts were situated in each participant. Indeed, while we carry out analysis on a region level, due to a limited number of contacts and therefore a limited ability to differentiate between subregions (e.g., amygdala subnuclei), it is possible that the contacts considered varied across participants in terms of location and possibly function and resulted in the differences in timing as well as sensitivity to positive and negative valence that is seen across subjects.

Finally, while the current paper sheds some new light on the processing of valenced stimuli, a few points are worth considering in the future. First, given that previous studies have shown enhanced processing in the amygdala when explicit emotion judgments are being made (e.g., Krolak-Salmon et al. 2004), it would be interesting to observe how the effects seen here are modulated by task, in other words, to what extent explicit assessment of pleasantness (in contrast to the timbral deviant counting task used here) may influence the system's behavior. Secondly, while Granger causality analysis is proving a useful tool in inferring causality in neural systems, it remains simply a statistical approach and, indeed, as argued by Pearl (1999), causality effects can be proven only by observing the consequences of disturbing the system. One way to confirm the source of modulatory influences on particular regions in a network may be to combine neuroimaging and lesion patient approaches. Using this method, Vuilleumier et al. (2004) were able to show a parametric relationship between the extent of damage of the amygdala and the level of activation in the visual areas involved in face processing. A similar approach taken here would allow further verification of the extent to which processing in the amygdala does influence activity in both the orbitofrontal cortex and auditory cortex.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

Funding

This study has received funding from the French Ministry (ANR-09-BLAN-0310-02) and the Institut Universitaire de France to S.S, the Regional Council of Nord-Pas de Calais (PhD scholarship) to D.D., and the program “Investissements d'avenir” ANR-10-IAIHU-06.

Notes

Conflict of Interest: None declared.

References

Adam
C
Clemenceau
S
Semah
F
Hasboun
D
Samson
S
Aboujaoude
N
Samson
Y
Baulac
M
.
1996
.
Variability of presentation in medial temporal lobe epilepsy: a study of 30 operated cases
.
Acta Neurol Scand
 .
94
:
1
11
.
Aftanas
LI
Varlamov
AA
Pavlov
SV
Makhnev
VP
Reva
NV
.
2001
.
Affective picture processing: event-related synchronization within individually defined human theta band is modulated by valence dimension
.
Neurosci Lett
 .
303
:
115
118
.
Amaral
DG
Price
JL
.
1984
.
Amygdalo-cortical projections in the monkey (Macaca fascicularis)
.
J Comp Neurol
 .
230
:
465
496
.
Anderson
KL
Rajagovindan
R
Ghacibeh
GA
Meador
M
Ding
M
.
2010
.
Theta oscillations mediate interaction between prefrontal cortex and medial temporal lobe in human memory
.
Cereb Cortex
 .
20
:
1604
1612
.
Asada
H
Fukuda
Y
Tsunoda
S
Yamaguchi
M
Tonoike
M
.
1999
.
Frontal midline theta rhythms reflect alternative activation of prefrontal cortex and anterior cingulate cortex in humans
.
Neurosci Lett
 .
274
:
29
32
.
Blood
AJ
Zatorre
RJ
.
2001
.
Intensely pleasurable responses to music correlate with activity in brain regions implicated in reward and emotion
.
Proc Natl Acad Sci
 .
98
:
11818
11823
.
Blood
AJ
Zatorre
RJ
Bermudez
P
Evans
AC
.
1999
.
Emotional responses to pleasant and unpleasant music correlate with activity in paralimbic brain regions
.
Nat Neurosci
 .
2
:
382
387
.
Bollimunta
A
Chen
Y
Schroeder
CE
Ding
M
.
2008
.
Neuronal mechanisms of cortical alpha oscillations in awake-behaving macaques
.
J Neurosci
 .
28
:
9976
9988
.
Bollimunta
A
Chen
Y
Schroeder
CE
Ding
M
.
2009
.
Characterizing oscillatory cortical networks with granger causality
. In:
Rubin
J
Josic
K
Matias
M
Romo
R
, editors.
Coherent behavior in neuronal networks
 .
New York
:
Springer
p.
169
189
.
Bollimunta
A
Mo
J
Schroeder
CE
Ding
M
.
2011
.
Neuronal mechanisms and attentional modulation of corticothalamic alpha oscillations
.
J Neurosci
 .
31
:
4935
4943
.
Bressler
SL
Richter
CG
Chen
Y
Ding
M
.
2007
.
Cortical functional network organisation from autoregressive modeling of local field potential oscillations
.
Statist Med
 .
26
:
3875
3885
.
Brovelli
A
Ding
M
Ledberg
A
Chen
Y
Nakamura
R
Bressler
SL
.
2004
.
Beta oscillations in a large-scale sensorimotor cortical network: directional influences revealed by Granger causality
.
Proc Natl Acad Sci
 .
101
:
9849
9854
.
Buzsaki
G
.
2006
.
Rhythms of the brain
 .
Oxford University Press
,
New York
.
Cohen
MX
Bour
L
Mantione
M
Figee
M
Vink
M
Tijssen
MAJ
van Rootselaar
A
van den Munckhof
P
Schuurman
PR
Deny
D
.
2012
.
Top-down-directed synchrony from medial frontal cortex to nucleus accumbens during reward anticipation
.
Hum Brain Mapp
 .
33
:
246
252
.
Cohen
MX
van Gaal
S
.
2013
.
Dynamic interactions between large-scale brain networks predict behavioral adaptation after perceptual errors
.
Cereb Cortex
 .
23
:
1061
1072
.
Costa
M
Ricci Bitti
PE
Bonfiglioli
L
.
2000
.
Psychological connotations of harmonic musical intervals
.
Psychol Music
 .
2
:
4
22
.
Davis
M
Whalen
PJ
.
2001
.
The amygdala: vigilance and emotion
.
Mol Psychiatr
 .
6
:
13
34
.
Dellacherie
D
Pfeuty
M
Hasboun
D
Lefevre
J
Hugueville
L
Schwartz
DP
Baulac
M
Adam
C
Samson
C
.
2009
.
The birth of musical emotion: a depth electrode case study in a human subject with epilepsy
.
Ann NY Acad Sci
 .
1169
:
336
341
.
Delorme
A
Makeig
S
.
2004
.
EEGLAB: an open source toolbox for analysis of single trial EEG dynamics
.
J Neurosci Methods
 
134
:
9
21
.
Ding
M
Bressler
SL
Yang
W
Liang
H
.
2000
.
Short-Window Spectral Analysis of Cortical Event-Related Potentials by Adaptive MultiVariate AutoRegressive (AMVAR) Modeling: Data Preprocessing, Model Validation, and Variability Assessment
.
Biol Cybern
 .
83
:
35
45
.
Ebeling
U
von Cramon
D
.
1992
.
Topography of the uncinate fascicle and adjacent temporal fiber tracts
.
Acta Neurochir
 .
115
:
143
148
.
Eger
E
Jedynak
A
Iwaki
T
Skrandies
W
.
2003
.
Rapid extraction of emotional expression: Evidence from evoked potential fields during brief presentation of face stimuli
.
Neuropsychologia
 .
41
:
808
817
.
Eimer
M
Holmes
A
.
2002
.
An ERP study on the time course of emotional face processing
.
Neuroreport
 .
13
:
427
431
.
Engel
A
Singer
W
.
2001
.
Temporal binding and the neural correlates of sensory awareness
.
Trends Cogn Sci
 .
5
:
16
25
.
Fries
P
.
2005
.
A mechanism for cognitive dynamics: Neuronal communication through neuronal coherence
.
Trends Cog Sci
 .
9
:
474
480
.
Gallagher
M
Chiba
AA
.
1996
.
The amygdala and emotion
.
Curr Opin Neurobiol
 .
6
:
221
227
.
Garavan
H
Pendergrass
JC
Ross
TJ
Stein
EA
Risinger
RC
.
2001
.
Amygdala response to both positively and negatively valenced stimuli
.
Neuroreport
 .
12
:
2779
2883
.
Gothard
KM
Battaglia
FP
Erickson
CA
Spitler
KM
Amaral
DG
.
2007
.
Neural responses to facial expression and face identity in the monkey amygdala
.
J Neurophysiol
 .
97
:
1671
1683
.
Granger
C
.
1969
.
Investigating causal relations by econometric models and cross-spectral methods
.
Econometrica
 .
37
:
424
438
.
Hamann
SB
Ely
TD
Grafton
ST
Kilts
CD
.
1997
.
Amygdala activity related to enhanced memory for pleasant and aversive stimuli
.
Nat Neurosci
 .
2
:
289
293
.
Hoerzer
GM
Liebe
S
Schloegl
A
Logothetis
NK
Rainer
G
.
2010
.
Directed coupling in local field potentials of macaque V4 during visual short-term memory revealed by multivariate autoregressive models
.
Front Comput Neurosci
 .
4
:
14
.
Iwai
E
Yukie
M
.
1987
.
Amygdalofugal and amygdalopetal connections with modality-specific visual cortical areas in macaques (Macaca fuscata, M. mulatta, and M. fascicularis)
.
J Comp Neurol
 .
261
:
362
387
.
James
CE
Britz
J
Vuilleumier
P
Hauert
CA
Michel
CM
.
2008
.
Early neuronal responses in right limbic structures mediate harmonic incongruity processing in musical experts
.
Neuroimage
 .
42
:
1597
1608
.
Knyazev
G
.
2007
.
Motivation, emotion, and their inhibitory control mirrored in brain oscillations?
Neurosci Biobehav Rev
 .
31
:
377
395
.
Knyazev
G
Bocharov
AV
Levin
EA
Savostyanov
AN
Slobodskoj-Plusnin
.
2008
.
Anxiety and oscillatory responses to emotional facial expressions
.
Brain Res
 .
1227
:
174
188
.
Koelsch
S
Fritz
T
von Cramon
DY
Muller
K
Friederici
AD
.
2006
.
Investigating Emotion With Music: An fMRI Study
.
Hum Brain Mapp
 .
27
:
239
250
.
Krolak-Salmon
P
Hénaff
M
Vighetto
A
Bertrand
O
Mauguière
F
.
2004
.
Early amygdala reaction to fear spreading in occipital, temporal, and frontal cortex: a depth electrode ERP study in human
.
Neuron
 .
42
:
665
676
.
Kumar
S
von Kriegstein
K
Friston
K
Griffiths
TD
.
2012
.
Features versus Feelings: Dissociable Representations of the Acoustic Features and Valence of Aversive Sounds
.
J Neurosci
 .
32
:
14184
14192
.
Lane
RD
Chua
RJ
Dolan
R
.
1999
.
Common effects of emotional valence, arousal and attention on neural activation during visual processing of pictures
.
Neuropsychologia
 .
37
:
989
997
.
Lang
PJ
Bradley
MM
Fitzsimmons
JR
Cuthbert
BN
Scott
JD
Moulder
B
Nangia
V
.
1998
.
Emotional arousal and activation of the visual cortex: an fMRI analysis
.
Psychophysiology
 .
35
:
199
210
.
LeDoux
JE
Cicchetti
P
Xagoraris
A
Romanski
LM
.
1990
.
The lateral amygdaloid nucleus: sensory interface of the amygdala in fear conditioning
.
J Neurosci
 .
10
:
1062
1069
.
Lerner
Y
Papo
D
Zhdanov
A
Belozersky
L
Hendler
T
.
2009
.
Eyes wide shut: amygdala mediates eyes-closed effect on emotional experience with music
.
PLoS One
 .
4
:
e6230
.
Lewis
MD
.
2005
.
Bridging emotion theory and neurobiology through dynamic systems modeling
.
Behav Brain Sci
 .
28
:
169
245
.
Maratos
FA
Mogg
K
Bradley
BP
Rippon
G
Senior
C
.
2009
.
Coarse threat images reveal theta oscillations in the amygdala: A magnetoencephalography study
.
Cogn Affect Behav Neurosci
 .
9
:
133
143
.
Meletti
S
Cantalupo
G
Benuzzi
F
Mai
R
Tassi
L
Gasparini
E
Tassinari
CA
Nichelli
P
.
2012
.
Fear and happiness in the eyes: An intra-cerebral event-related potential study from the human amygdala
.
Neuropsychologia
 .
50
:
44
54
.
Mirz
F
Gjedde
A
Sødkilde-Jrgensen
H
Pedersen
CB
.
2000
.
Functional brain imaging of tinnitus-like perception induced by aversive auditory stimuli
.
Neuroreport
 .
11
:
633
637
.
Mitchell
DG
Greening
SG
.
2012
.
Conscious perception of emotional stimuli: brain mechanisms
.
Neuroscientist
 .
18
:
386
398
.
Mormann
F
Kornblith
S
Quiroga
RQ
Kraskov
A
Cerf
M
Fried
I
Koch
C
.
2008
.
Latency and selectivity of single neurons indicate hierarchical processing in the human temporal lobe
.
J Neurosci
 .
28
:
8865
8872
.
Morris
JS
Friston
KJ
Büchel
C
Frith
CD
Young
AW
Calder
AJ
Dolan
RJ
.
1998
.
A neuromodulatory role for the human amygdala in processing emotional facial expressions
.
Brain
 .
121
:
47
57
.
Naccache
L
Gaillard
R
Adam
C
Hasboun
D
Clemenceau
S
Baulac
M
Dehaene
S
Cohen
L
.
2005
.
A direct intracranical record of emotions evoked by subliminal words
.
Proc Natl Acad Sci
 .
102
:
7713
7717
.
Nunez
PL
Wingeier
BM
Silberstein
RB
.
2001
.
Spatial-temporal structures of human alpha rhythms: theory, microcurrent sources, multiscale measurements, and global binding of local networks
.
Hum Brain Mapp
 .
13
:
125
164
.
Oya
H
Kawasaki
H
Howard
MA
III
Adolphs
R
.
2002
.
Electrophysiological responses in the human amygdala discriminate emotion categories of complex visual stimuli
.
J Neurosci
 .
22
:
9502
9512
.
Paré
D
Collins
DR
Pelletier
JG
.
2002
.
Amygdala oscillations and the consolidation of emotional memories
.
Trends Cogn Sci
 .
6
:
306
314
.
Parvaz
M
MacNamara
A
Goldstein
R
Hajcak
G
.
2012
.
Event related induced frontal alpha as a marker of lateral prefrontal cortex activation during cognitive appraisal
.
Cogn Affect Behav Neurosci
 .
12
:
730
740
.
Pearl
J
.
1999
.
Causality: models, reasoning, and inference
 .
Cambridge, UK
:
Cambridge University Press
.
Peretz
I
Blood
AJ
Penhune
V
Zatorre
R
.
2001
.
Cortical deafness to dissonance
.
Brain
 .
124
:
928
940
.
Petrides
M
Pandya
DN
.
2007
.
Efferent association pathways from the rostral prefrontal cortex in the macaque monkey
.
J Neurosci
 .
27
:
11573
11586
.
Plichta
MM
Gerdes
AB
Alpers
GW
Harnisch
W
Brill
S
Wieser
MJ
Fallgatter
AJ
.
2011
.
Auditory cortex activation is modulated by emotion: a functional near-infrared spectroscopy (fNIRS) study
.
Neuroimage
 .
55
:
1200
1207
.
Plomp
R
Levelt
WJ
.
1965
.
Tonal consonance and critical bandwidth
.
J Acoust Soc Am
 .
38
:
548
560
.
Pourtois
G
Dan
ES
Grandjean
D
Sander
D
Vuilleumier
P
.
2005
.
Enhanced extrastriate visual response to bandpass spatial frequency filtered fearful faces: Time course and topographic evoked-potentials mapping
.
Hum Brain Mapp
 .
26
:
65
79
.
Pourtois
G
Grandjean
D
Sander
D
Vuilleumier
P
.
2004
.
Electrophysiological correlates of rapid spatial orienting towards fearful faces
.
Cereb Cortex
 .
14
:
619
633
.
Pourtois
G
Vocat
R
N'diaye
K
Spinelli
L
Seeck
M
Vuilleumier
P
.
2010
.
Errors recruit both cognitive and emotional monitoring systems: simultaneous intracranial recordings in the dorsal anterior cingulate gyrus and amygdala combined with fMR1
.
Neuropsychologia
 .
48
:
1144
1159
.
Rutishauser
U
Ross
IB
Mamelak
AN
Schuman
EM
.
2010
.
Human memory strength is predicted by theta-frequency phase-locking of single neurons
.
Nature
 .
464
:
903
907
.
Sammler
D
Grigutsch
M
Fritz
T
Koelsch
S
.
2007
.
Music and emotion: electrophysiological correlates of the processing of pleasant and unpleasant music
.
Psychophysiology
 .
44
:
293
304
.
Sarnthein
J
Petsche
H
Rappelsberger
P
Shaw
GL
von Stein
A
.
1998
.
Synchronisation between prefrontal and posterior association cortex during human working memory
.
Proc Natl Acad Sci
 .
95
:
7092
7096
.
Sauseng
P
Klimesch
W
Schabus
M
Doppelmayr
M
.
2005
.
Fronto-parietal EEG coherence in theta and upper alpha reflect central executive functions of working memory
.
Int J Psychophysiol
 .
57
:
97
103
.
Seth
AK
.
2010
.
A MATLAB toolbox for Granger causal connectivity analysis
.
J Neurosci Methods
 .
186
:
262
273
.
Streit
M
Dammers
J
Simsek-Kraues
S
Brinkmeyer
J
Wolwer
W
Ioannides
A
.
2003
.
Time course of regional brain activations during facial emotion recognition in humans
.
Neurosci Lett
 .
342
:
101
104
.
Summerfield
C
Mangels
JA
.
2005
.
Coherent theta-band activity predicts item-context binding during encoding
.
Neuroimage
 .
24
:
692
703
.
Taylor
SF
Liberzon
RA
Koeppe
RA
.
2000
.
The effects of graded aversive stimuli on limbic and visual activation
.
Neuropsychologia
 .
38
:
1415
1425
.
Thiebaut de Schotten
M
Dell'acqua
F
Valabregue
R
Catani
M
.
2012
.
Monkey to human comparative anatomy of the frontal lobe association tracts
.
Cortex
 .
48
:
82
96
.
Trost
W
Ethofer
T
Zentner
M
Vuilleumier
P
.
2011
.
Mapping aesthetic musical emotions in the brain
.
Cereb Cortex
 .
22
:
2769
2783
.
Truccolo
WA
Ding
MZ
Knuth
KH
Nakamura
R
Bressler
SL
.
2002
.
Trial-to-trial variability of cortical evoked responses: Implications for the analysis of functional connectivity
.
Clin Neurophysiol
 .
113
:
206
226
.
van Heijnsbergen
CC
Meeren
HK
Grezes
J
de Gelder
B
.
2007
.
Rapid detection of fear in body expressions, an ERP study
.
Brain Res
 .
1186
:
233
241
.
Varela
F
Lachaux
JP
Rodriguez
J
Martinerie
J
.
2001
.
The brainweb: phase synchronization and large-scale integration
.
Nat Rev Neurosci
 .
2
:
229
239
.
Viinikainen
M
Katsyri
J
Sams
M
.
2011
.
Representation of perceived sound valence in the human brain
.
Hum Brain Mapp
 .
33
:
2295
2305
.
Vuilleumier
P
Richardson
MP
Armony
JL
Driver
J
Dolan
RJ
.
2004
.
Distant influences of amygdala lesion on visual cortical activation during emotional face processing
.
Nat Neurosci
 .
7
:
1271
1278
.
Wang
X
Ding
M
.
2011
.
Relation between P300 and event-related theta-band synchronization: A single-trial analysis
.
Clin Neurophysiol
 .
122
:
916
924
.
Wedin
L
.
1972
.
A multidimensional study of perceptual-emotional qualities in music
.
Scand J Psychol
 .
13
:
241
257
.
Zald
DH
.
2003
.
The human amygdala and the emotional evaluation of sensory stimuli
.
Brain Res
 .
44
:
81
123
.
Zald
DH
Pardo
JV
.
2002
.
The neural correlates of aversive auditory stimulation
.
Neuroimage
 .
16
:
746
753
.
Zhang
Y
Chen
S
Bressler
L
Ding
M
.
2008
.
Response preparation and inhibition: the role of the cortical sensorimotor beta rhythm
.
Neuroscience
 .
156
:
238
246
.