Successful behavior depends on effective communication between distant brain regions. Moreover, disturbance of effective communication can cause neurological symptoms like apraxia, dyslexia or object agnosia. Interregional communication can be assessed by coherence analysis of synchronized neuronal oscillations, and has been referred to as synchrony or ‘binding’. The concept of synchrony as a means of information coding is attractive, but its functional relevance has been challenged. We hypothesized that if synchrony is functionally relevant in humans, then more synchrony should determine better behavioral performance. Here, we show in a visuotactile integration task that the amount of low-frequency (7–13Hz), long-range electroencephalographic coherence between visual and sensorimotor cortex is significantly correlated with the level of performance. Trials with highest coherences were the most successful ones and vice versa in the absence of differences in regional activation measured as task-related spectral power. In summary, quantitatively linking the amount of long-range synchrony with the degree of behavioral success in humans, the present data suggest that the ability to generate topographically specific synchrony of high amplitude is functionally relevant for behavioral success. They also raise the possibility that the magnitude of regional activation is less representative of the efficacy of brain functioning than interregional synchrony.
Communication of distant brain areas provides the basis for integration of complex information to adapt to changes in the environment, to process this information, and to generate appropriate behavioral responses necessary for successful behavior in daily life. Hebb (1949) developed a theory with three basic assumptions that were speculative at that time, but have since been substantiated by empirical evidence: (i) cortical neurons strengthen their connections by repeated synchronous activation (Ahissar et al., 1992) — this is called ‘Hebb's rule’; (ii) the cortex is an apparatus in which extensive associative memory can be established by synaptic strengthening between adjacent neurons and/or neurons in distant brain areas (Braitenberg and Schüz, 1998); and (iii) repeated synchronous neuronal activation and subsequent synaptic strengthening has important functional implications (Singer and Gray, 1995). On the basis of this theory, it has been assumed that perceptions or actions are represented in the brain by large numbers of distributed neurons firing in synchrony, i.e. by coupled oscillators (Singer, 1994). In the visual system, this has been referred to as ‘binding’ (Singer and Gray, 1995). From correlated single unit discharges, the ‘binding’ concept has been extended to coherent oscillations of local field potentials (Singer, 1993) and from high-frequency (γ) (Pulvermuller et al., 1995; Miltner et al., 1999; Rodriguez et al., 1999) to oscillatory activity in lower frequency bands (below 30 Hz; α, β) (Classen et al., 1998) in the surface electroencephalogram (EEG). It has been shown in laboratory animals (Bressler et al., 1993; Singer and Gray, 1995; Gail et al., 2004) and in humans (Miltner et al., 1999; Classen et al., 1998; Rodriguez et al., 1999) that the presence of interregional coherent oscillations can represent certain aspects of behavior, e.g. expectation of stimuli after associative learning (Miltner et al., 1999), perception of faces (Rodriguez et al., 1999), binocular rivalry (Gail et al., 2004) or coordination of motor reactions in reaction time experiments (Bressler et al., 1993). In neurological and psychiatric patients, it has been speculated that disturbances of long-range ‘binding’ might be a pathogenetic factor in so-called ‘disconnection syndromes’, e.g. associative object agnosia (Carlesimo et al., 1998), limb apraxia (Tanaka et al., 1996; Marangolo et al., 1998; Leiguarda and Marsden, 2000), dyslexia (Paulesu et al., 1996), alien hand syndrome (Geschwind et al., 1995) or schizophrenia (Merrin et al., 1989; Koenig et al., 2001).
Nevertheless, the relevance of the concept of ‘binding’ in terms of synchrony in large-scale networks has been challenged (Farid and Adelson, 2001; for review, see Ghose and Maunsell, 1999; Shadlen and Movshon, 1999). Is coherent activity truly instrumental in generating successful behavior or is it an epiphenomenon of any type of anatomical connection without functional significance?
In the time domain, Gevins et al. (1987) (Gevins AS et al., 1987) showed that interregional covariance of event-related potentials (ERP) in a Go/No-Go task was positively correlated with the occurrence of correct responses. With respect to the concept of frequency coding, comparable results are not available, and spectral data cannot be predicted from time-averaged data as the two aspects of neuronal activity represent different elements of information processing (Pfurtscheller and Lopes da Silva, 1999; Gruber et al., 2004; Muller and Keil, 2004). If long-range synchrony of oscillatory neuronal activity is functionally relevant, then, we hypothesized, there should be a positive correlation of interregional coherence between the areas involved and the behavioral output.
We addressed the question of whether the magnitude of interregional coherence of oscillatory neuronal activity could be linked to the success of complex integrative behavior in a visuotactile matching task. The paradigm consisted of presentation of sets of three Braille letters as typical embossed dot patterns (tactile domain) and, simultaneously, as sets of point patterns on a computer screen (visual domain). Volunteers were instructed to compare these tactile and visual signals in order to decide if the information presented in the two sensory domains was a match or not (‘same’ or ‘different’; Match). To solve the matching task correctly, successful communication between neuronal assemblies representing the tactile stimulus (somatosensory cortex) and neuronal assemblies representing the visual stimulus (occipital cortex) is critical. The experimental setup is illustrated in Figure 1. As a control condition, subjects swept their right index finger (as in Match) across a random dot pattern, whose texture contained no tactile information (Control). In this task, there was similar tactile and visual input as in Match, but no communication between somatosensory and visual areas was necessary.
Materials and Methods
We studied eight right-handed, healthy subjects (four female, four male; mean age ± SD = 28.5 ± 8.7 years). Subjects gave their written informed consent for the study. The EEG study was approved by the local medical ethics committee.
During the recording session, subjects were seated comfortably in an armchair in front of a video screen in a light-dimmed room with the right arm relaxed. The right hand was positioned palm down on an armrest attached to a custom-made apparatus, which was designed for standardized presentation of tactile stimuli, so that the fingers could be moved freely (Fig. 1). Match and Control were presented in blocks of 10. A total of 12 blocks for both conditions were recorded for each subject. In Match, ‘non-matching’ and ‘matching’ trials were randomly presented; subjects were not forewarned of the type of trial. After the end of the exploration, in the break period between two trials, subjects were asked by the experimenter whether the visual and tactile stimuli were a match or not. Subjects were instructed to answer with the word ‘same’ or the word ‘different’. There was no hand motor response required. Feedback about accuracy of the response was not provided to avoid confounding variables such as frustration or changes in motivation. The period in which the subjects indicated their response was not included in data analysis. Each experimental block contained periods of unconstrained rest (Rest) between the string presentations.
One-hundred and twenty randomly generated combinations of three Braille letters were used as stimulus material. Each of the Braille letters consisted of four dots, every Braille letter combination consisted of the same amount of dots.
Continuous EEG was recorded from 28 (Ag/AgCl) surface electrodes (Fig. 1). Impedance was kept below 5 kOhm. Data were sampled at 250 Hz, upper cutoff was 50 Hz, and the time constant was set to DC (Synamp amplifiers and software by NeuroScan® Inc.). Linked earlobe electrodes served as reference. Two bipolar EMG channels were recorded from the right first and second dorsal interosseus (FDI and SDI) muscles. The high-pass filter for EMG was set to 30 Hz. Horizontal and vertical EOGs were registered to detect eye movement artifacts. EMG-onset was marked offline and was used for stimulus-locked averaging.
For statistical analysis, Match was divided in consecutive blocks of trials with excellent (Good) and poor performance (Bad) on an individual subject basis. Excellent performance was defined individually for each subject as correctness ≥ individual median performance; poor performance was defined as correctness < median performance. The main contrast for further analysis was the comparison of Good and Bad.
The EEG data were analyzed with four different types of spectral power analysis: (i) task-related coherence (TRCoh); (ii) task-related power (TRPow); (iii) event-related coherence (ERCoh); and (iv) event-related power (ERPow).
For analysis of TRCoh, EEG signals were digitally filtered off-line (1–50 Hz, slope 24 dB/octave) and, for each experimental condition separately, segmented into non-overlapping epochs of 1024 ms. The time window for the task epoch was determined from 700 to 1724 ms after the FDI EMG peak chosen on the basis of EEG time-average data obtained in a similar experiment indicating that a significant difference of activation occurs only later in the course of the exploration movement, i.e. with a maximum between 700 and 1724 ms after EMG onset (Koyama et al., 1997). The time window for Rest was chosen from 3000 ms to 1976 ms before visual stimuli onset. Single sweeps were visually inspected and trials with eye movement artifacts were rejected. On average, n = 90 ± 32 artifact-free trials were obtained per subject in Match, n = 56 ± 10 in Good, n = 40 ± 19 in Bad and n = 82 ± 21 Control. The slight, but not significant difference [one-way analysis of variance (ANOVA), Good versus Bad, n.s.] in the average number of trials between Good and Bad is related to a different number of trials, which had to be discarded from the analysis due to artifacts. Each sweep was Hamming-windowed to minimize spectral leakage. For spectral power analysis, a discrete Fourier transform was computed. Coherence values were calculated for each frequency bin λ according to equation (1) (implemented in commercial software by NeuroScan® Inc.).
For TRPow analysis, EEG signals were filtered, segmented, baseline-corrected, inspected for artifacts, Hamming-windowed, and Fourier-transformed as described for TRCoh. The power spectrum from 1 to 50 Hz was calculated for each single epoch and then averaged across epochs. TRPow was expressed as the percentage of spectral power during activation (Powactivation) compared with the spectral power during the rest condition (Powrest). The method has been discussed in detail previously (Pfurtscheller and Klimesch, 1991).
Since the main effect in TRCoh was in the α-range, the ERCoh analysis of the EEG data was performed in the frequency range of 7 to 13 Hz. In addition, we also analyzed the evolution of ERCoh over time in the θ- (4–7 Hz), β- (14–30 Hz) and γ-bands (30–50 Hz) to determine possible differences between the main conditions in adjacent frequencies. Preprocessing (e.g. artifact rejection) and calculating coherence were performed as for the task-related coherence (TRCoh) data; filter roll-off was 24dB/oct. In ERCoh, ‘power’ was computed using a complex demodulation. Filtering and complex demodulation (as described previously by Otnes and Enochson, 1978; Thatcher et al., 1994) occur as part of the same operation, as implemented in commercial software (Scan, Version 4.3). ERCoh was computed from epoched EEG data using the coherence formulas given in the TRCoh section. However, the frequency of interest was preselected (α, θ-, β- and γ-frequency), and the results are a function of time with respect to the event at time zero (EMG onset). Coherence values were calculated for each time point and analyzed for the POIs in the time window from 3000 ms pre- to 2000 ms post-EMG-onset. For better visualization data were smoothed. The first and the last 100 ms were not included in the ERCoh analysis because of trimming. Although the evolution of coherence over time (ERCoh) was calculated, which might imply that ERCoh is an instantaneous value, it is of note that ERCoh cannot be instantaneous, because coherence analysis is always based on defined epochs with a certain time-window (ms range), even for event-related analysis.
ERPow analysis was performed in the frequency range of 7 to 13 Hz in order to test whether differences in regional activation could be a confounding factor for the α-coherence data in the target conditions. Preprocessing (e.g. artifact rejection) and calculating power were performed as for the task-related power (TRPow) data. ERPow was computed from epoched EEG data. In ERPow, ‘power’ was computed using a complex demodulation. Filtering and complex demodulation occur as part of the same operation, as implemented in commercial software (Scan, Version 4.3). The results are a function of time with respect to the event at times zero (EMG onset). ERPow values were calculated for each time point and analyzed for the two regions of interest, the left central region (FC3, C3 and CP3) and the occipital cortex (O1, O2) separately in the time window from 3000 ms pre- to 2000 ms post-EMG-onset.
For analysis of muscle activity, bipolar EMG from right FDI was digitally filtered off-line (30–50 Hz, slope 12 dB/octave) and rectified for each experimental condition separately. The results are a function of time (−3000 to 2000 ms) with respect to the event at time zero corresponding to EMG onset.
On the basis of prior anatomical and physiological knowledge (Homan et al., 1987; Classen et al., 1998), a set of electrode pairs of interest (POI) was defined, left somatosensory cortex paired with visual cortex, represented by electrode pairs FC3–O1/2, C3–O1/2, CP3–O1/2 (Fig. 2, upper left). To analyze the coherence data statistically a factorial ANOVA design with contrast analysis and a correlation analysis (Pearson's R, two-tailed; with ‘performance’ (success rate in %) versus ‘centro-occipital α-coherence magnitude’ were employed. As the main factor of the ANOVA design, we defined Condition (good, bad, control). The coherence data were statistically analyzed in the θ- (4–7 Hz), α- (7–13 Hz), β- (13–30 Hz) and γ-band (30–50 Hz). An ANOVARM design with repeated measures FREQ (θ, α, β, γ) and Condition (good, bad) was used to further determine differential effects between frequencies. To stabilize the variance of the coherence data, an inverse hyperbolic tangent (arctanh) transformation was used. Another factorial ANOVA design with contrast analysis was employed for statistical analysis of the TRPow data. As factors, we defined Condition (good, bad) and Region (occipital, left central). To stabilize the variance of the spectral power data, a log transformation was used. To account for interindividual variability, the factor Subject was included in both ANOVAs. Differences were considered significant, if P < 0.05. Multiple comparisons on the same data pool were Bonferroni-corrected.
EEG Coherence Data
As a measure of ‘binding’-like interregional coupling, coherence was computed between channels overlying the occipital cortex and the lateral central region including the primary sensorimotor cortex (Fig. 2A, upper left). The experimental data of each individual subject during Match were divided into blocks of trials with excellent (Good) and poor (Bad) performance. During Good, the proportion of correct responses was 92.5 ± 5.5% and during Bad 67.3 ± 13.9% (Wilcoxon matched pairs test, P < 0.05). All trials pooled, during Match subjects were able to solve the matching task in 78.9 ± 16.1% (mean ± SD). Task performance was stable over the whole experiment.
Good performance was associated with higher centro-occipital α-coherence (7–13 Hz) than poor performance (Good versus Bad; ANOVA, P < 0.05, corrected for multiple comparisons). Furthermore, there was a significant positive correlation between degree of performance and magnitude of centro-occipital α-coherence in the group data analysis (Pearson's r = 0.55; P < 0.05). As expected on the basis of previous studies (Classen et al., 1998), in Match (Good and Bad pooled), centro-occipital coherence was higher than in Control (Match versus Control; ANOVA, P < 0.05, corrected). The quantitative coherence results are illustrated in Figure 2.
To exclude the possibility that the differential findings in the magnitude of centro-occipital α-coherence between Good and Bad were a general and topographically non-specific effect, we performed additional statistical analyses. Six electrode pairs were randomly chosen and a factorial ANOVA with the factor Condition (good, bad) was performed. This analysis showed small α-coherence magnitudes for both conditions as expected with no significant difference between Good and Bad.
The success-related increase of coherence was topographically specific and maximal for the functional links between occipital cortex bilaterally (electrodes O2, O1) and left central regions (electrodes C3, CP3). This can be appreciated in a differential coherence plot, in which α-coherence during Bad was subtracted from α-coherence during Good (Fig. 3). The topographic specificity gains further support from the fact that α-coherence between the right somatosensory and the occipital cortex was not different between Good and Bad as found between left somatosensory and occipital cortex.
In the θ- (4–7 Hz), β- (14–30 Hz) and γ-bands (30–50 Hz), interregional task-related coherence between occipital and central regions did not change in relation to performance (Good versus Bad, ANOVA, n.s.; Fig. 4). An ANOVARM revealed a significant interaction of performance (Good versus Bad) and frequency band (θ, α, β, γ), further underlining the finding that the differences between Good and Bad were predominantly in the α-band. ERCoh demonstrated a clear temporal evolution of α-coherence during Match. A slow gradient before EMG onset was followed by a steep increase of ERCoh after EMG onset. The maximum level was reached at ∼800 ms into the discrimination process and ERCoh slowly decreased afterward. The coherence time course unequivocally differentiated between Good and Bad. Beginning at the same coherence level right after EMG-onset (∼150 ms), ERCoh during Bad increased only mildly, reached its maximum level at ∼550 ms, then showed an early decline and reached baseline level at ∼1500 ms after EMG-onset. In contrast, during Good ERCoh increased with a steeper slope and a larger maximum magnitude occurring at ∼835 ms into the discrimination process, remained at a high level until ∼1300 ms, and slowly declined afterward. Of note, the coherence changes that differentiated between Good and Bad appeared after EMG-onset in the early part of the discrimination process. During Control, time course analysis revealed stable ERCoh values near baseline with no substantial increases after EMG-onset. The ERCoh data are illustrated in Figure 2.
EEG Power Data
Regional activation (task-related power decreases) was no different over the left lateral central region and the occipital cortex for Good and Bad (ANOVA, interaction Region × Condition, n.s.). TRPow and ERPow analyses revealed no difference for the amplitude and temporal evolution of occipital and left lateral central α-power in the Good and Bad conditions (Fig. 5). This indicates that neuronal long-range synchronization during the task was achieved by modulating the temporal structure of the oscillations rather than by changing the regional (local) power levels in the 7–13 Hz frequency range.
Rectified EMG before and after movement onset was no different between the three conditions (ANOVA, main effect Condition, n.s.). In particular, the time-course of EMG activity suggests that there was no relevant difference in motor behavior between the target conditions Good and Bad.
In a visuotactile integration task requiring effective communication between visual and somatosensory areas, the present data show highest levels of synchrony between these areas in trials with excellent task performance. Even more noteworthy, the efficacy of task performance was directly correlated with the magnitude of synchrony. Thus, significantly lower levels of synchrony were predictive of trials with poorer performance. To our knowledge, this is the first demonstration of a quantitative link between interregional synchrony, measured in the frequency domain as EEG coherence, and the efficacy of behavior in humans. These data support the concept that synchronizing neuronal events across large distances is instrumental for information coding in the human brain.
The biological significance of synchronous activity recorded from scalp EEG has been challenged because of the assumption of artificial synchronization induced by volume conduction (Menon et al., 1996). Volume conduction induces a pattern of synchronization that decays rapidly as the distance between electrodes increases over 2 cm and volume conduction does not covary with experimental variables. We found synchronized activity over brain areas far away from each other (somatosensory and occipital cortex); in addition, we applied an analytical approach that is based on relative coherence changes rather than absolute values; therefore, effects of volume conduction or influences of the reference (Nunez PL et al., 1997) cannot explain the data. Further arguments against a spurious origin of the observed coherence patterns are (i) topographic specificity; (ii) temporal evolution relative to EMG onset; and (iii) differential coherence magnitudes depending on the level of task performance. Of course, the first point (i) has to be viewed within the limits of spatial resolution of the EEG, which confines our conclusions to the level of regions. Therefore, we refer mostly to the left central region including the primary somatosensory and motor cortices and to the occipital cortex. Within these limits, the success-related increase of long-range coherence (Good > Bad) was restricted to regions known to be major components involved in visuotactile matching tasks. The major network nodes were occipital cortex bilaterally and left lateral central region, covering visual and somatosensory areas. Accordingly, the maximum of the difference between task-related α-coherences (Good minus Bad) occurred in connections from these occipital to left lateral central areas (C3, CP3). The specificity of the coherence findings is further supported by the fact that the differential effect between Good and Bad was absent in (i) task-related α-coherences in randomly chosen pairs of interest and (ii) in POIs representing connections between the right central region (ipsilateral to the exploring hand) and occipital cortex. In the Control task, there was no coherent activity between occipital and lateral central regions detectable.
Information Coding in Different Frequency Bands
The performance-related modulation of somatosensory-occipital coherence was only present in the α-band (7–13 Hz). α-Band power is known to be particularly sensitive to modulations in perceptual tasks and sensory processing (Pfurtscheller and Klimesch, 1991; Salmelin and Hari, 1994; Schurmann et al., 1997), whereas β-band oscillations are probably more sensitive to modulation of motor parameters (Conway et al., 1995; Pfurtscheller and Klimesch, 1991; Salmelin and Hari, 1994), θ-band oscillations to (working) memory and preparational processes (Gevins and Smith, 2000; Gevins et al., 1987; Klimesch, 1999), and γ-band oscillations to higher cognitive functions (Rodriguez et al., 1999). For excellent task performance in the present paradigm, the ongoing sensory (visuotactile) integration was the critical component. This is supported by the fact that the differential effects between Good and Bad appear after EMG-onset, i.e. during the process of sensory discrimination. Further substantiating the validity of these observations, the ERCoh time course showed no comparable increase of coherence after EMG-onset in the Control condition. Demands to the motor system and motor output (as indicated by the temporal evolution of the EMG) were similar in Good and Bad trials and cannot account for the observed differences in coherence. In this regard, the present data agree with the previous view of the differential functional role of α- and β-frequencies (Andres and Gerloff, 1999; Lopes da Silva, 1991). The differential effect in the α- versus the θ-, β- and γ-band lends further support to our interpretation that the greater coherence during Good was not related to a global (broad-band) phenomenon. However, this does not mean that information coding is restricted to or particularly effective in the α-band. The detailed relations between modulations of neuronal oscillations in different frequency bands (θ, α, β, γ) and their functional impact on information coding need to be determined in future studies.
It is known that, depending on the experimental setup, broad-band changes or coherence modulations restricted to certain frequency bands can be observed. For example, Bressler et al. (1993) studied large-scale synchronization in macaque monkeys during visuomotor tasks and demonstrated increased broad-band coherence among local field potentials. With conditioning tasks in dogs, Dumenko (1970) demonstrated coherence of macropotentials between visual and motor cortices specifically in the θ-frequency band (4–7 Hz). Earlier reports of coherent neuronal activity in the γ-range (Miltner et al., 1999; Rodriguez et al., 1999) have now been complemented by studies focusing on lower frequency ranges. In tasks with high motor demands in humans (Andres et al., 1999; Classen et al., 1998), task-related modulations of coherence were largest in (but not restricted to) the β-band (13–21 Hz). During object recognition (Mima et al., 2001) and in a working memory task (Sarnthein et al., 1998), coherence occurred in the α- (8–12 Hz) and θ-bands (4–7 Hz). In cats, von Stein et al. (2000) showed synchronization in the θ-, α-, β- and γ-frequency ranges depending on top-down or feed-forward processing during perception of stimuli. Recent findings in monkeys demonstrating low-to-medium frequency LFP coherence (<30 Hz) associated with top-down processing further substantiate this view (Gail et al., 2004). Finally, there is evidence for an inverse correlation between the spatial scale of information integration and the frequency of interaction (Schanze and Eckhorn, 1997; von Stein et al., 2000). For example, local interactions during visual processing frequently involve γ-frequency changes (Singer W and CM Gray, 1995), while very long-range interactions as, e.g. during working memory retention, tend to involve low-frequency interactions in the θ- or α-band (Sarnthein J et al., 1998). Taken together, the emerging picture is that a complex interaction of task, topographic extent of regions and functional systems engaged, and cognitive processing determines the modulated frequency range of oscillatory activity. The particular role of the α-band in our experiment might relate to the combination of a visuotactile task (as the α-band is particularly sensitive to modulations in perceptual tasks and sensory processing) and the requirement of long-range cortical interactions (somatosensory, visual cortex) to solve this task.
Influence of Attention and Motor Output
Could different levels of global α-desynchronization (α-block) during the Good and Bad trials, e.g. due to different levels of global attention, confound the present coherence findings? There are several facts that argue against this. First and most importantly, quantitative spectral analysis of α-desynchronization (TRPD) and its time course during the visuotactile matching task did not reveal significant differences between the two target conditions (Good, Bad). Further, the subjects did not know that the results were stratified into Good and Bad responses. They were instructed to perform each matching trial as accurately as possible (i.e. identical setting and instructions for Good and Bad) and did not receive any feedback about the quality of their responses. Finally, the matching and non-matching trials were randomly presented. Therefore, taking all argumentative points and the additional analyses together, changes in α-power (α-block) cannot explain the differences in the magnitude of task-related α-coherence between Good and Bad. Even though more trials were discarded during Bad compared with Good due to artifacts, the difference was not significant and therefore cannot explain the present findings.
Behavioral Relevance of Interregional Synchrony
The present data provide evidence for the behavioral relevance of low-frequency oscillations (α-band) in human large-scale cortical networks. They support the concept that slowly oscillating cell assemblies comprise large numbers of neurons (Roelfsema et al., 1997; Kopell et al., 2000) and constitute spatially extended and functionally connected neuronal representations (von Stein et al., 2000). Our findings relate to the general question about different modes of information coding in the brain. There is an open controversy about whether information coding is implemented in neuronal structures by synchrony and asynchrony of cell firing rather than by localized changes of mean spike firing rates (Shadlen and Movshon, 1999). The present data argue in favor of synchrony (‘binding’) as a behaviorally relevant means of information processing in large-scale networks. Based on time-domain data, Gevins et al. (1987, 1989a,b) have arrived at a similar conclusion. They found that interregional covariance of event-related potentials in the preparational phase of a bimanual Go/No-Go task predicted whether subjects responded correctly to the Go or No-Go signal. EEG coherence addresses a different aspect of information processing, i.e. frequency coding. Spectral data cannot be predicted from time-averaged data and it is known that evoked potentials and event-related power changes reflect different aspects of neuronal activity (Gruber et al., 2004; Muller and Keil, 2004; Pfurtscheller and Lopes da Silva, 1999). Therefore, our data in the frequency domain represent a different aspect of information coding and might help to bring forward our understanding of synchrony as a mechanism underlying the neuronal code.
At the level of cell assemblies, Fries et al. (2001) studied neuronal activity in V4 of macaque monkeys. During presentation of attended visual stimuli in the receptive field, α-coherence was increased compared with presentation of unattended stimuli. Fries and coworkers clearly demonstrated that these changes in coherence occurred in the absence of changes in local mean firing rates. Similar differential results were found in a visual rivalry task in strabismic cats (Fries et al., 1997). This corresponds well with the present data in which coherence changes occurred depending on the quality of task performance (Good > Bad), but local spectral power values (sensorimotor, visual cortex) were no different for Good and Bad performance.
The present data do not allow for further inferences regarding the neural mechanisms underlying the observed patterns of interregional synchrony. Coherence as a self-emerging property of intracortical pathways is an attractive concept. However, from simultaneous intracerebral thalamic and cortical recordings in cats (Contreras and Steriade, 1995; Steriade, 1997; Steriade et al., 1996) and humans (Slotnick et al., 2002), it has become clear that low-frequency and fast cortical rhythms are subject to influences of thalamo-cortical connections. It is likely that both cortico-cortical and cortico-subcortical mechanisms are important. Our main conclusion (that higher functional coupling predicts better performance) is independent of this matter.
In summary, in the present study a direct link was established between the magnitude of task-specific coherent oscillatory EEG activity and the degree of behavioral success. Coherence was significantly larger in trials with excellent performance and there was significant positive correlation between coherence and performance. This finding fills an important gap in our understanding of the role of interregional synchrony and its relevance as a mechanism for implementation of successful human complex behavior.
The authors wish to thank Rolf Kirsammer and Rüdiger Berndt for technical assistance, Guido Nolde and Zoltan Mari for help in data processing, and Devera Schoenberg for skillful editing. This research was supported by a grant from the Humboldt Foundation (Feodor-Lynen) to F.H. and by a grant to C.G. (SFB 550/C5) from the Deutsche Forschungsgemeinschaft (DFG).
1Cortical Physiology Research Group, Department of Neurology and Hertie Institute for Clinical Brain Research, Eberhard-Karls University Tübingen, 72076 Tübingen, Germany and 2Human Cortical Physiology Section, Medical Neurology Branch, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD 20892, USA