Rolandic alpha and beta EEG rhythms' strengths are inversely related to fMRI-BOLD signal in primary somatosensory and motor cortex

Subjects

Fifteen healthy subjects (mean age 28 years, range 25-30, 4 males/11 females) with no history of neurological or psychiatric disorder participated in the study. All subjects gave written informed consent prior to the investigation. The experiments were performed in compliance with the relevant laws and institutional guidelines and were approved by the local ethics committee.

Experimental Task

To modulate Rolandic rhythms, subjects performed a motor task. This task consisted of self-paced effortless synchronous opening and closing of the hands at a frequency of about 1 Hz (trained before the experiment) for 20 s alternating with rest periods of equal duration. In previous studies, such grip tasks have been shown to induce robust activation in fMRI [Jancke et al.,1998]. The task blocks were acoustically cued by “start” and “stop” commands presented via MR compatible headsets using the software Presentation v0.71 (Neurobehavioral Systems, Albany, NY). As Rolandic rhythms are known to be enhanced by complex visual input [Koshino and Niedermeyer,1975], the subjects were instructed to keep their eyes open during the entire experiment to maintain a low level of posterior alpha rhythm strength and to focus on a fixation cross in the middle of the screen that was embedded in a colored picture (N = 9) or in a cartoon (N = 6) during the experiment. All subjects were asked neither to move their eyes beyond normal eye-blinking nor their bodies beyond the motor task.

Simultaneous EEG-fMRI Recordings

Simultaneous EEG-fMRI recordings were performed with a 1.5-T scanner (Magnetom Vision, Siemens, Erlangen, Germany) and a 32-channel MR-compatible EEG (Brain Products, Munich, Germany). Functional imaging was performed using a T2*-weighted BOLD [Bandettini et al.,1992; Frahm et al.,1992; Kwong et al.,1992; Ogawa et al.,1992] sensitive gradient echo planar imaging sequence (repetition time TR = 2.5 s, acquisition time TA = 2.1 s, echo time TE = 60 ms, flip angle 90°, matrix 64 × 64, voxel size 3.3 × 3.3 × 6 mm³, interslice distance 1.5 mm, 20 slices, 200 scans). For EEG recordings, a customized 32-channel EEG cap was used including six electromyography (EMG) electrodes for bipolar recordings from chin and bilateral hand muscles, three electrooculogram (EOG) electrodes for vertical and horizontal EOG, and two electrocardiogram (ECG) electrodes (Easy cap, Falk Minow Services, Herrsching-Breitbrunn, Germany). The resolution of the MR-compatible 16-bit EEG amplifier was set to 0.5 μV giving a large dynamic range of ±16.384 mV to capture both low-amplitude EEG/EMG signals in the range of 50 μV and large MR-artifacts in the range of 5,000 μV. The sampling frequency was 5,000 Hz. The amplifier was equipped with an analog 1,000 Hz low-pass/0.1 Hz high-pass hardware filter. The EEG was recorded from 21 electrode positions (International 10-20 system) referenced to a point at the vertex at position FCz. This reference was selected as others seemed less appropriate for the following reasons: “Neutral” nose reference in the fMR setup proved to be unreliable since the nose-electrode happened to become loose, possibly due to vibrations during MR acquisition. “Mastoid” reference seemed unsuitable due to the likely occurrence of laterality effects. The average reference proved to be more artifact afflicted than a single-electrode reference, possibly due to residual artifacts in single channels included in the average reference. After acquisition, EEG data were MR-artifact corrected using artifact-template subtraction according to the approach proposed by Allen et al. [2000] using BrainAnalyzer Software (V. 1.03 BrainProducts, Munich, Germany). This approach has been shown to be capable of recovering physiological signal content from artifact-afflicted EEG periods [Becker et al.,2005] and to restore EEG rhythms in the spectral alpha and beta range with sufficient quality [Ritter et al., 2007]. Subsequently, a 70 Hz low-pass filter was applied.

Analysis of EEG Data

Identification of Rolandic alpha and beta rhythms (see Fig. 1)

Three features were used to identify Rolandic rhythms: (i) frequency, (ii) suppression during motor task, and (iii) location:

• (i)

  Rolandic alpha (“mu”) and beta rhythms have peak frequencies of around 10 and 20 Hz respectively [Tiihonen et al.,1989].

• (ii)

  During finger movements, Rolandic rhythms are known to be suppressed [Arroyo et al.,1993; Jasper and Penfield,1949]. To identify Rolandic alpha and beta rhythms, we performed frequency analysis of the EEG recordings and searched for components that were suppressed during hand movements.

• (iii)

  Both Rolandic alpha and beta rhythms occur most prominently in pericentral regions.

Validation of blind source separation

Before applying the TDSep algorithm to experimental EEG data, the algorithm was tested on surrogate data. Artificial EEG data were created as follows (see Fig. 2): 21 data channels with individual random noise plus “posterior alpha rhythm” consisting of a 10-Hz sinus wave with highest strength in posterior channels and “bilateral Rolandic rhythms”, consisting of a 11-Hz and a 19-Hz sinus wave and being modulated in strength every 20 s. Both posterior alpha and Rolandic rhythms gradually declined in channels adjacent to the site of maximum strength. As shown in Figure 2, by means of TDSep, Rolandic rhythms were reliably identified. The temporal and spatial characteristics of these rhythms were more accurately reconstructed than by any single EEG electrode signal, demonstrating the superiority of this method.

This superiority was confirmed on the experimental data. Rolandic alpha and beta power in sources identified by TDSep mirrored the motor task more closely than the power of both rhythms measured by the C3 or C4 EEG channels (see Results). This indicates higher functional specificity of signals from TDSep sources. Furthermore, regressors obtained from TDSep sources accounted for more BOLD signal variance than regressors obtained from EEG channels (see Results).

Analysis of fMRI Data

SPM2 was used for fMRI data analysis. Functional MR images were realigned, normalized to the MNI standard brain, smoothed with double voxel size, and high-pass filtered (512 s). Serial correlations were estimated using an autoregressive model to subsequently correct for non-sphericity during inference [Friston et al.,1995]. Because of the covariance between the two rhythms' power and the motor task, a fraction of the BOLD signal variance cannot be ascribed unambiguously to one particular regressor [Andrade et al.,1999]. In such a situation, a variety of model configurations can be proposed, each addressing distinct hypotheses and hence allowing for different conclusions. There is no single “perfect” model to unambiguously explain all effects. We employed three different modeling strategies each addressing different questions. In our opinion each of the approaches has its validity, as we will explicate below. Figure 3 schematically illustrates differences between the three approaches based on a hypothetical data set.

In the first approach (a), a separate model was estimated for each regressor. In this case, the regressors' weights reflect correlation of a single, distinct effect with the BOLD signal. However, because there are redundancies between the regressors, the identified BOLD signal changes could also be caused by a correlating covariate. With this approach, however, we were interested in (1) exploratively describing differences resulting from different definitions of Rolandic rhythms, i.e. rhythms directly obtained from electrodes, rhythms obtained from TDSep sources, rhythms at individual peak-frequencies and rhythms at fix spectral bands, and (2) the differences of the results for Rolandic alpha versus beta rhythms as assessed in a second level analysis. A correlating covariate may diminish the estimated differences of any true underlying effect, but not create spurious differences (hence avoiding type I error). Subsequent inferences regarding the differences of the results for Rolandic alpha and beta rhythms were exclusively based on significant differences between the fMRI correlates of the respective regressors. A disadvantage of this analysis strategy is that the model accounts only for the BOLD signal variance explained by a single effect. The BOLD signal variance due to other effects not within the model leads to residual error. An advantage, however, is that no bias from other regressors (see below) is introduced into the model, i.e. the “pure” and physiological rhythm strength, as measured by EEG (Fig. 3a,b), is used as a predictor for the BOLD signal.

In the second approach (b), both Rolandic alpha and beta regressors were integrated in a single model. An advantage of this approach over approach (a) is that the individual weights reflect BOLD signal changes that are specific for the respective regressor. Effects caused by the redundant parts of the regressors (thin red lines in Fig. 3a,b), however, will be missed. In approach (b) the motor-task is not included into the model due to the imprecision of its timing. The motor-task is a block-function defined by the start and stop markers recorded when acoustic commands are presented (Fig. 3c, bold red line). Thus, the “real” timing of motor action can slightly deviate from the task function (Fig. 3c, thin red line), which would lead to additional error variance where the motor-task function to be integrated into the model. Given that the Rolandic rhythms are weak and, therefore, the BOLD signal changes will be small, we sought to avoid such confounding effects in approach (b).

In the third approach (c), redundancies between the motor-task, Rolandic alpha and beta regressors were removed by a Gram-Schmidt orthogonalization procedure (SPM2 MATLAB code), which computes the projection of a unit vector x onto another vector y and then subtracts the projection from x, such that what remains is orthogonal to y. For two orthogonal vectors, the sum of element-wise multiplications of both vectors equals zero. Subsequently, one model was estimated including all orthogonalized regressors. As with approach (b), individual weights in approach (c) reflect BOLD signal changes that are specific for the respective regressor. However, in approach (c) effects caused by the redundant parts of the regressors will be attributed to the hierarchically higher regressor (in terms of the orthogonalization order). Since we were interested in specific effects of the two Rolandic rhythms, we first removed redundant variance caused by the motor-task, i.e. the rhythms were orthogonalized to the motor task. The order of the orthogonalization of the regressors of the Rolandic alpha and beta rhythm is slightly arbitrary and cannot be justified in terms of functionality. Therefore we estimated models for both orders and chose the one with the smallest residual error. An advantage of approach (c) is that including all three regressors in a single model would be expected to minimize residual BOLD signal variance. As mentioned earlier, however, the motor-task function may not perfectly reflect the real motor-action. In Figure 3d we show how orthogonalization based on a slightly imprecise motor-task function may artificially change regressors representing Rolandic rhythms. Figure 3e–h illustrates how the orthogonalization order slightly modifies regressors for Rolandic alpha and beta rhythms.

Group effects were computed for all models by a random effects analysis (one-sample Student's t-test), which allows for assignment of the observed effects to the general population. Contrasts were calculated for single effects as well as for the differences between effects of Rolandic alpha and beta rhythms for analysis strategy (a) and (b).

Depending on our hypotheses two types of statistical evaluations were performed. In order to test our anatomically specified hypotheses “right versus left pericentral cortex” and “precentral versus postcentral cortex” statistical maps were small volume corrected (SVC). For anatomically open hypotheses whole-brain false discovery rate (FDR) correction was performed on voxel and cluster levels. Coordinates of activation sites are given in MNI (Montreal Neurological Institute) space.

Rolandic Rhythms in the EEG Data

Two representative examples of the EEG before and after gradient artifact correction are shown in Figure 4. The correlation between Rolandic alpha as well as beta rhythms with the motor task was much stronger for rhythms identified by blind source separation as compared to Rolandic rhythms extracted from single electrode signals (see Fig. 5) in each individual subject as well as for the entire group. Furthermore, Figure 5 illustrates that the Rolandic beta rhythm is stronger modulated by the motor task than the Rolandic alpha rhythm.

Rolandic rhythms derived from selected electrodes correlate better with the motor task at individual peak frequencies (except for the β-2 band), whereas Rolandic rhythms obtained from blind source separation (TDSep) correlate best in the beta range when defined by spectral bands (see Fig. 5).

Modulation of Rolandic rhythms by the motor task was more pronounced over the right hemisphere than over the left hemisphere, indicated by higher mean t-scores as well as higher mean TDSep weights at right versus left positions C4/C3: α-2 band: 0.83 (left), 2.62 (right); β-2 band: 1.36 (left), 1.85 (right); paired Student's t-test: P = 0.0004/P = 0.26 (see Fig. 6).

FMRI Correlates of “Rolandic Alpha Rhythm” and “Rolandic Beta Rhythm”

Subjects performed a bimanual motor task in a block design to modulate Rolandic rhythms. In all 15 subjects, a clear modulation of the 10- and 20-Hz Rolandic rhythms “during bilateral hand movement versus rest” was detected. As expected, both rhythms were suppressed during hand movement and reappeared during rest. This approach, however, implies that the three parameters “motor task”, “Rolandic alpha rhythm”, and “Rolandic beta rhythm” are highly intercorrelated. Specifically, the power of the two pericentral rhythms is most likely inversely correlated to the motor task. Because of this presumable inverse covariance between the rhythms' power and the motor paradigm, the BOLD signal variance, which correlates with the motor task, is also inversely correlated with Rolandic rhythms. As a consequence, voxels that are “activated” during hand movement will tend to be “deactivated” during periods of high “rhythm activity”. We addressed this issue by using the three different approaches as described above. The first approach implied calculation of one separate model for each regressor, thus testing whether each regressor correlates with the BOLD signal irrespective of the others. In this type of analysis, differences between the respective functional maps can indicate characteristics of the respective regressor. For example, in addition to movement-induced modulations of Rolandic rhythms, there are also spontaneous fluctuations. These fluctuations differ slightly between the hemispheres [Niedermeyer and Koshino,1975]. While it is not entirely clear whether such fluctuations represent the same phenomena as the movement-correlated events, they offer a continuous “additional signal”, which goes beyond the mere on-off dichotomy of the bilateral movement task. The EEG recordings at C3 versus C4 positions, respectively, may be able to pick up such lateralized fluctuations of Rolandic rhythms.

fMRI Correlates of Left Versus Right (C3/C4) Rolandic Beta Rhythms During a Motor Task

Whether BOLD correlates are specific for left versus right Rolandic rhythms was tested for C3/C4 Rolandic beta rhythms. TDSep sources were not used for this question since they are spatially distributed over both hemispheres. Rolandic beta rhythm was chosen since it was most intensely modulated by the motor task (see Fig. 5) and since it was clearly identified over both hemispheres (see Fig. 6). Statistical t-maps of fMRI correlates of Rolandic beta rhythm at left central electrode position C3 are depicted in Figure 7. Negative correlations between Rolandic beta rhythms and the BOLD signal in the pericentral cortex were maximal in the left hemisphere (MNI x, y, z: −53, −26, 55; P = 0.0001 voxel level; P = 0.001 SVC voxel level FDR-corrected) for Rolandic rhythms derived from the left central electrode position C3 and in the right hemisphere (MNI x, y, z: 39, −23, 66; P = 0.00003 voxel level; P = 0.002 SVC voxel level FDR-corrected) for Rolandic rhythms derived from the right central electrode position C4. For a better illustration, a lower statistical threshold of 0.01, uncorrected for multiple comparisons was chosen in Figure 7

Thus, in general, there is an inverse correlation between the power of the Rolandic beta rhythm and the BOLD signal in cortical bilateral sensorimotor structures; however, there is a clear hemispheric dominance corresponding to the side of the electrode, i.e. the Rolandic beta rhythm obtained from the left hemisphere (“C3 Rolandic beta”) has a much stronger (inverse) correlation with the BOLD signal in the left hemisphere, and vice versa for the right side (see Fig. 7).

Pre- Versus Postcentral Distribution of fMRI Correlates of Rolandic Alpha and Beta Rhythms

There is no single optimal analysis addressing the question of whether there is a differential distribution within pericentral cortex between fMRI correlates of Rolandic alpha and beta rhythms. Therefore, different approaches were chosen as described in the Methods section. Results are listed in Table I and illustrated in Figure 8.

In addition to the results of our anatomically restricted hypothesis, additional loci within the entire brain are reported in Tables II and III.

Since we were interested in topographic differences of the BOLD correlates of both Rolandic rhythms in primary sensorimotor cortex, we looked for maximal (de-) activation sites (the tip of the iceberg) within the right pericentral cortex where rhythms had their strongest correlates in the EEG. Results are similar throughout all analyses. Maximum negative fMRI correlates of the Rolandic alpha rhythm were located posterior to the maximum negative fMRI correlates of the Rolandic beta rhythm throughout all analyses. Representative results yielded by two different methodological approaches are shown in Figure 8.

Figure 8a shows fMRI correlates of TDSep broad-band Rolandic alpha and beta rhythms based on separate models. Negative correlation in the right pericentral region was noted for the TDSep alpha rhythm to be located more posterior (MNI x, y, z: 46, −33, 61; P = 0.00016 voxel level uncorrected, P = 0.004 SVC voxel level FDR-corrected) as compared to the negative correlate of the TDSep beta rhythm (MNI x, y, z: 49, −20, 55; P = 0.0000056 voxel level uncorrected, P = 0.00025 SVC voxel level FDR-corrected. When contrasting alpha and beta effects in a group analysis on the second level (paired t-test), negative BOLD correlates of the TDSep Rolandic beta rhythms are significantly stronger in a precentral area (MNI x, y, z: 49, −16, 55; P = 0.00016 voxel level uncorrected; P = 0.005 SVC voxel level FDR-corrected) as compared to the effects of the TDSep Rolandic alpha rhythm (Fig. 8b).

As described under point “c” in “Analysis of fMRI data”, we also employed a fMRI-data modeling approach in which the three orthogonalized functional regressors “motor task”, “TDSep peak-frequency Rolandic alpha”, and “TDSep peak-frequency Rolandic beta” were included in one model. Two models were calculated with differing orthogonalization order of Rolandic alpha and beta rhythm regressors. Mean residual error was 857 when the Rolandic alpha rhythm regressor preceded the Rolandic beta rhythm regressor. In the converse case mean residual error was 872. Anatomical correlates were exactly the same for both models with slightly different statistical z-values. In the following we report the results of the model with the lower residual error. Results for the “motor task” (which are not the focus of this work) changed only slightly as compared to the separate-models approach given above (Table I). The posterior-anterior topographical distribution of FMRI responses to Rolandic alpha versus beta rhythms persisted, with the response to the Rolandic alpha rhythm, however, shifting backwards to the posterior parietal cortex (Fig. 8c). Pericentral negative fMRI correlates of the orthogonalized TDSep Rolandic alpha rhythm (MNI x, y, z: 46, −59, 55; P = 0.006 voxel level uncorrected; P = 0.237 SVC voxel level FDR-corrected) were located posterior to those of the orthogonalized TDSep Rolandic beta rhythm (MNI x, y, z: 59, −26, 50; P = 0.001 voxel level uncorrected; P = 0.033 SVC voxel level FDR-corrected). For this modeling approach, single subject analyses revealed topographically similar effects, i.e. precentral fMRI correlates versus posteroparietal fMRI correlates, in 14 subjects for the Rolandic alpha rhythm and in 12 subjects for the Rolandic beta rhythm (all P < 0.05, uncorrected).

fMRI correlates of the posterior (classical) alpha rhythm during a motor task

For both the broad-band (8-12 Hz) alpha rhythm of the right occipital electrode position O2 and the TDSep posterior alpha rhythm, we found negative fMRI correlates in the right occipital cortex (Table II; Fig. 9c,d). For the O2 posterior alpha rhythm, additional negative correlates were located in the left frontal and bilateral supramarginal/inferior parietal cortex whereas for the TDSep posterior alpha rhythm additional negative fMRI correlates were found in the left inferior parietal lobule and the bilateral medial frontal gyrus. No significant positive fMRI correlates have been found for the posterior alpha rhythm during a motor task at the given conservative statistical threshold.

Rolandic EEG Rhythms

Rolandic rhythms were first described by Jasper and Andrews [Jasper and Andrews,1936], and since then the nomenclature has varied among researchers, e.g. the term “mu rhythm” has sometimes been applied to the Rolandic alpha rhythm, and sometimes to both rhythms [Tiihonen et al.,1989]. Here, we use the terms “Rolandic alpha rhythm” and “Rolandic beta rhythm” respectively.

Both rhythms are suppressed during tactile stimulation [Cheyne et al.,2003] or voluntary phasic muscle activity [Jasper and Andrews,1936; Jasper and Penfield,1949; Pfurtscheller and Aranibar,1980; Pfurtscheller and Berghold,1989] and are enhanced by visual stimulation [Brechet and Lecasble,1965; Jasper and Andrews,1938; Koshino and Niedermeyer,1975]. However, Rolandic alpha and beta rhythms are not always modulated in parallel and their spatial distribution differs, with the beta rhythm being spatially more widespread distributed than the alpha rhythm [Magnus,1954]. In MEG studies during phasic motor activity [Salmelin and Hari,1994a] and tactile stimulation [Cheyne et al.,2003], the reactive sources of Rolandic beta rhythm were localized in MI and for Rolandic alpha rhythm in SI [Salmelin et al.,1995].

The precise function of these rhythms is not clear. Similar to the posterior alpha rhythm they are frequently referred to as “idle rhythms” indicating a “resting state” of the respective (sensory) system [Pfurtscheller,1992]. Apart from the suppression during somatosensory stimulation or during a motor action, this notion is in agreement with modulations of motor cortex excitability with Rolandic rhythms' strength as shown in transcranial magnetic stimulation (TMS) studies: Motor responses evoked by TMS have been shown to be facilitated during event-related desynchronization (ERD), i.e. during a decrease in Rolandic rhythm strength of up to 200 ms after somatosensory stimulation [Deletis et al.,1992; Deuschl et al.,1991; Hirashima and Yokota,1997; Komori et al.,1992; Maertens de Noordhout et al.,1992; Rossini et al.,1991; Terao et al.,1995] and reduced motor cortex excitability has been shown during event-related synchronization (ERS), i.e. an increased power of Rolandic rhythms following 200-1,000 ms after somatosensory stimulation [Chen et al.,1999].

EEG Rhythms Obtained During MRI Data Acquisition: Technical/Methodological Issues

The relationship of brain electrical activity and perfusion/metabolism has been studied by combined EEG and positron emission tomography (PET) [Leuchter et al.,1999; Oakes et al.,2004]. FMRI has the advantage of having a faster time resolution than PET, but artifacts due to electro-magnetic interference are inherent with this technique. Several studies have now shown that EEG can be recovered from recordings obtained within the MR environment [Bonmassar et al.,2002; Lemieux et al.,2001] and even from segments obtained during MRI data acquisition [Becker et al.,2005; Salek-Haddadi et al.,2003]. Few studies, however, have systematically evaluated the quality of the EEG after artifact removal [Becker et al.,2005; Goncalves et al.,2007; Grouiller et al.,2007; Ritter et al.,2007; Sammer et al.,2005], and the issue of its full validity is still not finally resolved. It clearly would constitute a further argument for the validity of the EEG-fMRI approach if even very subtle EEG features can be recovered that behave in a “meaningful” way, e.g. showing the expected (previously described) reaction to certain stimuli. Previous studies have shown that the posterior alpha rhythm can be obtained from EEG recorded during and/or between MRI data acquisition, and that the signal-to-noise ratio of the EEG is sufficient to use the power of the alpha rhythm as a regressor for modeling of the fMRI data [Feige et al., 2005]. In this study, we extend this approach to Rolandic rhythms with amplitudes markedly smaller than those of the posterior alpha rhythm (8-25 μV when compared to 20-60 μV [Jasper and Andrews,1936]. Nevertheless, gradient artifact correction made it possible to recover these rhythms with their major features, such as pericentral topography, motor-task reactivity and spectral predominance in the alpha and beta band. Further the EEG signal was strong enough to act as a regressor in BOLD correlation analysis, i.e. it explained a significant portion of the BOLD-signal variance.

There Is Distinct Laterality of the fMRI Response to Right Versus Left Rolandic Rhythms

There are strong and clearly lateralized correlations between Rolandic rhythms that were recorded from the right or from the left hemisphere (“C3/C4 Rolandic rhythms”) and the BOLD signal. The lateralization of the BOLD signal, particularly for the Rolandic areas (MI, SI), clearly corresponded to the site of the EEG electrode, i.e. there was a left hemispheric dominance in fMRI signal for C3 and right hemispheric dominance for C4 respectively. These findings indicate that spontaneous fluctuations of the Rolandic rhythms, occurring in addition to the block-wise modulations by the bimanual motor task, have distinct fMRI correlates in the respective hemisphere (see Fig. 7).

Rolandic Alpha and Beta Rhythms Differ with Respect to Their fMRI Correlates

With all employed modeling approaches, we found spatial differences in the sites of fMRI correlation between the Rolandic alpha versus beta rhythm. The fMRI correlates of the Rolandic alpha rhythm meet our strict significance criteria only in the case of the blind-source separated broad-band Rolandic alpha rhythm. The consistency of the anatomical locus in the postcentral cortex (as compared to the precentral correlate of the Rolandic beta rhythm, see Table I), adds to the credibility of this finding and exemplifies the utility of separating various EEG signatures.

Our findings are in agreement with previous MEG studies in which reactive sources of the Rolandic alpha rhythms were localized in the SI and the Rolandic beta rhythms in the MI [Salmelin et al.,1995] during phasic motor activity [Salmelin and Hari,1994a] and tactile stimulation [Cheyne et al., 2003].

There are electrocorticographic studies reporting rather widespread and individually variable cortical desynchronization during movement in both the precentral and postcentral cortex [Crone et al., 1998]. These authors are in favor of the view that oscillatory suppression reflects a synaptic network of a distributed cortical system rather than a locally isolated phenomenon, such that activation of a subset of the network will result in suppression of oscillations elsewhere in the network. Also in intracortical recordings, both the Rolandic alpha and beta rhythms are found in pre- and postcentral areas in human [Jasper and Penfield,1949] and in monkey [Rougeul et al.,1979]. Thus, the association of the Rolandic beta rhythm with motor cortex and the Rolandic alpha with somatosensory cortex is not definitive. Nevertheless, in these intracranial recordings, the maximal beta rhythm tends to be found more anterior and the alpha rhythm more posterior to the Rolandic fissure. These findings are supported by our results. In a recent combined EEG-fMRI study a temporary post-movement beta synchronization (“beta rebound”) was paralleled by a BOLD signal increase in the primary sensorimotor cortex [Parkes et al.,2006]. In this study the most significant “β-rebound” related BOLD signal change was located in the postcentral sulcus. Another recent study employing synthetic aperture magnetometry localized the post-movement beta rebound to the precentral gyrus whereas movement related desynchronization of Rolandic alpha (mu) and beta rhythms was localized to the postcentral gyrus [Jurkiewicz et al., 2006]. In our present study the continuous time course of Rolandic beta and alpha band desynchronization and synchronization across a motor-task was correlated to the BOLD signal. Thus, we identified regions where the BOLD signal followed the time course of both motor-task related synchronization and desynchronization of Rolandic alpha and beta rhythms.

Interestingly, although our results show a maximal cortical fMRI response to the (bimanual) motor task in the left SMI (Fig. 9a, Table III), the maximum fMRI correlate of the “TDSep Rolandic rhythms” was throughout in the right sensorimotor cortex. This is in concordance with our EEG findings of stronger motor-task related modulations of Rolandic rhythms in the right hemisphere. A slightly stronger Rolandic alpha rhythm in the right hemisphere has also been found independent of the handedness of the subject [Niedermeyer and Koshino,1975]. There are EEG reports showing stronger desynchronization in the left sensorimotor cortex during unimanual hand movements irrespective of the side of movement. However, these EEG data are associated with a stronger β-rebound in the right hemisphere [Stancak and Pfurtscheller,1996,1997]. Probably the latter effect outweighs the former resulting in a stronger modulation in the right hemisphere in our data.

The direct comparison of the maximal negative BOLD response to the Rolandic alpha rhythm versus beta rhythm indicates a motor cortex association of the beta rhythm (precentral activations) and a somatosensory association (postcentral/parietal activation) for the Rolandic alpha rhythm (Fig. 8, Table I).

In addition we found a positive correlation between the Rolandic beta rhythm and the BOLD signal in the bilateral paracentral lobule, i.e. an increasing strength of the rhythm was associated with an fMRI signal increase in this region. This contradictory behavior of the BOLD signal in the bilateral paracentral lobule and the primary sensorimotor hand area might be linked with the finding of bilateral increase of oscillations in hand sensorimotor cortices during foot movements [Pfurtscheller and Neuper,1994]. The hand movement task may induce short lasting idling states in the neuronal representation of other body parts as discussed by Pfurtscheller and Neuper and thus also lead to an inverse hemodynamic behavior in respect to somatomotor hand areas.

Functional Implications for Rolandic Rhythms

While above we emphasized the maximal inverse correlation between the Rolandic beta rhythm and the BOLD signal in MI, it is equally notable that when looking at the whole brain, the inverse correlation to Rolandic beta is seen to be almost identical to the activation pattern for the bimanual motor task. Since it is not possible to distinguish between only correlating brain activations and those which are directly attributable to beta synchronization, one should nevertheless mention that previous work has suggested a role of beta oscillations in widespread somatosensory and motor areas during premovement maintenance behavior [Donoghue et al.,1998; MacKay and Mendonca,1995; Rougeul et al.,1979; Sanes and Donoghue,1993]. A beta synchronized large-scale network that links pre- and postcentral areas during premovement motor maintenance behavior has been identified in monkey [Brovelli et al.,2004]. It has also been proposed that voluntary movement depends on a corticoperipheral cortical sensorimotor loop [Favorov et al.,1988] and that this loop is supported by oscillatory neuronal activity [MacKay,1997]. Different rhythms are known to employ different dynamic mechanisms to synchronize based on different ionic currents [Bibbig et al.,2002; Kopell et al.,2000] and beta rhythm tends to indicate synchronization over long conduction delays. Thus, oscillations may bind sensorimotor areas into a functional loop during premovement motor maintenance behavior [Brovelli et al.,2004]. A dynamic beta range (16–28 Hz) synchronization has also been demonstrated between activity in primary and premotor cortex and muscle activity, starting after termination of phasic voluntary movement, during a period of tonic muscle activity [Feige et al.,2000]. Feige et al. reported a low frequency (2–14 Hz) synchronization of the same cortical areas and medial premotor cortex with the EMG signal during movement execution. They also suggested that beta synchronization may reflect transition of the motor network into a new equilibrium state, while the low frequency synchronization may be related to movement execution.

In addition to inverse correlations to the BOLD signal, there are also positive correlations of the TDSep Rolandic beta rhythm in the bilateral paracentral lobule. It seems likely that these represent an attentional network which (at first glance rather counter-intuitively) is activated during the rest period. An explanation may be derived from a previous study [MacKay and Crammond,1987] in which the posterior parietal cortex, BA 5, bilaterally has been shown to be activated during expectation of an event. As described above, the positive hemodynamic behavior in the bilateral paracentral lobule could also be induced by short lasting idling states in the neuronal representation of other body parts than the functionally activated hand area. Lastly, a potential explanation is given by Gusnard et al. [Gusnard et al.,2001] who proposed an involvement of these areas in self-referential mental activity that becomes attenuated during an attention demanding motor task itself.

We also investigated the fMRI correlates of the posterior (classical) alpha rhythm as a control. Confirming previous findings [Feige et al.,2005; Goldman et al.,2002; Moosmann et al.,2003], the BOLD-fMRI signal in the occipital cortex was inversely related to the strength of the posterior alpha rhythm (Fig. 9c,d, Table II). The inverse relationship is more pronounced (see also Table II) for the blindly separated (TDSep) posterior alpha rhythm, which may be attributed to the separation of the posterior alpha rhythm from other rhythms and EEG signatures. The clear anatomical separation of fMRI correlates between Rolandic rhythms on the one side and occipital alpha on the other side confirms the regional specificity of the EEG-fMRI approach.

Lauf's group has proposed that MR correlates of the occipital alpha and beta band also reflect global mental states [Laufs et al.,2003a,b,2006]. While this seems to be plausible given the fact that alpha and beta spectral content indexes mental activities, it also highlights the challenge to functionally separate the multiplicity of cortical rhythms superimposed in the EEG. In this study, the finding that the strength of each of the three different background rhythms (Rolandic alpha, Rolandic beta, and posterior alpha) inversely correlates with the fMRI signal in “its cortical area” indicates that (1) functional segregation of the rhythms was sufficient and (2) this relation between rhythms and blood oxygenation in the underlying cortex is possibly a general feature presumably due to similar fundamental physiological mechanisms.