spectral changes of interhemispheric crosstalk during movement instabilities

Cerebral Cortex November 2010;20:2605--2613doi:10.1093/cercor/bhq008Advance Access publication February 22, 2010

Spectral Changes of Interhemispheric Crosstalk during Movement Instabilities

Sanne Houweling, Peter J. Beek and Andreas Daffertshofer

Research Institute MOVE, VU University Amsterdam, 1081 BT Amsterdam, the Netherlands

Address correspondence to S. Houweling. Email: [email protected].

Bimanual coordination requires the functional integration of theactivity in various cortical, subcortical, spinal, and peripheral neuralstructures. We challenged this functional integration by destabiliz-ing bimanual 5:8 tapping through an increase in movement tempo,while measuring brain and muscle activity using magnetoencepha-lography and electromyography. Movement instabilities werecharacterized by a drop in frequency locking. Time--frequencyanalysis revealed movement-related beta amplitude modulation inbilateral motor areas as well as movement-related corticospinalentrainment. Both of these synchronization patterns depended onmovement tempo suggesting that the timescale needed for theupregulation and downregulation of beta synchrony in rhythmictapping poses constraints on motor performance. Bilateral phaselocking over movement cycles appeared to be mediated by beta-frequency oscillations and constrained by its phase dynamics. Thetimescale of beta synchrony thus seems to play a key role inachieving timed phase synchrony in the motor cortex and along theneural axis. Once event-related desynchronization--synchronizationcycles cannot be build up properly, inhibition may becomeinadequate, resulting in a reduction of the stability of performance,which may eventually become unstable.

Keywords: beta modulation, interhemispheric inhibition, MEG, motorcortex, motor timing

Introduction

Unraveling the mechanisms that transfer information across thehuman nervous system forms one of the greatest challenges inneuroscience. Intra- and interhemispheric interactions be-tween bilateral motor areas provide an expedient window intothe neural underpinnings of motor control. Interhemisphericinteraction comprises both inhibitory and excitatory influencesthat are mediated through transcallosal pathways (Ferbert et al.1992). There is accumulating evidence that connectionsof premotor areas (PM1s) through the anterior portion of thecorpus callosum support skilled movements, including bi-manual coordination. In addition, ipsilateral primary motorcortex (M1) appears to continuously interact with itscontralateral counterpart (Daffertshofer et al. 1999). In fact,many studies have shown that during unimanual movement notonly contralateral but also ipsilateral motor areas are active.This ipsilateral activity may map onto its contralateral cortico-spinal pathway and thus induce bilateral crosstalk, which maybecome manifest as mirror movements (e.g., Shibasaki andNagae 1984; Britton et al. 1991). This line of argumenthighlights the importance of effectively inhibitory interhemi-spheric interactions as they may serve to suppress unwantedcrosstalk (Daffertshofer et al. 1999; Gerloff and Andres 2002;Serrien and Brown 2002).

In the present study, we investigate the role of balancingexcitatory and inhibitory interactions in the context of

movement instabilities. As movement instabilities, we specifi-cally consider the loss of bimanual coordination duringrhythmic performances (Kelso 1995). These can be interpretedas phase transitions and captured mathematically by (simple)bifurcation schemes. Daffertshofer et al. (2005) proposeda mathematical model for the neural dynamics accompanyingrhythmic motor performance. In a nutshell, this model de-scribes how effectively inhibitory interhemispheric interac-tions may result from excitatory transcallosal fibers projectingto M1 and PM1, and PM1 inhibiting M1 via intrahemisphericcortico-cortical connections. The model incorporates oscilla-tors to represent local activity and the interregional inter-actions (intra and interhemispheric) are given as couplingbetween these oscillators. Coupling is either excitatory orinhibitory, which causes in-phase or antiphase oscillations,respectively. If intrahemispheric excitation and inhibition areproperly balanced, these oscillations cancel each other so thatthe local power drops and the interhemispheric crosstalk iseffectively suppressed. Changes in local activity as measured bylocal power (or local event-related [de]synchronization, seebelow) can indicate that intrahemispheric excitation andinhibition are not balanced. The underlying oscillatory pro-cesses might be too slow to provide the proper balance, thatis, the oscillators’ phases cannot be adjusted in time(Daffertshofer et al. 2005). Put differently, the intrahemisphericinhibition fails whenever the phase coupling between M1 andPM1 becomes less stringent. This may yield a greater amount ofinterhemispheric crosstalk through the aforementioned, excit-atory transcallosal fibers and can result in a destabilization ofmotor performance.

Many empirical studies on neural activity accompanyingmovement instabilities have focused on in-phase versusantiphase coordination or on-the-beat versus off-the-beattapping (Jirsa and Fuchs 1998; Kelso et al. 1998; Daffertshoferet al. 2000a, 2000b; Fuchs et al. 2000; Mayville et al. 2001;Serrien and Brown 2002; Aramaki et al. 2006). However, suchisofrequency performances barely allow for studying theaforementioned crosstalk as the activities associated with the2 fingers (or hands) are very hard to distinguish in encephalo-graphic signals. To overcome this drawback, we investigatedpolyrhythmic tapping, that is, a task in which left and rightfingers move rhythmically in a p:q frequency relation (p and qare integers and with p:q in reduced form both p and q arelarger than 1). We induced instabilities in the performance ofthis task by increasing movement tempo (Peper et al. 1995;Meyer-Lindenberg et al. 2002). In polyrhythmic tapping, onecan uniquely identify neural activities that correspond to eitherthe left or the right finger (Lang et al. 1990). The left--rightcrosstalk may hence be seen as the ratio between activitiesassociated with the right and left fingers (Kristeva et al. 1990).In accordance with the model’s prediction, we hypothesizedthat a destabilization of behavior is accompanied by

! The Author 2010. Published by Oxford University Press. All rights reserved.For permissions, please e-mail: [email protected]

at Vrije U

niversiteit- Library on May 18, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

a disbalancing of intrahemispheric excitation/inhibition yield-ing a loss of effective interhemispheric inhibition. Besidesa pronounced bilateral interaction as mere reflection of theincreased crosstalk, we expected to find an increased,movement-related activity in the motor cortices, in particularin the hemisphere ipsilateral to the finger displaying themovement instability.

Materials and Methods

SubjectsThirteen right-handed percussionists (2 female, age 20--25 years), allstudents or graduates from the Conservatory of Amsterdam, partici-pated in the experiment, which was conducted in full compliance withthe guidelines of the Ethical Committee of the Faculty of HumanMovement Sciences of VU University Amsterdam. All subjects signed aninformed consent form prior to participation. They had completed thecourse ‘‘Advanced rhythms’’ in which several polyrhythms wereextensively practiced. They were paid for their participation.

ProcedureSubjects were invited to tap a 5:8 polyrhythm with their left and rightindex fingers. Starting with a steady 5:8 performance, movement tempowas increased over 23 plateaus in steps of 0.2 Hz starting with the rightfinger at xR = 2.5 Hz (plateau 1) and ending at 6.9 Hz (plateau 23); inplateaus 1--13, a single rhythmic cycle was produced, whereas fromplateau 14 onward, 3 rhythmic cycles were produced (Fig. 1). Theinitial frequency of the left finger was xL = 5/8 3 2.5 = 1.56 Hz. The beatwas acoustically paced; the fast beat was permanently present at theright ear (pitch 660 Hz, tone duration 50 ms), whereas only a singletone of the slow beat (left ear,, pitch 440 Hz, tone duration 50 ms) waspresented to indicate the start of a rhythmic cycle at the beginning ofa plateau (i.e., always coincident with the fast beat).

To guarantee steadiness of performance, all trials started with 3plateaus at xR = 2.5 Hz with rhythmic 5:8 binaural pacing. Theseplateaus were not analyzed further as they merely served to tune in onthe required polyrhythm. For the magnetoencephalographic (MEG)recordings, subjects sat in an upright position with their arms on anarmrest on which force transducers were mounted. The experimentconsisted of 5 identical blocks. Between-blocks subjects had a fewminutes break during which they stayed in the seat. A block consistedof 2 times 2 initial unimanual tapping trials (first right, then left)followed by 12 trials of bimanual polyrhythmic tapping. In total, we had5 3 12 = 60 polyrhythmic trials (~80 s each), 532 = 10 unimanual trials

for right and left movements (~80 s each), and 6 resting state trials (30 seach). The unimanual trials fully agreed with the bimanual one, withthe only difference that one hand was not moving. In the unimanualand resting state trials, the nonmoving hand(s) was (were) alwaysplaced in a relaxed position on the armrest. During all trials, subjectswere asked to fixate their gaze at a stationary spot marked as a red crossand located about 2 m in front of them.

Their musical training notwithstanding, all subjects had to followa 30-min training scheme to properly perform the 5:8 polyrhythm:tapping to audio files with unimanual and bimanual repetitions of serieswith constant (2.5, 3.3, 4.1, and 4.9 Hz) and increasing (2.7, . . ., 5.1 Hz)movement tempos at the required frequency ratio. This trainingscheme was repeated 4 times during the week prior to theexperimental session.

Data Acquisition and PreprocessingBrain activity was recorded using a 151-channel MEG system (CTFSystems Inc., Vancouver) with third-order synthetic gradiometers. Theelectromyogram (EMG) was assessed from bilateral extensor and flexordigitori with a reference electrode at the left wrist (Ag--AgCl electrodes:Ø 1 cm and about 2 cm interelectrode distance for off-line bipolar mon-tage). Tapping forces were recorded using MEG-compatible forcesensors that were mounted on the 2 armrests (Boonstra et al. 2005). Allsignals, including acoustic stimuli, were online low-pass filtered at 200Hz prior to digitization at a rate of 625 Hz. All signals were mean-centered and high-frequency noise was eliminated off-line usinga second-order bidirectional Butterworth low-pass filter at 200 Hz. Inaddition, EMG signals were high-pass filtered at 20 Hz to remove move-ment artifacts and rectified using the Hilbert transform (Myers et al.2003). To minimize head movement, small pads were positioned be-tween the helmet-shaped opening in the dewar and the subjects’ head.

Data Analysis

Source AnalysisThe cortical motor areas were identified via local event-relateddesynchronization (ERD) and synchronization (ERS) power differences(i.e., ERD--ERS) in the beta band (20--30 Hz) in the unimanual conditionsusing a synthetic aperture magnetometry (SAM) beamformer approach(Vrba and Robinson 2001; Cheyne et al. 2006). In brief, weightdistributions of sensors were determined to estimate activities in thevoxels of a probabilistic brain template (The International Consortiumfor Brain Mapping [ICBM], see, e.g., Mazziotta et al. 1995, for details).MEG recordings were positioned and scaled to that template usingintermittent head position registrations (3 reference coils at nasion andleft--right ears, see, e.g., de Munck et al. 2001). SAM was used to computein every voxel the variance-normalized power in a specific frequencyband (pseudo-t values). Subsequently, we computed condition-baseddifference by subtracting a so-called ‘‘active’’ state from a ‘‘control’’ orbaseline state (we note that here these differences were only used toenhance contrast and not to pinpoint active state and the ‘‘real’’ baseline).In detail, the interval from 200 ms (or 300 ms when tapping with the leftfinger given the longer movement cycle) before the tap to the tap onset(active state) was contrasted with the interval from tap onset to 200 ms(or 300 ms) after the tap onset (control state). The cluster size of voxelswith maximal pseudo-t differences was defined as the number ofsurrounding voxels exceeding 95% of the peak value; the significance ofthese differences between active and control states was assessed viaa paired permutation test across subjects (see below for details). Clusterswith less than 10 voxels were discarded. The MEG signals weremultiplied and summed using the weight distributions corresponding tothe beamformers showing the maximal, significant power differences.Given their location, we denote these projections of the neural signals asM1left and M1right, which thus yielded maximal power differences duringright and left finger tapping, respectively. During bimanual movementbeta activity in M1left may generally covary with beta activity in M1right(e.g., because of coinciding ERD--ERS cycles in coinciding taps); hence,we used the unimanual controls to uniquely localize sources. We notethat we did not find strong ipsilateral beta modulation during unimanualtrials.

Figure 1. Transition trial scheme. Vertical dashed lines indicate the start of therhythmic cycles; red solid lines indicate the plateaus (top axis) corresponding to thetarget tempos (left axis). Each trial lasted for about 80 s (bottom axis); the first 3rhythmic cycles that served to tune in on the rhythm are not shown.

2606 Spectral Changes of Interhemispheric Crosstalk d Houweling et al.

at Vrije U



ownloaded from

Spectral AnalysisPower spectra were computed for force signals and M1 projections foreach subject, trial, channel, and plateau, using Welch’s periodogrammethod with Hamming windows as tapers (size = one rhythmicalcycle). These power spectra were upsampled (via cubic splines) togenerate an identical frequency resolution across plateaus. (Thisapproach allows for using a fixed number of tapers for all estimates.We note that here the more conventional padding technique failsbecause padding size would exceed the data size because themovement frequency spans a fair range.) To remove an overall scalingin amplitude, the power spectra were normalized so that the totalpower (per subject and trial) was equal to 1. For every subject, trial,channel, and plateau the frequency containing the most power in thespectra of the force signals was identified as the movement frequency(xR for right tapping and xL for left tapping). The maximal tappingfrequency was defined at the group level as the target frequencycorresponding to the last plateau in which the movement frequencymatched the target tempo. To this end, the target frequency wascompared with the movement frequency across subjects, and tested forequality by a permutation test; if no significant (P < 0.001) differencewas found, this was considered a match. These maximal tappingfrequencies served to test whether the expected behavioral instabilitieswere indeed related to coordinative effects, that is, differences inmaximal frequencies between bimanual and unimanual conditionsimply an effect of bimanual coordination on the critical plateau (seebelow), which relates to the emergence of movement instabilities.The power at the movement tempos xR and xL and the overlap of the

normalized power spectra were used to assess the dynamics around thebehavioral instability. The latter provides the similarity Wy

x between 2spectra Px and Py after rescaling the frequency axis with q=p : q

Wyx ðqÞ=

2RPx ðxÞPy ðqxÞdx

R hP 2x ðxÞ +P 2

y ðqxÞidx

see Daffertshofer et al. (2000a, 2000b) for more details. We wereprimarily interested in the values at the target frequency ratio, that is, atp:q = 5:8. In addition, we also found large spectral power at harmonics(i.e., multiples) of the movement frequency yielding finite spectraloverlaps at the corresponding harmonic ratios. The frequency plateau,in which the behavioral transition from steady to unstable performanceoccurred, was defined as the plateau where the strength of 5:8frequency locking in the EMGs dropped substantially (referred to as‘‘critical plateau’’ P0). More specifically, the first plateau in which theactual ratio (given at the highest spectral overlap) differed more than0.1 from the target ratio 5:8, was considered the transition plateau. Toassess neural and behavioral changes during the transition, weevaluated the 6 plateaus preceding the critical plateau P0 with the6 plateaus following P0, that is, we considered P–6. . .P–1 as steady andP1. . .P6 as unstable performance, respectively, and compared theirproperties (referred to as ‘‘relative plateaus’’).

Event-Related AnalysisWe filtered the MEG projections and the EMG in frequency bands 0--3,3--6, . . ., 42--45 Hz, and computed amplitudes and phases using theHilbert transform. These signals were divided into 1-s windowscentered around motor events (taps). Per plateau and frequency band,the dispersion over motor events of the relative phase between M1leftand right--left EMG, M1right and right--left EMG, and M1right and M1leftwas quantified by phase uniformity (Mardia and Jupp 1972). For adiscrete set of phases fu1;u2; . . . ;uN g, their uniformity is defined as 1-circular variance of the phases or 1 – cv with

cv=1 –ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiC2 + S2

p; C=

1

N+N

k=1cosuk and S=

1

N+N

k=1sinuk :

(Uniformity agrees with the more recently employed phase coherence;see, e.g. [Mormann et al. 2000]). As phase uniformity is, by definition,biased by the number of phases N, the uniformity values weretransformed to the Rayleigh statistic 2#(1 – cv)2N. To assess the betamodulation and uniformity during the transition, the event-relatedsignals were averaged per plateau Pj (j = –6. . .6) over all trials andsubjects. Because the timing of the local ERD--ERS turned out to depend

on the movement frequency, we normalized amplitudes and relativephases by scaling the window such that each window contained 2movement cycles, and all epochs were of equal length. To identify ERD/ERS, maxima and minima of the amplitude were selected from aninterval of 70 ms before to 200 ms after the event (aERS and aERD,respectively), and we computed the amplitude deviations DA = aERS –

aERD. The temporal difference DT = tERS – tERD was quantified by thelag between the ERD and ERS. For the corticospinal synchronization,we integrated epochs of uniformity over its time interval and computedDU. Interestingly, these differences dropped with increasing tempo asquantified via an exponential decay of the form e–bx estimated for allthe grand averaged measures in bimanual and unimanual trials. Todecide whether the data were in agreement with the exponentialmodel, we assessed the goodness of fit by the squared correlation r2 =SSR/(SSE + SSR), where SSR is the sum of the squared residuals and SSEthe sum of the squared errors.

StatisticsAll values were subjected to a permutation test across subjects (Nicholsand Holmes 2002) to assess the hypothesis of a difference in thevariable T (e.g., spectral power) between stable and unstableperformance. This nonparametric test is based on resampling. In brief,the subjects’ difference values for stable and unstable performanceTu;k–Ts;k are labeled with random signs ri ;k , where k indexes allsubjects 1; . . . ;N and i indexes all possible permutations acrosssubjects of labelings with +1 or –1. For the ith permutation ri themean of the differences across all subjects yields the statistic

ti=1

N+kri ;k

"Tu;k – Ts;k

#:

Under the null hypothesis ‘‘no difference,’’ this distribution of means issymmetric (and has mean zero). Small values of the mean suggest thatthe null hypothesis should be rejected. The P value of this test is givenby the portion of ti greater or equal to the test statistic correspondingto the correctly labeled data (with all signs positive).

Results

Source Analysis

Beamformers yielded consistent power differences in the betaband (Fig. 2 and Supplementary Fig. 1). In the 12 left tappingtrials, M1right showed the largest event-related decrease (powerdifference –0.5, P < 0.001, cluster size 34), and in the 12 righttapping trials, M1left showed the largest event-related decrease

Figure 2. SAM sources. Axial view of the differential statistical images after setting thethreshold to 50% of the largest differences in beta power around the tap onsets overlaidon the brain template (an average of more than 150 MRI scans from different individualsfrom the ICBMMRI database). Differences in beta power between 200 ms (or 300 ms forperformance with the left finger) after (active state) and before (control state) the tapsfrom the unimanual right (left image) and left (right image) conditions. Talairachcoordinates of the peak sources were ($29.6,$20.6,52.8) and (30.7,$16.5,51.4),respectively. Both sources were located in the hand--arm area of the contralateral primarymotor cortex (Brodmann area 4, see also Supplementary Material).

Cerebral Cortex November 2010, V 20 N 11 2607

at Vrije U



ownloaded from

(power difference –0.1, P = 0.04, cluster size 22). To anticipate,the modulation of beta activity decreased with increasingfrequencies, which may explain the weaker result for M1leftcorresponding to the faster rhythm.

Spectral Changes around Transition

All subjects experienced the task as challenging, which wasalso apparent from the rapid decrease in frequency lockingwhen tempo increased (Fig. 3). One subject (#8) consistentlyshowed transitions at lower movement tempos than the othersubjects: The deviation of this subjects’ mean transitionfrequency (over trials) exceeded 2 standard deviations of thegroup mean; it was therefore considered an outlier. Foursubjects (5, 6, 10, and 11) completed the tapping sequencewithout deviating from the 5:8 frequency relation in severaltrials; analyses were performed over the remaining 582 trials.On average, performance became unstable between plateaus12 and 20 (mean: plateau 17, standard deviation: 2). Thiscorresponds to movement tempos around 5.7Hz, which agreeswith the previous findings of Peper et al. (1995). The transitionwas accompanied by a drop in power at both target tempos butmost pronounced in the right tapping data at the fastmovement tempo.

Figure 4 shows the power spectral densities of taps and M1s,and spectral overlap between tapright and tapleft, and between

M1left and M1right. In general, a loss of stability is reflected asa weaker frequency locking as expressed by the drop inspectral overlap, here at 5:8 (see Fig. 4, left panel). Thesecomparisons of spectral densities further show that theperformance of the right finger became unstable after thecritical plateau as its movement frequency spread around lowerfrequencies. The performance of the left finger also seemed tobecome less accurate during the transition. In the M1s, bothmovement frequencies were found to be present, in particularbefore the critical plateau P0 (same figure, inner panels). Thefrequency locking was not confined to 5:8, but the spectraloverlap was also large at 1:1 and the harmonics 5:4 (samefigure, right panel). We note that the locking at 5:4 can beexplained by the presence of harmonics of the movementfrequencies in the M1s (Daffertshofer et al. 2000a, 2000b). Bycomparing the spectral overlap in steady (P–1) and unstable (P1)states, we found that the frequency locking at 5:8 wassignificantly weaker after P0 in the tapping data (from 0.90 to0.55, P < 0.001) but not in M1s (P = 0.87). The frequencylocking at 1:1 was slightly enhanced after P0 in both tappingsignals (from 0.09 to 0.20, P < 0.001) and cortical activity (from0.92 to 0.94, P = 0.02).

We further assessed the relative ‘‘contribution’’ of the move-ment frequencies to the transition by comparing the power inM1 at xR and xL (Fig. 5). Between P–6 to P–5, there were no

Figure 3. Performance data. (A) Tapping profiles in a typical trial. The panel shows the left (dark gray) and right (light grea) force signals (ordinate) from 41.5 to 46.5 s during 5 sof steady performance of subject 9. (B) Frequency locking per plateau (y-axis) and frequency ratio (x-axis) in the same trial showing steady performance up to 5.9 Hz (or plateau18). (C) Box plot of critical plateau P0 (ordinate) for each subject (abscissa) as determined from the spectral overlap at 5:8 in the left:right tapping data (13 subjects, 60 transitiontrials). The boxes indicate lower and upper quantiles, the midlines indicate medians, and plus signs are outliers. The latter trials were excluded from the analysis; in total, 582transitions were identified.


at Vrije U



ownloaded from

differences, and this trend continued up to the comparisonbetween P–2 and P–1. From P–1 to P0, there was a drop in powerat the contralateral movement frequency in both M1s (P = 0.02for M1left and P = 0.004 for M1right). This was followed bya sharp power increase in M1right at the ipsilateral movementfrequency xR (P = 0.02). (We note that a power drop in M1rightafter P1 does not imply the coordination in P2 was more or less

stable than in P1; the performance in P2 just deviated morefrom auditory target than in P1. In general, P0 defines the pointat which a coordination pattern looses stability. That point isoften accompanied by an increase in fluctuations, whichreduce as soon as a new, stable coordination pattern emerges.Admittedly, we could not find clear-cut switches to other long-lasting, stable coordination, which we did in earlier studies[e.g., Peper et al. 1995; Daffertshofer et al. 2005]; nonetheless,we interpret the local increase in power as indicative fora phase transition.) Power also increased at xR in M1left (P =0.06). That is, the power of the right movement frequencydominated more after the transition, especially in M1right.Hence, the increase in 1:1 frequency locking was caused by therelative increase at the fast movement frequency in M1right.

Event-Related Changes: Beta Amplitude and CorticospinalUniformity

The event-related amplitude at different frequency plateausrevealed a pronounced modulation, particularly in the beta band(see Fig. 6, left panels and Supplementary Fig. 3). Event-relatedpatterns were also visible in the uniformity of the relative phasesof M1 and EMG, and were again most pronounced in the betaband. This corticospinal beta uniformity showed increases anddecreases comparable with the M1 amplitude modulation (seeFig. 6, right panels). Patterns in uniformity between M1left andM1right displayed also event-related structures but varied sub-stantially over subjects. That is, local beta synchronization (i.e.,ERS--ERD cycles) was supplemented by more global synchroni-zation (between M1s or between M1 and EMG) that revealedsimilar event-related modulations. In the alpha band (~10 Hz),a high uniformity was consistently found, but here, the changes

Figure 4. Spectra and spectral overlap in tapping and contralateral motor cortices. Inner 4 panels: time--frequency plots of normalized spectral power in tapright (top left), tapleft(bottom left), M1left (top right), and M1right (bottom right). Displayed are the averages per channel at different frequencies (ordinate) relative to the critical plateau (abscissa).Note that the timescale of the signals was adjusted with factor 2.5/xR or 5/8#2.5/xL in each relative plateau resulting in common target frequencies 2.5 and 5/8#2.5 Hz as aneffect of interpolation. Outer panels: frequency locking at different ratios (ordinate) in tapping data (left panel) and cortical sources (right panel) relative to the critical plateau(abscissa). Averages per channel combination over subjects are shown. We note that here we used 8 instead of 3 rhythmical cycles as size of the Hamming taper to improvelegibility and guaranteed that the total number of cycles still agreed over time (plateaus) by interpolating the time series prior to estimating the power so that all plateauscontained an identical number of samples (5 number during the slowest tempo).

Figure 5. Cortical power at the movement frequencies. The mean (±standard errorof the mean) of the power (ordinate) at the contralateral (dark gray) and ipsilateral(light gray) movement frequencies in the motor cortices (over all transitions) isdisplayed relative to the critical plateau (abscissa). Left panel: power at xR (dark) andxL (light) in M1left. Right panel: power at xR (dark) and xL (light) in M1right.


at Vrije U



ownloaded from

were not time locked to the motor event; M1left/right uniformitywas therefore not further analyzed. As anticipated above, thebeta amplitude modulation in M1 was clearly dependent onmovement frequency.

In the unimanual trials, the maximal tapping frequency was onaverage 0.8 Hz higher than in the bimanual trials. This differencebecame most apparent from plateau 14 (5.1 Hz) onward. Therethe movement frequency did not match the target frequency inthe bimanual trials (P < 0.01), whereas in the unimanual trials,that mismatch occurred at plateau 18 (5.9 Hz). By contrast, forthe performance with the left finger, the movement frequencymatched the target frequency across plateaus.

With respect to movement tempo, the timescale of themodulation as well as its amplitude tended to decrease. Thetemporal difference DT between ERD and ERS decreased fromabout 300 ms in the slowest movement frequency (1.6 Hz) toabout 90 ms in the fastest movement frequency (5.1 Hz) forbimanual and unimanual conditions (Fig. 7, left panels). Thatdecrease was nearly linear for the initial 10 frequencies, but therate of decrease decreased as movement tempo increased. Theextent of amplitude modulation DA also dropped; however, thisdecrease was clearly nonlinear (Fig. 7, right panels). Theparameters resulting from fitting the exponential functions toDA and DT did not differ between M1left and M1right. We alsocould not find any difference between parameters in bimanualand unimanual trials. Lateralization effects in the strength ofamplitude modulation were not present in the bimanualcondition. Interestingly, beta modulation was almost absent inthe ipsilateral M1 throughout unimanual trials in agreementwith the absence of significant power changes in the SAManalysis. Although the patterns in corticospinal uniformityappeared movement tempo dependent when comparing leftand right tapping (e.g., Fig. 6, right panels), their minima/maxima did not necessarily coincide with ERD/ERS. Themodulation of corticospinal uniformity DU, however, de-creased with increasing frequency because all exponents werepositive: b = 0.02 (r = 0.32) and b = 0.01 (r = 0.17) in thebimanual condition and b = 0.007 (r = 0.12) and b = 0.007 (r =0.12) in the unimanual condition for right EMG/M1left and leftEMG/M1right, respectively.

Averaged over subjects, trials, and steady or unstableperformance, DA in M1right was greater during steady perfor-

mance (P < 0.001, see Fig. 8). This decrease in beta modulationfrom steady to unstable performance correlated with anincrease in movement frequency. The modulation of cortico-spinal uniformity DU decreased also from steady to unstableperformance for both right EMG versus M1left (P = 0.01) andleft EMG versus M1right (P = 0.24), although only the formerdecrease was significant.

Discussion

Spectral power in bilateral motor areas at both movementfrequencies exposed the expected interhemispheric crosstalkin bimanual performance. A loss of coordination was primarily

Figure 6. Event-related modulation and corticospinal uniformity. (A) The amplitudes of M1left (left) and M1right (right) after narrow-band filtering for each frequency (ordinate)and time relative to the event (abscissa). Grand mean--removed average over subjects, trials, and plateaus. Note the difference in scale between the panels. (B) Uniformity of therelative phases of M1left and right EMG (left) and the relative phase of M1right and left EMG (right). Grand mean--removed averages over subjects, trials, and plateaus. Eachamplitude and uniformity epoch was upsampled per frequency plateau to match the largest movement cycle.

Figure 7. Frequency-dependent event-related beta modulation in M1s. Amplitudedifference DA 5 aERS $ aERD (left panels) and temporal difference DT 5 tERS $ tERD(right panels) in ERD and ERS per frequency plateau in M1left (3-markers) and M1right(O-markers). The light shaded markers indicate the plateaus following the meantransition plateau. The solid lines indicate the exponential function of the forme$bx resulting from the optimal fitting parameters in the bimanual condition in M1left(red line) and M1right (blue line), which were similar to those in the unimanualcondition: for DA, these parameters (mean over conditions) were b 5 0.23 (range0.19--0.32) and r5 0.96. For DT, these parameters (mean over conditions) were b50.05 (range 0.04--0.06) and r 5 0.93. Please note that the graphs are plotted insemilog scale.


at Vrije U



ownloaded from

visible as a power increase in the motor area ipsilateral to thefinger (predominantly) undergoing the motor instability. Asexplained in the ‘‘Introduction,’’ this power increase can beinterpreted as a reflection of an improper intrahemisphericphase locking, rendering the interhemispheric inhibition lesseffective (cf. Daffertshofer et al. 2005). The motor system isbilaterally activated during both bimanual and unimanualmovements (Swinnen 2002). The present results suggest whythis might be so: Our motor system appears to be inherentlydesigned for bilateral movement, and, as a result, unilateralmovements require the suppression contralateral corticalactivity (Kristeva et al. 1991; Mayston et al. 1999; Ghacibehet al. 2007). Homologous motor areas and premotor areas arereciprocally connected through the corpus callosum (Hoferand Frahm 2006), whereas premotor areas may inhibit theactivity from the homologous motor cortex (Daffertshofer et al.2005). When this inhibition falters, residual activity from theopposite hemisphere can be detected. This can also result inthe occurrence of mirror movements (Daffertshofer et al.1999), which are particularly pronounced when fatigued(Duque et al. 2005) or in the case of callosal damage (Dennis1976; Bonzano et al. 2008). The current study, however, did notallow for discriminating primary and premotor area, so that thisinterpretation of specific cortical areas remains speculative andshould be addressed in future studies.

The ERD--ERS amplitude difference, that is, the strength oflocal beta modulation was generally larger in M1right than inM1left. This lateralization might be readily attributed to theslower movement tempo (Toma et al. 2002; Houweling et al.2008). In addition, we found that the temporal relation as wellas the amplitude difference between the onsets of ERD and ERSchanged with movement tempo, but that change saturatedafter the tempo that marked the change in performance. That

point depended on coordination, as it occurred earlier for thebimanual case than for the unimanual case (5.1 Hz or plateau14 vs 5.9 Hz or plateau 18, respectively). Admittedly, theseneural activities are strictly speaking mere correlates of alteredmovement tempo and not direct evidence for causes of alteredperformance. In particular, in view of the well-establishedrelevance of ERD for voluntary movements (e.g., Pfurtschellerand Berghold 1989), however, we here interpret the change inperformance as consequences of the change in neural activity.

Complementary modalities (positron emission tomographyand functional magnetic resonance imaging) reported a posi-tive, linear relationship between metabolism and movementfrequency (Hayashi et al. 2008), whereas metabolism and betaactivity and metabolism are inversely related (Ritter et al.2009). We can support these findings as we also found a dropof overall beta power with increasing frequency (see Supple-mentary Fig. 2), unlike other, dissonant reports in the magneto-and electroencephalographic literature (Mayville et al. 2001;Serrien and Brown 2002; Toma et al. 2002).

Controlling movement via changes in distributed spectralprocessing in the beta band may be beneficial in view of itscomparably high temporal resolution. The production of a betacycle requires rapid changes in synchronization within a func-tional cluster of neuronal ensembles (Pfurtscheller and da Silva1999). We found that the upregulation and downregulation ofbeta synchrony was directly related to the movement frequency.Moreover, as frequency increased, both ERD/ERS shifted in timetoward the motor event. This finding challenges the notion ofERS as a simple response to recalibrate the underlying network(‘‘rebound’’). As the target frequency increases, ERD and ERSstart overlapping each other (see Fig. 9), which evidently limitsperformance. The time required to establish interactionsbetween neural populations poses a very natural constraint on

Figure 8. Event-related beta amplitude modulation and adjusted corticospinal uniformity. (A) Grand average of amplitude (ordinate) in M1left (left) and M1right (right) duringsteady (dark gray) and unstable performance (light gray) showing a decrease in modulation in M1right during the movement interval (abscissa). (B) Grand average of adjusteduniformity (ordinate) of the relative phase of M1left and right EMG (left) and the relative phase of M1right and left EMG (right) during steady (dark) and unstable performance (light).

Figure 9. Event-related beta amplitude modulation. With increasing tempo (left to right), the upregulation and downregulation of movement-related beta activity is challenged,and if ERD/ERS cycles get too close beta modulation vanishes, the movement phase can no longer be stabilized and an instability emerges (right panel).


at Vrije U



ownloaded from

motor performance. On this account, local and global betamodulations not only facilitate but also limit the ability toperform rapid rhythmic tasks. To what degree these timeconstraints apply to intra and interhemispheric interactions,respectively, remains to be seen. Given that here the ERD/ERSchanges coincided with coordinative instability, however, we aretempted to suggest that these time constraints also affectbilateral coordination. Very recent finding of Rizzo et al. (2009)do indeed support the notation of short-term plasticity in(effective) interhemispheric inhibition.

In addition to the cortical interactions, long-distant betasynchrony between contralateral M1 and EMG recorded fromextensor muscles also revealed a marked entrainment. In-terestingly, the corticospinal phase uniformity was stabilitydependent in that it varied less during unstable performancebetween both M1left and right EMG and between M1right andleft EMG (Boonstra et al. 2007). It appears that the controlcommands from primary motor cortex to muscle fibers areconveyed via beta modulation of afferent neural signals (whichis effective for long-range corticospinal connections), whereasthe interhemispheric interactions between M1s occur ona slower timescale as we found transition-related changes inM1--M1 phase coupling at the (ratio of the) movementfrequencies. That is, the interhemispheric communication maybe mediated by the phase dynamics at the movement frequen-cies. The latter can be motivated to be closely related to othercognitive functions needed for proper motor control (e.g.,rhythm perception), and as such, the actual task propertiescan be identified from its neural component. On the other hand,the performance needs to be very fast and to a high degreeautomatic so that cognition appears less important. It might thusalso be possible that subcortical structures—the basal gangliaand the cerebellum—are involved, and perhaps, even somelearning at the level of the spinal cord could take place(Houweling et al. 2008). Whether this implies that the processesrecordedby theMEG reflect predictionof sensory consequencesof the movement or directly the sensory feedback remains to beseen (see Baker 2007, for a recent discussion).

In summary, our results strongly underscore the functionalrelevance of modulations of oscillatory activity in specificfrequency ranges. In particular, the timescale of beta synchronyappears to play a key role in achieving timed phase synchronyin the motor cortex and along the neural axis and thus in thestabilization of motor output in general and of bimanualcoordination patterns in particular.

Supplementary Material

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

Funding

Netherlands Organisation for Scientific Research (NWO grant #452-04-344 awarded to A.D.).

Notes

Conflict of Interest : None declared.

References

Aramaki Y, Honda M, Okada T, Sadato N. 2006. Neural correlates of thespontaneous phase transition during bimanual coordination. CerebCortex. 16:1338--1348.

Baker SN. 2007. Oscillatory interactions between sensorimotor cortexand the periphery. Curr Opin Neurobiol. 17:649--655.

Bonzano L, Tacchino A, Roccatagliata L, Abbruzzese G, Mancardi GL,Bove M. 2008. Callosal contributions to simultaneous bimanualfinger movements. J Neurosci. 28:3227--3233.

Boonstra TW, Clairbois HE, Daffertshofer A, Verbunt J, van Dijk BW,Beek PJ. 2005. MEG-compatible force sensor. J Neurosci Methods.144:193--196.

Boonstra TW, Daffertshofer A, Breakspear M, Beek PJ. 2007. Multivariatetime-frequency analysis of electromagnetic brain activity duringbimanual motor learning. NeuroImage. 36:370--377.

Britton TC, Meyer BU, Benecke R. 1991. Central motor pathways inpatients with mirror movements. J Neurol Neurosurg Psychiatry.54:505--510.

Cheyne D, Bakhtazad L, Gaetz W. 2006. Spatiotemporal mapping ofcortical activity accompanying voluntary movements usingan event-related beamforming approach. Hum Brain Mapp.27:213--229.

Daffertshofer A, Peper CE, Beek PJ. 2000a. Spatio-temporal patterns ofencephalographic signals during polyrhythmic tapping. Hum MovSci. 19:475--498.

Daffertshofer A, Peper CE, Beek PJ. 2000b. Spectral analyses of event-related encephalographic signals. Phys Lett A. 266:290--302.

Daffertshofer A, Peper CE, Beek PJ. 2005. Stabilization of bimanualcoordination due to active interhemispheric inhibition: a dynamicalaccount. Biol Cybern. 92:101--109.

Daffertshofer A, van den Berg C, Beek PJ. 1999. A dynamical model formirror movements. Physica D. 132:243--266.

de Munck JC, Verbunt JP, Van ’t Ent D, Dijk BWV. 2001. The use of anMEG device as 3D digitizer and motion monitoring system. PhysMed Biol. 46:2041--2052.

Dennis M. 1976. Impaired sensory and motor differentiation withcorpus callosum agenesis: a lack of callosal inhibition duringontogeny? Neuropsychologia. 14:455--469.

Duque J, Mazzocchio R, Dambrosia J, Murase N, Olivier E, Cohen LG.2005. Kinematically specific interhemispheric inhibition operatingin the process of generation of a voluntary movement. CerebCortex. 15:588--593.

Ferbert A, Priori A, Rothwell JC, Day BL, Colebatch JG, Marsden CD.1992. Interhemispheric inhibition of the human motor cortex.J Physiol. 453:525--546.

Fuchs A, Jirsa VK, Kelso JA. 2000. Theory of the relation betweenhuman brain activity (MEG) and hand movements. NeuroImage.11:359--369.

Gerloff C, Andres FG. 2002. Bimanual interaction and interhemisphericinteraction. Acta Psychol. 110:161--186.

Ghacibeh GA, Mirpuri R, Drago V, Jeong Y, Heilman KM, Triggs WJ.2007. Ipsilateral motor activation during unimanual and bimanualmotor tasks. Clin Neurophysiol. 118:325--332.

Hayashi MJ, Saito DN, Aramaki Y, Asai T, Fujibayashi Y, Sadato N. 2008.Hemispheric asymmetry of frequency-dependent suppression in theipsilateral primary motor cortex during finger movement: a func-tional magnetic resonance imaging study. Cereb Cortex. 18:2932--2940.

Hofer S, Frahm J. 2006. Topography of the human corpus callosumrevisited--comprehensive fiber tractography using diffusion tensormagnetic resonance imaging. NeuroImage. 32:989--994.

Houweling S, Daffertshofer A, van Dijk BW, Beek PJ. 2008. Neuralchanges induced by learning a challenging perceptual-motor task.Neuroimage. 41:1395--1407.

Jirsa V, Fuchs. 1998. Connecting cortical and behavioral dynamics:bimanual coordination. Neural Comput. 10:2019--2045.

Kelso JAS. 1995. Dynamic patterns: the self-organization of brain andbehavior (complex adaptive systems). Cambridge (MA): MIT Press.

Kelso JAS, Fuchs A, Lancaster R, Holroyd T, Cheyne D, Weinberg H.1998. Dynamic cortical activity in the human brain reveals motorequivalence. Nature. 392:814--818.

Kristeva R, Cheyne D, Deecke L. 1991. Neuromagnetic fieldsaccompanying unilateral and bilateral voluntary movements: topog-raphy and analysis of cortical sources. Electroencephalogr ClinNeurophysiol. 81:284--298.


at Vrije U



ownloaded from

Kristeva R, Cheyne D, Lang W, Lindinger G, Deecke L. 1990. Movement-related potentials accompanying unilateral and bilateral fingermovements with different inertial loads. Electroencephalogr ClinNeurophysiol. 75:410--418.

Lang W, Obrig H, Lindinger G, Cheyne D, Deecke L. 1990.Supplementary motor area activation while tapping bimanuallydifferent rhythms in musicians. Exp Brain Res. 79:504--514.

Mardia KV, Jupp PE. 1972. Statistics of directional data. London:Academic Press.

Mayston MJ, Harrison LM, Stephens JA. 1999. A neurophysiologicalstudy of mirror movements in adults and children. Ann Neurol.45:583--594.

Mayville MM, Fuchs A, Ding M, Cheyne D, Deecke L, Kelso JAS. 2001.Event-related changes in neuromagnetic activity associated withsyncopation and synchronization timing tasks. Hum Brain Mapp.14:65--80.

Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J. 1995. A probabilisticatlas of the human brain: theory and rationale for its development.The International Consortium for Brain Mapping (ICBM). Neuro-image. 2:89--101.

Meyer-Lindenberg A, Ziemann U, Hajak G, Cohen L, Berman KF. 2002.Transitions between dynamical states of differing stability in thehuman brain. Proc Natl Acad Sci U S A. 99:10948--10953.

Mormann F, Lehnertz K, Davis P, Elger CE. 2000. Mean phasecoherence as a measure for phase synchronization and itsapplications to the EEG of epilepsy patients. Physica D. 144:358--369.

Myers LJ, Lowery M, O’Malley M, Vaughan CL, Heneghan C, Gibson ASC,Harley YXR, Sreenivasan R. 2003. Rectification and non-linear pre-processing of EMG signals for cortico-muscular analysis. J NeurosciMethods. 124:157--165.

Nichols TE, Holmes AP. 2002. Nonparametric permutation tests forfunctional neuroimaging: a primer with examples. Hum Brain Mapp.15:1--25.

Peper CE, Beek PJ, van Wieringen PC. 1995. Frequency-induced phasetransitions in bimanual tapping. Biol Cybern. 73:301--309.

Pfurtscheller G, Berghold A. 1989. Patterns of cortical activation duringplanning of voluntary movement. Electroencephalogr Clin Neuro-physiol. 72:250--258.

Pfurtscheller G, da Silva FHL. 1999. Event-related EEG/MEG synchro-nization and desynchronization: basic principles. Clin Neurophysiol.110:1842--1857.

Ritter P, Moosmann M, Villringer A. 2009. Rolandic alpha and beta EEGrhythms’ strengths are inversely related to fMRI--BOLD signal inprimary somatosensory and motor cortex. Hum Brain Mapp.30:1168--1187.

Rizzo V, Siebner HS, Morgante F, Mastroeni C, Girlanda P, Quartarone A.2009. Paired associative stimulation of left and right human motorcortex shapes interhemispheric motor inhibition based on a Heb-bian mechanism. Cereb Cortex. 19:907--915.

Serrien DJ, Brown P. 2002. The functional role of interhemisphericsynchronization in the control of bimanual timing tasks. Exp BrainRes. 147:268--272.

Shibasaki H, Nagae K. 1984. Mirror movement: application ofmovement-related cortical potentials. Ann Neurol. 15:299--302.

Swinnen SP. 2002. Intermanual coordination: from behavioural princi-ples to neural-network interactions. Nat Rev Neurosci. 3:348--359.

Toma K, Mima T, Matsuoka T, Gerloff C, Ohnishi T, Koshy B, Andres F,Hallett M. 2002. Movement rate effect on activation and functionalcoupling of motor cortical areas. J Neurophysiol. 88:3377--3385.

Vrba J, Robinson SE. 2001. Signal processing in magnetoencephalog-raphy. Methods. 25:249--271.


at Vrije U



ownloaded from

spectral changes of interhemispheric crosstalk during movement instabilities

Documents