Modulations in oscillatory activity with amplitudeasymmetry can produce cognitively relevant event-related responsesHanneke van Dijka,b,1, Jurrian van der Werfa, Ali Mazaheria,c, W. Pieter Medendorpa, and Ole Jensena

aDonders Institute for Brain, Cognition, and Behaviour, Radboud University Nijmegen, NL-6500 HE Nijmegen, The Netherlands; bInstitute for ClinicalNeuroscience and Medical Psychology, Heinrich-Heine University, D-40225 Düsseldorf, Germany; and cCentre for Mind and Brain, University of California,Davis, CA 95618

Edited by Nancy J. Kopell, Boston University, Boston, MA, and approved November 18, 2009 (received for review September 16, 2009)

Event-related responses and oscillatory activity are typicallyregarded as manifestations of different neural processes. Recentwork has nevertheless revealed a mechanism bywhich slow event-related responses are created as a direct consequence of modula-tions in brain oscillations with nonsinusoidal properties. It remainsunknown if this mechanism applies to cognitively relevant event-related responses. Here, we investigated whether sustainedevent-related fields (ERFs) measured during working memorymaintenance can be explained by modulations in oscillatorypower. In particular, we focused on contralateral delayed activity(CDA) typically observed in working memory tasks in whichhemifield specific attention is manipulated. Using magnetoence-phalography, we observed sustained posterior ERFs followingthe presentation of the memory target. These ERFs were sys-tematically lateralized with respect to the hemisphere in whichthe target was presented. A strikingly similar pattern emergedfor modulations in alpha (9–13 Hz) power. The alpha powerand ERF lateralization were strongly correlated over subjects.Based on a mechanistic argument pertaining to the nonsinusoidalproperties of the alpha activity, we conclude that the ERFsmodulated by working memory are likely to be directly producedby the modulations in oscillatory alpha activity. Given that pos-terior alpha activity typically reflects disengagement, we concludethat the CDA is not attributable to an additive process reflectingmemory maintenance per se but, rather, is a consequence of howattentional resources are allocated.

cognition | evoked-responses | electroencephalography | magnetoencepha-lography | oscillations

Noninvasive electrophysiological recordings in the humanbrain are typically carried out by using either electroenceph-

alography (EEG) or magnetoencephalography (MEG). In cogni-tive studies, data from these techniques are studied by means ofevent-related responses or bymodulations in oscillatory power (1).Event-related responses are recorded by averaging 30 or moretrials time-locked to a given stimulus, assuming that oscillatorybrain activity that is not phase-locked to the stimulus is “averagedout.” To determine task-related modulations in oscillatory activ-ity, the power is calculated for each trial separately and thenaveraged. This allows for quantification of stimulus-inducedchanges in oscillatory activity. The relation between event-relatedresponses and oscillatory activity remains a fundamental questionin human electrophysiological research (1).Recently, it was reported that human oscillatory activity in the

alpha band has “amplitude fluctuation asymmetry” (AFA) (2, 3).This refers to a nonsinusoidal property implying that amplitudechanges are reflected stronger in the peaks than in the troughs ofthe ongoing oscillations (or vice versa). As a consequence, thesignal distribution is skewed such that the mean covaries withmagnitude. This finding challenges the dogma that oscillations areaveragedoutwhen event-related responses are calculated. Indeed,Mazaheri and Jensen (3) demonstrated empirically that stimulus-

induced asymmetrical amplitude modulations of alpha band os-cillations could produce slow event-related responses. Althoughthisfinding provided a proof-of-principle, we test in this study if theproposed mechanism can explain the emergence of event-relatedresponses reflecting cognitive processing.Event-related potentials (ERPs) with a slow time course have

been reported to be modulated by a wide range of cognitive tasks(4). Although there are a few proposals for how slow responsesare generated, the underlying physiological mechanism remainslargely untested (5). Research on visuospatial working memoryhas revealed important slow sustained event-related responsesthat reflect memory maintenance and load (6). These responsesare lateralized depending on whether the remembered stimulus ispresented in the left or right hemifield and called contralateraldelayed activity (CDA). Other studies have shown hemifield-specific lateralized activity in the alpha band in both attention andworking memory tasks. In these tasks, the alpha power is sup-pressed in the hemisphere contralateral to the attended stimulus(7–9). Here, we hypothesize that such lateralizations are themechanism underlying sustained event-related responses becauseof the fact that the alpha activity during retention has AFA. Forexample, when the stimulus is presented in the left hemifield,there is a slight increase in alpha activity in a left posterior sensor(Fig. 1A). When the stimulus is presented to the right, the alphareduction is strong (Fig. 1B). Given that only the peaks of theoscillations are suppressed, whereas the troughs remain un-changed, the resulting sustained evoked responses would show astronger magnitude for right compared with left stimuli (Fig. 1 Cand D). To test this hypothesis, we analyzed MEG data from aspatial working memory task in which subjects remembered thelocation of stimuli presented in the left or right visual field. Theevent-related fields (ERFs), modulations in alpha power, andamplitude asymmetry were characterized during the memory in-terval. We found that both alpha power modulation and ampli-tude asymmetry correlated with the ERFs. This demonstrates acase in which a cognitively relevant evoked response, the CDA, islikely to be produced by changes in oscillatory brain activity.

ResultsTo study the relation between posterior alpha activity and ERFs,we conducted a standard working memory experiment in whichsubjects had to memorize a spatial target location for a short

Author contributions: J.v.d.W., W.P.M., and O.J. designed research; J.v.d.W. performedresearch. H.v.D., A.M., and O.J. contributed new reagents/analytic tools; J.v.d.W., H.v.D.analyzed data; and H.v.D. and O.J. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed at: Institute for Clinical Neuroscience andMedical Psychology, Heinrich-Heine University, Universitätsstr. 1, D-40225 Düsseldorf,Germany. E-mail: hanneke.vandijk@med.uni-duesseldorf.de.

This article contains supporting information online at www.pnas.org/cgi/content/full/0908821107/DCSupplemental.

900–905 | PNAS | January 12, 2010 | vol. 107 | no. 2 www.pnas.org/cgi/doi/10.1073/pnas.0908821107

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

period of time (10). We recorded the electrophysiological signalsusing MEG from 18 subjects performing this task. The targetstimulus was presented in either the left or right hemifield. After1.6 s, subjects were instructed to make a saccade toward the re-membered location (Fig. 2).

ERFs Are Lateralized by Working Memory. Based on the electro-oculogram (EOG) recordings, we selected the trials in whichsubjects followed the task instructions (i.e., saccades were madecorrectly after the target stimuli). Reaction times for the saccadeswere 212 ± 72 ms (SEM). Initially, we characterized the ERF ofthe combined planar gradient in response to the target stimuli.The grand average demonstrated strong sustained ERFs in pos-terior sensors (Fig. 3). In sensors overlying the left hemisphere,the magnitude of the fields was higher for targets presented in theleft hemifield than in the right hemifield (Fig. 3A). The reversewas observed in sensors overlying the right hemisphere (Fig. 3B).The topography of the lateralized ERFs revealed a significantdifference in left (increase) and right (decrease) parietooccipitalsensors (P= 0.016 and P= 0.022, respectively; Fig. 3C). Thus, wefound that our paradigm induced sustained working memory-related ERFs, which are lateralized with respect to the pre-sentation of the target—the CDA.

Alpha Band Power Is Lateralized by Working Memory. Next, weinvestigated modulations in oscillatory power with respect to thetarget stimuli. Time-frequency analysis of power revealed a strongmodulation in alpha activity over parietooccipital sensors duringthe delay period (Fig. 4). When the target was presented in theleft hemifield, the left sensors showed an increase in alpha band

power, whereas the right sensors showed a decrease (Fig. 4A).The reverse pattern was observed when the target was presentedto the right (Fig. 4B). Again, we defined laterality as the differ-ence between power for right and left hemifield stimuli (Fig. 4Cand Fig. S1). We found a posterior distribution with significantpower enhancements in the ipsilateral hemisphere (left cluster:P = 0.033, right cluster: P = 0.007). These results demonstrate astrong lateralization in alpha band activity with respect to thehemifield of the target stimulus.

Lateralizations of the ERFs and Alpha Activity Are Highly Correlated.The sustained ERFs and the alpha power enhancements are bothlateralized during the retention of a visuospatial memory item,with strikingly similar scalp topographies (compare Fig. 3C andFig. 4D). Additionally, the sustained ERFs and alpha powermodulations seemed to follow a similar time course (compare Fig.3A and Fig. S2). We quantified the relation between modulationsin ERF amplitudes and alpha power by a correlation analysis (seeExperimental Procedures). This revealed a highly significant cor-relation over subjects between ERFs and alpha powermodulationover posterior areas (Fig. 5A; r=0.92, P< 10-6). Subsequently, wecompared the time course of the alpha band and ERF amplitudefor the entire period of−0.2 to 1.2 s by correlating the values for 15time points in all subjects (see Experimental Procedures). Thecorrelation was highly significant in representative left and rightsensors (P < 10-6; Fig. 5B, Left and Right). The strong correlationbetween lateralized sustained ERFs and lateralized alpha bandpower strongly suggests that they reflect the same mechanism.

Lateralized Amplitude Asymmetry. We then proceeded to quantifythe degree of AFA in the alpha band by calculating the AFAindexduring retention. This measure estimates the amplitude modu-lations of the peaks of oscillatory activity relative to the modu-lations of the troughs (seeExperimental Procedures). Note that thismeasure is independent of the dc offset of the signal (3) (Fig. 2).Asshown in Fig. 6A, the AFAindex spectra have strong peaks in thealpha and beta bands over posterior sensors. This clearly demon-strates that AFA is present in the alpha band during the retentionof the spatial target. The alpha band AFAindex was strongest in leftsensors when the target was presented in the left hemifield,whereas the reverse pattern was observed in right sensors (Fig.6A). This was confirmed by the topographies of the AFAindex val-ues (Fig. 6B and Fig. S1C). For the difference between the twoconditions (left minus right target), the AFAindex shows a lateral-ized posterior parietooccipital distribution (left cluster: P=0.012,right cluster: P = 0.002; Fig. 6B) similar to both the alpha bandoscillatory activity (Fig. 4D) and the ERFs (Fig. 3C).

Amplitude Asymmetry Correlates with the Production of ERFs.Finally, we quantified the correlation between the AFAindex andERF modulations over subjects. Fig. 7 A and C shows a correla-tion for left and right sensors (MLO22: r = 0.63, P = 0.008;MRO33: r = 0.47, P = 0.046). The topography of the correlationcoefficients revealed a posterior distribution (Fig. 7B). Note thatthe sensors in which the correlations are significant correspond to

t = 0 t = 0

Left stim Right stim

t = 0

Left stimRight stim

A B

Left sensor

C

D

Fig. 1. The working hypothesis explaining how slow ERFs are created bymodulations in alpha power with asymmetrical amplitude properties. (A)Target stimulus is presented in the left hemifield at t = 0 s. Only MEG signalsmeasured over the left hemisphere are considered. Alpha activity is slightlyincreasedwith respect to the stimulus onset overmultiple trials. (B) Stimulus ispresented in the right hemifield, resulting in strong alpha suppression overmultiple trials. (C) ERFs. Because of the asymmetrical amplitude modulations,the oscillatory alpha activity is not averaged out; only the peaks of the os-cillations are modulated and not the troughs. As a consequence, the sys-tematic modulation in alpha power produces a slow sustained field. (D) Thestronger the modulation, the stronger is the magnitude of the field. In short,we hypothesized that working memory retention resulting in lateralized al-pha power modulation would produce sustained lateralized ERFs—the CDA.

Fig. 2. Description of the working memory task. After a baseline period of1.5 s, a visual stimulus (○) was presented in the peripheral left or righthemifield for 0.1 s. Subjects were instructed to remember the position of thevisual target. At 1.6 s after stimulus onset, the fixation cross (+) disappearedand the subjects were instructed to make a saccade (arrow) toward the re-membered target position.

van Dijk et al. PNAS | January 12, 2010 | vol. 107 | no. 2 | 901

NEU

ROSC

IENCE

PSYC

HOLO

GICALAND

COGNITIVESC

IENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

the group of sensors in which there was the strongest correlationbetween alpha and ERF lateralization (Fig. 5A). The correlationbetween sustained lateralized ERFs and the amplitude asymme-try even further supports the case that the ERFs are produced bymodulations in alpha band power.Considering the additional peak of the AFAindex in the beta band

(15–25Hz;Fig. 6A),weexplored the relationbetweenERFsandbetapower modulations (Fig. S3). We did identify systematic changes inbeta lateralizationwith regard to thehemifieldof the target; however,themagnitude ofAFAindexmodulationwasmuch less than that of thealpha band (Fig. S3B). The beta band activity could potentially beexplained as a harmonic frequency resulting from the alpha bandactivity. While such interactions can contribute to the AFAindex theywould not produce ERFs (11). Using a partial correlation analysis,we found no correlation between beta band power and ERF later-alization when controlling for the alpha band power (Fig. S3C).However, the correlation between alpha band power and ERF lat-eralization when controlling for the beta power (Fig. S3B) remained.We conclude that the lateralization in ERFs is primarily caused bypower modulations in the alpha band and is not likely explained byeither harmonics or physiological effects in the beta band.

DiscussionIn this study, we demonstrated that sustained ERFs related toworking memory maintenance are likely to be explained bymodulations of oscillatory activity in the alpha band. We identi-fied CDA in posterior ERFs with respect to target stimuli pre-sented in the left or right hemifield, which was strongly correlated(spatially and temporally) with hemispheric lateralization ofalpha band power. A measure of the amplitude asymmetry(AFAindex) revealed AFA in the alpha band over posterior areas.

The AFAindex correlated with ERF lateralization as well. It waspreviously proposed that systematic modulations in oscillatoryactivity with amplitude asymmetry constitute a mechanism forgenerating ERFs/ERPs (2, 3). We have now found even strongersupport for this mechanism and demonstrate that it can accountfor the generation of cognitively relevant ERFs.

Sustained ERFs Are Likely to Reflect Functional Inhibition. Althoughthe results of the present study strongly suggest that the measuredslow ERFs are produced by modulations in oscillatory power, thisdoes not bring into question the results of previous ERP/ERFresearch. Rather, our findings provide a unique interpretation forthe function of slow ERFs/ERPs. Conventionally, ERFs or ERPsare seen as additive brain responses that reflect engagement ofworking memory networks. Our findings suggest that the slowERFs/ERPs reflect a modulation in background oscillatory power,predominantly in the alpha band. Alpha activity has recently beeninterpreted to reflect functional disengagement of task-irrelevantregions, thereby allocating computational resources to task-relevant regions (12, 13). This is based on the observation that alphaactivity increases with working memory load during retention in

ER

F (fT

/cm

)

Left sensors

Time (s)0 1

Left stimRight stim

0

-0.08

Left stim - Right stim0.08

Time (s)0 1

ERF

(fT/c

m)

0

0 Time (s)1

Left stim - Right stim0.08

-0.08

Time (s)0

Right sensors

1

Left stimRight stim

0

0.1

0

0.1

P = 0.016

-0.1 0.1ERF

(fT/cm)

Left stim - Right stim

P = 0.022

C

A

B

Fig. 3. ERFs during the retention interval. (A) Grand average ERFs for targetstimuli (stim) presented in the left (blue) and right (red) hemifields. The ERFswere averaged for the left sensors, showing a significant difference betweenleft and right target stimuli. (B) Difference in ERFs (left minus right stimulus)for the left and right significant sensors. (C) Topography of the difference inERFs during the retention period (0.2–1.2 s). Clusters of sensors significantlydifferent are marked by dots (Left: P = 0.016, Right: P = 0.022).

A

B

C

D

P P

Fig. 4. Time-frequency representations (TFRs) of power during the retentioninterval. (A) TFRs for the left target stimuli (stim). Data are shown for left andright sensors (LeftandRight, respectively) asmarked inD. (B) TFRs for right targetstimuli. Data are shown for left and right sensors (Left and Right, respectively) asmarked in D. (C) Difference in TFRs (left minus right stimuli) for left and rightsensors (LeftandRight, respectively). (D) Topographyof thealphaband (9–13Hz,0.2–1.2 s)powerwith respect to leftminus right target stimuli. Thedotsmarktwoclusters of sensors with significant differences (Left: P = 0.0333, Right: P = 0.007).

902 | www.pnas.org/cgi/doi/10.1073/pnas.0908821107 van Dijk et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

areas not required for working memory maintenance (13, 14).Furthermore, as also demonstrated in this study, alpha band powerincreases in the hemisphere that is less relevant for the task at hand(7, 9). Intracranial animal recordings have demonstrated that neu-ronalfiring is constrained toa fractionof thealphacycle (15, 16).Weconclude that the slow ERFs producing the CDA observed in thecurrent study are reflecting functional inhibition/disengagementandnot an activeworkingmemory process. The same interpretationmighthold for the spatialworkingmemory-related lateralizedERPsreported by Vogel and Machizawa (6) and Vogel et al. (17). Inter-estingly,Drew andVogel (18) recently observed a very similar CDAduring a spatial tracking task. These findings support the notionthat CDA reflects a general mechanism for allocation of resourcesrather than memory maintenance per se.

Physiological Mechanism. Previously, it has been proposed that theevoked response was generated either by an additive mechanismor by phase resetting of ongoing oscillations (1, 19, 20). Themechanism we here embrace explains ERFs by modulations inpower of oscillatory activity that has amplitude asymmetry. Thephysiological mechanism proposed by Mazaheri and Jensen (3)explains asymmetrical oscillations based on the accepted notionthat themagneticfieldsmeasured byMEGare producedmainly byintracellular dendritic currents in pyramidal cells (21). The keyargument is that the dominating currents measured by MEG aredendritic, running from the distal synapses to the soma (i.e., in onedirection). The measured alpha band activity might therefore be aconsequence of bouts of synaptic activity producing inward den-dritic currents every 100 ms. It is the modulation of the magnitudeof these bouts that produces the AFA, which yields the CDA.Thus, the current findings provide important insight into thephysiological substrate of the CDA.

Do Power Modulations of Oscillatory Activity with AmplitudeAsymmetry Causally Predict the Emergence of Slow EvokedComponents? One might ask if the generation of slow evokedresponses is a causal consequence of modulations in oscillatoryactivity or if it couldbeexplainedbya comodulationofa thirdorigin.First, it should be stressed that systematic modulations in oscil-lations with amplitude asymmetry inevitably will result in the gen-

eration of evoked responses (Fig. 1). Because the oscillatory activitywe observe has amplitude asymmetry and is systematically modu-lated, slow evoked components are bound to emerge. Second, thecorrelationover subjects between lateralization inevokedresponsesand alpha activity was very strong (r= 0.92, P < 10-6; Fig. 5). Thus,the simplest explanation is that the slow lateralized ERFs are a di-rect consequence of oscillatorymodulations in the alpha band. Thiscausal explanation could be tested by pharmacological manipu-lations modulating the posterior alpha activity (22). The laterali-zation in alpha activity is most likely a consequence of a slow top-down drive involving regions in the intraparietal sulcus and frontaleye field (23). This drive might be produced by the entrainment ofslow delta oscillations determining the excitability of parietoocci-pital regions in relation to the components of the task (24, 25). Inthis study, we did not investigate the electrophysiological correlateof such a top-down drive beyond the alpha power modulation.

Does the Mechanism Generalize to Other Frequency Bands orCognitive Functions? Although the current study has focused onamplitude asymmetry of posterior alpha activity, the phenomenonmight generalize toother brain regionsand frequency bands (3). Forinstance, amplitude asymmetry of the ∼10-Hz somatosensory murhythm has been reported (2). Other candidates could be theta os-cillations [5–9Hz,working and long-termmemory (11, 26–28)], beta

?

A

BP

P

P

PP

P

Fig. 5. Relation between the sustained ERF and the modulation in alphapower. (A) Correlation over subjects of differences in ERFs and alpha powermodulations [left minus right target stimuli (stim)] for a left sensor and a rightsensor (Left and Right, respectively). The correlations were highly significant(Left, Right: P < 0.0001; Spearman test, Bonferroni-corrected with respect tonumber of sensors). The topography (Middle) of the correlation coefficients. (B)Correlation over subjects between the time course-resolved ERFs and alphapower modulations (left minus right target stimuli) for a left sensor and a rightsensor (LeftandRight, respectively). The topography (Middle) of the correlationbetween the difference in time courses for ERFs and alpha power modulation.

A

Frequency (Hz)

AFA

index

Left stimRight stim

20 30 4010

0.06

0.04

0.02

MLO11

Frequency (Hz)10 20 30 40

Left stimRight stim0.06

0.04

0.02AFA

index

MRO12

-0.04 0.04AFAindex

P = 0.012 P = 0.002

Left stim - Right stimB

Fig. 6. Amplitude asymmetry calculated during the retention interval (0.2–1.2 s). (A) Frequency spectra of the AFAindex for the left (blue) and right (red)targets for a left posterior sensor (MLO11) with maximum 10-Hz AFAindex anda right sensor (MRO12). (B) Topography of the AFAindex in the alpha band forthe difference between the conditions [left minus right target stimuli (stim)].The dots depict the clusters of sensors with significant differences (Left: P =0.012, Right: P = 0.002). These sensors were used for the spectra in B.

0

0.04

0

MLO22

-0.06 0.08Δ AFAindex

r = 0.63p = 0.008

0 0.8Correlation

Δ E

RF

(fT/c

m)

Δ AFAindex0 0.04-0.04

0

0.06

-0.1

r = 0.47p = 0.046

Δ E

RF

(fT/c

m)

MRO33A B C

Fig. 7. Correlation over subjects between the differences (left minus rightstimuli in sustained ERFs and the differences in absolute alpha amplitudeasymmetry (AFAindex) during retention. Correlations for a left sensor (A) and aright sensor (C) (Left: r = 0.63, P < 0.008; Right: r = 0.47, P < 0.046). (B) Top-ography of the correlation coefficients.

van Dijk et al. PNAS | January 12, 2010 | vol. 107 | no. 2 | 903

NEU

ROSC

IENCE

PSYC

HOLO

GICALAND

COGNITIVESC

IENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

band oscillations [15–25Hz,motor (29) and language (30) tasks], orgamma band activity [40–200 Hz, visual attention (31, 32) andworking memory (8, 14, 33)]. Future work is required to investigatewhich frequency bands have oscillatory activity with amplitudeasymmetry that might explain the generation of evoked responses.Many of the reported amplitude modulations in ERPs/ERFs in

cognitiveparadigmsareassociatedwith slowevokedcomponents.Thisis the case for long-term memory encoding (34) and retrieval (35). Inthedomainof language research, slowERPeffects have been found inrelation to syntactic violations (36, 37). In motor research, slow ERPsincreasing in anticipation of movements have been reported (38). Itwould be of great interest to investigate to what extent modulations ofoscillatory power can account for some of these phenomena.

ConclusionWe have shown that modulations of oscillatory power with ampli-tude asymmetry in the alpha band can produce cognitively modu-lated ERFs in a working memory task. Our findings point to ageneral mechanism for producing evoked responses that is likely toextendbeyond the realmofworkingmemoryandalphabandactivity.

Experimental ProceduresThedatausedinthisstudyhavebeenpublishedpreviously (8)butforadifferentpurpose. Here, we mainly describe the experimental details relevant for thecurrent analysis.

Participants. Eighteenhealthy subjects (3 femaleand15male,meanage: 26±3years) free of any known sensory, perceptual, or motor disorders volunteeredto participate in the experiment. All subjects provided written informedconsent according to institutional guidelines of the local ethics committee(CMO Committee on Research Involving Human Subjects, region Arnhem-Nijmegen, The Netherlands).

Stimuli and Design. Visual stimuli were generated with Presentation 9.10software (Neurobehavioral Systems, Inc.), presented using a liquid crystal dis-play video projector (SANYO PROxtraX multiverse; 60-Hz refresh rate), andback-projected onto the screen using two front-silveredmirrors. Subjectswereinstructedtofixateonacentrallypresentedwhitecrossonablack screenduringa period of 1.5 s. As a target, a peripheral white dot was presented for 0.1 s tothe left or right of the visual fixation cross at a random visual eccentricitybetween 9° and 18° and restricted between a −36° and 36° elevation anglerelative to the central fixation cross. After a delay of 1.6 s, the fixation crossdisappeared, which instructed the subjects to make a saccade toward the re-membered location of the target (Fig. 2). The fixation cross appeared againafter 0.3 s as an indication for the subject to fixate at the center of the screenuntil the end of the trial. Each trial lasted 4 s.

Data Recording and Analysis. Data were recorded using a whole-head MEGsystem (151 axial gradiometers; VSM/CTF Systems). Subjects were seatedupright in the MEG system and instructed to sit comfortably without moving.Before and after the recording session, the head position was measured withrespect to the MEG sensor array using localization coils fixed at anatomicallandmarks (the nasion and at the left and right ear canals). Bipolar EOGs wererecorded to check subjects’ gaze behavior and discard trials with eye blinksoffline. MEG and EOG data were low-pass-filtered at 300 Hz and digitized at1,200 Hz. Before the actual measurements, the EOG signal was calibratedusing a nine-point calibration grid. Eyemovement recordings in all 18 subjectsconfirmed that they performed the task correctly. Trials in which subjectsbroke fixation, made saccades in the wrong direction, or blinked the eyesduring the trial were excluded from further analysis. Additionally, trials con-taminated with muscle activity or sensor jumps were excluded from furtheranalysis. In total, we included 120 ± 18 trials for left hemifield targets and 118± 20 trials for right hemifield targets in the analysis.

ForERFanalysis, thedatawere low-pass-filteredat80Hz,anda0.2-sbaselineperiod prior to target onset was used. Subsequently, the trials were averagedfor each condition separately.

Oscillatory power was estimated using a multitaper spectral estimationmethod (40). A frequency-dependent sliding time window was applied. The

lengthof thewindowwasfive cycles (i.e.,ΔT=5/fo,where fo is the frequencyofinterest). The data from each sliding time windowwere multiplied with threeorthogonal Hanning tapers. Subsequently, the data were Fourier-trans-formed and the power spectral densities were averaged. This procedure re-sulted in estimates of oscillatory power with plus or minus ∼0.3-Hz fofrequency smoothing. Thedatawere thenbaseline-correctedusingdata in thetime window centered 0.25 s before target onset (i.e., no posttarget 10-Hzpower was bleeding into the baseline).

Sustained fields and alpha band activity in the delay period (0.2–1.2 s) werecompared statistically for the left and right target conditions. Tonormalize forintersubject variances, we computed the z values for each sensor with respectto left and right targets. This was done by initially comparing the conditionswithin subjects and calculating t values, which then were transformed into zvalues (using SPM2; the FIL methods group. These z values represented thenormalized difference between left and right targets (and not statistical val-ues). The z values were then tested against the null distribution over subjectsusing a nonparametric cluster-randomization routine in Fieldtrip (41). Thisroutine controls the type 1 error rate in a situation involving multiple com-parisons over sensors (see SI Methods).

The correlation between ERF and alpha band activity in the delay periodwasassessed by subtracting the delay activity with respect to right and left targets.The Spearman correlation was computed over subjects for each sensor. Sensorswith a P value smaller then 0.0003 were accepted as being significantly cor-related (Bonferroni-corrected for 151 comparisons). Additionally, we calcu-lated the correlation of the time evolution of the alpha power and the timeevolution of the ERF amplitude. This yielded 15 alpha power values and 15 ERFamplitude values for each subject. The alpha band values of all subjects wereconcatenated into one pool, which then contained 270 power values for eachsensor. We performed the same analysis for the ERF amplitude values andthen correlated the amplitude values as described previously.

To investigate whether the sustained ERF in the delay period could be a con-sequenceof theamplitudeasymmetry in theongoingoscillations,wedeterminedthe AFAindex as proposed byMazaheri and Jensen (3). To compute this index, thedatawere initially bandpass-filtered in the full trials at the frequencyof interest±1Hz.Thetimepointsof thepeaksandtroughsof theoscillatoryactivitywere thenidentified in thefiltered data (0.2–1.2 s). These timepointswere used to estimatethe amplitude values of the peaks (Speak) and troughs (Strough) in the non-band-pass-filtereddataafterapplyinga10-msboxcarsmoothingkernel. Theamplitudeasymmetry was estimated by quantifying the normalized difference in the var-iance for the peaks and troughs [as described by Mazaheri and Jensen (3)]:

AFAindex ¼�var

�Speak

�− var

�Strough

����var

�Speak

�þ var�Strough

��

An AFAindex close to zero would mean that the peaks and troughs aremodulated similarly (as for a sinusoidal “symmetrical” signal). When theAFAindex is larger than zero, this would mean that the peaks are modulatedstronger than the troughs, and vice versa for negative values. The AFAindex

spectrumwas calculated for the rangebetween5and50Hz (in 1-Hz steps). Thedifference in the AFAindex for 10 Hz between the left and right target con-ditions was statistically evaluated using a cluster-randomization routine sim-ilar to that described previously.

When analyzing the ERFs, power spectra, and AFAindex, we computed aplanar gradiometer representation of the data (42). The planar field gradientapproximates the signals measured by planar gradiometers. This is often ad-vantageous in MEG signal processing, because the strongest field of theplanar gradient usually is situated above the neural sources (43). The verticaland horizontal components were estimated using the fields of each sensorand its neighboring sensors. The resulting components were then combinedusing the rms after spectral or ERF analysis. By means of this approach, thestrongest signal will be directly above the source. However, information aboutsource orientation will be lost.

The Matlab package Fieldtrip was used for data analysis. This is an open-source toolbox for neurophysiological data analysis, which was developed atthe Donders Institute for Brain, Cognition, and Behaviour.

ACKNOWLEDGMENTS. We thank Ingrid Nieuwenhuis for providing helpwiththe statistical analysis of the ERF and frequency data. Thisworkwas supportedby Volkswagen Foundation Grant I/79876; a Netherlands Organization forScientific Research Rubicon scholarship; and theNetherlandsOrganization forScientific Research Grants 864.03.007, 400.04.186, and 452/03.307.

1. Makeig S, Debener S, Onton J, Delorme A (2004) Mining event-related brain dynamics.

Trends Cogn Sci 8:204–210.

2. Nikulin VV, et al. (2007) A novel mechanism for evoked responses in the human brain.

Eur J Neurosci 25:3146–3154.

904 | www.pnas.org/cgi/doi/10.1073/pnas.0908821107 van Dijk et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

3. Mazaheri A, Jensen O (2008) Asymmetric amplitude modulations of brain oscillationsgenerate slow evoked responses. J Neurosci 28:7781–7787.

4. Niedermayer E, Lopes Da Silva F (2004) Electroencephalography: Basic Prin-ciples, Clinical Applications and Related Fields. (Williams & Wilkins, Balti-more) 5th Ed.

5. Birbaumer N, Elbert T, Canavan AG, Rockstroh B (1990) Slow potentials of the cerebralcortex and behavior. Physiol Rev 70 (1):1–41.

6. Vogel EK, Machizawa MG (2004) Neural activity predicts individual differences invisual working memory capacity. Nature 428:748–751.

7. Medendorp WP, et al. (2007) Oscillatory activity in human parietal and occipital cortexshows hemispheric lateralization and memory effects in a delayed double-step saccadetask. Cereb Cortex 17:2364–2374.

8. VanDerWerf J, JensenO, Fries P,MedendorpWP (2008)Gamma-bandactivity inhumanposterior parietal cortex encodes the motor goal during delayed prosaccades andantisaccades. J Neurosci 28:8397–8405.

9. Worden MS, Foxe JJ, Wang N, Simpson GV (2000) Anticipatory biasing of visuospatialattention indexed by retinotopically specific alpha-band electroencephalographyincreases over occipital cortex. J Neurosci 20:RC63.

10. Goldman-Rakic PS (1996) Regional and cellular fractionation of working memory.Proc Natl Acad Sci USA 93:13473–13480.

11. Nikulin VV, Linkenkaer-Hansen K, Nolte G, Curio G (2009) Non-zero mean andasymmetry of neuronal oscillations have different implications for evoked responses.Clin Neurophysiol, [Epub ahead of print doi:10.1016/j.clinph.2009.09.028].

12. Klimesch W (1999) EEG alpha and theta oscillations reflect cognitive and memoryperformance: A review and analysis. Brain Res Brain Res Rev 29:169–195.

13. Palva S, Palva JM (2007) New vistas for alpha-frequency band oscillations. TrendsNeurosci 30:150–158.

14. Jensen O, Gelfand J, Kounios J, Lisman JE (2002) Oscillations in the alpha band (9-12Hz) increase with memory load during retention in a short-term memory task. CerebCortex 12:877–882.

15. Jokisch D, Jensen O (2007) Modulation of gamma and alpha activity during a workingmemory task engaging the dorsal or ventral stream. J Neurosci 27:3244–3251.

16. Bollimunta A, Chen Y, Schroeder CE, Ding M (2008) Neuronal mechanisms of corticalalpha oscillations in awake-behaving macaques. J Neurosci 28:9976–9988.

17. Lorincz ML, Kékesi KA, Juhász G, Crunelli V, Hughes SW (2009) Temporal framing ofthalamic relay-mode firing by phasic inhibition during the alpha rhythm. Neuron 63:683–696.

18. Vogel EK, McCollough AW, Machizawa MG (2005) Neural measures reveal individualdifferences in controlling access to working memory. Nature 438:500–503.

19. Drew T, Vogel EK (2008) Neural measures of individual differences in selecting andtracking multiple moving objects. J Neurosci 28:4183–4191.

20. Shah AS, et al. (2004) Neural dynamics and the fundamental mechanisms of event-related brain potentials. Cereb Cortex 14:476–483.

21. Penney TB, Maess B, Busch N, Derrfuss J, Mecklinger A (2003) Cortical activityreduction with stimulus repetition: a whole-head MEG analysis. Brain Res Cogn BrainRes 16:226–231.

22. Hämäläinen M, Hari R, Ilmoniemi RJ, Knuutila J, Lounasmaa OV (1993) Magnetoen-cephalography—Theory, instrumentation, and applications to noninvasive studies ofthe working human brain. Rev Mod Phys 65:413–497.

23. Ahveninen J, et al. (2007) MRI-constrained spectral imaging of benzodiazepinemodulation of spontaneous neuromagnetic activity in human cortex. NeuroImage 35:577–582.

24. Capotosto P, Babiloni C, Romani GL, Corbetta M (2009) Frontoparietal cortex controlsspatial attention through modulation of anticipatory alpha rhythms. J Neurosci 29:5863–5872.

25. Lakatos P, Karmos G, Mehta AD, Ulbert I, Schroeder CE (2008) Entrainment ofneuronal oscillations as a mechanism of attentional selection. Science 320:110–113.

26. Schroeder CE, Lakatos P (2009) Low-frequency neuronal oscillations as instruments ofsensory selection. Trends Neurosci 32 (1):9–18.

27. Raghavachari S, et al. (2006) Theta oscillations in human cortex during a working-memory task: Evidence for local generators. J Neurophysiol 95:1630–1638.

28. Jensen O, Tesche CD (2002) Frontal theta activity in humans increases with memoryload in a working memory task. Eur J Neurosci 15:1395–1399.

29. Fell J, et al. (2003) Rhinal-hippocampal theta coherence during declarative memoryformation: interaction with gamma synchronization? Eur J Neurosci 17:1082–1088.

30. Hari R, Salmelin R (1997) Human cortical oscillations: A neuromagnetic view throughthe skull. Trends Neurosci 20:44–49.

31. Bastiaansen M, Hagoort P (2006) Oscillatory neuronal dynamics during languagecomprehension. Prog Brain Res 159:179–196.

32. Fell J, Fernández G, Klaver P, Elger CE, Fries P (2003) Is synchronized neuronal gammaactivity relevant for selective attention? Brain Res Brain Res Rev 42:265–272.

33. Fries P, Reynolds JH, Rorie AE, Desimone R (2001) Modulation of oscillatory neuronalsynchronization by selective visual attention. Science 291:1560–1563.

34. Tallon-Baudry C, Bertrand O, Peronnet F, Pernier J (1998) Induced gamma-bandactivity during the delay of a visual short-term memory task in humans. J Neurosci 18:4244–4254.

35. Takashima A, et al. (2006) Successful declarative memory formation is associated withongoing activity during encoding in a distributed neocortical network related toworking memory: A magnetoencephalography study. Neuroscience 139:291–297.

36. Rugg MD, et al. (1998) Dissociation of the neural correlates of implicit and explicitmemory. Nature 392:595–598.

37. Hagoort P, Brown CM (2000) ERP effects of listening to speech compared to reading:The P600/SPS to syntactic violations in spoken sentences and rapid serial visualpresentation. Neuropsychologia 38:1531–1549.

38. van den Brink D, Hagoort P (2004) The influence of semantic and syntactic contextconstraints on lexical selection and integration in spoken-word comprehension asrevealed by ERPs. J Cognit Neurosci 16:1068–1084.

39. Walter WG, Cooper R, Aldridge VJ, McCallum WC, Winter AL (1964) Contingentnegative variation: An electric sign of sensorimotor association and expectancy in thehuman brain. Nature 203:380–384.

40. Percival DB, Walden AT (1993) Spectral analysis for physical applications: multitaperand conventional univariate techniques (Cambridge, UK, Cambridge UP).

41. Maris E, Oostenveld R (2007) Nonparametric statistical testing of EEG- and MEG-data.J Neurosci Methods 164:177–190.

42. Basitiaansen MC, Knosche TR (2000) Tangential derivative mapping of axial MEGapplied to event-related desynchronization research. Clin Neurophysiol 111 (7):1300–1305.

43. Ahonen AI, et al. (1993) Sampling theory for neuromagnetic detector arrays. IEEETrans Biomed Eng 40 (9):859–869.

van Dijk et al. PNAS | January 12, 2010 | vol. 107 | no. 2 | 905

NEU

ROSC

IENCE

PSYC

HOLO

GICALAND

COGNITIVESC

IENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0