www.elsevier.com/locate/ynimg

NeuroImage 39 (2008) 707–716

Influence of ongoing alpha rhythm on the visual evoked potential

Robert Becker,⁎ Petra Ritter,1 and Arno Villringer1

Berlin NeuroImaging Center, Department of Neurology, Charité-Platz 1, 10117 Berlin, Germany

Received 23 May 2007; revised 21 August 2007; accepted 3 September 2007Available online 19 September 2007

The relationship between ongoing occipital alpha rhythm (8–12 Hz)and the generation of visual evoked potentials (VEPs) has beendiscussed controversially. While the “evoked theory” sees no interac-tion between VEP generation and the alpha rhythm, the “oscillatorytheory” (also known as “phase-reset theory”) postulates VEPgeneration to be based on alpha rhythm phase resetting. Previousexperimental results are contradictory, rendering a straightforwardinterpretation difficult. Our approach was to theoretically model theimplications of the evoked and oscillatory theory also incorporatingstimulus-induced alpha-rhythm desynchronization. As a result, themodel based on the oscillatory theory predicts alpha-band dependentVEP amplitudes but constant phase locking. The model based on theevoked theory predicts unaffected VEP amplitudes but alpha-banddependent phase locking. Subsequently, we analyzed experimentaldata in which VEPs were assessed in an “eyes open” and “eyes closed”condition in 17 subjects. For early components of the VEP, findings arein agreement with the evoked theory, i.e. VEP amplitudes remainunaffected and phase locking decreases during periods of high alphaactivity. Late VEP component amplitudes (N175 ms), however, aredependent on pre-stimulus alpha amplitudes. This interaction iscontradictory to the oscillatory theory since this VEP amplitudedifference is not paralleled by a corresponding difference in alpha-band amplitude in the affected time window. In summary, by using amodel-based approach we identified early VEPs to be compatible withthe evoked theory, while results of late VEPs support a modulatory butnot causative role – the latter implied by the oscillatory theory – ofalpha activity for EP generation.© 2007 Elsevier Inc. All rights reserved.

Keywords: EEG; Alpha rhythm; Models; Additive; Phase reset; Interaction

Introduction

The human electroencephalogram (EEG) is dominated byspontaneous rhythms, one of the most prominent being the alpharhythm (8–12 Hz). Evoked potentials (EPs) being of small

⁎ Corresponding author.E-mail address: robert.becker@charite.de (R. Becker).

1 These authors contributed equally.Available online on ScienceDirect (www.sciencedirect.com).

1053-8119/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.neuroimage.2007.09.016

amplitudes are often obscured by such rhythms and can berevealed by averaging several epochs. This effect is reconcilablewith linear summation of ongoing and evoked activity (Arieli et al.,1996), i.e. by an “evoked theory” or “additive theory” of EPgeneration. In this theory, ongoing activity, e.g. the alpha rhythmhas no functional significance for EP generation. Another theory ofEP generation is the “oscillatory theory” also known as “phase-reset theory”, assuming the EP to be generated by alpha rhythmphase resetting (Sayers et al., 1974; Makeig et al., 2002). Althoughboth theories apparently differ in their assumed relationshipbetween ongoing and evoked activity, there is controversy onhow to validate the theories (Makeig et al., 2002; Makinen et al.,2005; Klimesch et al., 2006; Fuentemilla et al., 2006; Mazaheri andJensen, 2006; Hanslmayr et al., 2006). Shah et al. (2004) suggestedtwo criteria for differentiation: Sufficient ongoing alpha amplitudeis required for EP generation in the oscillatory theory, while anevent-related increase in signal power supports the evoked model.However, the conclusion, that in case of no increase or of adecrease in average post-stimulus alpha-band power the oscillatorymodel – i.e. an alpha rhythm phase reset – holds, may bepremature. Such a situation can be caused either by veridical alphaphase resetting but also by a “masking” of an evoked process by adominant but desynchronizing alpha rhythm (Hanslmayr et al.,2006). Without further positive evidence of a functional sig-nificance of the alpha rhythm for EP generation, the oscillatorytheory can neither be discarded nor be proven. While severalstudies on the impact of the ongoing alpha amplitude on theevoked response have been reported (Makeig et al., 2002; Basar,1980; Jasiukaitis and Hakerem, 1988; Rahn and Basar, 1993;Makinen et al., 2005), findings were contradictory or not related tomodel predictions and so far have not led to a general consensus onthe mechanisms of EP generation.

In this study, we combine theoretical modeling and experi-mental data analysis to examine the functional role of the ongoingalpha rhythm for VEP generation.

In theoretical modeling, we evaluate the relationship betweenongoing activity and evoked activity for both theories. Model-specific predictions are derived for varying pre-stimulus alphaamplitudes. Initially, both models contain a common subset of data,where an EP is accompanied by a dominating and desynchronizingalpha rhythm. We aim to demonstrate that by variation of pre-

708 R. Becker et al. / NeuroImage 39 (2008) 707–716

stimulus alpha amplitude both models are clearly differentiatedwith respect to their predictions. Subsequently, in experimentaldata the influence of pre-stimulus alpha amplitude on EPgeneration is investigated. Based on the comparison of theseempirical findings and the simulated EP parameters, the influenceof alpha rhythm on EP generation is defined in the light of the twoopposing models.

Materials and methods

Models

In order to model the evoked and the oscillatory theory of EPgeneration, we simulated EEG single trial activity composed oftwo types of subunit signals as detailed below and shown in Fig. 1.The evoked model assumes “dual generators” for alpha rhythm andEP generation whereas the oscillatory model assumes “sharedgenerators” producing both the alpha rhythm and the evokedresponse (Mazaheri and Jensen, 2006). In a first step, a subset ofdata is generated for each model which assumes an EP beinggenerated during spontaneous alpha activity of slightly largeramplitude than the EP itself. This demarcates the starting point ofour actual approach of alpha-variation to differentiate the twotheories of EP generation.

Fig. 1. Schematic view of the two competing models of EP generation. Thin colorblack colored lines indicate the resulting average. (A–D) Constituents of the evorhythm (A), the evoked response is generated (B). An event-related alpha rhythm dal., 2006) was included into the models (C), resulting in the compound single trials aalpha rhythm (E) as a generator of the EP via phase resetting (F). In the oscillatory mphase reset. The partial phase reset results in a minor fraction of non-phase-lockedalpha, i.e. the EP. The resulting EP amplitude is slightly smaller than the amplitudedifferentiation between the evoked and oscillatory (i.e. alpha-rhythm phase resettinAlso Yeung et al. (2004) raised concerns about the general suitability of previousshowing that surrogate data modeled by an evoked mechanism yielded same resultsof the alpha rhythm.

Evoked modelThe EP was created by a 10-Hz sine wave lasting for 100 ms

with constant amplitude of 7.5 μV (Fig. 1B). The EP isindependent from the simultaneously occurring 10 Hz simulatedalpha rhythm which amounted to 10 μV (further referred to as“medium” alpha group) and exhibited a random phase in each trial(superimposed single trials are shown in Fig. 1A). A simple event-related desynchronization of the alpha rhythm (alpha-ERD;Pfurtscheller and Aranibar, 1977) was integrated into the model.Alpha ERD strongly depends on the type of visual stimulation andcan vary from study to study (typical alpha ERD can vary from20% to 90%, e.g. see Klimesch et al., 2004, but see alsoPfurtscheller, 1989). Here, alpha ERD was modeled by decreasingthe simulated alpha rhythm amplitude after EP onset at 0 ms to alevel of 25% (i.e. an alpha ERD of 75%) of its initial amplitude(Fig. 1C), lasting for the entire EP time window (100 ms). Fig. 1Dshows the sum of the simulated EP and concurrent alpha activity.

Oscillatory modelIn contrast to the evoked model, the EP was simulated by

resetting and synchronizing the random phase of the ongoing alpharhythm from 0 to 100 ms post-stimulus time across trials (Figs.1E–H). The amplitude of the pre-stimulus alpha rhythm amountedto 10 μV (further referred to as “medium” alpha). Since in expe-

ed lines indicate single-trial signals being superimposed on each other; boldked model. Independent from simultaneously occurring spontaneous alphaesynchronization (alpha-ERD; Pfurtscheller and Aranibar, 1977; Klimesch ets depicted in panel D. (E–H) Scheme of the oscillatory model considering theodel, the phenomenon of alpha-ERD is accounted for by modeling a partialalpha activity after stimulation (G) and in a major fraction of phase-lockedof previous spontaneous alpha activity (F). In the situation depicted here, a

g) theory is problematic (see Hanslymayr et al., 2006, Klimesch et al., 2006).ly proposed methods for identification of the mechanism of EP generation,for conventional analyses as data believed to be generated by phase resetting

709R. Becker et al. / NeuroImage 39 (2008) 707–716

rimental data complete alpha-band phase reset has not beenobserved, a major fraction of the simulated alpha rhythm was phasereset (Fig. 1F) while a minor fraction preserved its phase (Fig. 1G).The ratio between the reset and non-phase-reset simulated alpharhythm was determined such that the oscillatory model and theevoked model generated the same data. The such determined ratioof the ‘partial phase reset’ is used in all subsequent simulations.The resulting sum of phase-reset and non-phase-reset-simulatedalpha rhythm is shown in Fig. 1H.

Impact of pre-stimulus alpha amplitude variation on the modeldata

Starting from the initial data subset as described above, we nowsystematically varied the amplitudes of the simulated alpha rhythmin both models which additionally resulted in a “low” and “high”alpha-amplitude group (Fig. 2). The aim was to test the hypothesisthat variation of the amplitude of the ongoing rhythm leads todivergent model predictions for both theories. The “medium”

group (see Figs. 2B, J) comprises the initial data subsets, asdepicted in Fig. 1, with alpha amplitude of 10 μV, while the “low”and “high” alpha amplitudes amount to 5 μV and 20 μV,respectively.

Fig. 2. Predictions of the two models on the impact of pre-stimulus alpha amplitudelow (A), medium (B) and high (C) alpha amplitude, superimposed on each other wit(D) No effect of alpha variation on the EP amplitude. (E) Relative alpha-band eveAlpha-band phase-locking index (PLI). (I–O) Corresponding analysis for the osci

Evoked modelBecause of the definition of the EP as being independent from

the simultaneous rhythm in the evoked theory, the same additiveEP was used for the “low” and the “high” alpha group as in the“medium” alpha group (Figs. 2A–C). Correspondingly in both the“high” and “low” alpha group, alpha rhythm desynchronizationwas modeled as an alpha rhythm decrease to 25% of the initial pre-stimulus alpha rhythm amplitude lasting for the entire time windowof the EP (Figs. 2A–C).

Oscillatory modelAccording to the oscillatory theory, the EP (or at least part of

the EP) is directly generated by the alpha rhythm via phaseresetting. Based on the previously determined ratio of partial phasereset, the here modeled amplitude of the resulting EP is 75% of thepre-stimulus alpha amplitude. In all groups (high, medium, lowpre-stimulus alpha rhythm), 25% of the pre-stimulus alpha rhythmis not phase reset, representing the non-phase-locked fraction of thealpha rhythm (Figs. 2I–K).

Subsequently, the effect of pre-stimulus alpha-variation on thefollowing parameters was analyzed: (1) average EP amplitude, (2)relative and absolute alpha-band event-related spectral perturbation(ERSP) and (3) alpha-band phase-locking index, PLI (Tallon-

variation on EP parameters. Single trials of the evoked model are depicted forh the average EP in bold lines (blue, black, red for varying alpha amplitudes).nt-related spectral perturbation (ERSP). (F) Absolute alpha-band ERSP. (G)llatory model.

710 R. Becker et al. / NeuroImage 39 (2008) 707–716

Baudry et al., 1996; Makeig et al., 2002); for detailed descriptionof these indices please refer to the following, see Experimental datasection. Due to the intrinsic differences of the two models in therelation between ongoing and evoked activity, differential predic-tions were expected. In a next step, these differential predictionsserve as reference for the analysis of experimental data and for thedetermination of underlying EP mechanisms.

Experimental data

In order to compare the model predictions with real data, wechose a visual stimulation protocol with two conditions: eyes openand eyes closed, allowing for statements on the validity of the twotheories of EP generation under different conditions.

SubjectsSeventeen healthy subjects (8 female/9 male; mean age 28.1±

2.7 years) participated in the study. Subjects had given writteninformed consent prior to the investigation. All experiments wereperformed in compliance with the relevant laws and institutionalguidelines and were approved by the ethics committee of theCharité, University Medicine Berlin. Data from subjects reaching avigilance level below sleep state 1 (Rechtschaffen and Kales, 1968)during the experiment were not used in the analysis which resultedin rejection of 2 subjects.

Stimulation protocolThe experiment consisted of a passive viewing task with two

conditions, eyes open and eyes closed. As stimulus during eyesopen condition, a circular checker-board with central fixation crosswas shown, for the eyes closed condition a white uniform flash waspresented. Duration of stimulation was 1.1 s, with an inter-stimulusinterval (ISI) ranging from 9.25 to 11.5 s with uniform distributionshowing a flat gray screen. Subjects were instructed to attentivelyperceive stimuli keeping their eyes fixated on the center of thescreen throughout the eyes open condition of the experiment. Theywere informed of the end of each block acoustically. Theexperiment was programmed using the Cogent toolbox (developedby the Cogent 2000 team at Functional Imaging Laboratory (FIL)and Institute of Cognitive Neuroscience (ICN) and CogentGraphics by John Romaya at the Laboratory of Neurobiology(LON) at the Wellcome Department of Imaging Neuroscience).Each condition consisted of four blocks containing 30 stimuli eachsumming to 120 trials per condition. Eight blocks of restingperiods of a duration of 75 s, half of the blocks with eyes closedand half with eyes open were randomly interspersed withstimulation periods. In order to achieve comparable conditionswhen transferring this experiment into an MR environment,subjects lay on an examination couch and received the stimulivia a mirror reflecting the picture located on a plexi-glass screen.Stimuli were presented by a modified LCD projector (NECMultisync MT 800, Japan).

EEG preprocessingA 32-electrode cap (Easy-Cap; FMS, Herrsching-Breitbrunn,

Germany) was used for EEG recordings following the international10/20 system. All electrodes were referenced against FCz position.Electrooculogram of vertical (EOGv) and horizontal (EOGh) eyemovements was recorded by one electrode placed below the righteye and two electrodes at the outer canthi. Impedances weremaintained below 5 kΩ by applying an abrasive electrode paste

(ABRALYT 2000; FMS, Herrsching-Breitbrunn, Germany). Ahigh dynamic range EEG amplifier with a sampling rate of 5 kHz(BrainAmp; BrainProducts GmbH, Munich, Germany) and VisionRecorder Software v1.02 was used to record EEG data. Data wereband pass-filtered between 0.5 and 70 Hz. A notch filter of 50 Hzwas applied. For further processing, data were down-sampled to200 Hz. Offline analysis was performed with EEGLab 4.515(Delorme and Makeig, 2004) and Matlab v7.0 (The MathworksInc., Natick, USA). Artifact rejection was performed in two steps.First, the following EEGLAB artifact rejection methods were used:an exclusion threshold of 100 μV for EEG/EOG channel data andimprobability of data as estimated by joint-probability andkurtosis-of-activity analysis using EEGLAB preset defaults.Second, visual inspection was performed for double-checking ofproposed artifact removal and removal of then-remaining artifacts.For the eyes closed condition, in average 108 trials remained afterartifact rejection. Due to occasional eye blinking, on average 96trials remained for eyes open condition. For analysis of EPs, trialswere segmented in −1 s to 9 s epochs time locked to the stimulus.Subsequently a baseline correction of data using the time windowfrom −1 s to 0 s was applied.

ICA decompositionTo further increase signal-to-noise ratio, an independent

component analysis (ICA) was performed using Infomax ICA.This was done separately for each subject and each condition usingthe artifact corrected, segmented data of all EEG/EOG channels,excluding all ECG and EMG channels. The resulting 10 out of 24components explaining most of the variance were classified intomain clusters. Components of artifactual or ocular origin wereidentified using their topographical, spectral and time–domainproperties and only components of non-artifactual clusters wereback-projected to EEG channels for further analysis as proposed byJung et al. (2000a). The resulting corrected signal at the occipitalelectrode O2 was used for subsequent analysis to allow forcomparison with conventional EP data.

EEG postprocessingThe grand average EP and the time–frequency plot were

calculated for both eyes open and eyes closed condition separately.The grand average time–frequency plot includes the frequencyrange from 0.5 Hz to 40 Hz.

Impact of pre-stimulus alpha amplitude variation on experimentaldata

Analogous to the simulated data, the impact of ongoing alphaamplitude on the experimental data was examined for comparisonwith model predictions. Analysis of EP parameters was doneseparately for “eyes open” and “eyes closed” condition. By sortingtrials according to their pre-stimulus alpha amplitude, each trialwas assigned to either a “high-alpha” or “low-alpha” group. Trialswere sorted individually for each subject according to absolutealpha amplitude in a pre-stimulus temporal window of −800 ms to−100 ms using a wavelet-based analysis (3-cycle Hanning-windowed sinusoidal wavelets, frequency stepping of 1 Hz). Theapproach of a pre-stimulus window sorting was employed to avoidthe assumed confounding issue of sorting ongoing activity in apost-stimulus time window (Yeung et al., 2004; Makinen et al.,2005). The weighted sorting window of the 3-cycle waveletcentered at −100 ms ranged up to 50 ms into the post-stimulusepoch, however the earliest pronounced evoked components in the

711R. Becker et al. / NeuroImage 39 (2008) 707–716

present study were found to arise at a latency of N70 ms afterstimulus onset.

Analogous to the analysis of the modeled data, results of bothexperimental conditions were compared for the following para-meters in an analysis window from 0 to 400 ms after stimulation:

(1) The grand average of the broadband EP (0.5–70 Hz)(2) Alpha-band event-related spectral perturbation (alpha-band

ERSP) based on the 3-cycle wavelet analysis (as mentionedabove). The amplitudes in the frequency range of 8–12 Hz(in steps of 1 Hz) were extracted and the dominant frequencywithin this range was determined for each subject separatelyto be used for further analysis. Please note that the event-related phase-locked component, i.e. the EP waveform, wasnot removed beforehand, thus yielding an estimate of totalalpha-band amplitude. The alpha-band ERSP is shown bothas “relative” ERSP, i.e. as a relative change with respect to itsbaseline (the pre-stimulus time window) as well as “absolute”ERSP providing absolute alpha-band amplitude changes.

(3) The alpha-band phase locking index (alpha-bandPLI) accordingto Makeig et al. (2002) in order to estimate the degree of phaselocking of (total) post-stimulus alpha activity. Phase infor-mation (φ, see Eq. (1)) was extracted from the wavelet analysisin the frequency range of 8–12 Hz for each time point (t).

PLI tð Þ ¼ 1N

����XN

k¼1

eiðukðtÞÞ���� ð1Þ

To account for individual differences in alpha frequency, thepre-stimulus window was examined for the dominant alphafrequency, which was then analyzed further. Increasing valuessignify increasing phase coherence across trials (withN=number of trials), ranging from zero (completely randomdistribution of phases) to one (perfect alignment of phase acrossall trials).

For statistical testing of differences of the “high” and “low”alpha-amplitude groups, the pair-wise signed-ranking test was usedas a non-parametric Student’s t-test analogue to estimatesignificances of intra-individual alpha-amplitude dependent differ-ences of the above mentioned parameters (EP amplitude, relative/absolute alpha-band ERSP and alpha-band PLI). A time window of0 to 400 ms after stimulus onset was analyzed. Data wereBonferroni corrected for multiple comparisons.

Fig. 3. Visualization of typical ICA components explaining most of thevariance. (A) Visual component clusters. (B) Non-visual, ocular andartifactual sources.

Results

In the following, we first show the results of our theoreticalmodeling approach, thereby demonstrating the usefulness of theproposed pre-stimulus alpha-variation for the differentiation ofprevailing theories of EP generation. Secondly, we compare theoutcome of the models with experimental data in order to identifythe model that is more compatible with our experimental data.

Models

In the situation of an ongoing rhythm exhibiting, an amplitudeequal or larger (in the case of the simulated “medium” group10 μV) than the amplitude of the arising EP (7.5 μV for the“medium” alpha group) both models can explain the generation of

EPs equally well (see Fig. 1). While in the evoked model the EP isgenerated independently from and additive to the simultaneousalpha rhythm, in the oscillatory model the EP is generated bypartially phase resetting the ongoing alpha rhythm. Thus, thephase-locked part of the signal, i.e. the EP can be explained byboth theories. The non-phase-locked part of the post-stimulussignal is desynchronized alpha rhythm in the evoked model andnon-phase-reset alpha rhythm in the oscillatory model. Thus, bytheoretical modeling, we demonstrate a substantial intersection ofthe two EP theories despite their opposing assumptions on therelationship between ongoing and evoked activity.

Impact of variation of pre-stimulus alpha amplitude on model dataFor the evoked model, increasing pre-stimulus alpha amplitude

(Figs. 2A–C) was paralleled by a constant EP amplitude (Fig. 2D),a transition from a relative increase of total alpha-band amplitudeto a relative decrease (see relative alpha-band ERSP in Fig. 2E) anda decrease in alpha-band phase locking (Fig. 2G). In contrast, inthe oscillatory model, increasing pre-stimulus alpha amplitude(Figs. 2I–K) was associated with enhanced EP amplitudes (Fig.2L). The relative alpha-band amplitude and the degree of phaselocking remained constant (see relative alpha-band ERSP andalpha-band PLI in Figs. 2M, O). Despite the subset of data bothmodels share, they deliver distinct predictions on the effect of pre-stimulus alpha-variation on simulated EPs and simultaneouslyoccurring alpha background activity.

Experimental data

ICA decompositionBack-projection of non-ocular and non-artifactual sources

comprising the “visual” clusters among the first 10 independentcomponents yielded data with a clear occipital topography as inJung et al. (2000b). Excluded sources were either event related butof frontal origin comprising eye movements or their topography orraw time courses identified them as being of ocular origin. In allsubjects, each of the first 10 components could be assigned byvisual inspection to one of the spatial patterns shown in Fig. 3.

Grand averagesGrand average EPs and grand average time–frequency plots

(range from 0.5 to 40 Hz) are provided in Fig. 4 for both eyes openand eyes closed condition. Relative alpha-band amplitudes (Figs.4A, B) change similarly in both conditions (for absolute alpha-bandamplitudes, please refer to Figs. 5C, H), exhibiting an event-relateddecrease in amplitude (alpha rhythm desynchronization, alpha-ERD). Notably, these experimental data reflect the “ambiguous”situation as simulated in the “medium” alpha group (depicted inFig. 1), i.e. no clear post-stimulus alpha-band amplitude increase isobserved which implies the possible validity of both theories of EPgeneration.

Fig. 4. Time–frequency plots and grand average EPs of experimental data.(A, B) Time–frequency plots (0.5–40 Hz) of both eyes open and eyes closedvisual stimulation. Vertical horizontal lines indicate begin and end of visualstimulation. Event-related amplitude increases in frequenciesb8Hz suggest anevoked process. The alpha-band amplitude in both conditions does not show asignal increase, but a dominating desynchronization, i.e. an ambiguoussituation similar to the simulated situation in Fig. 1. Possible interpretations areas follows: Either an evoked process is “masked” by concurrent alphadesynchronization or the alpha-band component of the EP is created by phasereset of the ongoing alpha rhythm. (C, D) Corresponding grand average EPs ofboth conditions.

712 R. Becker et al. / NeuroImage 39 (2008) 707–716

Impact of variation of pre-stimulus alpha amplitude onexperimental data and comparison to models

Duration of alpha sorting as indicated by absolute alpha-bandERSP. In contrast to modeled data, the sorting effect of pre-stimulus alpha amplitudes in experimental data decays in the post-stimulus time window. Thus, it is necessary to identify the relevanttime window in the experimental data for comparison with ourmodel predictions. The duration of efficient sorting is indicated by

Fig. 5. Effects of pre-stimulus alpha amplitude variation on eyes open (A–D) andamplitude sorting is indicated by the light blue cones on the left side of panels A andFilled rectangles mean testing groups against each other, dark gray stripes (B, G) meindicates the end of the analysis window; the solid line indicates begin and end ofrelated spectral perturbations (relative alpha-band ERSP). (C, H) Absolute alpha-b

significant differences in the absolute alpha-band amplitude (oralpha ERSP) between the “low” and “high” alpha groups. For eyesopen condition, the resulting time window lasts from 0 to 158 msafter stimulation (Fig. 5C). For eyes closed condition, a significantdifference between absolute alpha-band amplitude remainsthroughout the entire analysis time window from 0 to 400 ms,indicating a slower sorting “decay” of ongoing alpha amplitudes(Fig. 5H). In general, absolute alpha amplitudes during eyes closedcondition are higher than during eyes open condition.

Results of EP analysis. For the eyes open condition, early EPcomponents were constant, i.e. unaffected by alpha sorting, despitethe enduring effect of alpha sorting within the EP time window(Fig. 5A, see also comments above). A late EP component (cf. Fig.5A), in turn, showed a significant (Bonferroni corrected, pb0.05)enhancement with a maximum of 2.5 μV difference at a timewindow of 220 to 310 ms. Notably, it is not paralleled by aconcurrent increase in absolute alpha amplitude (see absolutealpha-band ERSP analysis, Fig. 5C). This indicates an alreadydecayed effect of pre-stimulus alpha-amplitude sorting. Shape andtime course of the enhancement indicate a frequency componentslower than alpha frequency. One additional significantly differenttime point was found at 175 ms for a negative evoked componentwhose negativity decreased for high alpha.

For the eyes closed condition, alpha sorting did not result in anysignificant differences in the EP over the entire analysis window. Apositive trend for the high alpha group was observable from200 ms on (Fig. 5F). In order to exclude alpha-band contribution tothis positive trend, spectral alpha-band components of the EP wereremoved by filtering, not eliminating the trend (results not shown).Thus, the observed positive trend does not appear to be causeddirectly by phase-reset alpha activity.

Results of relative alpha-band ERSP analysis. In the eyes opencondition, a switch from an increase in the relative alpha-band

eyes closed (F–I) visual stimulation data. The time range used for alpha-F. In dark gray, Bonferroni corrected significant effects (pb0.05) are shown.an testing against the respective group pre-stimulus baseline. The dotted linevisual stimulation. (A, F) EPAmplitudes. (B, G) Relative alpha-band event-and ERSP (sorting effect). (D, I) Alpha-band phase-locking index (PLI).

713R. Becker et al. / NeuroImage 39 (2008) 707–716

amplitude to a relative decrease was found when comparing “low”and “high” alpha groups (Fig. 5B). The increase for the “low”alpha group paralleled the time course of the increase in alpha-band phase locking (see results of alpha-band phase lockingbelow).

For the eyes closed condition, the relative alpha-band ERSPalso showed a significant increase after stimulation for the “low”alpha group; however, this was smaller in amplitude and temporalextent (Fig. 5G).

Alpha-band PLI analysis. There were intervals of significantlyreduced phase locking of the high-alpha group in both eyes openand eyes closed conditions (for eyes open ranging from 60 to158 ms, Fig. 5E; for eyes closed ranging from 110 to 158 ms, Fig.5J) being paralleled by constant EP amplitudes.

Comparison of experimental results with model predictions. Forthe eyes open condition, the observed effect of alpha-variationduring the relevant time window is well predicted by the evokedmodel. Unaffected early EP components together with a decreasedphase locking for increasing alpha and also the relative alpha-bandamplitude, showing a transition from an increase to a decrease (fromthe “low” to the “high” alpha group), are predictions of the evokedmodel. The eyes closed early EP components exhibit a correspond-ing pattern and are equally well predicted by the evoked model.

The observed significant enhancement of the eyes open latepositive EP component in the “high” alpha group is not predictedby any model. Both models basically assume a difference inabsolute post-stimulus alpha-band amplitude which is not presentin the late EP component of the experimental data. Neither modelstates the possibility of an indirect effect of pre-stimulus alpha-sorting without a simultaneous alpha-amplitude difference in therelevant time window of the EP component of interest.

Discussion

We have demonstrated that variation of the ongoing alphaamplitude allows differentiation of the two theories of EPgeneration, even when a subset of the data is ambiguous withrespect to the two theories. Our experimental data show significantdifferences in EP parameters for varying pre-stimulus alphaamplitudes with early EP components being most consistent withthe evoked theory, while later components co-vary with pre-stimulus but are not generated by post-stimulus alpha rhythm.

Models

Both oscillatory and evoked theory are consistent with phase-locked and non-phase-locked components. The oscillatory modelonly requires the assumption of a partial phase reset in order to explainthe same data as the evoked theory (Figs. 1 and 2). Thus, givensufficient alpha activity, the absence of a single-trial post-stimulusalpha-band amplitude increase (model data in Figs. 3E, M,experimental data in Figs. 4A, B) does not exclude an evokedmechanism as previously assumed by some authors (Sayers et al.,1974; Fuentemilla et al., 2006) but presents an ambiguous situation.Accordingly, the origin of EPs can only be unveiled by assessing theinfluence of amplitude variations of ongoing alpha activity on the EP.Especially EPs recorded during visual stimulation are often paralleledby a spontaneous alpha rhythm having larger amplitudes than the EPitself (also reflected in our data, see Figs. 5C, H).

We compared the predictions of the evoked and the oscillatorymodel concerning the impact of pre-stimulus alpha variation on EPamplitude with our experimental data. Within eyes open and eyesclosed condition, higher pre-stimulus alpha activity was accompaniedby constant early EP amplitudes and a concurrent decrease in alpha-band phase locking. Also the total alpha-band amplitude changedfrom a relative decrease to a relative increase for low pre-stimulusalpha activity. All these phenomena are predicted by the evokedmodel and correspond to the assumption of dual and separategenerators of EP and alpha rhythm (Fig. 2). The change in the degreeof phase locking is explained by the superposition of alpha activity onthe evoked potential, i.e. “masking” of the single-trial EP by a strongalpha rhythm. With increasing alpha activity, the non-phase-lockedpart of the single-trial signal (i.e. post-stimulus alpha rhythm)increases, which is reflected by less phase locking across trials (seeFig. 5E, J, respectively). For periods of low alpha activity in turn, theadditional evoked EP causes an increase in total alpha-band amplitudeinstead of a decrease of alpha-band amplitude. Thus, for the datapresented, the combined approach of alpha-amplitude variation ofexperimental data and comparison with model predictions succeededin differentiation of early EP mechanisms.

Concerning other modeling approaches, Makinen et al. (2005)suggests a single parameter for differentiation: standard deviationacross trials (SDT). Straight-forward predictions of this approachwere that EPs generated by an evoked model show no decrease inSDT, while EPs yielded by an oscillatory model exhibit an SDTdecrease. Experimental auditory data of this study showed nodecrease in SDT, which was regarded as evidence against anoscillatory mechanism of EP generation. In a comment to Makinenet al., Klimesch et al. (2006) in turn argued that integrating event-related decreases in alpha amplitude, i.e. alpha desynchronization,questions the validity of these claims. Concerning the experimentaland model data of our study, we observed the following: (a) ourexperimental data do show an SDT decrease (see SupplementaryFig. 3), which is mainly caused by event-related decrease of thealpha rhythm amplitude (for an early report of decrease of visualdata SDT see Pfurtscheller et al., 1989), and (b) both our models, inconcordance with our experimental data, show an SDT decrease aswell. Thus, we discarded the SDT measure as a sufficientparameter for model differentiation and adopted the more complexapproach based on combined EP parameters.

For the models in our study, we used a noise-free approach,which has the advantage of visualizing the general principles of thetwo theories even in single trials that otherwise would easilydisappear under noisy conditions. We excluded, however, thepossibility of different model behavior under noisy conditions byadding noise to our simulated data. As can be seen in SupplementaryFig. 1, both models preserve their differential predictions despitesingle trials being obscured by intense noise. In order to test whetherthe chosen value of alpha rhythm desynchronization (or of residual,non-phase-locked alpha amplitude) influences the model outcome,we tested two additional values, i.e. 50% (weak ERD) and 12.5%residual alpha amplitude (strong ERD) for the evoked model, whichcorresponds to 50% and 12.5% of non-phase-reset alpha rhythm forthe oscillatory model. Again, differential model predictions werewell preserved (for results see Supplementary Fig. 2).

Experimental data

The enhancement of a late evoked component due to pre-stimulus alpha sorting was found to be incompatible with the

714 R. Becker et al. / NeuroImage 39 (2008) 707–716

oscillatory model. For an enhancement taking place, the oscillatorymodel requires a concomitant increase in post-stimulus alpha-bandamplitude (see models, Fig. 2F). This was not the case in the data,which excludes a direct causative role of alpha activity. Since theobserved enhancement is neither predicted by the pure evokedmodel, the results can only be reconciled with a model integratingan additional indirect interaction between pre-stimulus alphaactivity and the evoked potential. As the focus of this study wasto determine whether and how well the two dominant theories of EPgeneration in their strict interpretation would predict empiricalfindings, we did not change our models post-hoc-wise. In principle,the observed empirical interaction between alpha and EP amplitudecould be integrated, for example by an additional arousal orvigilance term which is related to the amplitude of spontaneousalpha activity. Another way of accounting for the empirical findingsin future modelling may be based on the recently proposedhypothesis of Nikulin et al. (2007) on alpha-dependent baselineshifts in the EEG. This hypothesis accounts for the empiricalphenomenon that each level of alpha activity appears to beassociated with a different EEG baseline level. This implies, thatduring alpha amplitude fluctuations also the mean baseline level ismodulated, which could explain modulations of EP components asa function of the amount of pre-stimulus alpha activity beingdesynchronized.

The view of the evoked response as being modulated – but notgenerated – by the alpha rhythm, is also supported by Jasiukaitisand Hakerem (1988), who, using pre-stimulus alpha-amplitudesorted auditory oddball data, also observed an exclusive enhance-ment of a late positive evoked component, the P300. They statedthat it is not the case “that the P300 is a phase constrained alphawave, but that the state of which alpha is a sign, predicts the largerP300”, giving support to a co-variation but not causationhypothesis of alpha activity. Similar positive correlations betweenongoing alpha activity and the P300 were reported by Polich(1997) and by Basar et al. (1984). Consistently with the results ofJasiukaitis and Hakerem (1988), Barry et al. (2000) reported adirect relationship between pre-stimulus alpha amplitude andfollowing EP amplitudes for auditory oddball data. Thut et al.(2003) reported an exclusive enhancement of a late positivecomponent using a visual stimulation setup with repetitive TMS(rTMS). The enhancement was observed after application of rTMS.No early component was modulated, despite a concurrent decreasein the amount of event-related alpha desynchronization after rTMSapplication. The authors concluded that ongoing and evokedactivity seem to be rather separate phenomena. In summary, whilealpha-dependent modulations on late EP components are fre-quently observed, no direct involvement of concurrent alphaactivity has been reported.

With respect to findings in early potentials, Makinen et al.(2005), reported early auditory potentials (N1m, ~100 ms) to beunaffected by increasing pre-stimulus alpha activity. Complemen-tary to this negative evidence supporting the evoked model is ourresult of a significant decrease in phase locking and at the sametime unaffected EP amplitudes for significantly higher (pre- andpost-stimulus) alpha activity, which provides additional positiveevidence for the evoked model and also argues for sufficientsignal-to-noise ratio of the data.

Also intracranial studies in monkeys support the evoked natureof early EPs (e.g. Schroeder, 1991; Rols et al., 2001). In thesestudies, early scalp EPs were linked to increased multi-unit activityin low-level sensory areas. Additionally, early EPs are also

reported to be relatively unaffected by attentional fluctuations(Mehta et al., 2000), which fits our observation of early EPs beingunaffected by alpha rhythm amplitude. Assuming that alpharhythm is an indicator of attentional and/or vigilance fluctuationone would expect attention-dependent EP components to be relatedto these fluctuations.

An inverse relation between pre-stimulus alpha amplitude andvisual EP amplitude (N1–P2) has been proposed by Rahn andBasar (1993). For stages of low pre-stimulus alpha activityrecorded at the vertex an enhancement of the N1–P2 peak-to-peakamplitude (100–250 ms) was demonstrated in fronto-central butnot in occipital electrodes. These results are not directlycomparable with our data, since the channels exhibiting this effect,the reference used and the topographical definition of the alpharhythm at the vertex differ from our study.

On the other hand, Basar et al. (1998) reported an inversecorrelation between a factor termed “EP enhancement” – whichwas defined as the ratio of the EP amplitude to pre-stimulus alphaamplitude – and pre-stimulus alpha amplitude. This finding wasneither accompanied by simulations nor interpreted in the light ofthe two opposing EP theories. Thus, we analyzed our modelsanalogously for this factor, which yielded a comparably clearinverse relation between “EP enhancement” and pre-stimulusalpha amplitude only for the evoked model, while the EPenhancement index for the oscillatory model remains constantfor varying alpha amplitudes (see Supplementary Fig. 4). Theresults for the evoked model also show the previously reportedinverse but more exactly “curvilinear” relationship between VEPenhancement and the pre-stimulus alpha amplitude as found byBrandt and Jansen (1991) when replicating the analysis of Basar etal. (1998). Also Barry et al. (2000) showed convincingly forauditory oddball data that a direct relationship between pre-stimulus alpha activity and EP component amplitude (as proposedby Brandt and Jansen) is compatible with the existence of aninverse relationship between pre-stimulus alpha and the mentioned“EP enhancement factor” (as proposed by the group of Basar).This implies, that the reported inverse correlation allows manifoldinterpretations on the relationship between EP amplitude andongoing alpha amplitude. Actually, any relationship which isweaker than directly proportional, even a clear independence of EPand ongoing alpha amplitude yields such inverse correlation (seeSupplementary Fig. 4). In the specific case of the pure evokedmodel the curve behaves even “curvilinear”, as demonstrated inSupplementary Fig. 4A. The results of Basar’s and Brandt’s studyon the “EP enhancement factor” do not contradict but rathersupport the evoked theory and its implications. However, at least astrict phase reset can be excluded, since such a behavior wouldprevent the consistently reported inverse relationship between pre-stimulus and post-stimulus alpha enhancement. In case of a purephase reset the relation between pre-stimulus alpha activity and thepost-stimulus enhancement remains constant, as can be seen forthe phase-reset model in Supplementary Fig. 4B.

In contrast to our results on early EP components, Brandt andJansen (1991) reported a positive enhancement of eyes closed earlyVEP components for higher pre-stimulus alpha activity. Again, forcomparison we applied the same kind of data analysis as Brandtand Jansen to our experimental “eyes closed” data yielding similarresults (data not shown). In contrast to our actual approach, Brandtand Jansen had analyzed sets of averages of 40 trials each andmaximum peak-to-peak amplitudes were individually determinedfor each such average. This analysis is susceptible to systematic

715R. Becker et al. / NeuroImage 39 (2008) 707–716

confounding effects of residual alpha activity being superimposedon the evoked potential, because estimated peak-to-peak ampli-tudes with variable latencies are affected especially by strongresidual ongoing alpha activity. In the study presented here, theanalysis approach was to average a large number of trials (N ~750) pooled over subjects and to analyze the grand average EPswith fixed latencies. Thus, residual effects of ongoing alphaactivity were canceled out more effectively. For this analysis thementioned early EP-enhancing effect disappeared.

It has been demonstrated by Makeig et al. (2002) that forincreased alpha activity measured in a post-stimulus time windowthere is an increase of EP amplitudes with accentuated alpha-activity and increased phase locking. By using a pre-stimulus timewindow, we observed the contrary effect, i.e. strong alphabackground activity was accompanied by weaker phase lockingand constant early EP amplitudes during the critical post-stimulustime window. On the other hand, sorting our data according to thepost-stimulus time window yielded an alpha-amplitude dependentincrease of EP amplitude (data not shown). Thus, differences inresults between the two studies can be explained by differences inthe definition of the sorting window. Since a post-stimulus sortingwindow tends to be biased by stimulus-related activity (Makinen etal., 2005; Yeung et al., 2004), a pre-stimulus time windowappeared more appropriate to us. In summary, while some previousstudies on the relation between ongoing alpha rhythm and earlyevoked potentials contradict our study with respect to theinterpretations regarding the prevailing mechanism of EP genera-tion, the experimental results of these studies can be reconciledwith our data.

Our combined simulated and empirical data approach onstudying the relationship between background alpha rhythm andthe VEP favors the evoked model for early components and at thesame time demonstrates an interaction between pre-stimulus alphaamplitude and VEP features occurring after 175 ms, arguing for amodulatory rather than direct causative role of alpha activity. Inconclusion, we propose to study empirical data in the context of amodeling-based approach to further shed light on mechanisms ofEP generation.

Acknowledgments

The authors thank Steven Lemm, Peter Brunecker, RobertSchmidt and Matthias Reinacher for helpful comments. This workwas supported by the German Federal Ministry for Education andResearch BMBF (Berlin NeuroImaging Center and BernsteinCenter for Computational Neuroscience Berlin) and the GermanResearch Foundation DFG (Berlin School of Mind and Brain).

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.neuroimage.2007.09.016.

References

Arieli, A., Sterkin, A., Grinvald, A., Aertsen, A., 1996. Dynamics ofongoing activity: explanation of the large variability in evoked corticalresponse. Science 273, 1868–1871.

Barry, R.J., Kirkaikul, S., Hodder, D., 2000. EEG alpha activity and the ERP

to target stimuli in an auditory oddball paradigm. Int. J. Psychophysiol.39, 39–50.

Basar, E., 1980. EEG-Brain Dynamics. Relation Between EEG and BrainEvoked Potentials. Elsevier, Amsterdam.

Basar, E., Basar-Eroglu, C., Rosen, B., Schütt, A., 1984. A new approach toendogenous event-related potentials in man—relation between EEG andP300-wave. Int. J. Neurosci. 24, 1–21.

Basar, E., Karakas, S., Rahn, E., Schürmann, M., 1998. Major operatingrhythms. In: Basar, E. (Ed.), Brain Function and Oscillations. Springer,Berlin, pp. 219–238.

Brandt, M.E., Jansen, B.H., 1991. The relationship between prestimulus-alpha amplitude and visual evoked potential amplitude. Int. J. Neurosci.61, 261–268.

Delorme, A., Makeig, S., 2004. EEGLAB: open source toolbox for analysisof single-trial EEG dynamics including independent componentanalysis. J. Neurosci. Methods 134, 9–21.

Fuentemilla, LI., Marco-Pallares, J., Grau, C., 2006. Modulation of spectralpower and of phase resetting of EEG contributes differentially to thegeneration of auditory event-related potentials. NeuroImage 30,909–916.

Hanslmayr, S., Klimesch, W., Sauseng, P., Gruber, W., Doppelmayr, M.,Freunberger, R., Pecherstorfer, T., Birbaumer, N., 2006. Alpha PhaseReset Contributes to the Generation of ERPs. Cereb. Cortex 17,1–8.

Jasiukaitis, P., Hakerem, G., 1988. The effect of prestimulus alpha activityon the P300. Psychophysiology 25, 157–165.

Jung, T.P., Makeig, S., Humphries, C., Lee, T.W., McKeown, M.J., Iragui,V., Sejnowski, T.J., 2000a. Removing electroencephalographic artifactsby blind source separation. Psychophysiology 37, 163–178.

Jung, T.P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E.,Sejnowski, T.J., 2000b. Removal of eye activity artifacts from visualevent-related potentials in normal and clinical subjects. Clin. Neuro-physiol. 111, 1745–1758.

Klimesch, W., Schack, B., Schabus, M., Doppelmayr, M., Gruber, W.R.,Sauseng, P., 2004. Phase locked alpha and theta oscillations generate theP1–N1 complex and are related to memory performance. Brain Res.Cogn. Brain Res. 19, 302–316.

Klimesch, W., Hanslmayr, S., Sauseng, P., Gruber, W.R., 2006. Distinguish-ing the evoked response from phase reset: a comment to Makinen et al.NeuroImage 29, 808–811.

Makeig, S., Westerfield, M., Jung, T.P., Enghoff, S., Townsend, J.,Courchesne, E., Sejnowski, T.J., 2002. Dynamic brain sources of visualevoked responses. Science 295, 690–694.

Makinen, V., Tiitinen, H., May, P., 2005. Auditory event-related responsesare generated independently of ongoing brain activity. NeuroImage 24,961–968.

Mazaheri, A., Jensen, O., 2006. Posterior {alpha} activity is not phase-resetby visual stimuli. Proc. Natl. Acad. Sci. U. S. A. 103, 2948–2952.

Mehta, A.D., Ulbert, I., Schroeder, C.E., 2000. Intermodal selectiveattention in monkeys. I: distribution and timing of effects across visualareas. Cereb. Cortex 10, 343–358.

Nikulin, V.V., Linkenkaer-Hansen, K., Nolte, G., Lemm, S., Muller, K.R.,Ilmoniemi, R.J., Curio, G., 2007. A novel mechanism for evokedresponses in the human brain. Eur. J. Neurosci. 25, 3146–3154.

Pfurtscheller, G., 1989. Spatiotemporal analysis of alpha frequencycomponents with the ERD technique. Brain Topogr. 2, 3–8.

Pfurtscheller, G., Aranibar, A., 1977. Event-related cortical desynchroniza-tion detected by power measurements of scalp EEG. Electroencephalogr.Clin. Neurophysiol. 42, 817–826.

Polich, J., 1997. On the relationship between EEG and P300: individualdifferences, aging, and ultradian rhythms. Int. J. Psychophysiol. 26,299–317.

Rahn, E., Basar, E., 1993. Enhancement of visual evoked potentials bystimulation during low prestimulus EEG stages. Int. J. Neurosci. 72,123–136.

Rechtschaffen, A., Kales, A., University of California LAB, InformationService., NINDB Neurological Information Network (U.S.), 1968. A

716 R. Becker et al. / NeuroImage 39 (2008) 707–716

manual of standardized terminology, techniques and scoring system forsleep stages of human subjects. Allan Rechtschaffen and Anthony Kales,editors.

Rols, G., Tallon-Baudry, C., Girard, P., Bertrand, O., Bullier, J., 2001.Cortical Mapping of Gamma oscillations in areas V1 and V4 of themacaque monkey. Vis. Neurosci. 18, 527–540.

Sayers, B.M.A., Beagley, H.A., Henshall, W.R., 1974. Mechanism ofauditory evoked Eeg responses. Nature 247, 481–483.

Schroeder, C.E., 1991. Striate cortical contribution to the surface-recordedpattern-reversal VEP in the alert monkey. Vision Res. 31, 1143–1157.

Shah, A.S., Bressler, S.L., Knuth, K.H., Ding, M.Z., Mehta, A.D., Ulbert,I., Schroeder, C.E., 2004. Neural dynamics and the fundamental

mechanisms of event-related brain potentials. Cereb. Cortex 14,476–483.

Tallon-Baudry, C., Bertrand, O., Delpuech, C., Pernier, J., 1996. Stimulusspecificity of phase-locked and non-phase-locked 40 Hz visualresponses in human. J. Neurosci. 16, 4240–4249.

Thut, G., Theoret, H., Pfennig, A., Ives, J., Kampmann, F., Northoff, G.,Pascual-Leone, A., 2003. Differential effects of low-frequency rTMS atthe occipital pole on visual-induced alpha desynchronization and visual-evoked potentials. NeuroImage 18, 334–347.

Yeung, N., Bogacz, R., Holroyd, C.B., Cohen, J.D., 2004. Detection ofsynchronized oscillations in the electroencephalogram: an evaluation ofmethods. Psychophysiology 41, 822–832.