International Journal of Psychophysiology 73 (2009) 350–354
Contents lists available at ScienceDirect
International Journal of Psychophysiology
j ourna l homepage: www.e lsev ie r.com/ locate / i jpsycho
Greater frontal-parietal synchrony at low gamma-band frequencies for inefficientthan efficient visual search in human EEG
Steven Phillips ⁎, Yuji TakedaNeuroscience Research Institute, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 2, 1-1-1 Umezono, Tsukuba, Ibaraki, 305-8568 JapanInstitute for Human Science and Biomedical Engineering, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 6, 1-1-1 Higashi,Tsukuba, Ibaraki, 305-8566 Japan
⁎ Corresponding author. Neuroscience Research InAdvanced Industrial Science and Technology (AIST), TsukTsukuba, Ibaraki, 305-8568 Japan. Tel.: +81 29 861 5165
E-mail address: [email protected] (S. Phillips).URL: http://staff.aist.go.jp/steven.phillips (S. Phillip
In a study on monkeys, [Buschman, T.J., Miller, E.K., 2007. Top-down versus bottom-up control of attention inthe prefrontal and posterior parietal cortices. Science 315 (5820), 1860–1862.] reported more frontal-parietalneuronal synchrony in a low gamma-band (22–34 Hz) for inefficient than efficient visual search, but a reverseeffect in a higher gamma-band (36–56 Hz). We examine whether this difference in top-down versus bottom-up influence on visual search also occurs in humans using scalp EEG. Tenparticipants identified the location of atarget item (coloured, oriented rectangular bar) in search displays also containing similar distractors. For theefficient search condition, in which response time was less dependent on set size (two or four items),distractors had no feature in common with the target. For the inefficient search condition, in which responsetime increased with set size, distractors shared one feature with the target. Analysis of phase-locking valuesrevealed significantly greater synchronization between frontal-parietal electrode pairs in the lower frequencyband around 160–480 ms post-stimulus for inefficient search. No significant difference was observed inthe higher frequency band. These results partly correspond to [Buschman, T.J., Miller, E.K., 2007. Top-downversus bottom-up control of attention in the prefrontal and posterior parietal cortices. Science 315 (5820),1860–1862.], suggesting that top-down control is mediated by neuronal synchrony at lower frequencies. Thefailure to observe a bottom-up effectmay be due to stimulus familiarity—monkeys requireweeks of training incontrast to the few minutes given to humans.
Visual search involves a variety of component processes, includingperception of visual features, attention to specific items, storage/maintenance of targets, matching of targets to items, and ultimatelyresponse selection. A common organizing principle for conceptualiz-ing relationships between these components is a general distinctionbetween bottom-up perceptual information processing versus top-down processing of contextual information by working memory(Corbetta and Shulman, 2002; Desimone and Duncan, 1995; Wood-man and Chun, 2006). Reflecting this distinction, some models ofvisual search include two types of computation based on the similarity
stitute, National Institute ofuba Central 2, 1-1-1 Umezono,.
s).
ll rights reserved.
between (1) display items and the currently searched for target; and(2) items in the search display independent of the target item(Bundesen, 1990; Duncan and Humphreys, 1989; Wolfe et al., 1989).Suchmodels have accounted for thewell-established effects of target–distractor and distractor–distractor similarity on search efficiency.Respectively, they are the increase and decrease in search time perdisplay item (i.e., search slope, or conversely efficiency) with moreitems sharing the same features (e.g., colour, or orientation). In anextreme case, where all distractor (i.e., non-target) items are the sameand have no features in common with the target, search time istypically independent of display set size (efficient search). By contrast,where the target is uniquely identifiable by a conjunction of features,search time typically increases with set size (inefficient search). Theway in which bottom-up (possibly parallel) and top-down (possiblyserial) processes interact to determine search efficiency, though,remains unclear despite decades of intensive behavioural research(Wolfe, 2003).
Recently, a study on monkeys using implanted electrodes iden-tified two characteristic differences between efficient and inefficientsearch in terms of the ordering and synchronization of the firing of
Fig. 1. Each trial consists of a fixation (1500 ms), target cue (1000 ms), delay (1000 ms), and search display (2500 ms). Example search displays are shown for (a) Efficient-2,(b) Efficient-4, (c) Inefficient-2, and (d) Inefficient-4 conditions.
Table 1Mean response error rates and times (ms) for each display type-size condition.
351S. Phillips, Y. Takeda / International Journal of Psychophysiology 73 (2009) 350–354
populations of neurons in posterior and frontal cortices (BuschmanandMiller, 2007).With regard to order, a signal discriminating targetfrom distractors proceeded from parietal to frontal areas duringefficient search, but from frontal to parietal areas during inefficientsearch. With regard to synchronization, greater synchrony betweenfrontal and parietal neurons occurred at a low gamma-band (22–34 Hz) for inefficient search, but at a higher gamma-band (36–56 Hz)for efficient search. These results suggested that efficient search isprimarily driven bottom-up by perception, and inefficient searchprimarily top-down by working memory. Higher frequency syn-chrony would facilitate the rapid response seen in popout search,whereas lower frequency synchrony would be more robust againsttiming delays that may be induced by the changes in visual attentionthat are more likely to accompany conjunctive search. This functionaldivision of labor may also apply to humans. We test this possibilitywith scalp EEG on human participants using a similar visual searchtask.
2. Materials and methods
2.1. Participants
Ten Japanese university students (male, right-handed) partici-pated in the experiment, aged 23±3.3 years (mean±SD). Partici-pants had normal, or corrected-to-normal vision and were paid fortheir time on the experiment.
2.2. Apparatus and stimuli
A standard desktop computer was used to present stimuli on a 17-inch CRTat about 57 cm from the participant (i.e., 1 cm equals about 1°of the participant's field of view). The field of view extendedapproximately 29° horizontally and 23° vertically. Stimuli wererectangular bars, subtending 2° in length and 0.4° in width of a singlesolid colour (either red, green, blue, or yellow) at an orientation of 0°,45°, 90°, or 135° from horizontal. The background color for all trialswas grey. The display was divided into four equal quadrants byinvisible horizontal and vertical centerlines. A quadrant contained atmost one stimulus item, jittered about its center so that the location ofthe target was clearly identifiable by the containing quadrant.Electrical potentials were collected using a digital electroencephalo-
graph system (Nihon Kohden Neurofax EEG-1100) with an Ag/AgClelectrode cap.
2.3. Conditions
There were two search display type conditions: efficient, wheredistractors shared no feature with the target so that response timesare independent of search set size; and inefficient, where distractorsshared one feature (either colour, or orientation) with the target so thatresponse times increasewith size. In both cases, set sizewas two or fouritems. The display type-size conditions are labeled, Eff-2, Eff-4, Ineff-2,and Ineff-4 for efficient and inefficient, two and four item conditions(i.e., one target plus either one or three distractors), respectively. Eff-4corresponds to a popout condition, and Ineff-4 corresponds to a con-junctive search condition.
2.4. Procedure
Each trial commenced with a fixation period (1500 ms) whereparticipants focused on a small white ring placed at the centerof the screen. The fixation was then replaced by a target cue(1000 ms), positioned at the screen center. The target was thenreplaced by a second fixation delay period (1000 ms). Followingthis delay period, a search set of items was displayed for 2500 ms,or until a key was pressed, whichever came first. Participantswere required to identify the target location within this timeperiod by pressing a key corresponding to the quadrant withinwhich the target was located. Speed and accuracy of response werestressed. Participants pressed either key ‘a’ (upper left), or ‘z’(lower left) with their left hand; or ‘k’ (upper right), or ‘m’ (lowerright) with their right hand to identify quadrants. The assignmentof stimulus items to quadrants was randomized and responses were
352 S. Phillips, Y. Takeda / International Journal of Psychophysiology 73 (2009) 350–354
counterbalanced across trials. Trial timing and example searchdisplays are shown in Fig. 1. Trials were blocked by display type.There were 64 trials per block. A prompt, displayed for 10 s,indicated the start of the next block. There were 3 blocks for eachdisplay type, totaling 6 blocks per session. Each participant did twosessions, separated by about 5 min of rest. Thus, each participantreceived 768 experiment trials (=2 sessions×6 blocks×64 trials).Response keys and times were recorded. Pressing an incorrectkey, or failure to respond within the maximum allotted time wasregarded as an error.
Electroencephalograms (EEG) were measured from the following19 electrode sites of the International 10-20 system: Fp1, Fp2, F7, F3,Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2, with AFz asthe ground electrode. To monitor possible artifacts due to eyemovements, a vertical electrooculogram (EOG) was recorded usingelectrodes placed above and below the right eye, and a horizontal EOGwas recorded from the outer left and right canthi. Electrodes wereattached with gel to reduce impedance to below 5 kΩ. EEG and EOGwere digitized at a rate of 1000 Hz, and were band-pass filtered at0.032 Hz and 300 Hz. The experiment was conducted inside an elec-trically shielded room.
Prior to conducting the experiment sessions, participants receivedgeneral instructions regarding the experiment and EEG/EOG proce-dure. After attaching the head cap and electrodes, participants werethen given specific instructions regarding the task, then a shortdemonstration, followed by a 3–4 min practice session for familiar-ization. The total time required for practice and experiment sessionswas about 1 h. All procedures were approved by the AIST Safety andEthics committee, and conducted after receiving informed consentfrom participants.
2.5. Analysis
Analyses of variance (ANOVAs) were conducted on response errorsand times. Response time analysis was done after removing errorand outlier trials, which were determined by the modified recursivemethod (Selst and Jolicoeur, 1994) on error-free trials. Data wereentered into 2 (display type)×2 (display size) repeated measuresANOVAs to infer significant effects. For response errors, where errorrates are bounded by 0 and 1, an arcsine transform (arcsin
ffiffiffix
p) was
applied to error rates to stabilize variances before conducting analysis(Sheskin, 2004).
EEG data were re-referenced offline to the mean of earlobepotentials A1 and A2. A data window was set at −200 ms to 1000 msrelative to search display onset. Trials containing artifacts (approxi-mately 25% of trials, identified by visual inspection) or response errorswere excluded from further analysis. Independent componentsanalysis, as implemented in EEGLab (Delorme and Makeig, 2004),was used to remove eye movement related components.
Phase-locking values (Lachaux et al., 1999) were used as measuresof synchrony. They have several advantages over coherences mea-sures: (1) The PLV is derived from wavelet decomposition, providingan instantaneous measure of phase, making them applicable to non-stationary signals. By contrast, coherence based on the discrete fouriertransform is essentially a stationary measure due to the temporalinvariance property of the transform. (2) Measures of PLV separatephase and amplitude components, but coherence does not. Since therelative contribution of these components is unknown, it is importantto keep them separate, particularly for cognitive models that assumean independent role for phase (see Discussion section). (3) For PLV,
Fig. 2. Time-frequency plots of PLV significance (z-score) relative to baseline (top parectangle) gamma-bands. Line graphs (lower panel) indicate mean PLVs for higher (top r(right) pairs.
the null hypothesis is derived from participant data (e.g., periodswhen not engaged in the task), rather than white noise. Since neuralsignals in the baseline condition are unlikely to be white noise, co-herence statistics may too easily reject this form of null hypothesis.These advantages make PLVs a more reliable measure of synchrony(Lachaux et al., 1999).
PLVswere computed from stimulus-locked EEG data for each trial asmeasures of synchronization between brain regions. The phase φ(t,f,n,ei)
at time (t), frequency (f), trial (n) and electrode (ei) was computed byfirst convolving the data with a complex Morlet wavelet, defined as:
w t;fð Þ = σ t
ffiffiffiπ
p� �− 12 e
− t2
2σ2t e2πift ;
where σt=FTR/2πf. Following (Lachaux et al., 1999), we set FTR=7.For the analysis, f ranged from 20 Hz to 60 Hz at intervals of 2 Hz. PLVfor electrode pair (ei,ej) was computed as:
PLV t;f ;ei ;ejð Þ =1NjXNn=1
ei /ðt;f ;n;eiÞ−/ðt;f ;n;ejÞ� �
j;where N is the number of trials. Thus, a value of 1 indicates that thetwo signals are synchronous (i.e., separated by 0° or 180°) at aparticular time point and frequency component, and a value of 0indicates that the signals are desynchronous (i.e., separated by 90°).PLVs were normalized (Rodriguez et al., 1999). The normalized values,PLV(t,f,ei,ej)norm, were computed as:
where μbase and σbase are the mean and standard deviation PLV over abaseline period from 200 ms to 0 ms prior to stimulus onset. Hence,normalized PLVs are no longer bounded between 0 and 1. Twofrequency bands were defined as: 22–34 Hz (lower gamma band) and36–56 Hz (higher gamma band), following (Buschman and Miller,2007). ANOVAs were computed for PLVs averaged over regions ofinterest for frontal-parietal electrode pairs, F3–P3, Fz–Pz, and F4–P4,approximating the ipsilateralized positioning of frontal and parietalelectrodes in the (Buschman and Miller, 2007) study (see Resultssection for details).
3. Results
3.1. Behaviour
An ANOVA revealed a significant effect of display type on errors,F(1,9)=8.28, pb02, but not display size, F(1,9)=5.01, pN .08. Theerror rate was greater in the inefficient than efficient condition. Theinteraction of type and size was not significant, F(1,9)=2.87, pN .12.An ANOVA also revealed significant effects of type, F(1,9)=82.86,pb .0001, and size, F(1,9)=76.09, pb .0001, on response times.There was also a significant type by size interaction, F(1,9)=45.62,pb .0001. Post hoc analysis (Newman–Keuls) revealed significantdifferences between Ineff-2 and Ineff-4, pb .002, Eff-2 and Ineff-2,pb .006, and Eff-4 and Ineff-4, pb .0002, but not between Eff-2 andEff-4, pN .9. Mean untransformed error rates and response times areshown in Table 1.
nel), including higher (36–56 Hz, dotted rectangle) and lower (22–34 Hz, solidow) and lower (bottom row) bands for the F3–P3 (left), Fz–Pz (middle), and F4–P4
353S. Phillips, Y. Takeda / International Journal of Psychophysiology 73 (2009) 350–354
Table 2Mean PLVs at each region for each display type-size condition.
354 S. Phillips, Y. Takeda / International Journal of Psychophysiology 73 (2009) 350–354
3.2. Phase-locking
Time-frequency plots of PLV significance (z-score, one-samplet-test) revealed significant increases in frontal-parietal synchronyfor all three electrode pairs in all four conditions from approxi-mately 200 ms post-stimulus onset. Synchrony was most significantin the lower frequency band and was more extended in time forthe inefficient conditions (Fig. 2, top panel). Mean higher and lowerfrequency PLVs for each electrode pair in each display conditionrevealed peaks at approximately 200–250 ms post-stimulus onsetfor all conditions and electrode pairs in both frequency bands (Fig. 2,bottom panel). Accordingly, we defined a temporal region of interestto be 160–320 ms post-stimulus. The plots also revealed a secondregion of interest in the lower frequency band where mean PLVexceeded baseline. This second region was defined as 320–480 mspost-stimulus.
A three-way ANOVA ([type: Eff, Ineff]×[size: 2, 4]×[pair: left,central, right]) was conducted for each frequency band for the 160–320 ms region of interest. For the higher frequency band, there wereno main effects and no interactions on PLV (pN .1). For the lowerfrequency band, there was a main effect of Type, F[1,9]=11.25,pb .01, but no other main effects, or interactions (pN .4). For thesecond region of interest (320–480 ms), an ANOVA revealed asignificant effect of Type, F[1,9]=19.24, pb .01, but no other maineffects, or interactions (pN .15). Mean PLVs for each frequency-timeband are shown in Table 2.
4. Discussion
Our results correspond to one of the main findings of (Buschmanand Miller, 2007): humans, like monkeys, show greater frontal-parietal synchrony for inefficient than efficient visual search. In regardto response time, the difference in search efficiency is consistent withthe numerous studies showing that search is independent of set sizefor targets with unique features, but increases with set size whentargets share features with distractors (Duncan and Humphreys,1989;Treisman and Gelade, 1980; Wolfe et al., 1989). Although the highererror rate for the two-item set size appears anomalous, this differencewas primarily due to the size means for the efficient conditions.Repeated distractors in the efficient-four condition provide anadditional cue making misidentification (e.g., from failing to maintainan accurate memory trace) of the target less likely. In any case, therewas no effect of set size on PLV, so our results are not influenced bythis anomaly.
In contrast to (Buschman and Miller, 2007), we did not observe aneffect of search type for the higher frequency band. One possiblereason for this difference is the amount of training. Monkeys generallyrequire weeks of training over many sessions before being able tosuccessfully perform a task. In our case, participants were given verbalinstructions and a few minutes for familiarization. Thus, monkeyshad far more familiarity with the stimuli than our participants. Thedifference between short-term and long-termmemory influences hasbeen revealed in dual-task and lesioning studies (Woodman and
Chun, 2006). For example, a concurrent working memory loadinterferes with visual search when the targets are varied from trialto trial, but not when the target is constant (Woodman et al., 2007). Alaterality effect was observed for monkeys with lesions to the corpuscollusum and unilateral prefrontal cortex with varied targets, but noeffect for constant target (Rossi et al., 2001). Given that long-termmemory influences bottom-up control of visual search, we expecthumans to exhibit greater synchrony in the higher frequency bandwith more training.
One possible function of low frequency phase-locking in visualsearch is to align the target inmemorywith a display item on the basisof the item's position. In this way, an item can be compared to thetarget while minimizing cross-talk from other items. This sort ofmechanism is central to some models of analogy (Hummel andHolyoak, 1997), where source and target items are compared onthe basis of their common relational roles, which are tagged by thephase of a carrier signal. Thus, being in-phase corresponds to havingthe same role. Alignment could also apply in search, where spatialposition is encoded as phase. More items may need to be compared tothe target before a match is found in inefficient than efficient search.Hence, there will be more periods of phase synchrony, culminating inan overall increase for inefficient search.
Acknowledgements
We thank Eri Sugawara for assistance in collecting the data andArchana Singh for advice on statistical methods. This work wassupported by a Grant-in-aid (19300092) from the Japanese Society forthe Promotion of Science (JSPS). We thank the reviewers for theirthoughtful comments.
References
Bundesen, C., 1990. A theory of visual attention. Psychological Review 97 (4), 523–547.Buschman, T.J., Miller, E.K., 2007. Top-down versus bottom-up control of attention in the
prefrontal and posterior parietal cortices. Science 315 (5820), 1860–1862.Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven
attention in the brain. Nature Reviews. Neuroscience 3 (3), 201–215.Delorme, A., Makeig, S., 2004. EEGLab: an open source toolbox for analysis of single-trial
EEG dynamics including independent component analysis. Journal of NeuroscienceMethods 134 (1), 9–21.
Desimone, R., Duncan, J., 1995. Neural mechanisms of selective visual attention. AnnualReview of Neuroscience 18, 193–222.
Hummel, J.E., Holyoak, K.J., 1997. Distributed representations of structure: a theory ofanalogical access and mapping. Psychological Review 104 (3), 427–466.
Lachaux, J.P., Rodriguez, E., Martinerie, J., Varela, F.J., 1999. Measuring phase synchronyin brain signals. Human Brain Mapping 8 (4), 194–208.
Rodriguez, E., George, N., Lachaux, J.P., Martinerie, J., Renault, B., Varela, F.J., 1999.Perception's shadow: long-distance synchronization of human brain activity. Nature397 (6718), 430–433.
Rossi, A.F., Bichot, N.P., Desimone, R., Ungerleider, L.G., 2001. Top-down, but not bottom-up: deficits in target selection inmonkeyswith prefrontal lesions [Abstract]. Journalof Vision 1 (3), 18a.
Selst, M.V., Jolicoeur, P., 1994. A solution to the effect of sample size on outlierelimination. The Quarterly Journal of Experimental Psychology 47A (3), 631–650.
Sheskin, D.J., 2004. 3rd Edition. Handbook of parametric and non-parametric statisticalprocedures, vol. 3. Chapman and Hall/CRC Press, Boca Raton: FL.
Treisman, A.M., Gelade, G., 1980. A feature-integration theory of attention. CognitivePsychology 12 (1), 97–113.
Wolfe, J.M., 2003. Moving towards solutions to some enduring controversies in visualsearch. Trends in Cognitive Sciences 7 (2), 70–76.
Wolfe, J.M., Cave, K.R., Franzel, S.L., 1989. Guided search: an alternative to the featureintegration model for visual search. Journal of Experimental Psychology: HumanPerception and Performance 15 (3), 419–433.
Woodman, G.F., Chun, M.M., 2006. The role of working memory representations invisual search. Visual Cognition 14 (4–8), 808–830.
Woodman, G.F., Luck, S.J., Schall, J.D., 2007. The role of working memory representationsin the control of attention. Cerebral Cortex 17, 118–124.
