functional parcellation of attentional control regions of the brain

Functional Parcellation of Attentional ControlRegions of the Brain

Marty G Woldorff1 Chad J Hazlett1 Harlan M Fichtenholtz1Daniel H Weissman1 Anders M Dale2 and Allen W Song1

Abstract

amp Recently a number of investigators have examined theneural loci of psychological processes enabling the control ofvisual spatial attention using cued-attention paradigms incombination with event-related functional magnetic reso-nance imaging Findings from these studies have providedstrong evidence for the involvement of a fronto-parietalnetwork in attentional control In the present study we buildupon this previous work to further investigate these atten-tional control systems In particular we employed additionalcontrols for nonattentional sensory and interpretative aspectsof cue processing to determine whether distinct regions inthe fronto-parietal network are involved in different aspectsof cue processing such as cuendashsymbol interpretation andattentional orienting In addition we used shorter cuendashtarget

intervals that were closer to those used in the behavioraland event-related potential cueing literatures Twenty partic-ipants performed a cued spatial attention task while brainactivity was recorded with functional magnetic resonanceimaging We found functional specialization for differentaspects of cue processing in the lateral and medial sub-regions of the frontal and parietal cortex In particular themedial subregions were more specific to the orienting ofvisual spatial attention while the lateral subregions wereassociated with more general aspects of cue processing suchas cuendashsymbol interpretation Additional cue-related effectsincluded differential activations in midline frontal regionsand pretarget enhancements in the thalamus and early visualcortical areas amp

INTRODUCTION

A number of neuroimaging studies of attentionally de-manding tasks have shown activity in areas of the frontaland parietal cortex including dorsal midline frontalregions such as the anterior cingulate (eg MacDonaldet al 2000 Milham Banich Webb amp Barad 2001Hopfinger Buonocore amp Mangun 2000 Shulmanet al 1999 Wojciulik amp Kanwisher 1999 drsquoEsposito etal 1998 Nobre et al 1997 Vandenberghe et al 1997Corbetta Miezin Shulman amp Petersen 1993 CorbettaKincade Ollinger McAvoy amp Shulman 2000 Posner ampPetersen 1990 for a review see Corbetta and Schulman2002) Until recently however these findings werebased on hemodynamic studies using block designapproaches which precluded the ability to delineatewhich of the activity observed in these frontal andparietal cortical regions was related to the control ofattention such as processes related to attentional ori-enting and which was due to stimulus or target pro-cessing and any differential effects of attention on thatprocessing The separation of the brain activity relatedto these two sets of processes requires that an approachother than block design be used such as event-relatedfunctional magnetic resonance imaging (fMRI ) (drsquoEspo-

sito Zarahn amp Aguirre 1999 Rosen Buckner amp Dale1998)

Several of the more recent studies (eg Corbetta et al2000 Hopfinger et al 2000 Shulman et al 1999) haveapplied event-related fMRI to cued-attention paradigmsto separate cue-induced activity (which would includeactivity related to attentional orienting) from target-induced activity To address the severe overlap of thecue and target hemodynamic responses that wouldoccur at the typical stimulus onset asynchronies (SOAs)used in the behavioral (or event-related potential [ERP])literature these studies either focused on trials withlong cuendashtarget SOAs (sup18 sec) (Hopfinger et al 2000)or have used a combination of special trial types signalextraction techniques and moderate SOAs (4ndash5 sec)(eg Ollinger Corbetta amp Shulman 2001 OllingerShulman amp Corbetta 2001 Corbetta et al 2000 Shul-man et al 1999 2002) Several of these studies (egHopfinger et al 2000 see also Kastner et al 1999)reported that spatial attention-directing cues triggeredenhanced activity in the visual sensory cortices contra-lateral to the cued direction of attention prior to theoccurrence of the target stimulus This cue-relatedactivity was proposed to reflect a biasing of sensorycortical activity in favor of the expected target stimulus(see Desimone amp Duncan 1995)1Duke University 2Harvard University

copy 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 161 pp 149ndash165

The present experiment builds on the methods ofthese earlier studies for distinguishing cue- and target-related activity to further determine the contributionsof various regions in the fronto-parietal network iden-tified as active during attention-orienting tasks This wasdone by extending these approaches in mainly twoways (1) using SOAs (900ndash1900 msec) that are closerto those used in behavioral studies and (2) addingcontrol conditions that allow the decomposition ofcue-triggered activity into activity more specifically re-lated to orienting and activity related to general aspectsof cue processing

First regarding the SOAs long delays between cueand target may make it more likely that subjects engageother cognitive processes such as working memoryduring the delay period Thus it is possible that dueto the relatively long cuendashtarget SOAs in these severalprevious studies (eg 8 sec in Hopfinger et al 2000 4ndash5 sec in Corbetta et al 2000 5ndash6 sec in Shulman et al1999) the broad network of areas observed may haveincluded some regions more related to working memoryrather than attentional orienting per se Indeed thefrontal and parietal areas previously observed to beactivated by spatial attentional cues included areas sim-ilar to those activated in working memory tasks (eg Jhaamp McCarthy 2000 Nystrom et al 2000 LaBar GitelmanParrish amp Mesulam 1999 drsquoEsposito et al 1998) Toreduce this possibility we used a substantially shortercuendashtarget SOA (randomly 900 and 1900 msec) Thischange not only reduces the likelihood of invokingworking memory processes but also better controlsthe timing of attention orienting (subjects learn thattargets may come as early as 900 msec so that theywould be more likely to begin those processes rightaway after the cue) In addition using shorter SOAsmakes the results more directly comparable to the widebody of behavioral and ERP work using similar SOAs inendogenous cueing tasks (eg 800ndash1200 msec in Pos-ner Snyder amp Davidson 1980 700 msec in Downing1988 1100 msec in Hopf amp Mangun 2000 700 msec inEimer 2000)

Second as noted above previous studies have suc-cessfully separated cue-related activity from target-related activity (eg Corbetta et al 2000 Hopfingeret al 2000 Shulman et al 1999) However within thebrain activity that is triggered by cues there is a need tofurther distinguish activity due to the actual orienting ofattention from activity due to sensory and semanticoperations required for processing and interpretingthe cues themselves (eg cue-interpretation processes)Both Hopfinger et al (2000) and Shulman et al (1999)included a control condition for the cues (lsquolsquocue controlrsquorsquoin Hopfinger et al 2000 lsquolsquopassive cuesrsquorsquo in Shulmanet al 1999) but these trials were in a separate passivelyviewed run and thus were not intermixed with the othertrial types and were not behaviorally relevant Whilethese runs may have controlled for the most basic

sensory stimulation of the cues they did not requireany sort of processing for meaning or interpretation andthus they were limited in their effectiveness as controlsfor the general cue-processing operations that would belikely to occur during the experimental attention runsThus in the present study we included a lsquolsquocue-interpre-tationrsquorsquo control trial type that was randomly intermixedwith the other attention-directing trial types On thesetrials participants needed to perceive process andinterpret a behaviorally relevant cue which instructedthem that orienting of their spatial attention was unnec-essary on that particular trial These cues which we havetermed lsquolsquointerpret-cuesrsquorsquo could not be ignored as theywere intermixed with all other trial types in the sameruns and needed to be interpreted and processed toachieve correct behavior Thus interpret-cue trials re-quired both sensory and cue-interpretion processessimilar to those engaged by attend-cues but did notrequire attentional orienting to a particular spatial loca-tion in the visual field

Thus our paradigm and analysis structure is lsquolsquohierar-chicalrsquorsquo in nature At the first level activity can beseparated into cue- and target-related activity This sep-aration and assessment of target-triggered activity allowsus to determine which sensory regions would be ex-pected to be the site of any prestimulus biasing duringcued orienting as well as to assess and separate outactivity due to motor-related processes At the secondlevel cue-related activity can be further split into activityrelated to attentional orienting and activity more relatedto general cue-interpretation processes allowing us tobetter isolate attention-orienting networks while deter-mining the functional roles of different subregions of thefronto-parietal executive control network

RESULTS

Subjects were presented with a series of trials each ofwhich began with an instructional cue presented atfixation (Figure 1) Active-attend (ie attention-direct-ing) cues consisted of either the letter L or R whichinstructed the subject to attend to a location in the leftor right lower visual field to detect a possible faint targetthat might occur in that location Some of the instruc-tional cues consisted of the letter P These were thelsquolsquointerpret-cuersquorsquo control trials instructing the subject tonot orient their attention and to not attend for targetsIn some active-attend trials a target would occur at arandomized time (900 or 1900 msec) after the onset ofthe cue (attend-cue-plus-target trials) In other active-attend trials (attend-cue-only trials) as well as in theinterpret-cue trials no target was presented so that thebrain response would be due to the cue only Thesevarious trial types were randomized with lsquolsquono-stimrsquorsquo trials(ie periods of fixation only) so that the full event-related responses to the various trial types could beextracted using selective averaging (Buckner et al 1998

150 Journal of Cognitive Neuroscience Volume 16 Number 1

Burock Buckner Woldorff Rosen amp Dale 1998 Dale ampBuckner 1997 see Methods) Each trial (including thelsquolsquono-stimrsquorsquo trials) lasted 4500 msec

Note the hierarchical structure of the trial types no-stims interpret-cue left- and right-attend-cue-only andleft- and right-attend-cue-plus-target (Table 1a) Thishierarchical structure was designed to allow variouskey contrasts to be performed that could isolate differ-ent brain responses associated with specific processing

components (Table 1b) while also subtracting outthe overlapping hemodynamic responses All stimulustrial types were also statistically contrasted with thelowest condition level the no-stim trials The results ofthe voxelwise analysis of these various contrasts willfirst be presented followed by the regions-of-interest(ROI) analyses

Figure 2 displays horizontal sections (at the Talairachz coordinate of +44 Talairach amp Tournoux 1988)showing areas whose activation varied significantly be-tween the various trial types in the voxelwise analysesContrasts between each of the different trial types(interpret-cue left and right-attend-cue-only left- andright-attend-cue-plus-target) versus the no-stims areshown in Figure 2A These contrasts reveal the fullevent-related activation maps for these trial types rela-tive to a low-level baseline Figure 2B shows thecorresponding hierarchical subtraction images relativeto the level closer below it in the hierarchy (eg left-attend-cue-only vs interpret-cue) Also shown are mapsderived for the activations collapsed over side of atten-tional focus (eg the average response to left-attend-cue-only and right-attend-cue-only vs interpret-cues)

Cue-Related Activity Effects in the Lateral versusMedial Regions of the Frontal and Parietal Cortex

Figure 2A reveals that relative to the no-stim trials alltrials with instructional cues (including interpret-cuetrials) activated the parietal and frontal areas These

Time 0Cue

ISI

Time 900 or 1900Faint Target(Duration 100)

ISI

Time 2700End of Trial Signal(Duration 100)

ITI

Time 4500End of Trial

Time(milliseconds)

Visual Cued Spatial Attention Task

Figure 1 Schematic diagram of a compound event cued-attentiontrial At the beginning of the trial single-letter cues were presented(lsquolsquoLrsquorsquo lsquolsquoRrsquorsquo or lsquolsquoPrsquorsquo) at central fixation instructing the subject to eithercovertly attend to a location in the left or right lower visual field todetect a possible faint dot target there or to interpret the cue but thennot attend for a target In some trials there was no target so that thebrain response would be due only to the cue At 2700 msec a briefvisual stimulus noting the end of the trial (the letters lsquolsquoREPrsquorsquo) waspresented in the midline lower visual field at which point the subjectwas to press a button if they had detected a target ISI = interstimulusinterval ITI = intertrial interval

Table 1a Hierarchical Design of Trial Types

Attend-left-cue-plus-target

lsquolsquoLrsquorsquo Target EOT

Attend-right-cue-plus-target

lsquolsquoRrsquorsquo Target EOT

Attend-left-cue-only(not followed by target)

lsquolsquoLrsquorsquo EOT

Attend-right-cue-only(not followed by target)

lsquolsquoRrsquorsquo EOT

Interpret-cue lsquolsquoPrsquorsquo EOT

No-stim

Table 1b Functional Contrasts and Associated Processes

ContrastAssociated Processing

Component(s)

Interpret-cueversus no-stim

Visual stimulation by fovealpresentation of a letter

Processing and interpretinga symbolic cue

Left-attend-cue-onlyversus interpret-cue

Attentional orienting to theleft visual field

Prestimulus biasing of right(contralateral) visualsensory cortex

Right-attend-cue-onlyversus interpret-cue

Attentional orienting to theright visual field

Prestimulus biasing of left(contralateral) visualsensory cortex

Left-attend-cue-plus-target versus left-attend-cue-only

Processing of an attendedtarget in the left visual field

Right-attend-cue-plus-target versus right-cue-only

Processing of an attendedtarget in the right visual field

Woldorff et al 151

areas are in the vicinity of those reported previouslyas being involved in the orienting of visual spatialattention (eg Corbetta et al 2000 Hopfinger et al2000) Notably however even the interpret-cues forwhich subjects did not need to orient their spatialattention activated a portion of these putative atten-tional control areas in the frontal and parietal cortexHowever the active-attention trial types appeared toelicit a greater and more extensive activation in someof these regions

The hierarchical contrasts shown in Figure 2B re-vealed important details about these activation pat-ternsmdashnamely that the frontal and parietal areasactivated by the interpret-cues and attend-cues did nothave the same spatial distribution Specifically the acti-vations elicited by the interpret-cues were quite lateralwhereas the active attend-cues triggered this lateralactivity plus additional activation in more medial por-tions of the frontal and parietal regions This pattern isreflected in the hierarchical contrast maps in that aftersubtracting out the interpret-cue response from theattend-cue-only responses the residual activation isconsiderably more medial than the activation for inter-pret-cues versus no-stims (see Figure 2)

Cue-Related Activity Effects in the MidlineFrontal Areas

All of the instructional cues (including interpret-cues)also activated several midline dorsal regions of thefrontal cortex (Figure 2A) As before the hierarchicalcontrasts (Figure 2B) revealed additional specificityThe interpret-cues activated an anterior medial dorsalarea that appeared to be in the anterior cingulatecortex (ACC) In the activation map of the contrastof attend-cue-only versus interpret-cue there was noadditional activity apparent in this region (althoughROI analyses presented below revealed that there wasa small degree of additional activity) This responsepattern differed sharply from that of the slightlymore posterior supplementary motor area (SMA)which was not responsive to any of the cue-only trials(see below)

Target-Related Activity

The hierarchical contrast between attend-cue-plus-tar-get and attend-cue-only trials was aimed at isolatingactivity related to target processing by subtracting out

Figure 2 Activation maps for various contrasts between the trial types (A) Contrasts of each of the trial types versus the no-stim at thelevel of superior cortex showing the full event-related activation maps relative to a low-level baseline The contrast images are overlaid on the T1-weighted structural images from a single subject that was normalized into the same Talairach space Note that relative to the no-stim trials all ofthe instructional cues (including the interpret-cues) activated parietal and frontal areas with the active-attention trial types appearing to elicit agreater and more extensive activation in these regions (B) The corresponding hierarchical subtraction images relative to a level closerbelow it in the hierarchy of trial types revealed greater detail concerning the distributions of these effects The frontal and parietal regionsactivated by the interpret-cues were quite lateral (green arrows) whereas the regions activated by the attend-cues relative to the interpret-cueswhich were more specific to attentional orienting were more medial (blue arrows) At this level of the brain through superior cortex the target-related processing derived from the attend-cue-plus-target versus attend-cue-only contrast (right panels) revealed significant activity in leftsensorimotor cortex and the supplementary motor area (SMA) (red arrows) For both (A) and (B) the color scale for these t value contrast imagesrange from 325 to 8 (dark red to yellow) for the contrasts that are collapsed across the leftright factor (columns 4 and 7) and from 275 to 8 (darkred to yellow) for the uncollapsed contrasts (columns 1ndash3 5ndash6)


the overlapping cue-related activity To encourage sub-jects to begin attending immediately for these trialstargets were sometimes presented very soon after thecue (900 msec) although on other trials they werepresented later (1900 msec) However since no targetwould come in the attend-cue-only trials subjects wouldattend continuously from the cue until the end of trial(EOT 2700 msec) for these trials Thus the bestcontrast for identifying (attended) target-related activitywas attend-cue-plus-late-target minus attend-cue-onlysince the duration of attentional maintenance was mostsimilar in these two conditions The results of thesecontrasts are shown in Figure 2B (right panels) (see alsoTable 2)

In contrast to the cue-sensitive ACC mentioned abovethe attend-cue-plus-target versus attend-cue-only con-trast (a subtraction specific for isolating activity associ-ated with processing the target and performing theright-hand button-press motor response) yielded a

slightly more posterior medial dorsal region that seemslikely to be the SMA Target-related activity was alsoobserved in the left posterolateral inferior frontal cortex(likely SII) in the visual sensory cortex contralateral tothe target and in the thalamus (see below) Strongtarget-related activity was also observed in various addi-tional motor structures including the left sensorimotorcortex (Figure 2B) right cerebellum and the caudateconsistent with the task requirement of pressing abutton with the right hand upon detection of a target

Region-of-Interest Analyses Frontal and ParietalExecutive Control Areas

The main ROIs that were selectively activated in thehierarchical contrasts described above were analyzedfurther More specifically the event-related hemody-namic response functions (HRFs) generated by time-locked averaging for the various event types in those

Table 2 Talairach Coordinates and Cluster Sizes of Key ROIs Activated in the Voxelwise (SPM) Analyses along with Results of theStatistical Contrasts between the Raw Event-Related Responses in These ROIs

Talairach Coordinates

ROI Number Regions x y z Defining ContrastNumberof Voxels

P value for ROI(One-Tailed)

1 Left lateral frontal cortex iexcl48 3 43 interpret-cue versus no-stim 71 25Eiexcl05

2 Left medial frontal cortex iexcl23 iexcl4 46 attend-cue-only versus interpret-cue 48 0051

3 Left lateral parietal cortex iexcl28 iexcl62 41 interpret-cue versus no-stim 79 62Eiexcl06

4 Left medial parietal cortex iexcl18 iexcl58 48 attend-cue-only versus interpret-cue 72 017

5 Right medial frontal cortex 27 1 46 attend-cue-only versus interpret-cue 73 00089

6 Right lateral frontal cortex 46 6 43 interpret-cue versus no-stim 58 69Eiexcl05

7 Right medial parietal cortex 20 iexcl57 50 attend-cue-only versus interpret-cue 66 00071

8 Right lateral frontal cortex 32 iexcl61 45 interpret-cue versus no-stim 106 17Eiexcl07

9 ACC iexcl1 11 47 interpret-cue versus no-stim 24 63Eiexcl07

10 SMA iexcl2 0 43 attend-cue-plus-target versusattend-cue-only

32 37Eiexcl08

11 Left sensorimotor cortex iexcl44 iexcl23 50 attend-cue-plus-target versusattend-cue-only

242 21Eiexcl09

12 Left SII iexcl55 iexcl25 20 attend-cue-plus-target versusattend-cue-only

75 27Eiexcl08

13 Left TPJ iexcl52 iexcl52 14 attend-cue-plus-target versus no-stim 17 11Eiexcl06

14 Right TPJ 58 iexcl45 16 attend-cue-plus-target versus no-stim 48 20Eiexcl08

Right cerebellum 21 iexcl57 iexcl17 attend-cue-plus-target versusattend-cue-only

124 88Eiexcl07

Left thalamus iexcl12 iexcl21 0 attend-cue-plus-target versus no-stim 112 56Eiexcl09

Right thalamus 12 iexcl21 0 attend-cue-plus-target versus no-stim 106 15Eiexcl08

Left dorsal occipital cortex iexcl21 iexcl90 14 (all) left-attend-cue-plus-target versus(all) right-attend-cue-plus-target

57 0012Eiexcl08

Right dorsal occipital cortex 30 iexcl85 15 (all) left-attend-cue-plus-target versus(all) right-attend-cue-plus-target

119 25Eiexcl06

Woldorff et al 153

ROIs were extracted by subtracting the response to eachof these types relative to the no-stims yielding HRFtime-courses (with the adjacent-trial overlap subtractedaway) for each of the event types in these ROIs Theoverlaying of these ROI response curves for the variousevent types enabled a closer examination of the relativespecificity of the various cognitive processing functionsin these various brain areas

The ROIs in the superior cortical areas that wereanalyzed in this way are indicated in Figure 2B (ar-rows) They include the lateral areas in the frontal andparietal cortex (derived from the interpret-cue vs no-stim contrast) the medial areas in the frontal andparietal cortex (derived from the attend-cue-only vsinterpret-cue contrast) the medial dorsal ACC (derivedfrom the interpret-cue vs no-stim contrast) and theSMA and left motor cortex (derived from the attend-cue-plus-target vs attend-cue-only contrasts) In eachof these ROIs statistical analyses were performed on

the peak amplitudes of the raw time-locked averageddata which confirmed that there were significantamplitude differences at the regional level betweenthe relevant trial types (Table 2)

Figure 3 shows the HRF time-courses for the variousevent types relative to the no-stims in the lateral andmedial areas of frontal and parietal cortex The figureshows that in the lateral areas the responses to theinterpret-cues were roughly as large as those for theattend-cue-only and attend-cue-plus-target trials Incontrast in the medial areas the responses to theinterpret-cues were considerably smaller than thosefor the attend-cue trials These differential responseamplitude patterns are reflected in the maps ofFigure 2B More specifically the similar level of activityin the lateral areas for the different types of cuesresulted in those regions subtracting out in the mapsfor the attend-cue-only versus interpret-cue contrastHowever in these same maps the differential activity

Figure 3 ROIs and hemodynamic time-courses in the superior frontal and parietal cortex The main superior cortex ROIs that were selectivelyactivated in the hierarchical contrasts are shown in different colors on a horizontal section (z = +44) The HRF time-courses for the various eventtypes relative to the no-stims are shown for the lateral and medial ROIs in the frontal and parietal cortex (ROIs 1ndash8 green and blue areas) Bysubtracting off the responses to the no-stims the overlap from responses to adjacent trials in the sequence was removed The time-courses showthat in the lateral areas the responses to the interpret-cues were almost as large as those for the attend-cue-only and attend-cue-plus-target trialsbut in the medial areas the interpret-cue responses were considerably smaller than for the attend-cue trials (black arrows)


more medially for these trial types resulted in robustdifferences in the medial regions

To statistically assess whether lateral and medialregions of parietal and frontal cortex responded differ-ently to attend-cues and interpret-cues we tested thedata for an interaction between cue type (attend cue-only interpret-cue) and region (medial lateral) sepa-rately for the left frontal right frontal left parietal andright parietal ROIs This was accomplished by enteringthe peak amplitudes of the raw event-related re-sponses to interpret-cues trials and attend-cue-onlytrials from these ROIs into a random effects analysisof variance (ANOVA) (Figure 4A) The statistical inter-action confirming this differential distribution of activ-ity was significant in both left and right parietalcortices ( p lt 01 in both cases) and in left frontalcortex ( p lt 05) this interaction did not hold in theright frontal cortex Note that since the magnitude ofinterpret-cue and attend-cue-only activations is nearlythe same in lateral regions there should be nodifference in spatial extent based on magnitude aloneand yet the interpret-cue activity drops off much moredramatically in the more medial regions Thus theinteraction effect is most likely due to a differentialdistribution of attend-cue and interpret-cue activity andis not an artifact based on spatial extent or thresh-olding effects in the t maps

Because there were differences in overall activationlevels for the lateral and medial areas (the medialregions tended to be more weakly activated) theseinteraction analyses were also performed after normal-izing the activation amplitudes to the mean level(across subjects) of the stronger conditionmdashie theattend-cue conditionmdashin both lateral and medial brainregions (Figure 4B) After such normalization thesignificance of these interactions became even stron-ger and more highly significant (left parietal = 01 rightparietal = 005 left frontal = 005 right frontal = still notsignificant) The statistical interactions in these frontaland parietal subregions therefore provide statisticalsupport for the view that lateral and medial regionshave a differential selectivity for the spatial orienting ofattention

Region-of-Interest Analyses Midline DorsalFrontal Areas

Figure 5A shows the time-courses of the time-locked-averaged response functions for the ACC SMA and leftsensorimotor cortex ROIs For the ACC (ROI 9) thereappeared to be a gradation of responses for the interpret-cue attend-cue-only and attend-cue-plus-target trialtypes ROI analyses of the peak amplitudes confirmedthese relationships in the ACC indicating some sig-nificant activity for interpret-cue trials alone (relative tono-stims) ( p lt 0001) significantly more activity forattend-cue-only trials than for interpret-cues ( p lt 01)

and significantly more still for attend-cue-plus-targettrials than for attend-cue-only trials ( p lt 0002) In sharpcontrast just posterior to this ACC region in what wouldappear to be SMA (ROI 10) the response was almostcompletely selective for those trials with targets withlittle response for the interpret-cues or the attend-cue-only trials

Figure 4 Lateralmedial interactions in the superior frontal andparietal cortex for cue-related activity Peak amplitudes for the time-locked average responses to interpret-cue and attend-cue-only trials(after subtracting out the no-stim averages) were entered into repeatedmeasures ANOVAs These measures and the statistical analysis of themassessing for a significant interaction are shown both for the rawamplitude values and for the amplitude values after being normalizedto the mean level (across subjects) of the stronger condition (attend-cue) This interaction was significant for both the parietal lobes and theleft frontal cortex The statistical interactions in these frontal andparietal subregions provide additional statistical support for thedifferential selectivity of these areas for the spatial orienting ofattention

Woldorff et al 155

Figure 5 ROIs andhemodynamic time-courses in themedial dorsal cortex SII and TPJAs in Figure 3 various ROIs selec-tively activated in the hierarchicalcontrasts are shown on two hor-izontal sections(z = +44 and z = +14) with theHRF time-courses for the variousevent types relative to the no-stims in these ROIs (A) Time-courses in the ACC SMA and leftsensorimotor cortex Notice thegraded response across trial typesin the ACC and the high selectivityin the SMA and in the left sensor-imotor cortex for trials with targets(B) Time-courses in left SII and inleft and right TPJ Notice the highselectivity in left SII for target trialsSII = secondary somatosensorycortex TPJ = temporal-parietaljunction


Region-of-Interest Analyses Target-RelatedActivations

Like the medial dorsal SMA region the activations in leftsensorimotor cortex were present almost exclusively forthose trials with targets ( Figure 5A) consistent with thetask requiring a button-press only on those trials Sim-ilarly a left inferior parietal region (likely the left sec-ondary somatosensory area SII) also was active only fortrials with targets ( Figure 5B) with flat responses for theother trial types

Because of previous reports indicating target-selectiveactivity in the left and right temporalndashparietal junctions(TPJs) (eg Corbetta et al 2000) additional analyseswere performed on these brain areas Figure 5B showsthese regions as activated by the attend-cue-plus-targetminus no-stim contrast The time-courses from theseROIs reveal that all the trial types appeared to elicit someactivation in these regions (relative to no-stims) In theleft TPJ neither the SPM maps nor the ROI analysesindicated any significant differences in the activation levelbetween the various trial types including those withtargets In the right TPJ the ROI analysis indicated thatthere was slightly more activity for attend-cue-only thanfor interpret-cue trials ( p lt 02) and also slightly moreactivity for attend-cue-plus-target than for attend-cue-only trials ( p lt 05)

Some previous studies have also found that thefrontal or parietal regions activated by cues were alsoactivated by the targets themselves (ie intraparietalsulcus in Corbetta et al 2000 precentral sulcussupe-rior frontal sulcus in Shulman et al 1999) As notedabove t contrasts at the voxelwise level do not indi-cate this in the present experiment At the ROI levelhowever the time-courses shown in Figure 3 suggestthat cue-plus-target trials elicited slightly more activitythan cue-only trials in some of these regions Althoughthis additional activity for the targets in these regionswas fairly small (on the order of 10ndash15 of the totalactivity in these regions) it did reach significance inthe ROI analyses for three of the eight regions ( leftlateral frontal left medial frontal and left lateralparietal cortex) based on a one-tailed random effectst test of two time points around the response peakHowever it is clear that the bulk of activity observedin these areas at least in the present experiment wastriggered by the cues rather than the targets It ispossible that target activity was much less pronouncedin our study relative to previous ones because we usedfaint small targets

Effects in the Visual Cortex and Thalamus

Both attend-cue-plus-target and attend-cue-only trialsenhanced activity in the occipital cortex contralateralto the direction of attention an effect that is best seen ina comparison of left-attend versus right-attend condi-tions (Figure 6A) The contrast for left-attend-cue-plus-

target versus right-attend-cue-plus-target (Figure 6Atop) showed particularly strong contralateral effects inthe dorsal occipital cortex presumably in part becausethe targets in these trials were unilateral stimuli in thelower visual field To assess for possible pretarget biasingeffects in visual sensory cortex in response to theattention-directing cues this contrast of left-attend-cue-plus-target versus right-attend-cue-plus-target was usedto define ROIs for these low-level retinotopically orga-nized visual sensory areas The HRF time-courses inthese ROIs (after subtracting out the no-stim responses)were then extracted for these conditions as well as forthe left- and right-attend cue-only conditions ( Figure 6A)The ROI time-courses show the clear relative enhance-ment of the response level in the dorsal occipital areascontralateral to the direction of attention for both theattend-cue-only trials and attend-cue-plus-target trialsThe contralaterality of attend-cue-plus-target activitywas confirmed statistically by ANOVA of the peak ampli-tudes of the raw time-locked averages in these dorsaloccipital ROIs ( p lt 000003) In addition the sameANOVA test applied to the peak amplitudes for theattend-cue-only trials in these dorsal occipital ROIsconfirmed that the centrally presented attention-direct-ing cues also produced contralateral effects in thesesame target processing areas even when no targets werepresented ( p lt 0003)

Robust activations were also seen bilaterally in thethalamus particularly for the attend-cue-plus-targettrials (Figure 6B) ROIs defined from the attend-cue-plus-target versus no-stim contrast were used to extracttime-courses for all the event types in these areas and toperform ROI analyses The time-course response func-tions indicated that all the event types including theinterpret-cues activated the thalamus The attend-cue-plus-target trials showed a particularly large responsebut even the attend-cue-only trials appeared to haveenhanced activity relative to the interpret trials (arrowsin Figure 6B) A statistical analysis of the peak ampli-tudes in these thalamic ROIs indicated that this differ-ence (ie attend-cue-only vs interpret-cue) wassignificant ( p lt 025 and p lt 01 for the left and rightthalamus respectively)

DISCUSSION

In this study event-related fMRI was used to study thebrain networks involved in the voluntary control ofspatial attention A spatial cueing paradigm was em-ployed in which an initial cue instructed subjects as towhether and where to direct their attention for apossible upcoming target stimulus In contrast to previ-ous event-related neuroimaging studies the parametersused for this study including the cuendashtarget SOAs andintertrial intervals were very similar to previous behav-ioral and ERP studies using cued-attention paradigmswith endogenous cueing (eg Hopf amp Mangun 2000

Woldorff et al 157

Posner et al 1980) Furthermore we extended themethodologies for separating cue- and target-relatedactivity introduced in prior studies (cf Corbetta et al2000 Hopfinger et al 2000) in ways that allowed us todistinguish not only between target- and cue-relatedactivity but also between cue-related processes thatenable attentional orienting versus those related tomore general aspects of cue processing such as cuendashsymbol interpretation Using this method activity relat-ed to attentional orienting presumably lsquolsquopretarget bias-ingrsquorsquo was found in the early visual cortical areas andthalamus The ACC an area with a controversial role inattentional orienting showed a graded response acrosscue types with some degree of activity during cue-interpretation processes and additional activity aboveand beyond this in response to attention-directing cuesMost importantly the data revealed functional parcella-tion within the frontal and parietal control regions withthe more medial regions being more specific for atten-tional orienting and the more lateral regions beinginvolved in more general aspects of cue processingsuch as cuendashsymbol interpretation

Executive Control and Functional Specificity inthe Frontal and Parietal Regions

A key result in this study was that all of the instructionalcues (including the interpret-cues) activated the parietaland frontal areas that were near the same areas previ-ously implicated as being involved in the orienting ofspatial attention (eg Corbetta et al 2000 Hopfingeret al 2000) However in both frontal and parietalcortex the activity elicited by the interpret-cues wasfound in the more lateral of the frontal and parietalregions whereas the attend-cues triggered this activityPLUS additional activation in adjacent subregions thatwere more medial This activity pattern was explored ingreater detail by extracting the time-locked averagedtime-courses in these ROIs for the attend-cue-only andinterpret-cue trial types and analyzing their peak ampli-

Figure 6 ROIs and response time-courses in the occipital cortex andthalamus (A) ROIs activated in the dorsal occipital cortex in the left-attend-cue-plus-target versus right-attend-cue-plus-target conditionsalong with the corresponding HRF time-courses in these ROIs for thevarious event types relative to the no-stims Note the larger responsescontralateral to the direction of attention For target trials this wouldinclude an enhanced response to the unilateral targets at attendedpositions However for attend-cue-only trials no target is presentedand therefore the activation of these contralateral regions suggests aretinotopically based lsquolsquobiasingrsquorsquo signal in expectation of a possible targetin the attended region (B) ROIs activated in the thalamus in thehierarchical contrasts along with the corresponding HRF time-coursesfor the various event types relative to the no-stims The response forattend-cue-plus-target trials was substantially larger than for the othertrial types but even the attend-cue-only responses were enhancedrelative to the interpret-cue responses (arrows) The t value colormapping is identical to that used in Figure 2


tudes These additional analyses revealed statisticallysignificant lateralmedial interactions for these ROIsand confirmed that the differential patterns observedin the SPM activation maps were not an artifact of theparticular threshold at which these maps were createdNote also that the attend-cue activity found in themedial regions was likely not due to motor preparationfor two reasons First motor preparation for a right-handed button-press would likely be lateralized andthese activations were not Second the lack of cue-period activation in areas much more sensitive to motorpreparation (eg SII left sensorimotor cortex and evenSMA) strongly argues that activation seen in other areasis unlikely to be motor preparation

The similar activation for interpret-cue and attend-cue-only trials in the lateral frontal and parietal regionssuggests that these areas are activated by the engage-ment of some common cognitive process (or process-es) for these different cue types One clear candidateprocess is cuendashsymbol interpretation All the cues inthis study including the lsquolsquointerpretrsquorsquo ones were rele-vant and needed to be attended and processed bysubjects to interpret the meaning of the cue stimulusand decide what to do on that trial In contrastattend-cue-only trials produced substantially greateractivity than interpret-cue trials in the medial frontaland parietal regions suggesting that these regions playa relatively specific role in the orienting of visualspatial attention

Note that the present design allowed distinctionsbetween lateral and medial regions of the frontal andparietal cortex that would be more difficult without anintermediate cuendashcondition level that was randomizedinto the sequence (eg the lsquolsquointerpret-cuersquorsquo trials in thepresent study) Hemodynamic responses triggered bythe instructional cues themselvesmdashfor example relativeto the prestimulus baseline rather than to an interme-diate condition levelmdashcould reflect processes that arenot specific to visual attentional orienting Moreoverthe inclusion of interpret-cue trials in separate sensorycontrol blocks (eg Hopfinger et al 2000) without atask is not sufficient to control for such processes Whenpresented in a separate block these cues are much lesslikely to be fully processed and to engage cue-interpre-tation processes By using a randomized trial sequencethat included both attend-cue and interpret-cue trialswe were able to distinguish functionally specific activa-tion patterns that likely correspond to distinct aspects ofcue processing

As noted above HRF time-courses for the variousevent types relative to the no-stims provided furtherevidence for a functional dissociation between lateraland medial regions of the frontal and parietal cortexThese time-courses however revealed that althoughthe interpret-cue response amplitudes in the medialareas were considerably lower than those for attend-cue-only trials they were not flat This may be due to

some spatial smearing of the hemodynamic responsesbetween the regions (due eg to the averaging of thedata across subjects) although it could simply reflectthat the activations in these adjacent executive controlareas are not lsquolsquoall or nonersquorsquo In either case it is clear thatthe medial areas likely perform a function that is rela-tively specific for the orienting of visual spatial attentionwhile the lateral areas likely perform a more generalfunction such as cuendashsymbol interpretation

Given the present evidence that lateral and medialregions of the frontal and parietal cortex perform some-what different functions one might wonder why theywould be so physically near to each other One possi-bility is that the cuendashsymbol interpretation required forall the cue types involves higher level executive process-es which includes evaluating the cue meaning andmaking the decision of what to do in response to thatcue even if it does not involve the spatial orienting ofattention It may be that there is a functional advantage(eg speed of communication) in having neuroanatom-ical proximity between higher level brain areasperforming interpretive and decisional processes relatedto meaningful environmental stimuli and higher levelareas that perform orienting of attention in response tosuch stimuli when appropriate

The frontal areas activated by cues appeared to over-lap with the frontal eye fields Such results are consistentwith previous studies reporting neuroanatomical over-lap of the frontal areas involved in the covert focusing ofattention (as used here) and those areas involved inovert control of eye position (Nobre Gitelman Dias ampMesulam 2000 Corbetta 1998) Note however thisresult does not suggest that subjects were moving theireyes as it is known that the frontal eye fields can beactivated without concomitant eye movements (Gitel-man Parrish LaBar amp Mesulam 2000)

Finally the current findings of cue-triggered activityin the frontal and parietal cortex occurred at relativelyshort SOAs that are comparable to those used in thebehavioral and ERP literatures This helps to rule outthe possibility that cue-triggered activity observed inthese areas in some previous studies (eg Hopfingeret al 2000) occurred mainly because long SOAsengaged working memory processes to a greater de-gree than is normally the case However the Hop-finger et al (2000) study reported cue-related frontalactivation that included more anterior regions of leftlateral prefrontal cortex that were not activated in thepresent experiment Thus it is possible that activationof these anterior regions was associated with highworking memory demands (due to the long SOAs)rather than visual spatial attention per se We acknowl-edge however that attention-directing cues may en-gage working memory processes even at short SOAsIndeed some evidence indicates that working memorymay be used to store representations of task-relevant(ie to-be-attended) stimuli (eg de Fockert Rees

Woldorff et al 159

Frith amp Lavie 2001) and spatial locations (eg AwhAnllo-Vento amp Hillyard 2000)

Executive Control and Functional Specificity forResponse Selection and Target Processing

By contrasting cue-plus-target trials with cue-only trialswe were able to extract target-related activity As ex-pected targets selectively activated classical motor-relat-ed areas associated with a right-handed button-pressmdashnamely the left sensory-motor cortex the right cerebel-lum the caudate and the SMA (see Figure 5 and Table 2)HRF time-courses for each of the trial types (after sub-tracting out the overlap by contrasting to the no-stims)revealed that each of these ROIs was selectively activatedby target processing In particular activation occurredwhen there was a target (and hence a button-press) onattend-cue-plus-target trials while the HRFs for attend-cue-only and interpret-cue trials in which no motorresponse was required were relatively flat

Target effects in the midline dorsal regions of thefrontal cortex were somewhat more complex Two adja-cent regions were robustly activated but distinguishedthemselves by behaving differently as a function of trialtype The more anterior of the two which appeared to bedorsal ACC showed a gradated response to the differenttrials types with some response for the interpret-cuesmore for attend-cue-only and still more for attend-cue-plus-target In sharp contrast a slightly more posteriorregion which seems likely to be SMA was activatedstrongly when there were targets (and button-presses)but exhibited little response in the cue-only conditions

The more anterior area is likely to include the sameACC area that has been activated in numerous experi-ments in which various cognitive processes are engaged(reviewed in Paus 2001) Some recent findings areconsistent with the view that this area is importantmainly for detecting andor resolving response conflict(eg Botvinick Braver Barch Carter amp Cohen 2001Van Veen Cohen Botvinick Stenger amp Carter 2001MacDonald Cohen Stenger amp Carter 2000 CarterBotvinick amp Cohen 1999) However it has been shownthat this area is activated just by anticipation of targets(Murtha Chertkow Beauregard Dixon amp Evans 1996)Moreover in block-design attention experiments inwhich the task was to detect target stimuli in a streamof nontarget stimuli and the target frequency was ma-nipulated the ACC showed substantial activation relativeto passive viewing regardless of whether there weremany (16) or very few (2) targets but with littledifference in activity between the many- and few-targetconditions ( Woldorff Matzke Zamarripa amp Fox 1999)These studies suggest that the functional contribution ofthe ACC to selective attention may not be limited todetecting response conflict Consistent with this viewthe interpret-cue trials in the present event-relatedstudy which should not have evoked either motor

preparation or conflict detection processes activatedthe ACC (although not the slightly more posterior SMAregion) Importantly these cues were relevant stimulithat needed to be attended to and processed (eg likethe nontarget attended lsquolsquostandardsrsquorsquo in attentionalstream experiments such as Woldorff et al 1999)Active attentional orienting without targets in the pres-ent study activated this same ACC area somewhat moreand trials with targets produced still stronger activationin this area These findings especially those for theinterpret-cue trials suggest that the ACC contributes toselective attention at nonresponse levels of processing

It should also be noted that in the more posteriorSMA area trials with targets elicited strong responseswhereas both the attend-cue-only and interpret-cuetrials elicited little activity This pattern of results sug-gests that there was not substantial motor preparationoccurring for these other conditions or at the very leastit was not sufficient to produce much activation in thisputative motor-planning area

Lastly significant effects of trial type in the TPJ regionidentified by Corbetta et al (2000) were rather limited inthe present study One of the goals of Corbetta et alrsquosstudy was to examine the effects of manipulating cuevalidity The authors reported enhanced activity in boththe left and right TPJs for targets with the right TPJbeing more activated for invalidly cued than for validlycued targets In the present study there was someactivation in the region of the TPJ on both sides of thebrain However this activation was elicited by all the cuetrial types (relative to the no-stims) did not differ at allbetween the trial types in the left TPJ ROI and differedonly slightly between the trial types even for those trialshaving targets in the right TPJ ROI We speculate thatthis very limited target-related activation in the TPJ inthe present study might have been due to the targetsconsisting of very faint dots that were always validwhen they occurred and the task being just to detecttheir occurrence


Consistent with the results previously reported in Hop-finger et al (2000) the attend-cue-only trials enhancedactivity in the occipital cortex contralateral to the direc-tion of attention and diminished activity ipsilaterally Thisactivity may reflect biasing of activity in favor of expectedtarget stimuli in sensory cortices (also see Kastner et al1999) In the present study this enhancement wascontralateral to the direction of attention and mainlypresent in the dorsal occipital cortex as expected forattention directed laterally and to the lower visual fieldwhich is represented in that part of cortex (Sereno et al1995) In addition these cue-triggered attention effectswere in the same visual cortex locations activated by thecontralateral unilateral lower-visual-field targets addingsupport for the view that the effects of attention are


retinotopically organized (Noesselt et al 2002 Hop-finger et al 2000 Brefczynski amp DeYoe 1999 Martinezet al 1999 Tootell et al 1998 Woldorff et al 1997)Note also that the contralaterality of these effects verifiesthat subjects did not merely lsquolsquoexpand the spotlight ofattentionrsquorsquo to encompass the target areas but rathershifted the spotlight to the target positions as has alsobeen verified in similar experiments using ERPs (re-viewed in Hillyard Mangun Woldorff amp Luck 1995) Inaddition since the cue was centrally presented regionsthat are contralateral or ipsilateral to the instructed shiftof attention should have experienced the same activa-tion in response to the sensory properties of the cueThus the observed difference between contralateral andipsilateral time-courses must be due to an effect ofattentional orienting in response to the directional cuerather than the sensory properties of the cue itself

Although we and others (Hopfinger et al 2000 Kast-ner et al 1999) have interpreted cue-triggered activity invisual sensory cortices contralateral to the direction ofattention as reflecting attentional biasing of those corti-ces in favor of expected target stimuli other studies haveeither observed different results or have interpreted cue-triggered or cue-period activity in sensory cortices differ-ently For example Corbetta et al (2000) found transientactivation of visual sensory cortex in response to cuesbut they concluded that this effect might have been duesimply to the sensory encoding of cue stimuli Shulmanet al (1999) observed a pretarget transient increase inmotion area MT+ in response to informative versusnoninformative cues in a motion cueing task howeverthey interpreted this effect as possibly reflecting lsquolsquoencod-ing andor maintenance of instruction signalsrsquorsquo In a laterstudy Shulman et al (2002) observed transient cue-period activity in both sensory and nonsensory corticesthat appeared to be triggered by the EOT signal occur-ring 4ndash7 sec after the cue rather than by the cue itselfThis result led to the proposal that certain brain regionswere lsquolsquoreactivatedrsquorsquo by the EOT signal due to the lsquolsquoturn-ing-offrsquorsquo of the preparatory state that had been main-tained while attending In the present study it is possiblethat the later portions of activity in visual sensory corticesduring cue-only trials (Figure 6) could include someactivity due to this kind of lsquolsquoreactivationrsquorsquo responsetriggered by the EOT signal However the time-coursescontralateral and ipsilateral to the direction of attention(Figure 6) diverge as soon as hemodynamic activitybegins rather than being delayed 27 sec as would havebeen expected if it was associated with the EOT signal Inaddition it is difficult to understand how lsquolsquoreactivationrsquorsquoactivity due to the ending of a preparatory state would belarger contralateral to the direction of attention whilethe initial triggering of the preparatory state would notbe contralateral Thus in the present study we concludethat the centrally presented attend-cues triggered en-hanced activity in a visual sensory area that was contra-lateral to the instructed direction of attention (the same

visual sensory area that would process a possible upcom-ing visual target stimulus if it were to occur) and that thiscue-period activity is evident even in the absence of thetarget stimulus actually occurring In accord with Hop-finger et al (2000) and Kastner et al (1999) we interpretthis pattern as being consistent with an attention-in-duced pretarget biasing of the visual sensory cortices

Robust effects of attention were also seen in thethalamus The attend-cue-plus-target trials showed aparticularly large response but even the attend-cue-onlytrials had enhanced activity relative to the interpret-cuetrials Notably these effects in the thalamus were notcontralateral to the direction of attention for the attend-cue-only trials nor were they contralateral to thedirection of the unilateral targets in the attend-cue-plus-target trials In addition they were not contralateralto the right-hand button-press These results suggest anoverall enhancement of both sides of the thalamus whencued to attend with still additional activity bilaterallywhen processing a detected attended target even whenit is unilateral Since these effects were not left-sidedand thus not contralateral to the right-hand button-presses these activations seem unlikely to be due tomotor preparation although this cannot be completelyruled out

Conclusions and Summary

In the present study brain networks underlying thevoluntary control of visual spatial attention were studiedusing a recently developed event-related fMRI approachin combination with a cued-attention paradigm charac-terized by having additional control conditions and therelatively short cuendashtarget SOAs that are more similar tothose often used in behavioral (and ERP) work Evenwith these shorter cuendashtarget SOAs enhanced activitywas seen in the frontal and parietal brain regions inresponse to the cues This argues against the possibilitythat cue-related effects previously reported in theseareas in long-SOA experiments were actually only theresult of increased working memory demands or othercognitive operations induced by those long cuendashtargetSOAs Moreover by including interpret-cue trials in thepresent study that were randomized into the sequencewe observed a pattern of activation that supports theview that there is functional specialization within themedial and lateral subregions of the parietal and frontalcortex Activation in the more medial regions wasrelatively specific to the orienting of visual spatial atten-tion while that in the lateral regions was relativelygeneral in nature perhaps reflecting cognitive functionsthat occur for all cue stimuli such as cuendashsymbolinterpretation Accordingly we also hypothesize thatit is the more medial frontal and parietal areas thatcontrol the observed retinotopically based biasing ofactivity in visual sensory cortex prior to the occurrenceof an expected target

Woldorff et al 161

METHODS

Participants

Twenty young adults (13 men and 7 women mean age27 years) were paid US$20 per hour for their participa-tion Each one gave hisher informed consent prior toparticipating in accordance with the rules of the Insti-tutional Review Board at Duke University

Design Rationale

At the short SOAs and fast stimulus rates needed for thisstudy overlap of the hemodynamic responses fromadjacent events in the stimulus sequence is a majorissue To address this problem we used a combinationof trial-type randomization and hierarchical structuringof trial types

Randomization of trial types is particularly useful forfast-rate event-related fMRI in that it results in theoverlapping responses due to adjacent trials in thesequence being about the same for the different trialtypes Thus contrasts between the trial types subtractout the overlap leaving the event-related differentialresponse activity (Buckner et al 1998 Dale amp Buckner1997 Woldorff 1993) The fast-rate approach can befacilitated with the use of so-called no-stim eventspoints in time in the sequence that are randomized justas if they were a real stimulus event type but duringwhich no stimulus actually occurs (Buckner et al 1998Burock et al 1998) Because no-stims then also containthe same overlap from adjacent trials contrasts betweenthem and the other trial types also subtract out thisoverlap while also providing a low-level control condi-tion for the various trial types However neither thisrandomization approach in and of itself nor the use ofno-stims solves the separation of the cue- and target-triggered activity because the sequence of cues andtargets cannot be randomized

Thus to separate the cue- and target-related activitywe combined the randomization approach (includingno-stims) with a hierarchical structure of the trial typesin which cues sometimes occur without targets Thisapproach also provided a natural mechanism for includ-ing an interpret-cue trial type within the hierarchy as ameans of controlling for and assessing the contributionof cuendashsymbol processing and interpretation (For pre-viously reported alternative approaches using multipleregression see Ollinger Corbetta et al 2001 OllingerShulman et al 2001 Hinrichs et al 2000)

Stimuli and Task

Participants were presented with a series of trials each ofwhich began with an instructional cue letter presented atfixation (Figure 1) The attention-directing instructionalcues consisted of the letter L or R which instructed thesubject to attend for a possible visual target (a faint dot)

at a location in either the left or right lower visual field (38lateral and 38 below horizontal meridian) The lsquolsquointer-pret-cuersquorsquo instructional trials began with the letter lsquolsquoPrsquorsquoinstructing the subject to not attend for a target on thattrial In some of the attention directing trials (attend-cue-plus-target trials) a target would occur at a randomizedtime (900 or 1900 msec) after the onset of the cue Inattend-cue-only (ie attend-left-cue-only or attend-right-cue-only) trials as well as in interpret-cue trials no targetwas presented so that the brain response would be dueto the cue only Finally lsquolsquono-stimrsquorsquo trials (ie periods offixation only) were randomized with these other trialtypes so that the full event-related responses to thevarious trial types could be extracted using selectiveaveraging (Buckner et al 1998 Burock et al 1998) Inall trial types (other than no-stims) an end-of-trial (EOT)stimulus (the letters lsquolsquoREPrsquorsquo) was presented 2700 msecafter trial onset Participants were instructed to press abutton after the EOT stimulus appeared if a target hadbeen presented during that trial The EOT stimulus waspresented in all attend-cue-plus-target trials attend-cue-only trials and interpret-cue trials to equate sen-sory processing demands across conditions Each triallasted 4500 msec and the sequence of trial types wasrandomized

Note the hierarchical structure of the trial types asillustrated in Table 1a This hierarchical structure wasdesigned to allow various key contrasts to be performedthat could isolate different brain responses associatedwith specific processing components (Table 1b) whilesubtracting out the hemodynamic response overlapfrom adjacent trials in the sequence (cf Ollinger Cor-betta et al 2001 Ollinger Shulman et al 2001Corbetta et al 2000 Shulman et al 1999)

Image Acquisition

While the subjects were engaged in the visual attentiontask functional magnetic resonance images of theirbrains were recorded with the GE 15-T Signa LX MRIsystem at the Brain Imaging and Analysis Center at DukeUniversity Eighteen T2-weighted echo-planar imageslices each 5 mm thick with 1-mm skip were acquiredSlice direction was oblique axial (close to horizontal) setto be parallel to the anterior commisurendashposteriorcommisure line with the sixth slice through that lineImaging parameters were TR = 15 sec TE = 40 msecand flip angle = 908 with in-plane resolution of 64 pound 64(375 pound 375 mm) and a field of vision of 24 cm With aTR of 15 sec three brain volumes were acquired in each45-sec trial

For most subjects high-resolution T1-weighted 2-Dstructural scans were also acquired using the same slicethickness and orientation as the functional images Thesewere then used for overlaying the functional activationmaps These also had a 24-cm field of vision with an in-plane resolution of 256 pound 256 (094 pound 094 mm)


Image Analysis

Image preprocessing was performed using routines inSPM99 ( Friston et al 1995) The images were correctedfor subject motion and for slice acquisition timingvariation within the TR normalized to the MNI templateand spatially smoothed using an 8-mm FWHM isotropickernel Voxelwise regression analyses were performedusing SPM99 ( Friston et al 1995) with individual re-gressors containing the onset times for different trialtypes as well as regressors for subject motion

Voxelwise analyses

The procedures described above were followed byacross-subject random-effects voxel-level analyses whichincluded contrasts between the responses to the differ-ent trial types These included hierarchical contrasts (seeResults) as well as contrasts between each trial typerelative to the lowest level trial type the no-stimsAdditional contrasts included left versus right compar-isons (eg left-attend-cue-only vs right-attend-cue-on-ly) In some cases conditions were collapsed across forcontrasting to another condition or group of conditions(eg the average response to left-attend and right-attend-cues vs the response to interpret-cues) to in-crease statistical power

Threshold values for the voxelwise statistical paramet-ric maps were set at a t value of either 275 ( p lt 012one-tailed) or 325 ( p lt 006 one-tailed) the latterbeing for higher power contrasts in which two condi-tions (eg left and right) were collapsed together Toreduce false positives from noise these thresholds wereused in conjunction with a cluster extent threshold of 11contiguous voxels (Forman et al 1995 Xiong GaoLancaster amp Fox 1995) In addition results from thesevoxelwise analyses were confirmed by ROI analyses ofthe time-locked averaged responses derived from selec-tive averaging as described below

Region-of-Interest Analyses

ROIs were functionally defined based on the variouscontrasts and additional functional analysis of theseROIs was performed using custom in-house softwareIn each ROI we used selective averaging to extract theaverage hemodynamic response across all voxels in theROI that was elicited by the various trial types (in termsof percent change from the prestimulus baseline) Thetime-locked average response to no-stim trials was thensubtracted from the time-locked averaged responses tothe other trial types thereby subtracting off the hemo-dynamic overlap from adjacent responses and leavingthe lsquolsquopurersquorsquo event-related transient response to each trialtype (Burock et al 1998 Dale amp Buckner 1997 ) withinthese ROIs The average peak amplitudes (ie theaverage of two time points around the peak) of these

responses were contrasted with random effects t tests todetermine whether and how distinct trial types differen-tially activated the various ROIs In addition the peakamplitude values across ROIs in the frontal and parietalcortex were analyzed using repeated measures ANOVAto directly test whether the medial and lateral regions ofthe frontal and parietal cortex were differentially in-volved in the processing of interpret-cues and atten-tion-directing cue stimuli

Conversion of Coordinates from MNI toTalairach Space

The neuroanatomical locations of activated areas werelater converted from MNI space to Talairach space usinga nonlinear transformation that has been previouslydescribed (httpwwwmrc-cbucamacukImagingmnispacehtml) More specifically coordinates abovethe anterior commissure were transformed by x =099x y = 09688y + 00460z and z = iexcl00485y +09189z and coordinates below the anterior commisurewere transformed with x = 099x y = 09688y +00420z and z = iexcl00485y + 08390z Talairach coor-dinates in the Table 2 indicate the location(s) of peakactivity within each region

Acknowledgments

This work was supported by grants from NIMH (MH60415)NINDS (P01 NS41328 Proj 2) and the ARO (DAAD 19-00-1-0530) to MGW

Reprint requests should be sent to Marty G Woldorff Centerfor Cognitive Neuroscience Duke University Room B203LSRC Building Box 90999 Durham NC 27708-0999 USA or viae-mail woldorffdukeedu

The data reported in this experiment have been deposited inthe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2003-11421

REFERENCES

Awh E Anllo-Vento L amp Hillyard S A (2000) The role ofspatial selective attention in working memory for locationsEvidence from event-related potentials Journal of CognitiveNeuroscience 12 840ndash847

Botvinick M M Braver T S Barch D M Carter C S ampCohen J D (2001) Conflict monitoring and cognitivecontrol Psychological Review 108 624ndash652

Brefczynski J A amp DeYoe E A (1999) A physiologicalcorrelate of the lsquospotlightrsquo of visual attention NatureNeuroscience 2 370ndash374

Buckner R L Goodman J Burock M Rotte M KoutstaalW Schacter D Rosen B R amp Dale A M (1998)Functionalndashanatomic correlates of object priming in humansrevealed by rapid presentation event-related fMRI Neuron20 285ndash296

Burock M A Buckner R L Woldorff M G Rosen B R ampDale A M (1998) Randomized event-related experimentaldesigns allow for extremely rapid presentation rates usingfunctional MRI NeuroReport 9 3735ndash3739

Woldorff et al 163

Carter C S Botvinick M M amp Cohen J D (1999) Thecontribution of the anterior cingulate cortex to executiveprocesses in cognition Reviews in the Neurosciences 1049ndash57

Corbetta M (1998) Frontoparietal cortical networks fordirecting attention and the eye to visual locations Identicalindependent or overlapping neural systems Proceedings ofthe National Academy of Sciences USA 95 831ndash838

Corbetta M Kincade J M Ollinger J M McAvoy M P ampShulman G L (2000) Voluntary orienting is dissociatedfrom target detection in human posterior parietal cortexNature Neuroscience 3 292ndash297

Corbetta M Miezin S Shulman G L amp Petersen S E(1993) A PET study of visuospatial attention Journal ofNeuroscience 13 1020ndash1026

Corbetta M amp Shulman G L (2002) Control of goal-directedand stimulus-driven attention in the brain Nature ReviewsNeuroscience 3 215ndash229

Dale A M amp Buckner R L (1997) Selective averaging ofrapidly presented individual trials using fMRI Human BrainMapping 5 329ndash340

Desimone R amp Duncan J (1995) Neural mechanisms ofselective visual attention Annual Review of Neuroscience18 193ndash222

DrsquoEsposito M D Aguirre G K Zarahn E Ballard D ShinR K amp Lease J (1998) Functional MRI studies of spatialand nonspatial working memory Cognitive Brain Research7 1ndash13

DrsquoEsposito M D Zarahn E amp Aguirre G K (1999)Event-related functional MRI Implications for cognitivepsychology Psychological Bulletin 125 155ndash164

De Fockert J W Rees G Frith C D amp Lavie N (2001) Therole of working memory in visual selective attentionScience 291 1803ndash1806

Downing C J (1988) Expectancy and visualndashspatial attentionEffects on perceptual quality Journal of ExperimentalPsychology Human Perception and Performance14 188ndash202

Eimer M (2000) The time course of spatial orienting elicitedby central and peripheral cues Evidence from event-relatedbrain potentials Biological Psychology 53 253ndash258

Forman S D Cohen J D Fitzgerald M Eddy W F MintunM A amp Noll D C (1995) Improved assessment ofsignificant activation in functional magnetic resonanceimaging (fMRI) Use of a cluster-size threshold MagneticResonance in Medicine 33 636ndash647

Friston K J Holmes A P Worsley K J Poline J P Frith CD amp Frackowiack R S J (1995) Statistical parametric mapsin functional imaging A general linear approach HumanBrain Mapping 2 189ndash210

Gitelman D R Parrish T B LaBar K S amp Mesulam M -M(2000) Real-time monitoring of eye movements usinginfrared video-oculography during functional magneticresonance imaging of the frontal eye fields Neuroimage11 58ndash65

Hillyard S A Mangun G R Woldorff M G amp Luck S J(1995) Neural systems mediating selective attention In M SGazzaniga (Ed) The cognitive neurosciences (pp 665ndash681)Cambridge MIT Press

Hinrichs H Scholtz M Tempelmann C Woldorff M GDale A M amp Heinze H-J (2000) Deconvolution ofevent-related fMRI responses in fast-rate experimentaldesign Tracking amplitude variations Journal ofCognitive Neuroscience 12 S76ndashS89

Hopf J amp Mangun G R (2000) Shifting visual attention inspace An electrophysiological analysis using high spatialresolution mapping Clinical Neurophysiology 1111241ndash1257

Hopfinger J B Buonocore M H amp Mangun G R (2000)The neural mechanisms of top-down attentional controlNature Neuroscience 3 284ndash291

Jha A P amp McCarthy G (2000) The influence of memoryload upon delay-interval activity in a working-memory taskAn event-related functional MRI study Journal of CognitiveNeuroscience 12 90ndash105

Kastner S Pinsk M A De Weerd P Desimone R ampUngerleider L G (1999) Increased activity in human visualcortex during directed attention in the absence of visualstimulation Neuron 22 751ndash761

LaBar K S Gitelman D R Parrish T B amp Mesulam M-M(1999) Neuroanatomic overlap of working memory andspatial attention networks A functional MRI comparisonwithin subjects Neuroimage 10 695ndash704

MacDonald A W Cohen J D Stenger V A amp Carter C S(2000) Dissociating the role of the dorsolateral prefrontaland anterior cingulate cortex in cognitive control Science288 1835ndash1838

Martinez A Anllo-Vento L Sereno M I Frank L R BuxtonR B Dubowitz D J Wong E C Hinrichs H Heinze H Jamp Hillyard S A (1999) Involvement of striate andextrastriate visual cortical areas in spatial attention NatureNeuroscience 2 364ndash369

Milham M P Banich M T Webb A amp Barad V (2001) Therelative involvement of anterior cingulate and prefrontalcortex in attentional control depends on nature of conflictCognitive Brain Research 12 467ndash473

Murtha S Chertkow H Beauregard M Dixon R amp EvansA (1996) Anticipation causes increased blood flow to theanterior cingulate cortex Human Brain Mapping4 103ndash112

Nobre A C Gitelman D R Dias E C amp Mesulam M M(2000) Covert visual spatial orienting and saccadesOverlapping neural systems Neuroimage 11 210ndash216

Nobre A C Sebestyen G N Gitelman D R Mesulam MM Frackowiak R S amp Frith C D (1997) Functionallocalization of the system for visuospatial attention usingpositron emission tomography Brain 120 515ndash533

Noesselt T Hillyard S A Woldorff M Hagner T JaenckeL Tempelmann C Hinrichs H amp Heinze H-J (2002)Delayed striate cortical activation during spatial attentionNeuron 35 575ndash587

Nystrom L E Braver T S Sabb F W Delgado M R NollD C amp Cohen J D (2000) Working memory for lettersshapes and locations fMRI evidence against stimulus-basedregional organization in human prefrontal cortexNeuroimage 11 424ndash446

Ollinger J M Corbetta M amp Shulman G L (2001)Separating processes within a trial in event-related functionalMRI Neuroimage 13 218ndash229

Ollinger J M Shulman G L amp Corbetta M (2001)Separating processes within a trial in event-related functionalMRI Neuroimage 13 210ndash217

Paus T (2001) Primate anterior cingulate cortex Wheremotor control drive and cognition interface NatureReviews Neuroscience 2 417ndash424

Posner M I amp Petersen S E (1990) The attention system ofthe human brain Annual Review of Neuroscience 1325ndash42

Posner M I Snyder C R R amp Davidson B J (1980)Attention and the detection of signals Journal ofExperimental Psychology General 109 160ndash174

Rosen B R Buckner R L amp Dale A M (1998) Event-relatedfunctional MRI Past present and future Proceedings of theNational Academy of Sciences USA 95 773ndash780

Sereno M I Dale A M Reppas J B Kwong K K BelliveauJ W Brady T J Rosen B R amp Tootell R B H (1995)


Borders of multiple visual areas in humans revealed byfunctional magnetic resonance imaging Science 258889ndash893

Shulman G L Ollinger J M Akbudak E Conturo T ESnyder A Z Petersen S E amp Corbettta M (1999)Areas involved in encoding and applying directionalexpectations to moving objects Journal of Neuroscience19 9480ndash9496

Talairach J Tournoux P (1988) Co-planar stereotaxic atlasof the human brain Thieme New York

Tootell R B Hadjikhani N Hall E K Marrett S VanduffelW Vaughan J T amp Dale A M (1998) The retinotopy ofvisual spatial attention Neuron 21 1409ndash1422

Vandenberghe R Duncan J Dupont P Ward R PolineJ-B Bormans G Michiels J Mortelmans amp Orban G A(1997) Attention to one or two features in left and rightvisual field A positron emission tomography study Journalof Neuroscience 17 3739ndash3750

Van Veen V Cohen J D Botvinick M M Stenger V Aamp Carter C S (2001) Anterior cingulate cortex conflict

monitoring and levels of processing Neuroimage 141302ndash1308

Wojciulik E amp Kanwisher N (1999) The generality of parietalinvolvement in visual attention Neuron 23 747ndash764

Woldorff M G (1993) Distortion of ERP averages due tooverlap from temporally adjacent ERPs Analysis andcorrection Psychophysiology 30 98ndash119

Woldorff M G Fox P T Matzke M Lancaster J LVeeraswamy S Zamarripa F Seabolt M Glass T Goa JH Martin C C amp Jerabek P (1997) Retinotopicorganization of early spatial attention effects as revealedby PET and ERPs Human Brain Mapping 5 280ndash286

Woldorff M G Matzke M Zamarripa F amp Fox P T (1999)Hemodynamic and electrophysiological study of the roleof the anterior cingulate in target-related processingand selection for action Human Brain Mapping 8121ndash127

Xiong J Gao J Lancaster J L amp Fox P T (1995) Clusteredpixels analysis for functional MRI activation studies in thehuman brain Human Brain Mapping 3 287ndash301

Woldorff et al 165

The present experiment builds on the methods ofthese earlier studies for distinguishing cue- and target-related activity to further determine the contributionsof various regions in the fronto-parietal network iden-tified as active during attention-orienting tasks This wasdone by extending these approaches in mainly twoways (1) using SOAs (900ndash1900 msec) that are closerto those used in behavioral studies and (2) addingcontrol conditions that allow the decomposition ofcue-triggered activity into activity more specifically re-lated to orienting and activity related to general aspectsof cue processing

First regarding the SOAs long delays between cueand target may make it more likely that subjects engageother cognitive processes such as working memoryduring the delay period Thus it is possible that dueto the relatively long cuendashtarget SOAs in these severalprevious studies (eg 8 sec in Hopfinger et al 2000 4ndash5 sec in Corbetta et al 2000 5ndash6 sec in Shulman et al1999) the broad network of areas observed may haveincluded some regions more related to working memoryrather than attentional orienting per se Indeed thefrontal and parietal areas previously observed to beactivated by spatial attentional cues included areas sim-ilar to those activated in working memory tasks (eg Jhaamp McCarthy 2000 Nystrom et al 2000 LaBar GitelmanParrish amp Mesulam 1999 drsquoEsposito et al 1998) Toreduce this possibility we used a substantially shortercuendashtarget SOA (randomly 900 and 1900 msec) Thischange not only reduces the likelihood of invokingworking memory processes but also better controlsthe timing of attention orienting (subjects learn thattargets may come as early as 900 msec so that theywould be more likely to begin those processes rightaway after the cue) In addition using shorter SOAsmakes the results more directly comparable to the widebody of behavioral and ERP work using similar SOAs inendogenous cueing tasks (eg 800ndash1200 msec in Pos-ner Snyder amp Davidson 1980 700 msec in Downing1988 1100 msec in Hopf amp Mangun 2000 700 msec inEimer 2000)

Second as noted above previous studies have suc-cessfully separated cue-related activity from target-related activity (eg Corbetta et al 2000 Hopfingeret al 2000 Shulman et al 1999) However within thebrain activity that is triggered by cues there is a need tofurther distinguish activity due to the actual orienting ofattention from activity due to sensory and semanticoperations required for processing and interpretingthe cues themselves (eg cue-interpretation processes)Both Hopfinger et al (2000) and Shulman et al (1999)included a control condition for the cues (lsquolsquocue controlrsquorsquoin Hopfinger et al 2000 lsquolsquopassive cuesrsquorsquo in Shulmanet al 1999) but these trials were in a separate passivelyviewed run and thus were not intermixed with the othertrial types and were not behaviorally relevant Whilethese runs may have controlled for the most basic

sensory stimulation of the cues they did not requireany sort of processing for meaning or interpretation andthus they were limited in their effectiveness as controlsfor the general cue-processing operations that would belikely to occur during the experimental attention runsThus in the present study we included a lsquolsquocue-interpre-tationrsquorsquo control trial type that was randomly intermixedwith the other attention-directing trial types On thesetrials participants needed to perceive process andinterpret a behaviorally relevant cue which instructedthem that orienting of their spatial attention was unnec-essary on that particular trial These cues which we havetermed lsquolsquointerpret-cuesrsquorsquo could not be ignored as theywere intermixed with all other trial types in the sameruns and needed to be interpreted and processed toachieve correct behavior Thus interpret-cue trials re-quired both sensory and cue-interpretion processessimilar to those engaged by attend-cues but did notrequire attentional orienting to a particular spatial loca-tion in the visual field

Thus our paradigm and analysis structure is lsquolsquohierar-chicalrsquorsquo in nature At the first level activity can beseparated into cue- and target-related activity This sep-aration and assessment of target-triggered activity allowsus to determine which sensory regions would be ex-pected to be the site of any prestimulus biasing duringcued orienting as well as to assess and separate outactivity due to motor-related processes At the secondlevel cue-related activity can be further split into activityrelated to attentional orienting and activity more relatedto general cue-interpretation processes allowing us tobetter isolate attention-orienting networks while deter-mining the functional roles of different subregions of thefronto-parietal executive control network

RESULTS

Subjects were presented with a series of trials each ofwhich began with an instructional cue presented atfixation (Figure 1) Active-attend (ie attention-direct-ing) cues consisted of either the letter L or R whichinstructed the subject to attend to a location in the leftor right lower visual field to detect a possible faint targetthat might occur in that location Some of the instruc-tional cues consisted of the letter P These were thelsquolsquointerpret-cuersquorsquo control trials instructing the subject tonot orient their attention and to not attend for targetsIn some active-attend trials a target would occur at arandomized time (900 or 1900 msec) after the onset ofthe cue (attend-cue-plus-target trials) In other active-attend trials (attend-cue-only trials) as well as in theinterpret-cue trials no target was presented so that thebrain response would be due to the cue only Thesevarious trial types were randomized with lsquolsquono-stimrsquorsquo trials(ie periods of fixation only) so that the full event-related responses to the various trial types could beextracted using selective averaging (Buckner et al 1998








Time 0Cue

ISI


ISI


ITI


Time(milliseconds)













No-stim



Component(s)














Woldorff et al 151




























32 37Eiexcl08


242 21Eiexcl09


75 27Eiexcl08




124 88Eiexcl07




57 0012Eiexcl08


119 25Eiexcl06

Woldorff et al 153














Woldorff et al 155











DISCUSSION


Woldorff et al 157














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165







Time 0Cue

ISI


ISI


ITI


Time(milliseconds)













No-stim



Component(s)














Woldorff et al 151




























32 37Eiexcl08


242 21Eiexcl09


75 27Eiexcl08




124 88Eiexcl07




57 0012Eiexcl08


119 25Eiexcl06

Woldorff et al 153














Woldorff et al 155











DISCUSSION


Woldorff et al 157














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165




























32 37Eiexcl08


242 21Eiexcl09


75 27Eiexcl08




124 88Eiexcl07




57 0012Eiexcl08


119 25Eiexcl06

Woldorff et al 153














Woldorff et al 155











DISCUSSION


Woldorff et al 157














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165














Woldorff et al 155











DISCUSSION


Woldorff et al 157














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165











DISCUSSION


Woldorff et al 157














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165














Woldorff et al 159


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165


















Woldorff et al 161

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165

METHODS

Participants


Design Rationale




Stimuli and Task




Image Acquisition




Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165

Image Analysis


Voxelwise analyses








Acknowledgments




REFERENCES






Woldorff et al 163



















































Woldorff et al 165



















































Woldorff et al 165

functional parcellation of attentional control regions of the brain

Documents