A Parametric Manipulation of Factors AffectingTask-induced Deactivation in Functional Neuroimaging
Kristen A McKiernan Jacqueline N KaufmanJane Kucera-Thompson and Jeffrey R Binder
Abstract
amp Task-induced deactivation (TID) refers to a regionaldecrease in blood flow during an active task relative to alsquolsquorestingrsquorsquo or lsquolsquopassiversquorsquo baseline We tested the hypothesis thatTID results from a reallocation of processing resources byparametrically manipulating task difficulty within three factorstarget discriminability stimulus presentation rate and short-term memory load Subjects performed an auditory targetdetection task during functional magnetic resonance imaging(fMRI) responding to a single target tone or in the short-termmemory load conditions to target sequences Seven taskconditions (a common version and two additional levels foreach of the three factors) were each alternated with lsquolsquorestrsquorsquo in ablock design Analysis of covariance identified brain regions inwhich TID occurred Analyses of variance identified sevenregions (left anterior cingulatesuperior frontal gyrus leftmiddle frontal gyrus right anterior cingulate gyrus left and
right posterior cingulate gyrus left posterior parieto-occipitalcortex and right precuneus) in which TID magnitude variedacross task levels within a factor Follow-up tests indicated thatfor each of the three factors TID magnitude increased withtask difficulty These results suggest that TID representsreallocation of processing resources from areas in which TIDoccurs to areas involved in task performance Short-termmemory load and stimulus rate also predict suppression ofspontaneous thought and many of the brain areas showingTID have been linked with semantic processing supportingclaims that TID may be due in part to suspension ofspontaneous semantic processes that occur during lsquolsquorestrsquorsquo(Binder et al 1999) The concept that the typical lsquolsquorestingstatersquorsquo is actually a condition characterized by rich cognitiveactivity has important implications for the design and analysisof neuroimaging studies amp
INTRODUCTION
This study is a systematic investigation of the phenom-enon of task-induced deactivation (TID) TID refers torelative decreases in regional activity as measured byblood flow or the blood oxygenation level dependent(BOLD) signal (Ogawa Lee Kay amp Tank 1990) duringan active task compared to a lsquolsquorestingrsquorsquo baseline Thisdefinition can be restated Areas that show TID havehigher levels of blood flow during lsquolsquorestrsquorsquo states thanduring the task of interest While TID is a commonfinding in neuroimaging (Mazoyer et al 2001 Binderet al 1999 Shulman et al 1997 Bookheimer ZeffiroBlaxton Gaillard amp Theodore 1995 Howard et al1992 Frith Friston Liddle amp Frackowiak 1991) it isnot well understood and only recently became the focusof systematic study
Several brain regions consistently show TID acrossdifferent tasks stimuli and imaging modalities (Mazoyeret al 2001 Binder et al 1999 Shulman et al 1997 )(Figure 1) Evidence on this point comes from a largePET meta-analysis study by Shulman et al (1997) involv-ing 97 subjects in nine different visual processing tasksin which the authors searched for common brain
regions that showed TID relative to either a visualfixation or passive stimulation baseline Binder et al(1999) reported common areas of TID in 30 subjectsperforming an auditory tone discrimination task relativeto an eyes-closed resting baseline Finally Mazoyer et al(2001) described common areas of TID during PET in 63subjects performing a range of visual auditory andimagery tasks relative to an eyes-closed resting baselineThe pattern of regions showing TID is remarkablysimilar across these three studies Common regionsinclude the posterior cingulate cortex dorsomedialfrontal cortex in the middle and superior frontal gyrirostral anterior cingulate gyrus and angular gyrus(Mazoyer et al 2001 Binder et al 1999 Shulmanet al 1997) (Figure 1) The posterior cingulate focusoften extends dorsally into the precuneus (Mazoyeret al 2001 Shulman et al 1997) and the rostralanterior cingulate focus typically extends ventrally intothe gyrus rectus and the orbital frontal cortex (Mazoyeret al 2001 Binder et al 1999 Shulman et al 1997)The angular gyrus focus spreads anteriorly into thesupramarginal gyrus in the Shulman et al study andposteriorly into the superior occipital cortex in the studyby Mazoyer et al In all three studies left hemisphereregions showed more extensive deactivation (52 peakcoordinates reported including this study) than theMedical College of Wisconsin
copy 2003 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 153 pp 394ndash408
homologous regions in the right hemisphere (18 peaksreported see Figure 1A)
Several theories regarding the underlying basis forTID have been proposed As discussed by Shulmanet al (1997) and Gusnard and Raichle (2001) decreasesare unlikely to be the result of a redistribution ofcerebral blood flow to areas that are active from adjacentareas an idea commonly referred to as the lsquolsquovascular-stealrsquorsquo hypothesis There is a vast cerebral vascularreserve that can easily accommodate the relatively small(approximately 10) changes in blood flow produced bycognitive activity (Heistad amp Kontos 1983) Conse-quently there is no physiological need to lsquolsquostealrsquorsquo bloodfrom inactive areas to supply active regions Moreoverthe large cortical areas associated with TID showdecreased blood flow across a wide range of visual andauditory tasks that activate different groups of brainregions Thus there is little evidence that regions show-ing decreased activity are necessarily adjacent to activeareas as would be predicted by the lsquolsquostealrsquorsquo hypothesis
Decreases in activity may represent a reduction inneural activity in or to a brain region There are twocategories of decreases task-dependent and task-independent Gusnard and Raichle (2001) explaintask-dependent decreases as decreases that occur inbrain regions remote from areas that are activated by atask These remote decreases may be the result oflsquolsquogatingrsquorsquo input to areas that are not involved in taskperformance Thus the specific areas that decrease inactivity are dependent on the characteristics of the taskDeactivation of this type is most often seen in sensorycortices Task-independent decreases are consistentlyfound in specific brain regions across a wide variety oftasks This consistency in location regardless of the taskindicates that these decreases occur independent oftask characteristics Task-independent decreases aremost often found in cortical areas involved in higher-level cognitive processing
The theory put forth in this article focuses on explain-ing TIDs and proposes that TID represents a relative
Figure 1 Four studies illus-trating the consistency in brainregions that deactivate relativeto rest (A) Peak deactivationsfrom four studies and (B)current study data thresholdedat z = iexcl1 representing areasthat deactivate relative to restCrosshairs indicate stereotaxiccoordinates of x = 0 and y = 0z coordinate denoted next toeach slice In panel A the zcoordinate for each peak wasadjusted to fall on the nearestrepresentative slice
McKiernan et al 395
decrease in neural activity Specifically this theory positsthat decreases in cerebral blood flow are caused byinterruption of ongoing internal processing that occursin the passive or lsquolsquorestrsquorsquo state (Gusnard amp Raichle 2001Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) This model assumes that lsquolsquorestrsquorsquo is a state oforganized functional brain activity These organizedprocesses are suspended or interrupted when the brainperforms an exogenously generated task Such internalactivity could involve many processes including thefollowing noted by Gusnard and Raichle (2001) andShulman et al (1997) monitoring of the external envi-ronment monitoring of body image and state andmonitoring of emotional state An additional processnoted by Shulman et al and emphasized by Binder et al(1999) is the ongoing internal lsquolsquothoughtrsquorsquo processingthat humans experience during resting consciousnessThis ongoing verbal and visual imagery referred to byJames (1890) as a lsquolsquostream of consciousnessrsquorsquo alsoappears to be suspended in the presence of an exoge-nously generated task (McGuire Paulesu Frackowiak ampFrith 1996 Giambra 1995 Teasdale Proctor Lloyd ampBaddeley 1993 Pope amp Singer 1976 Antrobus Singeramp Greenberg 1966 Antrobus 1968) The left lateraliza-tion of TID areas observed in most studies (Figure 1A)lends some support to this idea Cognitive processesongoing during rest are posited to require attentionalresources this theory allows deactivation to occur via apassive mechanism of reallocation of these resourcesThus deactivation could be effected by removal ofexcitatory (eg thalamic or transcortical) input in muchthe same way that passive removal of cortical inputproduces cerebellar deactivation in cases of cerebellarlsquolsquodiaschisisrsquorsquo (Martin amp Raichle 1983)
Suspension of ongoing resting state processing whilein a task condition implies that there are finite brainresources available for information processing Imposinga task on a subject requires that processing resources bereallocated from internally generated information pro-cessing to processing of the exogenous task Shulmanet al (1997) noted that if the suspension of passiveprocesses during an active task is due to competition forgeneral processing resources then the magnitude of thedeactivation should depend on the degree to which theactive task requires these same resources They foundlittle evidence however for an effect of task difficulty onTID magnitude while noting that their manipulation oftask difficulty may not have been strong enough toproduce such an effect We hypothesize that task diffi-culty is indeed a factor in determining the extent ofresources to be reallocated and that resource realloca-tion is the basis of TID Our aim will be to provide a cleartest of this hypothesis through carefully controlledmanipulations of task difficulty
Factors that influence task processing demands for anonsemantic perceptual task can include but are notlimited to task familiarity perceptual discriminability
stimulus presentation rate working memory load andtarget density Based on prior research (McGuire et al1996 Giambra 1995 Teasdale et al 1993 Pope ampSinger 1976 Antrobus et al 1966 Antrobus 1968)we chose three of these factors stimulus presentationrate perceptual discriminability and short-term memoryload to manipulate in order to vary the processingdemands of the exogenous task The goal was to system-atically increase the task demands within each factorthus requiring different degrees of resource reallocationin order to perform the task To reduce the possibility oftask-related activation in the regions of interest (ROIs) anonsemantic task (auditory target detection) was used(Binder et al 1999) We predicted that at the moredifficult levels the task would require greater allocationof resources thereby resulting in a greater degree ofdeactivation in those areas that are hypothesized to beactive during rest and consistently show TID
The importance of better understanding the type andextent of cognitive processing that occurs during theresting periods of a neuroimaging task cannot beoverstated A common approach in neuroimaging re-search is to alternate blocks of task activity with blocksof rest or in event-related designs to treat the inter-stimulus rest periods as a neutral baseline The brainactivity that occurs during rest is then subtracted fromthe activity that is task-related The assumption of thisapproach is that the remaining activity will be mostlyrelated to the specific demands of the task If restperiods are actually as we suggest periods of complexcognitive work then the use of rest as a baselinecondition and the subtraction of resting neural activityfrom the task-related activity may be inappropriate andcould confound interpretation of the data
An auditory target detection task was presented tosubjects This task had three levels of task difficulty foreach of three factors The task involved discriminationbetween two sounds one of which was defined as thetarget The design was a parametric manipulation ofthe difficulty levels of each factor so conditions weredistinguishable on the basis of only one task variable(Table 1) Subjects heard blocks of a task conditionalternating with rest periods and responded to targetstimuli by pressing a button The two more difficultmemory conditions used trains of stimuli and requiredsubjects to press a button after trains that containedspecific combinations of the target and nontarget stimuli
RESULTS
Behavioral Results
Due to technical difficulties with the response recordingsoftware behavioral data (accuracy and reaction time)were recorded for only 19 of the 30 subjects Accuracydata are presented as a percentage of the number ofcorrect responses minus the number of false alarms
396 Journal of Cognitive Neuroscience Volume 15 Number 3
divided by the number of targets This calculation adds apenalty for incorrect responses and accounts for thevariation in the number of stimuli per block in differentstimulus presentation rate conditions Reaction timedata were used to compute the percentage of the trialduration used by the subject (time-on-task) This value iscomputed by adding stimulus duration (200 msec) tomean reaction time (for correct responses only) anddividing by the length (msec) of the trial This ratioallows us to assess the average percentage of total trialtime spent on-task across a wide range of presentationand response rates Since only correct responses areincluded in the analysis we assume that the subject wason-task for the 200-msec stimulus presentation thus theratio described above is valid for any condition of singlestimulus presentation Unlike the other conditionshowever the two more difficult levels of the short-termmemory factor employed trains of stimuli During thepresentation of a sound train there were two to fourinterstimulus intervals of 400 msec and there is nospecific measure of whether or not the subject wason-task during these periods Thus the time-on-taskratio cannot be calculated and comparison of responsetimes across such different trial types is not possibleTime-on-task data for these two conditions will there-fore not be presented All statistical analyses were con-ducted using SPSS 90 for Windows (SPSS Chicago) andevaluated at p lt 05 with a modified Bonferroni correc-tion (Keppel 1991)
Repeated-measures ANOVAs assessed differencesacross the conditions for accuracy and time-on-task dataIn both cases there was a significant effect of conditionF(6108) = 1107 p lt 001 for accuracy and F(472) =50185 p lt 001 for time-on-task Within each of thethree factors t tests compared responses across thethree levels of task difficulty (Figure 2)
Within the stimulus presentation rate factor accuracydecreased in the fastest condition (78) compared toboth the moderate (96) and slow conditions (95)[t(18) = 712 p lt 001 and t(18) = 521 p lt 001respectively] Similarly accuracy in the target discrim-inability factor significantly decreased in the difficultcondition (84) compared to both the moderate(93) and easy (96) conditions [t(18) = 292 p =009 and t(18) = 353 p = 002 respectively] In theshort-term memory load conditions accuracy was sig-nificantly higher in the easy condition (96) than ineither the moderate (88) [t(18) = 277 p = 013] orthe difficult condition (91) [t(18) = 280 p = 012]The accuracy data indicate that the manipulations hadthe expected effects on level of difficulty without pro-ducing a severe performance decrement (Figure 2A)
For the stimulus presentation rate factor a signifi-cantly higher percentage of trial time was spent on-taskduring the difficult (fastest) condition (69) as com-pared to either the moderate (50) or the slowestcondition (30) [t(18) = iexcl1980 p lt 001 and t(18) =iexcl3477 p lt 001 respectively] and more time was
Table 1 Parametric Manipulation of Task Parameters
used at the moderate pace than at the slower pace[t(18) = 2742 p lt 001] This result is inevitable ofcourse because the presentation rate manipulation is adirect manipulation of time-on-task Within the targetdiscriminability factor significantly more of the trialtime (54) was used at the most difficult level of thetask as compared to both the easy (50) and moderate(50) levels of discriminability [t(18) = iexcl473 p lt001 and t(18) = iexcl391 p = 001 respectively] There-fore manipulation of task difficulty level had theexpected effect on time-on-task (Figure 2B) Summarydata for both behavioral measures are presented inTable 2
fMRI Results
The overall map of areas that showed TID (Figure 1B) isan average of z-scores from all seven conditions acrossall subjects The 20 strongest foci representing areas thatconsistently show deactivation are listed in Table 3
To focus the study on specific brain regions data fromall 30 subjects were combined across task conditions todefine ROIs in stereotaxic space using the following
technique For each condition the correlation valuesbetween the MRI signal and the ideal response vectorwere converted to z-scores These z-score maps werethen smoothed with a 4-mm RMS gaussian filter andaveraged over conditions and subjects A threshold wasarbitrarily set at z = iexcl1 From this map (Figure 1B)11 stereotaxic ROIs were identified as large (1684to 25280 mm3) discrete volumes The 11 ROIs are(in order of decreasing volume) left posterior parieto-occipital (PPO) cortex which includes parts of theangular gyrus dorsolateral occipital lobe and cuneusright PPO cortex right precuneussuperior parietallobule left precuneussuperior parietal lobule left ante-rior cingulatesuperior frontal gyrus left posterior cin-gulate gyrus left middle frontal gyrus left middleoccipital gyrus right posterior cingulate gyrus rightanterior cingulate gyrus and left fusiform gyrus(Figure 3) The regions identified from this analysisinclude many of the same foci identified in other studies(Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) (Figure 1A)
The relative magnitude of deactivation within theseROIs was measured using fit coefficients which are
Figure 2 (A) Accuracy dataacross conditions (B) Time-on-task data across conditions
398 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
homologous regions in the right hemisphere (18 peaksreported see Figure 1A)
Several theories regarding the underlying basis forTID have been proposed As discussed by Shulmanet al (1997) and Gusnard and Raichle (2001) decreasesare unlikely to be the result of a redistribution ofcerebral blood flow to areas that are active from adjacentareas an idea commonly referred to as the lsquolsquovascular-stealrsquorsquo hypothesis There is a vast cerebral vascularreserve that can easily accommodate the relatively small(approximately 10) changes in blood flow produced bycognitive activity (Heistad amp Kontos 1983) Conse-quently there is no physiological need to lsquolsquostealrsquorsquo bloodfrom inactive areas to supply active regions Moreoverthe large cortical areas associated with TID showdecreased blood flow across a wide range of visual andauditory tasks that activate different groups of brainregions Thus there is little evidence that regions show-ing decreased activity are necessarily adjacent to activeareas as would be predicted by the lsquolsquostealrsquorsquo hypothesis
Decreases in activity may represent a reduction inneural activity in or to a brain region There are twocategories of decreases task-dependent and task-independent Gusnard and Raichle (2001) explaintask-dependent decreases as decreases that occur inbrain regions remote from areas that are activated by atask These remote decreases may be the result oflsquolsquogatingrsquorsquo input to areas that are not involved in taskperformance Thus the specific areas that decrease inactivity are dependent on the characteristics of the taskDeactivation of this type is most often seen in sensorycortices Task-independent decreases are consistentlyfound in specific brain regions across a wide variety oftasks This consistency in location regardless of the taskindicates that these decreases occur independent oftask characteristics Task-independent decreases aremost often found in cortical areas involved in higher-level cognitive processing
The theory put forth in this article focuses on explain-ing TIDs and proposes that TID represents a relative
Figure 1 Four studies illus-trating the consistency in brainregions that deactivate relativeto rest (A) Peak deactivationsfrom four studies and (B)current study data thresholdedat z = iexcl1 representing areasthat deactivate relative to restCrosshairs indicate stereotaxiccoordinates of x = 0 and y = 0z coordinate denoted next toeach slice In panel A the zcoordinate for each peak wasadjusted to fall on the nearestrepresentative slice
McKiernan et al 395
decrease in neural activity Specifically this theory positsthat decreases in cerebral blood flow are caused byinterruption of ongoing internal processing that occursin the passive or lsquolsquorestrsquorsquo state (Gusnard amp Raichle 2001Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) This model assumes that lsquolsquorestrsquorsquo is a state oforganized functional brain activity These organizedprocesses are suspended or interrupted when the brainperforms an exogenously generated task Such internalactivity could involve many processes including thefollowing noted by Gusnard and Raichle (2001) andShulman et al (1997) monitoring of the external envi-ronment monitoring of body image and state andmonitoring of emotional state An additional processnoted by Shulman et al and emphasized by Binder et al(1999) is the ongoing internal lsquolsquothoughtrsquorsquo processingthat humans experience during resting consciousnessThis ongoing verbal and visual imagery referred to byJames (1890) as a lsquolsquostream of consciousnessrsquorsquo alsoappears to be suspended in the presence of an exoge-nously generated task (McGuire Paulesu Frackowiak ampFrith 1996 Giambra 1995 Teasdale Proctor Lloyd ampBaddeley 1993 Pope amp Singer 1976 Antrobus Singeramp Greenberg 1966 Antrobus 1968) The left lateraliza-tion of TID areas observed in most studies (Figure 1A)lends some support to this idea Cognitive processesongoing during rest are posited to require attentionalresources this theory allows deactivation to occur via apassive mechanism of reallocation of these resourcesThus deactivation could be effected by removal ofexcitatory (eg thalamic or transcortical) input in muchthe same way that passive removal of cortical inputproduces cerebellar deactivation in cases of cerebellarlsquolsquodiaschisisrsquorsquo (Martin amp Raichle 1983)
Suspension of ongoing resting state processing whilein a task condition implies that there are finite brainresources available for information processing Imposinga task on a subject requires that processing resources bereallocated from internally generated information pro-cessing to processing of the exogenous task Shulmanet al (1997) noted that if the suspension of passiveprocesses during an active task is due to competition forgeneral processing resources then the magnitude of thedeactivation should depend on the degree to which theactive task requires these same resources They foundlittle evidence however for an effect of task difficulty onTID magnitude while noting that their manipulation oftask difficulty may not have been strong enough toproduce such an effect We hypothesize that task diffi-culty is indeed a factor in determining the extent ofresources to be reallocated and that resource realloca-tion is the basis of TID Our aim will be to provide a cleartest of this hypothesis through carefully controlledmanipulations of task difficulty
Factors that influence task processing demands for anonsemantic perceptual task can include but are notlimited to task familiarity perceptual discriminability
stimulus presentation rate working memory load andtarget density Based on prior research (McGuire et al1996 Giambra 1995 Teasdale et al 1993 Pope ampSinger 1976 Antrobus et al 1966 Antrobus 1968)we chose three of these factors stimulus presentationrate perceptual discriminability and short-term memoryload to manipulate in order to vary the processingdemands of the exogenous task The goal was to system-atically increase the task demands within each factorthus requiring different degrees of resource reallocationin order to perform the task To reduce the possibility oftask-related activation in the regions of interest (ROIs) anonsemantic task (auditory target detection) was used(Binder et al 1999) We predicted that at the moredifficult levels the task would require greater allocationof resources thereby resulting in a greater degree ofdeactivation in those areas that are hypothesized to beactive during rest and consistently show TID
The importance of better understanding the type andextent of cognitive processing that occurs during theresting periods of a neuroimaging task cannot beoverstated A common approach in neuroimaging re-search is to alternate blocks of task activity with blocksof rest or in event-related designs to treat the inter-stimulus rest periods as a neutral baseline The brainactivity that occurs during rest is then subtracted fromthe activity that is task-related The assumption of thisapproach is that the remaining activity will be mostlyrelated to the specific demands of the task If restperiods are actually as we suggest periods of complexcognitive work then the use of rest as a baselinecondition and the subtraction of resting neural activityfrom the task-related activity may be inappropriate andcould confound interpretation of the data
An auditory target detection task was presented tosubjects This task had three levels of task difficulty foreach of three factors The task involved discriminationbetween two sounds one of which was defined as thetarget The design was a parametric manipulation ofthe difficulty levels of each factor so conditions weredistinguishable on the basis of only one task variable(Table 1) Subjects heard blocks of a task conditionalternating with rest periods and responded to targetstimuli by pressing a button The two more difficultmemory conditions used trains of stimuli and requiredsubjects to press a button after trains that containedspecific combinations of the target and nontarget stimuli
RESULTS
Behavioral Results
Due to technical difficulties with the response recordingsoftware behavioral data (accuracy and reaction time)were recorded for only 19 of the 30 subjects Accuracydata are presented as a percentage of the number ofcorrect responses minus the number of false alarms
396 Journal of Cognitive Neuroscience Volume 15 Number 3
divided by the number of targets This calculation adds apenalty for incorrect responses and accounts for thevariation in the number of stimuli per block in differentstimulus presentation rate conditions Reaction timedata were used to compute the percentage of the trialduration used by the subject (time-on-task) This value iscomputed by adding stimulus duration (200 msec) tomean reaction time (for correct responses only) anddividing by the length (msec) of the trial This ratioallows us to assess the average percentage of total trialtime spent on-task across a wide range of presentationand response rates Since only correct responses areincluded in the analysis we assume that the subject wason-task for the 200-msec stimulus presentation thus theratio described above is valid for any condition of singlestimulus presentation Unlike the other conditionshowever the two more difficult levels of the short-termmemory factor employed trains of stimuli During thepresentation of a sound train there were two to fourinterstimulus intervals of 400 msec and there is nospecific measure of whether or not the subject wason-task during these periods Thus the time-on-taskratio cannot be calculated and comparison of responsetimes across such different trial types is not possibleTime-on-task data for these two conditions will there-fore not be presented All statistical analyses were con-ducted using SPSS 90 for Windows (SPSS Chicago) andevaluated at p lt 05 with a modified Bonferroni correc-tion (Keppel 1991)
Repeated-measures ANOVAs assessed differencesacross the conditions for accuracy and time-on-task dataIn both cases there was a significant effect of conditionF(6108) = 1107 p lt 001 for accuracy and F(472) =50185 p lt 001 for time-on-task Within each of thethree factors t tests compared responses across thethree levels of task difficulty (Figure 2)
Within the stimulus presentation rate factor accuracydecreased in the fastest condition (78) compared toboth the moderate (96) and slow conditions (95)[t(18) = 712 p lt 001 and t(18) = 521 p lt 001respectively] Similarly accuracy in the target discrim-inability factor significantly decreased in the difficultcondition (84) compared to both the moderate(93) and easy (96) conditions [t(18) = 292 p =009 and t(18) = 353 p = 002 respectively] In theshort-term memory load conditions accuracy was sig-nificantly higher in the easy condition (96) than ineither the moderate (88) [t(18) = 277 p = 013] orthe difficult condition (91) [t(18) = 280 p = 012]The accuracy data indicate that the manipulations hadthe expected effects on level of difficulty without pro-ducing a severe performance decrement (Figure 2A)
For the stimulus presentation rate factor a signifi-cantly higher percentage of trial time was spent on-taskduring the difficult (fastest) condition (69) as com-pared to either the moderate (50) or the slowestcondition (30) [t(18) = iexcl1980 p lt 001 and t(18) =iexcl3477 p lt 001 respectively] and more time was
Table 1 Parametric Manipulation of Task Parameters
used at the moderate pace than at the slower pace[t(18) = 2742 p lt 001] This result is inevitable ofcourse because the presentation rate manipulation is adirect manipulation of time-on-task Within the targetdiscriminability factor significantly more of the trialtime (54) was used at the most difficult level of thetask as compared to both the easy (50) and moderate(50) levels of discriminability [t(18) = iexcl473 p lt001 and t(18) = iexcl391 p = 001 respectively] There-fore manipulation of task difficulty level had theexpected effect on time-on-task (Figure 2B) Summarydata for both behavioral measures are presented inTable 2
fMRI Results
The overall map of areas that showed TID (Figure 1B) isan average of z-scores from all seven conditions acrossall subjects The 20 strongest foci representing areas thatconsistently show deactivation are listed in Table 3
To focus the study on specific brain regions data fromall 30 subjects were combined across task conditions todefine ROIs in stereotaxic space using the following
technique For each condition the correlation valuesbetween the MRI signal and the ideal response vectorwere converted to z-scores These z-score maps werethen smoothed with a 4-mm RMS gaussian filter andaveraged over conditions and subjects A threshold wasarbitrarily set at z = iexcl1 From this map (Figure 1B)11 stereotaxic ROIs were identified as large (1684to 25280 mm3) discrete volumes The 11 ROIs are(in order of decreasing volume) left posterior parieto-occipital (PPO) cortex which includes parts of theangular gyrus dorsolateral occipital lobe and cuneusright PPO cortex right precuneussuperior parietallobule left precuneussuperior parietal lobule left ante-rior cingulatesuperior frontal gyrus left posterior cin-gulate gyrus left middle frontal gyrus left middleoccipital gyrus right posterior cingulate gyrus rightanterior cingulate gyrus and left fusiform gyrus(Figure 3) The regions identified from this analysisinclude many of the same foci identified in other studies(Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) (Figure 1A)
The relative magnitude of deactivation within theseROIs was measured using fit coefficients which are
Figure 2 (A) Accuracy dataacross conditions (B) Time-on-task data across conditions
398 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
decrease in neural activity Specifically this theory positsthat decreases in cerebral blood flow are caused byinterruption of ongoing internal processing that occursin the passive or lsquolsquorestrsquorsquo state (Gusnard amp Raichle 2001Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) This model assumes that lsquolsquorestrsquorsquo is a state oforganized functional brain activity These organizedprocesses are suspended or interrupted when the brainperforms an exogenously generated task Such internalactivity could involve many processes including thefollowing noted by Gusnard and Raichle (2001) andShulman et al (1997) monitoring of the external envi-ronment monitoring of body image and state andmonitoring of emotional state An additional processnoted by Shulman et al and emphasized by Binder et al(1999) is the ongoing internal lsquolsquothoughtrsquorsquo processingthat humans experience during resting consciousnessThis ongoing verbal and visual imagery referred to byJames (1890) as a lsquolsquostream of consciousnessrsquorsquo alsoappears to be suspended in the presence of an exoge-nously generated task (McGuire Paulesu Frackowiak ampFrith 1996 Giambra 1995 Teasdale Proctor Lloyd ampBaddeley 1993 Pope amp Singer 1976 Antrobus Singeramp Greenberg 1966 Antrobus 1968) The left lateraliza-tion of TID areas observed in most studies (Figure 1A)lends some support to this idea Cognitive processesongoing during rest are posited to require attentionalresources this theory allows deactivation to occur via apassive mechanism of reallocation of these resourcesThus deactivation could be effected by removal ofexcitatory (eg thalamic or transcortical) input in muchthe same way that passive removal of cortical inputproduces cerebellar deactivation in cases of cerebellarlsquolsquodiaschisisrsquorsquo (Martin amp Raichle 1983)
Suspension of ongoing resting state processing whilein a task condition implies that there are finite brainresources available for information processing Imposinga task on a subject requires that processing resources bereallocated from internally generated information pro-cessing to processing of the exogenous task Shulmanet al (1997) noted that if the suspension of passiveprocesses during an active task is due to competition forgeneral processing resources then the magnitude of thedeactivation should depend on the degree to which theactive task requires these same resources They foundlittle evidence however for an effect of task difficulty onTID magnitude while noting that their manipulation oftask difficulty may not have been strong enough toproduce such an effect We hypothesize that task diffi-culty is indeed a factor in determining the extent ofresources to be reallocated and that resource realloca-tion is the basis of TID Our aim will be to provide a cleartest of this hypothesis through carefully controlledmanipulations of task difficulty
Factors that influence task processing demands for anonsemantic perceptual task can include but are notlimited to task familiarity perceptual discriminability
stimulus presentation rate working memory load andtarget density Based on prior research (McGuire et al1996 Giambra 1995 Teasdale et al 1993 Pope ampSinger 1976 Antrobus et al 1966 Antrobus 1968)we chose three of these factors stimulus presentationrate perceptual discriminability and short-term memoryload to manipulate in order to vary the processingdemands of the exogenous task The goal was to system-atically increase the task demands within each factorthus requiring different degrees of resource reallocationin order to perform the task To reduce the possibility oftask-related activation in the regions of interest (ROIs) anonsemantic task (auditory target detection) was used(Binder et al 1999) We predicted that at the moredifficult levels the task would require greater allocationof resources thereby resulting in a greater degree ofdeactivation in those areas that are hypothesized to beactive during rest and consistently show TID
The importance of better understanding the type andextent of cognitive processing that occurs during theresting periods of a neuroimaging task cannot beoverstated A common approach in neuroimaging re-search is to alternate blocks of task activity with blocksof rest or in event-related designs to treat the inter-stimulus rest periods as a neutral baseline The brainactivity that occurs during rest is then subtracted fromthe activity that is task-related The assumption of thisapproach is that the remaining activity will be mostlyrelated to the specific demands of the task If restperiods are actually as we suggest periods of complexcognitive work then the use of rest as a baselinecondition and the subtraction of resting neural activityfrom the task-related activity may be inappropriate andcould confound interpretation of the data
An auditory target detection task was presented tosubjects This task had three levels of task difficulty foreach of three factors The task involved discriminationbetween two sounds one of which was defined as thetarget The design was a parametric manipulation ofthe difficulty levels of each factor so conditions weredistinguishable on the basis of only one task variable(Table 1) Subjects heard blocks of a task conditionalternating with rest periods and responded to targetstimuli by pressing a button The two more difficultmemory conditions used trains of stimuli and requiredsubjects to press a button after trains that containedspecific combinations of the target and nontarget stimuli
RESULTS
Behavioral Results
Due to technical difficulties with the response recordingsoftware behavioral data (accuracy and reaction time)were recorded for only 19 of the 30 subjects Accuracydata are presented as a percentage of the number ofcorrect responses minus the number of false alarms
396 Journal of Cognitive Neuroscience Volume 15 Number 3
divided by the number of targets This calculation adds apenalty for incorrect responses and accounts for thevariation in the number of stimuli per block in differentstimulus presentation rate conditions Reaction timedata were used to compute the percentage of the trialduration used by the subject (time-on-task) This value iscomputed by adding stimulus duration (200 msec) tomean reaction time (for correct responses only) anddividing by the length (msec) of the trial This ratioallows us to assess the average percentage of total trialtime spent on-task across a wide range of presentationand response rates Since only correct responses areincluded in the analysis we assume that the subject wason-task for the 200-msec stimulus presentation thus theratio described above is valid for any condition of singlestimulus presentation Unlike the other conditionshowever the two more difficult levels of the short-termmemory factor employed trains of stimuli During thepresentation of a sound train there were two to fourinterstimulus intervals of 400 msec and there is nospecific measure of whether or not the subject wason-task during these periods Thus the time-on-taskratio cannot be calculated and comparison of responsetimes across such different trial types is not possibleTime-on-task data for these two conditions will there-fore not be presented All statistical analyses were con-ducted using SPSS 90 for Windows (SPSS Chicago) andevaluated at p lt 05 with a modified Bonferroni correc-tion (Keppel 1991)
Repeated-measures ANOVAs assessed differencesacross the conditions for accuracy and time-on-task dataIn both cases there was a significant effect of conditionF(6108) = 1107 p lt 001 for accuracy and F(472) =50185 p lt 001 for time-on-task Within each of thethree factors t tests compared responses across thethree levels of task difficulty (Figure 2)
Within the stimulus presentation rate factor accuracydecreased in the fastest condition (78) compared toboth the moderate (96) and slow conditions (95)[t(18) = 712 p lt 001 and t(18) = 521 p lt 001respectively] Similarly accuracy in the target discrim-inability factor significantly decreased in the difficultcondition (84) compared to both the moderate(93) and easy (96) conditions [t(18) = 292 p =009 and t(18) = 353 p = 002 respectively] In theshort-term memory load conditions accuracy was sig-nificantly higher in the easy condition (96) than ineither the moderate (88) [t(18) = 277 p = 013] orthe difficult condition (91) [t(18) = 280 p = 012]The accuracy data indicate that the manipulations hadthe expected effects on level of difficulty without pro-ducing a severe performance decrement (Figure 2A)
For the stimulus presentation rate factor a signifi-cantly higher percentage of trial time was spent on-taskduring the difficult (fastest) condition (69) as com-pared to either the moderate (50) or the slowestcondition (30) [t(18) = iexcl1980 p lt 001 and t(18) =iexcl3477 p lt 001 respectively] and more time was
Table 1 Parametric Manipulation of Task Parameters
used at the moderate pace than at the slower pace[t(18) = 2742 p lt 001] This result is inevitable ofcourse because the presentation rate manipulation is adirect manipulation of time-on-task Within the targetdiscriminability factor significantly more of the trialtime (54) was used at the most difficult level of thetask as compared to both the easy (50) and moderate(50) levels of discriminability [t(18) = iexcl473 p lt001 and t(18) = iexcl391 p = 001 respectively] There-fore manipulation of task difficulty level had theexpected effect on time-on-task (Figure 2B) Summarydata for both behavioral measures are presented inTable 2
fMRI Results
The overall map of areas that showed TID (Figure 1B) isan average of z-scores from all seven conditions acrossall subjects The 20 strongest foci representing areas thatconsistently show deactivation are listed in Table 3
To focus the study on specific brain regions data fromall 30 subjects were combined across task conditions todefine ROIs in stereotaxic space using the following
technique For each condition the correlation valuesbetween the MRI signal and the ideal response vectorwere converted to z-scores These z-score maps werethen smoothed with a 4-mm RMS gaussian filter andaveraged over conditions and subjects A threshold wasarbitrarily set at z = iexcl1 From this map (Figure 1B)11 stereotaxic ROIs were identified as large (1684to 25280 mm3) discrete volumes The 11 ROIs are(in order of decreasing volume) left posterior parieto-occipital (PPO) cortex which includes parts of theangular gyrus dorsolateral occipital lobe and cuneusright PPO cortex right precuneussuperior parietallobule left precuneussuperior parietal lobule left ante-rior cingulatesuperior frontal gyrus left posterior cin-gulate gyrus left middle frontal gyrus left middleoccipital gyrus right posterior cingulate gyrus rightanterior cingulate gyrus and left fusiform gyrus(Figure 3) The regions identified from this analysisinclude many of the same foci identified in other studies(Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) (Figure 1A)
The relative magnitude of deactivation within theseROIs was measured using fit coefficients which are
Figure 2 (A) Accuracy dataacross conditions (B) Time-on-task data across conditions
398 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
divided by the number of targets This calculation adds apenalty for incorrect responses and accounts for thevariation in the number of stimuli per block in differentstimulus presentation rate conditions Reaction timedata were used to compute the percentage of the trialduration used by the subject (time-on-task) This value iscomputed by adding stimulus duration (200 msec) tomean reaction time (for correct responses only) anddividing by the length (msec) of the trial This ratioallows us to assess the average percentage of total trialtime spent on-task across a wide range of presentationand response rates Since only correct responses areincluded in the analysis we assume that the subject wason-task for the 200-msec stimulus presentation thus theratio described above is valid for any condition of singlestimulus presentation Unlike the other conditionshowever the two more difficult levels of the short-termmemory factor employed trains of stimuli During thepresentation of a sound train there were two to fourinterstimulus intervals of 400 msec and there is nospecific measure of whether or not the subject wason-task during these periods Thus the time-on-taskratio cannot be calculated and comparison of responsetimes across such different trial types is not possibleTime-on-task data for these two conditions will there-fore not be presented All statistical analyses were con-ducted using SPSS 90 for Windows (SPSS Chicago) andevaluated at p lt 05 with a modified Bonferroni correc-tion (Keppel 1991)
Repeated-measures ANOVAs assessed differencesacross the conditions for accuracy and time-on-task dataIn both cases there was a significant effect of conditionF(6108) = 1107 p lt 001 for accuracy and F(472) =50185 p lt 001 for time-on-task Within each of thethree factors t tests compared responses across thethree levels of task difficulty (Figure 2)
Within the stimulus presentation rate factor accuracydecreased in the fastest condition (78) compared toboth the moderate (96) and slow conditions (95)[t(18) = 712 p lt 001 and t(18) = 521 p lt 001respectively] Similarly accuracy in the target discrim-inability factor significantly decreased in the difficultcondition (84) compared to both the moderate(93) and easy (96) conditions [t(18) = 292 p =009 and t(18) = 353 p = 002 respectively] In theshort-term memory load conditions accuracy was sig-nificantly higher in the easy condition (96) than ineither the moderate (88) [t(18) = 277 p = 013] orthe difficult condition (91) [t(18) = 280 p = 012]The accuracy data indicate that the manipulations hadthe expected effects on level of difficulty without pro-ducing a severe performance decrement (Figure 2A)
For the stimulus presentation rate factor a signifi-cantly higher percentage of trial time was spent on-taskduring the difficult (fastest) condition (69) as com-pared to either the moderate (50) or the slowestcondition (30) [t(18) = iexcl1980 p lt 001 and t(18) =iexcl3477 p lt 001 respectively] and more time was
Table 1 Parametric Manipulation of Task Parameters
used at the moderate pace than at the slower pace[t(18) = 2742 p lt 001] This result is inevitable ofcourse because the presentation rate manipulation is adirect manipulation of time-on-task Within the targetdiscriminability factor significantly more of the trialtime (54) was used at the most difficult level of thetask as compared to both the easy (50) and moderate(50) levels of discriminability [t(18) = iexcl473 p lt001 and t(18) = iexcl391 p = 001 respectively] There-fore manipulation of task difficulty level had theexpected effect on time-on-task (Figure 2B) Summarydata for both behavioral measures are presented inTable 2
fMRI Results
The overall map of areas that showed TID (Figure 1B) isan average of z-scores from all seven conditions acrossall subjects The 20 strongest foci representing areas thatconsistently show deactivation are listed in Table 3
To focus the study on specific brain regions data fromall 30 subjects were combined across task conditions todefine ROIs in stereotaxic space using the following
technique For each condition the correlation valuesbetween the MRI signal and the ideal response vectorwere converted to z-scores These z-score maps werethen smoothed with a 4-mm RMS gaussian filter andaveraged over conditions and subjects A threshold wasarbitrarily set at z = iexcl1 From this map (Figure 1B)11 stereotaxic ROIs were identified as large (1684to 25280 mm3) discrete volumes The 11 ROIs are(in order of decreasing volume) left posterior parieto-occipital (PPO) cortex which includes parts of theangular gyrus dorsolateral occipital lobe and cuneusright PPO cortex right precuneussuperior parietallobule left precuneussuperior parietal lobule left ante-rior cingulatesuperior frontal gyrus left posterior cin-gulate gyrus left middle frontal gyrus left middleoccipital gyrus right posterior cingulate gyrus rightanterior cingulate gyrus and left fusiform gyrus(Figure 3) The regions identified from this analysisinclude many of the same foci identified in other studies(Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) (Figure 1A)
The relative magnitude of deactivation within theseROIs was measured using fit coefficients which are
Figure 2 (A) Accuracy dataacross conditions (B) Time-on-task data across conditions
398 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
used at the moderate pace than at the slower pace[t(18) = 2742 p lt 001] This result is inevitable ofcourse because the presentation rate manipulation is adirect manipulation of time-on-task Within the targetdiscriminability factor significantly more of the trialtime (54) was used at the most difficult level of thetask as compared to both the easy (50) and moderate(50) levels of discriminability [t(18) = iexcl473 p lt001 and t(18) = iexcl391 p = 001 respectively] There-fore manipulation of task difficulty level had theexpected effect on time-on-task (Figure 2B) Summarydata for both behavioral measures are presented inTable 2
fMRI Results
The overall map of areas that showed TID (Figure 1B) isan average of z-scores from all seven conditions acrossall subjects The 20 strongest foci representing areas thatconsistently show deactivation are listed in Table 3
To focus the study on specific brain regions data fromall 30 subjects were combined across task conditions todefine ROIs in stereotaxic space using the following
technique For each condition the correlation valuesbetween the MRI signal and the ideal response vectorwere converted to z-scores These z-score maps werethen smoothed with a 4-mm RMS gaussian filter andaveraged over conditions and subjects A threshold wasarbitrarily set at z = iexcl1 From this map (Figure 1B)11 stereotaxic ROIs were identified as large (1684to 25280 mm3) discrete volumes The 11 ROIs are(in order of decreasing volume) left posterior parieto-occipital (PPO) cortex which includes parts of theangular gyrus dorsolateral occipital lobe and cuneusright PPO cortex right precuneussuperior parietallobule left precuneussuperior parietal lobule left ante-rior cingulatesuperior frontal gyrus left posterior cin-gulate gyrus left middle frontal gyrus left middleoccipital gyrus right posterior cingulate gyrus rightanterior cingulate gyrus and left fusiform gyrus(Figure 3) The regions identified from this analysisinclude many of the same foci identified in other studies(Mazoyer et al 2001 Binder et al 1999 Shulman et al1997) (Figure 1A)
The relative magnitude of deactivation within theseROIs was measured using fit coefficients which are
Figure 2 (A) Accuracy dataacross conditions (B) Time-on-task data across conditions
398 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
Table 2 Behavioral Data
Reaction Time (msec) Amount of Trial Used Accuracy
(N = 19) M M
Stimulus Presentation Rate
Slower 39581 (7423) 2980 (371) 9464 (1044)
Moderate 29579 (3789) 4960 (379) 9552 (633)
Fast 21156 (2986) 6860 (498) 7769 (983)
Target Discriminability
Easy 29579 (3789) 4960 (379) 9552 (633)
Moderate 30298 (4447) 5030 (445) 9347 (983)
Difficult 33532 (4181) 5350 (418) 8397 (1321)
Short-Term Memory Load
Easy 29579 (3789) ndash 9552 (633)
Moderate 79697 (5954) ndash 8759 (1474)
Difficult 58790 (9864) ndash 9123 (656)
Values in parentheses are standard deviations
Table 3 Talairach Coordinates for Areas that Show Maximal Deactivation
Atlas StructureApproximate
Brodmannrsquos Areas
Talairach Coordinates
x y z z-score
L anterior cingulate gyrus 32 iexcl6 44 1 iexcl167
L superior frontal gyrus 9 iexcl6 48 19 iexcl144
L superior frontal gyrus 89 iexcl12 43 41 iexcl133
L superior frontal sulcus 8 iexcl23 25 46 iexcl164
L middle frontal gyrus 6 iexcl26 11 51 iexcl135
L posterior cingulate gyrus 31 iexcl7 iexcl50 30 iexcl185
L posterior cingulate gyrus 31 iexcl7 iexcl42 iexcl45 iexcl147
R posterior cingulate gyrus 3123 7 iexcl58 19 iexcl142
R precuneus 7 7 iexcl58 61 iexcl190
L precuneus 7 iexcl6 iexcl63 50 iexcl188
R superior parietal lobule 7 29 iexcl45 57 iexcl174
L superior parietal lobule 7 iexcl24 iexcl48 63 iexcl126
L cuneus 19 iexcl16 iexcl84 42 iexcl202
R cuneus 19 9 iexcl85 39 iexcl159
L superior occipital gyrus 19 iexcl37 iexcl82 30 iexcl189
R middle occipital gyrus 1937 49 iexcl68 7 iexcl177
L middle occipital gyrus 1937 iexcl49 iexcl73 4 iexcl168
L angular gyrus 39 iexcl47 iexcl72 21 iexcl173
L fusiform gyrus 37 iexcl28 iexcl42 iexcl19 iexcl136
R central sulcus 34 14 iexcl27 73 iexcl123
McKiernan et al 399
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
computed by a least-squares fit of the ideal responsevector to the observed response in each voxel for eachcondition in each subject (see Methods) Examples ofgroup average fit coefficient maps are shown in Figure 4to illustrate anatomic details of the variation in fitcoefficient magnitude across conditions For each taskcondition average deactivation values for each ROIwere measured in each subject by averaging the fitcoefficients for voxels within the ROI Figure 4 alsoshows brain regions in which fit coefficients had pos-itive values indicating positive correlations betweenthe ideal response vector and the observed BOLDsignal These activated areas which included bilateralauditory cortices in the superior temporal gyrus SMAand the adjacent anterior cingulate cortex bilaterallypremotor cortex and anterior insula bilaterally rightprefrontal cortex in the middle frontal gyrus rightsupramarginal gyrus the bilateral thalamus and bilat-eral cerebellum were also modulated to varyingdegrees by the task difficulty manipulations Theseeffects on activation will be presented in detail in asubsequent report
An omnibus repeated-measures (Condition pound ROI)ANOVA for each of the three factors assessed differencesin the magnitude of deactivation as measured by fitcoefficients Interest was focused on the interactionterm which was significant for each factor [Presentationrate F(20580) = 216 p = 003 Perceptual discrim-inability F(20580) = 342 p lt 001 Short-term mem-ory F(20580) = 804 p lt 001] indicating differingeffects of the task manipulations on different ROIs Allfollow-up analyses were evaluated at p lt 05 with amodified Bonferroni correction (Keppel 1991) Theeffectiveness of the task difficulty manipulation withineach factor was evaluated using repeated-measuresANOVAs (one for each ROI) to identify ROIs in which
there were differences in the magnitude of deactivationacross the three difficulty levels of the task
Within each factor the effect of manipulating taskdifficulty on the magnitude of deactivation was signifi-cant in at least one ROI Also within each of the threefactors and across all 11 ROIs the average fit coefficientfor the easiest task level was significantly different fromzero (lsquolsquorestrsquorsquo) indicating that even the easiest level of thetask produced TID in every ROI
The rate of stimulus presentation affected the degreeof deactivation only in the left anterior cingulatesuperior frontal gyrus F(258) = 417 p = 02 Therewas a trend toward significance in the left PPO cortexF(258) = 303 p = 056
Increasing the difficulty of stimulus discriminabilityaffected the magnitude of deactivation in five ROIs leftmiddle frontal gyrus F(258) = 890 p lt 001 left anteriorcingulatesuperior frontal gyrus F(258) = 586 p = 005right anterior cingulate gyrus F(258) = 490 p = 011left posterior cingulate gyrus F(258) = 470 p = 013 andleft PPO cortex F(258) = 370 p = 031
The effect of altering short-term memory load on themagnitude of deactivation was significant in five ROIsright precuneussuperior parietal lobule F(258) = 110p lt 001 right anterior cingulate gyrus F(258) = 703p = 002 left anterior cingulatesuperior frontal gyrusF(258) = 676 p = 002 left middle frontal gyrusF(258) = 557 p = 006 and right posterior cingulategyrus F(258) = 425 p = 019 There was a trendtoward significance in the left posterior cingulate gyrusF(258) = 319 p = 048
In those ROIs in which TID varied across conditionsfollow-up tests defined the effects of the within-factortask difficulty manipulations on TID magnitude Inalmost every case TID magnitude increased with greatertask difficulty The left anterior cingulatesuperior frontal
Figure 3 Eleven ROIs based on averaged z-scores across all subjects and all conditions The z-scores were thresholded at iexcl1 and the ROIs weredefined as areas that consistently deactivated across subjects and conditions
400 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
gyrus was significantly affected by manipulations to allthree factors (Figure 5A) Within the stimulus presenta-tion rate manipulations the fastest presentation rateincreased the average degree of deactivation more thanthe slowest rate [t(29) = 270 p = 011] or the moderaterate [t(29) = 240 p = 023] For the perceptualdiscriminability factor the moderately difficult and diffi-cult levels of discrimination (target was 28 and 16 of
an octave different from the standard respectively)attenuated activation significantly more than did theeasiest level of discrimination (40 of an octave differ-ence) [t(29) = 258 p = 015 and t(29) = 299 p = 006respectively] Similarly the moderate and difficultlevels of the short-term memory manipulation (stimulipresented in trains of three to five sounds) producedgreater average deactivation than did the easiest level
Figure 4 Magnitude of deacti-vation as measured by fit coef-ficients averaged across subjectswithin each ROI by conditionROIs are outlined Each panelrepresents the magnitude ofdeactivation during a specificcondition (A) task level that iscommon to all three factors(easiest task level for discrimin-ability and short-term memoryand moderate level for presen-tation rate) (B) fastest stimuluspresentation rate (C) difficultlevel of stimulus discriminabil-ity (D) difficult level of short-term memory factor Slice loca-tions match those in Figure 3
McKiernan et al 401
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
of the task [t(29) = 236 p = 025 and t(29) = 366p = 001 respectively]
The right anterior cingulate gyrus showed effects ofboth stimulus discriminability and short-term memoryload The degree of deactivation during the mostdifficult level of stimulus discrimination was significantlylarger than during the easiest level [t(29) = 278p = 009] Effects of short-term memory load mirroredthose in the left anterior cingulatesuperior frontalgyrus The moderate and difficult levels of the task eachproduced larger average deactivations than the easiestlevel [t(29) = 387 p = 001 and t(29) = 325 p = 003respectively] (Figure 5B)
The left middle frontal gyrus also showed effects ofboth stimulus discriminability and short-term memory
load (Figure 5C) Within stimulus discriminability theaverage degree of deactivation was larger during boththe moderate and difficult levels of the task than duringthe easy level [t(29) = 281 p = 009 and t(29) = 419p lt 001 respectively] For short-term memory thedegree of deactivation was significantly stronger duringthe difficult condition compared to the easy condition[t(29) = 283 p = 008]
Deactivation in the left posterior cingulate gyrus variedwith stimulus discriminability The most difficult tasklevel produced significantly stronger deactivation thandid the easiest level [t(29) = 303 p = 005] (Figure 5D)An identical pattern was observed in the PPO region ofthe left hemisphere which included the angular gyrusthe dorsolateral occipital cortex and the cuneus
Figure 5 TID in six ROIs that show differences in magnitude of deactivation across levels of task difficulty Bars represent the magnitude of thedeactivation for each condition As task difficulty increases magnitude of deactivation also increases (except in the right precuneussuperior parietallobule) In every ROI the easiest task condition produces a significant degree of TID
402 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
(Figure 5E) where the most difficult level of the stimulusdiscriminability factor produced significantly strongerdeactivation than did the easiest level [t(29) = 260p = 015]
The right posterior cingulate gyrus showed sensitivityto short-term memory load (not shown in Figure 5)There was stronger deactivation during the moderatelevel of difficulty than during the easy level of the task[t(29) = 249 p = 019]
The right precuneussuperior parietal lobule alsoshowed sensitivity to the short-term memory load ma-nipulation but in a pattern opposite from the otherROIs (Figure 5F) Unlike the other ROIs the easiest levelof the task produced stronger deactivation than dideither the moderate or difficult levels of short-termmemory load [t(29) = iexcl294 p = 006 and t(29) =iexcl314 p = 004 respectively] This deviation from theusual pattern appears to reflect partial activation of thisROI by the more difficult memory levels of the task Ascan be seen by close inspection of Figure 4D a largeincrease in activation of the right parietal cortex (supra-marginal gyrus) just lateral to this ROI occurs duringthe most difficult memory condition In some subjectsthis activation extended into the superior parietal ROIdecreasing the average level of deactivation during themore difficult memory load conditions
DISCUSSION
These behavioral and fMRI results indicate that in manybrain regions increasing task processing demands re-sults in greater degrees of TID Behaviorally all three ofthe factors thought to affect allocation of informationprocessing resources had the intended effect Accuracydecreased and time-on-task increased with increasingtask difficulty indicating that more difficult levels of thetask required greater processing and attentional resour-ces The imaging results show that for most regions themagnitude of deactivation increased during the moredifficult conditions relative to the easiest conditionalthough this pattern was observed primarily with theshort-tem memory load and target discriminability ma-nipulations We propose that as task demands increaseprocessing resources are increasingly diverted from on-going internal processes occurring at lsquolsquorestrsquorsquo to areasthat are involved in the task If this is correct then thedeactivation seen in these areas can be thought of asindicating the extent to which lsquolsquorestingrsquorsquo state processingwas interrupted or suspended to accommodate task-related processing
The idea that the brain has a limited informationprocessing capacity is relatively axiomatic and has beenused to account for a range of task interference phe-nomena (Posner 1978) One example of task interfer-ence that is particularly relevant to models of TID is theeveryday experience of interference with an internallsquolsquotrain of thoughtrsquorsquo when unexpected external events
require attention Laboratory studies of subjects whowere asked to report thought content at random inter-vals during controlled task conditions demonstrate thatthere is a continuous and graded shifting of attentionbetween externally and internally generated sources ofinformation The degree of interference with ongoingthought processes depends on the perceptual andshort-term memory demands of the external taskindicating a direct competition between exogenousand endogenous signals for processing resources(McGuire et al 1996 Giambra 1995 Teasdale et al1993 Pope amp Singer 1976 Antrobus et al 1966Antrobus 1968) If this type of resource reallocationunderlies TID it is necessarily the case that increasingexternal task demands should produce greater interfer-ence with ongoing internal processes resulting in great-er TID Shulman et al (1997) found no evidence forsuch an effect of task difficulty but noted that theirdifficulty manipulation (novel versus practiced verb gen-eration) may have been too weak to produce an effectIn contrast the present experiment in which taskdifficulty was parametrically manipulated within threefactors provides clear confirmation of the predictedeffect for task difficulty for two of the independentfactor manipulations stimulus discriminability andshort-term memory load
Although the present results were generally positiveour experimental design may also have suffered fromlack of sensitivity Specifically the easiest task conditionsmay have been more difficult than we intended leavingrelatively little capacity for further deactivation and thusinducing a lsquolsquoceiling effectrsquorsquo on TID magnitude Note inFigure 5 for example that even the easiest task con-ditions produced extensive deactivation and furtherdeactivations were relatively small by comparison Alsosupporting the impression of a ceiling effect is the factthat there were four instances in which the moderatedifficulty condition produced greater TID than theeasy condition but only one case in which the hardestcondition produced greater TID than the moderatecondition (Figure 5) Such a ceiling effect could alsobe responsible for the relatively small effects of stimulusrate on TID It may be that maximal levels of TID (for thenonmemory easy discrimination version of the task)were attained even at the slowest presentation rate(05 Hz) and that larger rate effects would have beendetected had slower rate conditions been included
A second reason why this study may lack sensitivity todetect effects of task difficulty lies in the method of ROIaveraging Even though the ROIs were defined by neg-ative correlation with the reference vector (averagedacross task conditions and subjects) there still may besome number of individual voxels or cortical subregionswithin an ROI that are activated by the different tasklevels Depending on the degree and extent of thisactivation the average intensity value for a given ROImay be affected (ie made less negative) This possibility
McKiernan et al 403
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
could explain why a few of the ROIs did not show aneffect of task difficulty across any factor
As discussed in the Introduction and illustrated inFigure 1 several brain areas consistently show TIDacross varying tasks stimulus modalities and imagingmodalities (Mazoyer et al 2001 Binder et al 1999Shulman et al 1997) Regions showing TID in thepresent study were strikingly similar to those observedin previous studies and as in previous studies showedan overall left hemisphere preponderance (Figure 1Table 3) One notable exception was the finding ofTID in the lateral occipital lobe (middle occipital gyrus)which was not observed in previous studies Becauselateral occipital cortex is involved in visual functions thedecrease in activity in this region may be an example ofmodality-specific task-dependent deactivation (ie thisarea is activated in studies that use visual tasks thus withthose tasks it will not show less activation relative torest) (Lewis Beauchamp amp DeYoe 2000) Similarly theleft fusiform gyrus deactivation observed in this studyand in a previous study using auditory tasks (Binderet al 1999) may only occur when auditory tasks areused because visual tasks partially activate this regionNeither of these regions showed effects of task difficultyon TID
As noted by Gusnard and Raichle (2001) and Raichleet al (2001) the consistency in location of TID across awide variety of tasks supports the idea that thesedecreases in brain activity are task-independentThese authors propose that decreases in brain activity[as measured by the oxygen extraction fraction (OEF)during PET studies] occur relative to a lsquolsquodefaultrsquorsquo baselinestate of brain activity They define baseline as a state ofmetabolic equilibrium as determined by maximal spatialuniformity of OEF Thus the localized decreases inactivity during TID are viewed as changes from a lsquolsquotruersquorsquobaseline rather than a return to baseline from somestate of absolute activation The ideas proposed byRaichle et al are similar to our own theory that theareas that deactivate are not necessarily more active thanother areas during rest Rather we propose that areasdeactivate because they require attentional input to stayactive when those attentional resources are needed forprocessing other information and are reallocated thesebrain regions become deactivated
The variety of possible attention-dependent processesongoing during the conscious resting state has beendiscussed at length by a number of authors (Gusnard ampRaichle 2001 Mazoyer et al 2001 Binder et al 1999Shulman et al 1997 Andreasen et al 1995 Ingvar1985 Pope amp Singer 1976 James 1890) These includeverbal and visual imagery planning and problem-solvingmonitoring the external environment monitoring theinternal sensory state and body image monitoring emo-tional state episodic memory encoding and retrievaland working memory These processes are in no waymutually exclusive and it seems quite likely that differ-
ent components of the TID lsquolsquonetworkrsquorsquo may be involvedto different degrees in these and other processes Thereis converging evidence for example for an emotion-processing role of the ventromedial frontal structuresthat show TID suggesting that these areas may beprocessing information regarding emotional state duringrest (Bechara Damasio amp Damasio 2000) Despite alack of obvious external stimulation subjects experiencenot only emotional states but also random thoughts andbodily sensations during rest These episodes reachconscious awareness and can later be recalled indicatingsome degree of episodic memory encoding Anatomicneuropsychologic and neuroimaging data suggest a rolefor the posterior cingulate cortex in episodic memoryencoding suggesting that TID in this region may partlyreflect interruption of such processes (Cabeza amp Nyberg2000 Maddock 1999 Andreasen et al 1995 Rudge ampWarrington 1991 Valenstein et al 1987) Maddock(1999) additionally proposed that the posterior cingu-late and retrosplenial cortices may be activated by emo-tionally salient words and memories that are attendedto during rest
Simpson Drevets Snyder Gusnard and Raichle(2001) and Simpson Snyder Gusnard and Raichle(2001) investigated the effects of performance anxietyon medial prefrontal cortical activity Their results(ie suspension of activity in the medial prefrontalcortex during attention-demanding cognitive tasks) arein agreement with other published imaging data Theysuggested that the increase in anxiety experienced bysubjects during novel or more demanding cognitivetasks may also influence blood flow Specifically reduc-tions in activity in this region represent the combinedeffects of the attentional demands of the task and anyaccompanying performance anxiety They reported thatblood flow was maximally reduced when attentionaldemands were high and anxiety was low and the leastreduction occurred when attentional demands andanxiety were both either high or low This implies thatthe emotional state of the subject may be a modulatingfactor between the degree of observed TID and theprocessing demands of the task In order for thishypothesis to explain the findings in the current study[eg anterior cingulate gyrus ROIs (bilaterally) showedan effect of task difficulty] we would have to assumethat our subjects experienced little performance anxietyIt seems plausible however that the more difficultconditions that have higher attentional requirementswould evoke a stronger degree of performance anxietyBased on Simpson et al this combination of increasedattentional demand and stronger anxiety componentshould effectively cancel each other out resulting in aminimal decrease in blood flow As we did not collectsubjective mood ratings this is an unanswered issue It iscertainly a plausible idea that there are other factors(most likely related to the processes ongoing during theconscious resting state listed above) that contribute to
404 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
the relationship between cognitive task performanceand TID Given that the experimental manipulation inthis study was focused on task processing demands theimpact of other variables is only speculative at this pointThe regions that did not show an effect of task difficultymay be more sensitive to some of these other factors
Binder et al (1999) reported a striking similaritybetween the network of regions showing TID and thenetwork of regions associated with semantic processingin a study contrasting semantic and phonologic auditorytasks The lsquolsquosemantic networkrsquorsquo observed in this studyincluded the left superior and middle frontal gyri leftangular gyrus left posterior cingulate gyrus and leftfusiform gyrusmdashall regions that show TID This samegroup of regions has been observed repeatedly in otherstudies that compared semantic with phonologic tasks(Binder amp Price 2001 Roskies Fiez Balota Raichle ampPetersen 2001 Poldrack et al 1999 Mummery Patter-son Hodges amp Price 1998 Price Moore Humphreys ampWise 1997 Demonet et al 1992) Binder et al foundthese regions to be deactivated by a nonsemantic tonediscrimination task (similar to the memory tasks used inthe present study) but not by a semantic word catego-rization task suggesting that these regions are engagedboth by exogenously presented semantic tasks andduring rest They suggested that during resting con-sciousness these regions participate in retrieval andmanipulation of conceptual knowledge for the purposeof planning and problem-solving activities that manifestphenomenologically as verbal and visual images orthoughts (Binder et al 1999)
The present study along with others just discussedprovides evidence that there are specific brain regionsthat are more active during lsquolsquorestrsquorsquo than during anexogenously generated cognitive task The literatureand results reviewed here summarize what we knowabout TID It is not specific to any one sensory modality(ie both visual and auditory tasks produce TID) itoccurs in response to a wide variety of cognitive tasksit occurs consistently in specific brain regions and insome regions the magnitude of the deactivation issensitive to the difficulty level (ie processing demand)of the task These findings have significantly contrib-uted to our understanding of this phenomenon butthere are still many important unanswered questionsTwo of the most critical questions are (1) What are thespecific cognitive (or other noncognitive) processesthat occur during rest states and which are subject tosuspension in the presence of an external task and (2)What isare the physiological mechanism(s) that medi-ate this suspension of activity (or from our perspectivethe reallocation of processing resources) Without ananswer to the first question the rest state remainsundefined from a cognitive processing perspectiveThe use of an undefined state as a baseline or controlcondition could be very problematic for cognitive neu-roimaging studies A major goal of such studies is to
identify the neural correlates of specific cognitive pro-cesses A critical prerequisite is that the cognitiveprocesses engaged during the various experimentalconditions under study be defined with some reason-able level of confidence A more complete understand-ing of both the similarities and the differences betweenthe processes that are engaged during rest and thoseengaged by tasks of interest would therefore allow forstronger experimental designs (eg better matching ofcontrol and task states) in future neuroimaging studiesPerhaps a reasonable compromise at this time would bethe inclusion of both a simple baseline such as rest orfixation and a more complex control state specific tothe experiment
In this study we parametrically manipulated thedifficulty of an auditory target detection task to inves-tigate how altering processing demands impacts theallocation of available processing resources within thebrain We propose that there is ongoing internal infor-mation processing during the conscious resting stateand that the task-induced decreases in BOLD signalrepresent suspension of this ongoing processing Thisstudy supports the theory that the TIDs frequentlyfound in brain imaging studies are produced by areallocation of processing resources
METHODS
Subjects
Participants were 30 neurologically normal subjects(19 women and 11 men) ranging in age from 18 to 48years All subjects were right-handed as measured by alaterality quotient gt50 on the Edinburgh HandednessInventory (Oldfield 1971) Subjects performed an audi-tory target detection task identifying targets by pressinga button with the index finger of their right handSubjects received an hourly stipend for participatingThis study was approved by the Medical College ofWisconsin Human Research Review Committee
Stimulus Construction and Study Design
An auditory stimulus was constructed by phase-scram-bling the spectral components of a sample of 50 spokenwords This produced a tone with an average spectrumthat matched that of speech but with no temporalcoherence thus rendering it unrecognizable as speechA randomly selected 200-msec segment from this soundserved as the standard (nontarget) stimulus The targetstimulus was constructed by modulating over time allfrequencies in the standard stimulus using a linear rampfunction with an excursion equal to 40 of an octave(2 octavessec) We constructed the target stimulus inthis fashion to minimize the requirement for comparisonto the nontarget stimulus (eg higher or lower in pitchor loudness) thus minimizing the demand on short-termmemory For the experimental conditions in which target
McKiernan et al 405
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
discriminability was manipulated (moderate discrimina-bility and difficult discriminability) two alternative targetstimuli were constructed These stimuli were created asdescribed above but with frequency excursions equal to28 (14 octavessec) and 16 (08 octavessec) of anoctave respectively
There were three factors and seven conditions in theparametric design one condition that served as a com-mon level across all factors (the easy level in targetdiscriminability and short-term memory load and themoderate level in stimulus presentation rate) a moder-ate and difficult level of both target discriminability andshort-term memory load factors and a slower and fasterlevel of the stimulus presentation rate Due to therequirement for different instructions in the memoryload conditions a block design was employed Eachsubject completed two imaging runs for each of theseven conditions The order of the condition runs wascompletely randomized across subjects Within a con-dition the two runs differed only in the location of thetargets Within each condition there were 12 taskrestcycles (48 sec per cycle) The stimuli (target and stand-ard sounds) were semi-randomly ordered within eachpresentation the first two stimuli were always standardsounds and there were never more than two targets in arow Table 1 provides a summary of the defining taskparameters for each condition In the short-term mem-ory load conditions the stimuli were presented in trainsof three to five sounds In all other conditions thestimuli were presented one at a time
Except for the memory load conditions subjects wereinstructed to lsquolsquopress the button whenever you hear atarget soundrsquorsquo In the memory load conditions theinstructions were as follows (moderate memory load)lsquolsquopress the button after any train that contained exactlytwo target soundsrsquorsquo and (difficult memory load) lsquolsquopressthe button after any train that contained exactly twotarget sounds and two standard soundsrsquorsquo Before eachrun the subjects were told which condition they wereabout to hear All stimuli were presented and responseswere recorded using E-PRIME software (Beta 52 Psy-chology Software Tools)
Image Acquisition
Scanning was conducted at 15 Tesla on a General Electric(GE Medical Systems Milwaukee WI) Signa scannerequipped with a three-axis local gradient coil Functionaldata were collected via a multislice gradient-echo echo-planar sequence with repetition time of 3 sec and an echotime of 40 msec Other imaging parameters included afield-of-view of 24 cm a matrix size of 64 by 64 pixels anda voxel size of 375 pound 375 pound 6 mm Twenty-one or 23contiguous sagittal slices covering the whole brain wereacquired every 3 sec A total of 106 sequential images wereacquired at each slice location in each acquisition runEach 106-image EPI series began with four baseline
images to allow equilibrium of the magnetic resonancesignal to be reached followed by 100 images during sixtaskrest cycles and finally ending with two images withvariable echo time (used in reconstruction of the data)When the two runs for each condition were concatenateda total of 96 images were available for each task conditionas well as for the related rest periods An additional 12-secrest period was included at the end of the run the fourimages acquired during this time were not included in theanalysis High-resolution T1-weighted anatomic imageswere collected as a set of 124 contiguous sagittal slices(12 mm thick) using a 3-D spoiled gradient-echosequence (SPGR GE Medical Systems)
Image Analyses
All image analysis was completed using the AFNI soft-ware package (Cox 1996a) Subject data were volumetri-cally registered using an iterative procedure thatminimizes variance in voxel intensity differencesbetween images (Cox 1996b) Both time series for eachcondition were concatenated into a single dataset afterremoval of linear trends and means The time series of3-D images for each of the seven conditions was sub-jected to an analysis of covariance (ANCOVA) withvariability due to movement (as measured by six move-ment parameters) factored out The predicted responsevector was created by convolving a lsquolsquoboxcarrsquorsquo function ofunit height and period = 48 sec with a gamma functionrepresenting an average hemodynamic response Tensuch response vectors were created varying in phasefrom iexcl25 to 25 sec in 05-sec increments about theactual onset time of the block cycles to compensate forspatial variation in the hemodynamic response delayDuring the ANCOVA the program determined which ofthe 10 response vectors best represented the data ineach voxel Areas that were negatively correlated withthe response vector were defined as areas that deacti-vated during the task (Figures 1 and 3) In preparationfor the group analysis correlation values were convertedto z-scores The correlation analysis also provided fitcoefficients at each voxel for each condition Thesecoefficients represent the scaling value required to bestfit the reference vector to the observed BOLD responseThey represent a measure of the magnitude of thedeactivation Very large fit coefficients (absolute valuepara15) were assumed to correspond to large drainingveins and were ignored in the analyses
Data for group analyses were projected into standardstereotaxic space (Talairach amp Tournoux 1988) andsmoothed slightly with a 4-mm root-mean-square Gaus-sian kernel To identify ROIs z-scores for all sevenconditions across all subjects were averaged into a singlemap and thresholded to retain values less than z = iexcl1Eleven ROIs were identified Average fit coefficientswithin each ROI for each condition were subjected toa repeated-measures ANOVA to identify differences in
406 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
the magnitude of deactivation across the seven condi-tions Within each of the three factors follow-up testsidentified which levels of task difficulty produced sig-nificantly different degrees of deactivation
Acknowledgments
This research was supported by National Institute of MentalHealth Grant PO MH51358 National Institute of NeurologicalDisorders and Stroke Grant RO1 NS33576 and NationalInstitutes of Health General Clinical Research Center GrantM01 RR00058 We thank T Prieto T Thelaner B DrsquoAngeloand A Moths for technical assistance
Reprint requests should be sent to Kristen A McKiernanNeuropsychiatry Research Center Whitehall Building TheInstitute of Living 200 Retreat Avenue Hartford CT 06106USA or via e-mail kmckierharthosporg
The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-11323
REFERENCES
Andreasen N C OrsquoLeary D S Cizadlo T Arndt S Rezai KWatkins G L Boles Ponto L L amp Hichwa R D (1995)Remembering the past Two facets of episodic memoryexplored with positron emission tomography AmericanJournal of Psychiatry 152 1576 ndash1585
Antrobus J S (1968) Information theory and stimulus-independent thought British Journal of Psychology 59423 ndash430
Antrobus J S Singer J L amp Greenberg S (1966) Studies inthe stream of consciousness Experimental enhancementand suppression of spontaneous cognitive processesPerceptual and Motor Skills 23 399 ndash417
Bechara A Damasio H amp Damasio A R (2000) Emotiondecision making and the orbitofrontal cortex CerebralCortex 10 295 ndash307
Binder J R Frost J A Hammeke T A Bellgowan P S FRao S M amp Cox R W (1999) Conceptual processingduring the conscious resting state A functional MRI studyJournal of Cognitive Neuroscience 11 80ndash93
Binder J R amp Price C J (2001) Functional imaging oflanguage In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 187 ndash251)Cambridge MIT Press
Bookheimer S Y Zeffiro T A Blaxton T Gaillard T ampTheodore W (1995) Regional cerebral blood flow duringobject naming and word reading Human Brain Mapping 393ndash106
Cabeza R amp Nyberg L (2000) Imaging cognition II Anempirical review of 275 PET and fMRI studies Journal ofCognitive Neuroscience 12 1ndash47
Cox R W (1996a) AFNI Software for analysis andvisualization of functional magnetic resonance neuroimagesComputer and Biomedical Research 29 162 ndash173
Cox R W (1996b) Algorithms for image registration motiondetection and motion correction In R Savoy (Ed)fMRI2DAY A two day workshop on functional MRI(pp 25ndash43) Boston MGH-NMR Center
Demonet J-F Chollet F Ramsay S Cardebat DNespoulous J-L Wise R Rascol A amp Frackowiak R(1992) The anatomy of phonological and semanticprocessing in normal subjects Brain 115 1753 ndash1768
Frith C D Friston K J Liddle P F amp Frackowiak R S J(1991) A PET study of word finding Neuropsychologia 291137 ndash1148
Giambra L M (1995) A laboratory method for investigatinginfluences on switching attention on to task-unrelatedimagery and thought Consciousness and Cognition 41ndash21
Gusnard D A amp Raichle M E (2001) Searching for abaseline Functional imaging and the resting human brainNature Neuroscience 2 685ndash694
Heistad D D amp Kontos H A (1983) The cardiovascularsystem In J T Sheppard amp F M Abboud (Eds) Handbookof physiology (pp 137 ndash182) Maryland American Physiolo-gical Society
Howard D Patterson K Wise R Brown W D Friston KWeiller C amp Frackowiak R (1992) The cortical localizationof the lexicons Brain 115 1769 ndash1782
Ingvar D H (1985) Memory of the future An essay on thetemporal organization of conscious awareness HumanNeurobiology 4 127ndash136
James W (1890) Principles of psychology (vol 1) New YorkDover
Keppel J (1991) Design and analysis A researcherrsquoshandbook New Jersey Prentice-Hall
Lewis J W Beauchamp M S amp DeYoe E A (2000) Acomparison of visual and auditory motion processing inhuman cerebral cortex Cerebral Cortex 10 873ndash888
Maddock R J (1999) The retrosplenial cortex and emotionNew insights from functional neuroimaging of the humanbrain Trends in Neuroscience 22 310 ndash316
Martin W R amp Raichle M E (1983) Cerebellar blood flowand metabolism in cerebral hemisphere infarction Annals ofNeurology 14 168 ndash176
Mazoyer B Zago L Mellet E Bricogne S Etard O HoudeO Crivello F Joliot M Petit L amp Tzourio-Mazoyer N(2001) Cortical networks for working memory and executivefunctions sustain the conscious resting state in man BrainResearch Bulletin 54 287 ndash298
McGuire P K Paulesu E Frackowiak R S J amp Frith C D(1996) Brain activity during stimulus independent thoughtNeuroReport 7 2095 ndash2099
Mummery C J Patterson K Hodges J R amp Price C J(1998) Functional neuroanatomy of the semantic systemDivisible by what Journal of Cognitive Neuroscience 10766 ndash777
Ogawa S Lee T M Kay A R amp Tank D W (1990) Brainmagnetic resonance imaging with contrast dependent onblood oxygenation Proceedings of the National Academy ofSciences USA 87 9868 ndash9872
Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113
Poldrack R A Wagner A D Prull M W Desmond J EGlover G H amp Gabrieli J D E (1999) Functionalspecialization for semantic and phonological processingin the left inferior prefrontal cortex Neuroimage 10 15ndash35
Pope K S amp Singer J L (1976) Regulation of the stream ofconsciousness Toward a theory of ongoing thoughtIn G E Schwartz amp D Shapiro (Eds) Consciousness andself-regulation (pp 101 ndash135) New York Plenum
Posner M I (1978) Chronometric explorations of mindHillsdale NJ Erlbaum
Price C J Moore C J Humphreys G W amp Wise R J S(1997) Segregating semantic from phonological processesduring reading Journal of Cognitive Neuroscience 9727 ndash733
McKiernan et al 407
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
Raichle M E (2001) Functional neuroimaging A historicalperspective In R Cabeza amp A Kingstone (Eds) Handbookof functional neuroimaging of cognition (pp 3ndash26)Cambridge MIT Press
Raichle M E MacLeod A M Snyder A Z Powers W JGusnard D A amp Shulman G L (2001) A default mode ofbrain function Proceedings of the National Academy ofSciences USA 98 676ndash682
Roskies A L Fiez J A Balota D A Raichle M E ampPetersen S E (2001) Task-dependent modulation ofregions in the left inferior frontal cortex during semanticprocessing Journal of Cognitive Neuroscience 13829 ndash843
Rudge P amp Warrington E K (1991) Selective impairment ofmemory and visual perception in splenial tumours Brain114 349 ndash360
Shulman G L Fiez J A Corbetta M Buckner R L MiezinF M Raichle M E amp Pertersen S E (1997) Commonblood flow changes across visual tasks II Decreases incerebral cortex Journal of Cognitive Neuroscience 9648 ndash663
Simpson J R Drevets W C Snyder A Z Gusnard D Aamp Raichle M E (2001) Emotion-induced changes in thehuman medial prefrontal cortex II During anticipatoryanxiety Proceedings of the National Academy of ScienceUSA 98 688 ndash693
Simpson J R Snyder A Z Gusnard D A amp Raichle M E(2001) Emotion-induced changes in the human medialprefrontal cortex I During cognitive task performanceProceedings of the National Academy of Sciences USA 98683 ndash687
SPSS 90 for Windows [Computer programming language]Chicago SPSS
Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme
Teasdale J D Proctor L Lloyd C A amp Baddeley A D(1993) Working memory and stimulus-independentthought Effects of memory load and presentation rateEuropean Journal of Cognitive Psychology 5 417 ndash433
Valenstein E Bowers D Verfaellie M Heilman K MDay A amp Watson R T (1987) Retrosplenial amnesiaBrain 110 1631 ndash1646
408 Journal of Cognitive Neuroscience Volume 15 Number 3
