Localization of Executive Functions in Dual-TaskPerformance with fMRI

Andre J Szameitat1 Torsten Schubert2 Karsten Muller1 andD Yves von Cramon1

Abstract

amp We report a study that investigated the neuroanatomicalcorrelates of executive functions in dual-task performance withfunctional magnetic resonance imaging Participants performedan auditory and a visual three-choice reaction task eitherseparately as single tasks or concurrently as dual tasks In thedual-task condition two stimuli were presented in rapidsuccession to ensure interference between the componenttasks (psychological refractory period) The behavioral datashowed considerable performance decrements in the dual-taskcompared to the single-task condition Dual-task-relatedactivation was detected with two different neuroimagingmethods First we determined dual-task-related activationaccording to the method of cognitive subtraction For thatpurpose activation in the dual-task was compared directly withactivation in the single-task conditions This analysis revealed

that cortical areas along the inferior frontal sulcus (IFS) themiddle frontal gyrus (MFG) and the intraparietal sulcus (IPS)are involved in dual-task performance The results of thesubtraction method were validated with the method of para-metric manipulation For this purpose a second dual-task con-dition was introduced where the difficulty of the dual-taskcoordination was increased compared with the first dual-taskcondition As expected behavioral dual-task performancedecreased with increased dual-task difficulty Furthermorethe increased dual-task difficulty led to an increase of activationin those cortical regions that proved to be dual-task relatedwith the subtraction method that is the IFS the MFG and theIPS These results support the conclusion that dorsolateralprefrontal and superior parietal cortices are involved in thecoordination of concurrent and interfering task processing amp

INTRODUCTION

Executive functions are believed to be a prerequisitefor complex human behavior especially in situationswhere contradictory and interfering information has tobe processed in order to execute goal-oriented behav-ior (Badgaiyan 2000 Miller 2000 Knight Graboweckyamp Scabini 1995) One classical example of such situa-tions is dual-task performance in which two tasks haveto be carried out concurrently Here executive func-tions are needed in order to coordinate the concurrentprocessing of the different streams of informationAccordingly the investigation of dual-task performancegives insights into how the brain realizes complexhuman behavior

However the precise functioning and neural imple-mentation of these executive functions is still unre-solved Previous neuroimaging studies investigatingdual-task performance yielded rather contradictoryresults with respect to the functional neuroanatomy ofdual-task processing While some studies identified thelateral prefrontal cortex as related to dual-task perfor-

mance (Herath Klingberg Young Amunts amp Roland2001 Koechlin Basso Pietrini Panzer amp Grafman 1999Goldberg et al 1998 DrsquoEsposito et al 1995) otherstudies failed to show such results (Adcock ConstableGore amp Goldman-Rakic 2000 Bunge Klingberg Jacob-sen amp Gabrieli 2000 Klingberg 1998) One reason forthese contradictory findings might be the paradigmsemployed Previous studies often used rather complexparadigms which do not allow for control of the pro-cessing strategies applied by the participants Thereforeit is possible that in some of these studies the partic-ipants processed the tasks without interference Cru-cially interference between the processing of the tasks isthought to be the main cause for the need of executivefunctions in dual-task performance (Meyer amp Kieras1997 DeJong 1995 Umilta Nicoletti Simion Tagliabueamp Bagnara 1992 Baddeley 1990) Accordingly if pre-viously used paradigms enabled task processing withoutinterference executive functions would have beenabsent (Pashler 1994 Broadbent 1982 Craik 1947)This assumption is supported by the finding that someof the abovementioned studies which yielded no dual-task-related activation showed no (Bunge et al 2000)or rather small (Adcock et al 2000) performance dec-rements in the dual-task compared to the single-task

1Max-Planck-Institute of Cognitive Neuroscience Leipzig2Humboldt University Berlin

copy 2002 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 148 pp 1184ndash1199

conditions (see also Smith et al 2001) Thus the firstaim of the present study was to test for dual-task-relatedactivation by using a behavioral dual-task paradigmwhich ensures the presence of interference and there-fore of executive functions

A second issue in this study concerns the method-ology used to obtain dual-task-related activation Mostprevious studies used the method of cognitive sub-traction to obtain dual-task-related activation wheredual-task-related activation is detected by subtractingactivation in two single tasks from activation in a dualtask However this approach has been criticizedbecause of some problematic methodological presump-tions (Sartori amp Umilta 2000 Sidtis Strother Andersonamp Rottenberg 1999 Braver et al 1997 Friston Priceet al 1996 Sternberg 1969) As will be shown later theuse of the cognitive subtraction method might haveobscured dual-task-related activation so far Thereforein the present study we used the method of parametricmanipulation (Braver et al 1997 Cohen et al 1997) inaddition to the method of cognitive subtraction toassess dual-task-related activation

Interference in Dual-Task Situations

The typical indicators of interference in dual-taskperformance are performance decrements reflectedby prolonged reaction times (RTs) or increased errorrates as compared to single-task performance Theo-ries of dual-task performance have suggested that theinterference between two tasks results from a process-ing mechanism that is limited to processing only onetask at a time According to many authors such aprocessing mechanism constitutes a bottleneck duringthe concurrent processing of two tasks (see Figure 1Schubert 1999 Meyer amp Kieras 1997 Pashler 1994DeJong 1993) If the component tasks are processedclosely in time so that they compete for the process-ing by the bottleneck mechanism interference ariseswhich has to be resolved by additional executivefunctions In more detail empirical findings fromexperimental psychology suggest that in such situa-tions the processing of one task is interrupted as longas the bottleneck mechanism is processing the othertask (Pashler 1994)1 In this context executive func-tion is conceptualized as coordination of interfering

processing at the stage of the bottleneck by schedul-ing the order in which the tasks are processed byinterrupting one of the two tasks switching to theinterrupted task and reinstating the interrupted taskwhen bottleneck processing in the other task hasfinished (Meyer amp Kieras 1997 DeJong 1995) Becausethe executive functions are required solely in the dual-task but not in the single-task situation they shouldlead to increased neural activity which should bemeasurable by means of functional magnetic resonanceimaging (fMRI)

In the present study we used the dual-task paradigmof the psychological refractory period (PRP) because itpermits assessing the occurrence of a bottleneck mech-anism and accordingly of executive functions In thisparadigm two stimuli are presented in rapid successionseparated by a variable stimulus onset asynchrony(SOA) and the participants have to respond to eachstimulus in the order of their presentation The pres-ence of a bottleneck is indicated by the so-called PRPeffect which is reflected by an increase in RT on thesecond task with decreasing SOA and by a constant RTon the first task (eg Pashler 1994) When the SOA isshort both tasks temporally overlap considerably andbottleneck processing of the first task leads to aninterruption of second task processing which resultsin prolonged RTs for the second task In contrast whenthe SOA is long the temporal overlap of the componenttasks is small and accordingly the second task isinterrupted for a shorter period which results in shorterRTs for the second task In brief this pattern of PRP RTsindicates the presence of concurrent and interferingprocesses in both tasks

In the present experiment participants had to per-form a three-choice reaction task on the pitch of anauditory stimulus with their left hand and a three-choicereaction task on the position of a visual stimulus withtheir right hand respectively (Figure 6) The order ofthe component tasks in one of two dual-task conditionswas fixed throughout a block We call this the DT-fixedcondition (In addition to the DT-fixed condition asecond dual-task condition was also administered tothe participants which will be explained below) Besidesthe dual-task condition blocks participants performedthe component tasks separately (single tasks auditory(AUD) and visual (VIS)) and a resting baseline condition(BASE) All conditions were presented blockwise Basedon the cognitive subtraction method (Friston et al1995) dual-task-related activation was assessed by test-ing whether both tasks interact with each other whenperformed concurrently2 In other words we tested foroveradditive that is surplus activation in the dual-taskcondition as compared to the summed activationevoked by the single-task conditions (Koechlin et al1999 Friston et al 1995 Friston Price et al 1996) Thiscomparison takes into account that areas identified asdual-task-related may already be involved in single-task

task 1 t task 1

task 2task 2

Figure 1 Illustration of the dual-task situation according to thePRP paradigm Executive functions are involved because order ofinformation processing in both tasks has to be coordinated at thebottleneck (shaded gray)

Szameitat et al 1185

processing but are more strongly involved in dual-taskprocessing

Parametric Manipulation of Dual-Task-RelatedExecutive Functions

However the cognitive subtraction method is based ona problematic assumption that is the assumption ofpure insertion (Donders 18681969) This assumptionholds that the insertion of an additional process into atask does not change the remaining processes an issuethat has been questioned by a number of researchers(Sartori amp Umilta 2000 Sidtis et al 1999 Braver et al1997 Friston Price et al 1996 Sternberg 1969) Forexample the insertion of an additional task in order tocreate a dual-task situation may not only insert addi-tional executive functions but may also change thegeneral perceptual memory and motor processes ofthe first task as well Consequently dual-task-relatedactivation revealed by DT-fixed cannot unequivocallybe interpreted because it might be caused either byaltered component-task processing or by the involve-ment of additional executive functions in dual-taskperformance One way to circumvent such interpreta-tional problems is to use additional information yieldedby an alternative methodological approach the para-metric manipulation method (Braver et al 1997 Cohenet al 1997) According to this method one experimen-tal factor that affects the operation of a single process isvaried and it is determined whether this manipulationresults in systematic activation changes of certain cor-tical areas If this is the case then these cortical areascan be attributed to the manipulated process

To implement the parametric manipulation method inthe present experiment a second dual-task conditionwas introduced which incorporated generally higherdemands on the dual-task-related executive functionsthan DT-fixed In this dual-task condition the presenta-tion order of the component tasks changed randomlyacross trials (DT-random condition) DeJong (1995) hasshown that coordination demands differ between theconditions because the same scheduling strategy can beimplemented throughout a block in DT-fixed whereasin DT-random the task processes have to be rearrangedwhenever the presentation order changes This resultsin higher demands on the coordination of the process-ing streams in DT-random compared to DT-fixed Inother words the requirement for frequent rearrange-ment makes the coordination of task-order schedulingtask interruption switching and reinstatement moredemanding Accordingly we expected dual-task-relatedareas as identified by the cognitive subtraction methodin DT-fixed to be more strongly activated in DT-randomSuch a result of converging evidence by the cognitivesubtraction and the parametric manipulation methodwould greatly increase the credibility and validity of ourfinal interpretation

RESULTS

Behavioral Data

If not otherwise noted in the following analyses eitheran analysis of variance (ANOVA) with repeated meas-ures or two-sample paired t tests were used Signifi-cance for t tests was calculated two-tailed All erroranalyses were performed with arcsin-transformed rela-tive error frequencies

Analyses of RTs revealed a typical PRP effect in theDT-fixed condition (Figure 2) That is the RTs on the sec-ond task (RT2) increased with decreasing SOA [F(220) =2949 p lt 001] while the RTs on the first task (RT1)remained constant over the range of SOAs [F(220) =036 p gt 05] resulting in an interaction of the factorsResponse (first or second) and SOA [F(220) = 29430p lt 001] According to prior research (eg Pashler1994) the PRP effect points to concurrent drawing ofboth tasks on a bottleneck representing a limited capa-city mechanism in dual-task processing (cf Figure 1)

Further comparisons showed that RTs (Figure 3 sym-bols) increased significantly from VIS to AUD [t(10) =1083 p lt 001] from AUD to DT-fixed [t(10) = 270p lt 05] and from DT-fixed to DT-random [t(10) = 809p lt 001] For these comparisons only RT1 averagedacross the auditory and visual task was used for thedual-task conditions RT2s were further prolonged ascompared to the RT1s in DT-fixed [t(10) = 1007p lt 001] as well as in DT-random [t(10) = 977p lt 001] These results indicated reliable dual-task costsin both dual-task conditions compared to the single-taskconditions An error analysis indicated that dual-task-related RT increases were not due to a speedndashaccuracy

600

700

800

900

1000

reac

tion

time

(mse

c)

50 125 200SOA (msec)

task 1task 2

Figure 2 Behavioral data depicting the PRP effect RTs plotted againstthe SOAs for DT-fixed RTs are averaged across auditory and visual taskThe RTs for the second task significantly increase with decreasing SOAwhile the RTs of the first task are independent of the SOA Error barsdenote 95 confidence intervals (Loftus amp Masson 1994)

1186 Journal of Cognitive Neuroscience Volume 14 Number 8

trade-off Participants made more errors (Figure 3 bars)in DT-random than in any other condition [smallestt(10) = 433 all prsquos lt 01] while they performed equiv-alently in all other conditions [highest t(10) = 202 allprsquos gt 05] It has to be noted that dual-task trials in whichthe correct keys were pressed according to the stimulibut in the wrong order (reversals) were also scored as anerror and consequently they were included in the erroranalysis above However only 759 of the trials in theDT-random condition were reversal errors which indi-cates that the participants were able to process the tasksin the correct order in this condition

fMRI Data

Single Tasks

Activation in the single tasks was determined by com-paring the functional data of these conditions with theresting baseline (AUDndashBASE and VISndashBASE) In thefollowing we report prefrontal activations only Acomplete list of significantly activated areas is given inTable 1 Both conditions led to rather small and circum-scribed activation foci in the lateral prefrontal cortex(Figure 4c) In more detail the analysis of the AUDcondition revealed bilateral activation in the anteriorndashsuperior portion of the middle frontal gyrus (MFGBrodmannrsquos area [BA] 8 9) while the analysis of theVIS condition revealed activation in anterior parts of theleft MFG (BA 9) and right superior frontal gyrus (SFGBA 8)

Dual-Task Fixed Order

The first goal of the present study was to examine dual-task-related activation by using the method of cognitive

subtraction For this purpose we conducted an inter-action analysis composed of the following contrast((DT-fixed ndashAUD) ndash (VIS ndash BASE))2 This comparison(Figure 4b upper row Table 2) revealed a large lefthemispheric prefrontal activation in cortical areas liningthe inferior frontal sulcus (IFS BA 10 45 46) the MFG(BA 8 9 46) and the superior frontal sulcus (SFS BA 6)The activation extended in the anteriorndashposterior axisfrom frontopolar regions to the precentral sulcus (PCS)and in the superiorndash inferior axis from the SFS to theIFS This activation consisted of a chain of six inter-connected (local) peaks along the anteriorndashposterioraxis with the highest peak located in the MFG (seeTable 2 local peaks in italics) In the right hemispherecortical areas were activated in the mid portion of theMFG (BA 46) and in a region along the SFS (BA 6)These results show that performance of the DT-fixedcondition led to extended bilateral prefrontal activationwhich cannot be reduced to the summed effects ofsingle-task performance Further activation related tothe dual-task processing was found along the intra-parietal sulcus (IPS BA 7) bilaterally the left precuneus(BA 7m) the left middle temporal gyrus (BA 37) and theleft cerebellum

In the next step of the analysis we tested whether theprefrontal activation in the DT-fixed condition was dif-ferentially located compared to the single-task condi-tions For this purpose we compared the locations ofthe peak activation (ie Talairach coordinates) in theleft and right MFG in DT-fixed with the correspondinglocations of the peak activation in the single tasks (leftand right MFG in AUD and left MFG and right SFG inVIS3 respectively) With respect to the statistical para-metric maps (SPMs) of the group analysis the activationpeaks in the single-task conditions and dual-task con-ditions were separated by 17ndash30 mm (average 24 mmEuclidian distances) This indication of spatially differentactivation peaks was statistically confirmed by calculatingthe Euclidian distances for each subject individually Thisanalysis revealed Euclidian distances between 22 and32 mm (average 27 mm) which differed significantlyfrom zero (all prsquos lt 001 one-sample t tests) Furtheranalysis showed that the activation peaks in the single-task conditions were located more anteriorly ( y-axisaccording to Talairach coordinates average 19 mm)and medially (x-axis average 12 mm) than the activa-tion peaks in the DT-fixed condition (all prsquos lt 01two-sample t tests) This indicates that single- and dual-task performances recruited different regions of thelateral prefrontal cortex

Dual-Task Random Order

The second aim of the present study was to testwhether the results revealed by the cognitive subtrac-tion method could be validated by using a parametricmanipulation method For this purpose we introduced

Figure 3 Behavioral data as a function of the different taskconditions Left axis and symbols denote RTs in msec right axis andbars denote error rates including reversal errors For the dual-taskconditions RTs were averaged across the auditory and the visual tasksError bars denote 95 confidence intervals (Loftus amp Masson 1994)

Szameitat et al 1187

Table 1 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activation in theSingle-Task Conditions Compared with the Resting Baseline (AUDndashBASE and VISndashBASE)

AUD VIS

Anatomical Area H BA (x y z) z BA (x y z) z

Frontal

Middle frontal G R 89 28 42 32 730

Middle frontal G L 89 iexcl32 40 32 740 9 iexcl26 32 29 518

Sup frontal G R 8 13 41 42 425

Medial frontal G RL 6 1 5 56 1061 6 iexcl7 iexcl10 56 96

Precentral S R 46 37 iexcl4 52 74

Precentral S R 96 46 4 41 765 69 46 4 41 768

Precentral S L 6 iexcl47 2 32 782 6 iexcl47 2 32 573

Precentral G L 46 iexcl31 iexcl10 53 1089 46 iexcl31 iexcl10 56 1151

Central S R 4 34 iexcl21 59 1099

Central S L 4 iexcl35 iexcl25 56 1261

Lateral Sinsula L 45 iexcl26 23 13 560 4042 iexcl38 iexcl3 15 56

Lateral Sinsula R 44 43 12 6 763 44 43 12 6 521

Parietal

Inf parietal lobe (IPS) R 40 47 iexcl41 39 900 2240 50 iexcl44 27 692

Inf parietal lobe (IPS) L 40 iexcl47 iexcl42 36 974 41 iexcl53 iexcl26 23 83

Sup parietal lobe R 7 11 iexcl64 59 610 5 34 iexcl46 58 564

Sup parietal lobe (precuneus) L 7 iexcl13 iexcl66 53 576 7 iexcl16 iexcl70 60 441

Temporal

Sup temporal Glateral S L 41 iexcl44 iexcl26 11 1325

Sup temporal Glateral S R 4142 53 iexcl30 18 1338

Sup temporal S L 3739 iexcl47 iexcl58 18 776

Other

Calcarine S L 17 iexcl16 iexcl89 4 632 17 iexcl20 iexcl89 5 796

Occipital G R 18 14 iexcl94 8 760 17 iexcl20 iexcl89 4 1085

Globus pallidus L iexcl13 0 5 501

Globus pallidus R 20 iexcl4 10 474

Thalamus R 10 iexcl16 12 585

Thalamus L iexcl11 iexcl17 9 414 iexcl14 iexcl21 16 607

Cerebellum L iexcl14 iexcl55 iexcl9 946

Cerebellum R 26 iexcl67 iexcl9 604 23 iexcl52 iexcl12 874

Activation is thresholded at z gt 33 ( p lt 0005 uncorrected)

Abbreviations H = Hemisphere (L = left R = right) BA = Brodmannrsquos area G = gyrus S = sulcus Sup = superior Inf = inferior Ant = anteriorIPS = intraparietal sulcus IFS = inferior frontal sulcus MFG = middle frontal gyrus

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592

1188 Journal of Cognitive Neuroscience Volume 14 Number 8

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

conditions (see also Smith et al 2001) Thus the firstaim of the present study was to test for dual-task-relatedactivation by using a behavioral dual-task paradigmwhich ensures the presence of interference and there-fore of executive functions

A second issue in this study concerns the method-ology used to obtain dual-task-related activation Mostprevious studies used the method of cognitive sub-traction to obtain dual-task-related activation wheredual-task-related activation is detected by subtractingactivation in two single tasks from activation in a dualtask However this approach has been criticizedbecause of some problematic methodological presump-tions (Sartori amp Umilta 2000 Sidtis Strother Andersonamp Rottenberg 1999 Braver et al 1997 Friston Priceet al 1996 Sternberg 1969) As will be shown later theuse of the cognitive subtraction method might haveobscured dual-task-related activation so far Thereforein the present study we used the method of parametricmanipulation (Braver et al 1997 Cohen et al 1997) inaddition to the method of cognitive subtraction toassess dual-task-related activation

Interference in Dual-Task Situations

The typical indicators of interference in dual-taskperformance are performance decrements reflectedby prolonged reaction times (RTs) or increased errorrates as compared to single-task performance Theo-ries of dual-task performance have suggested that theinterference between two tasks results from a process-ing mechanism that is limited to processing only onetask at a time According to many authors such aprocessing mechanism constitutes a bottleneck duringthe concurrent processing of two tasks (see Figure 1Schubert 1999 Meyer amp Kieras 1997 Pashler 1994DeJong 1993) If the component tasks are processedclosely in time so that they compete for the process-ing by the bottleneck mechanism interference ariseswhich has to be resolved by additional executivefunctions In more detail empirical findings fromexperimental psychology suggest that in such situa-tions the processing of one task is interrupted as longas the bottleneck mechanism is processing the othertask (Pashler 1994)1 In this context executive func-tion is conceptualized as coordination of interfering

processing at the stage of the bottleneck by schedul-ing the order in which the tasks are processed byinterrupting one of the two tasks switching to theinterrupted task and reinstating the interrupted taskwhen bottleneck processing in the other task hasfinished (Meyer amp Kieras 1997 DeJong 1995) Becausethe executive functions are required solely in the dual-task but not in the single-task situation they shouldlead to increased neural activity which should bemeasurable by means of functional magnetic resonanceimaging (fMRI)

In the present study we used the dual-task paradigmof the psychological refractory period (PRP) because itpermits assessing the occurrence of a bottleneck mech-anism and accordingly of executive functions In thisparadigm two stimuli are presented in rapid successionseparated by a variable stimulus onset asynchrony(SOA) and the participants have to respond to eachstimulus in the order of their presentation The pres-ence of a bottleneck is indicated by the so-called PRPeffect which is reflected by an increase in RT on thesecond task with decreasing SOA and by a constant RTon the first task (eg Pashler 1994) When the SOA isshort both tasks temporally overlap considerably andbottleneck processing of the first task leads to aninterruption of second task processing which resultsin prolonged RTs for the second task In contrast whenthe SOA is long the temporal overlap of the componenttasks is small and accordingly the second task isinterrupted for a shorter period which results in shorterRTs for the second task In brief this pattern of PRP RTsindicates the presence of concurrent and interferingprocesses in both tasks

In the present experiment participants had to per-form a three-choice reaction task on the pitch of anauditory stimulus with their left hand and a three-choicereaction task on the position of a visual stimulus withtheir right hand respectively (Figure 6) The order ofthe component tasks in one of two dual-task conditionswas fixed throughout a block We call this the DT-fixedcondition (In addition to the DT-fixed condition asecond dual-task condition was also administered tothe participants which will be explained below) Besidesthe dual-task condition blocks participants performedthe component tasks separately (single tasks auditory(AUD) and visual (VIS)) and a resting baseline condition(BASE) All conditions were presented blockwise Basedon the cognitive subtraction method (Friston et al1995) dual-task-related activation was assessed by test-ing whether both tasks interact with each other whenperformed concurrently2 In other words we tested foroveradditive that is surplus activation in the dual-taskcondition as compared to the summed activationevoked by the single-task conditions (Koechlin et al1999 Friston et al 1995 Friston Price et al 1996) Thiscomparison takes into account that areas identified asdual-task-related may already be involved in single-task

task 1 t task 1

task 2task 2

Figure 1 Illustration of the dual-task situation according to thePRP paradigm Executive functions are involved because order ofinformation processing in both tasks has to be coordinated at thebottleneck (shaded gray)

Szameitat et al 1185

processing but are more strongly involved in dual-taskprocessing

Parametric Manipulation of Dual-Task-RelatedExecutive Functions

However the cognitive subtraction method is based ona problematic assumption that is the assumption ofpure insertion (Donders 18681969) This assumptionholds that the insertion of an additional process into atask does not change the remaining processes an issuethat has been questioned by a number of researchers(Sartori amp Umilta 2000 Sidtis et al 1999 Braver et al1997 Friston Price et al 1996 Sternberg 1969) Forexample the insertion of an additional task in order tocreate a dual-task situation may not only insert addi-tional executive functions but may also change thegeneral perceptual memory and motor processes ofthe first task as well Consequently dual-task-relatedactivation revealed by DT-fixed cannot unequivocallybe interpreted because it might be caused either byaltered component-task processing or by the involve-ment of additional executive functions in dual-taskperformance One way to circumvent such interpreta-tional problems is to use additional information yieldedby an alternative methodological approach the para-metric manipulation method (Braver et al 1997 Cohenet al 1997) According to this method one experimen-tal factor that affects the operation of a single process isvaried and it is determined whether this manipulationresults in systematic activation changes of certain cor-tical areas If this is the case then these cortical areascan be attributed to the manipulated process

To implement the parametric manipulation method inthe present experiment a second dual-task conditionwas introduced which incorporated generally higherdemands on the dual-task-related executive functionsthan DT-fixed In this dual-task condition the presenta-tion order of the component tasks changed randomlyacross trials (DT-random condition) DeJong (1995) hasshown that coordination demands differ between theconditions because the same scheduling strategy can beimplemented throughout a block in DT-fixed whereasin DT-random the task processes have to be rearrangedwhenever the presentation order changes This resultsin higher demands on the coordination of the process-ing streams in DT-random compared to DT-fixed Inother words the requirement for frequent rearrange-ment makes the coordination of task-order schedulingtask interruption switching and reinstatement moredemanding Accordingly we expected dual-task-relatedareas as identified by the cognitive subtraction methodin DT-fixed to be more strongly activated in DT-randomSuch a result of converging evidence by the cognitivesubtraction and the parametric manipulation methodwould greatly increase the credibility and validity of ourfinal interpretation

RESULTS

Behavioral Data

If not otherwise noted in the following analyses eitheran analysis of variance (ANOVA) with repeated meas-ures or two-sample paired t tests were used Signifi-cance for t tests was calculated two-tailed All erroranalyses were performed with arcsin-transformed rela-tive error frequencies

Analyses of RTs revealed a typical PRP effect in theDT-fixed condition (Figure 2) That is the RTs on the sec-ond task (RT2) increased with decreasing SOA [F(220) =2949 p lt 001] while the RTs on the first task (RT1)remained constant over the range of SOAs [F(220) =036 p gt 05] resulting in an interaction of the factorsResponse (first or second) and SOA [F(220) = 29430p lt 001] According to prior research (eg Pashler1994) the PRP effect points to concurrent drawing ofboth tasks on a bottleneck representing a limited capa-city mechanism in dual-task processing (cf Figure 1)

Further comparisons showed that RTs (Figure 3 sym-bols) increased significantly from VIS to AUD [t(10) =1083 p lt 001] from AUD to DT-fixed [t(10) = 270p lt 05] and from DT-fixed to DT-random [t(10) = 809p lt 001] For these comparisons only RT1 averagedacross the auditory and visual task was used for thedual-task conditions RT2s were further prolonged ascompared to the RT1s in DT-fixed [t(10) = 1007p lt 001] as well as in DT-random [t(10) = 977p lt 001] These results indicated reliable dual-task costsin both dual-task conditions compared to the single-taskconditions An error analysis indicated that dual-task-related RT increases were not due to a speedndashaccuracy

600

700

800

900

1000

reac

tion

time

(mse

c)

50 125 200SOA (msec)

task 1task 2

Figure 2 Behavioral data depicting the PRP effect RTs plotted againstthe SOAs for DT-fixed RTs are averaged across auditory and visual taskThe RTs for the second task significantly increase with decreasing SOAwhile the RTs of the first task are independent of the SOA Error barsdenote 95 confidence intervals (Loftus amp Masson 1994)

1186 Journal of Cognitive Neuroscience Volume 14 Number 8

trade-off Participants made more errors (Figure 3 bars)in DT-random than in any other condition [smallestt(10) = 433 all prsquos lt 01] while they performed equiv-alently in all other conditions [highest t(10) = 202 allprsquos gt 05] It has to be noted that dual-task trials in whichthe correct keys were pressed according to the stimulibut in the wrong order (reversals) were also scored as anerror and consequently they were included in the erroranalysis above However only 759 of the trials in theDT-random condition were reversal errors which indi-cates that the participants were able to process the tasksin the correct order in this condition

fMRI Data

Single Tasks

Activation in the single tasks was determined by com-paring the functional data of these conditions with theresting baseline (AUDndashBASE and VISndashBASE) In thefollowing we report prefrontal activations only Acomplete list of significantly activated areas is given inTable 1 Both conditions led to rather small and circum-scribed activation foci in the lateral prefrontal cortex(Figure 4c) In more detail the analysis of the AUDcondition revealed bilateral activation in the anteriorndashsuperior portion of the middle frontal gyrus (MFGBrodmannrsquos area [BA] 8 9) while the analysis of theVIS condition revealed activation in anterior parts of theleft MFG (BA 9) and right superior frontal gyrus (SFGBA 8)

Dual-Task Fixed Order

The first goal of the present study was to examine dual-task-related activation by using the method of cognitive

subtraction For this purpose we conducted an inter-action analysis composed of the following contrast((DT-fixed ndashAUD) ndash (VIS ndash BASE))2 This comparison(Figure 4b upper row Table 2) revealed a large lefthemispheric prefrontal activation in cortical areas liningthe inferior frontal sulcus (IFS BA 10 45 46) the MFG(BA 8 9 46) and the superior frontal sulcus (SFS BA 6)The activation extended in the anteriorndashposterior axisfrom frontopolar regions to the precentral sulcus (PCS)and in the superiorndash inferior axis from the SFS to theIFS This activation consisted of a chain of six inter-connected (local) peaks along the anteriorndashposterioraxis with the highest peak located in the MFG (seeTable 2 local peaks in italics) In the right hemispherecortical areas were activated in the mid portion of theMFG (BA 46) and in a region along the SFS (BA 6)These results show that performance of the DT-fixedcondition led to extended bilateral prefrontal activationwhich cannot be reduced to the summed effects ofsingle-task performance Further activation related tothe dual-task processing was found along the intra-parietal sulcus (IPS BA 7) bilaterally the left precuneus(BA 7m) the left middle temporal gyrus (BA 37) and theleft cerebellum

In the next step of the analysis we tested whether theprefrontal activation in the DT-fixed condition was dif-ferentially located compared to the single-task condi-tions For this purpose we compared the locations ofthe peak activation (ie Talairach coordinates) in theleft and right MFG in DT-fixed with the correspondinglocations of the peak activation in the single tasks (leftand right MFG in AUD and left MFG and right SFG inVIS3 respectively) With respect to the statistical para-metric maps (SPMs) of the group analysis the activationpeaks in the single-task conditions and dual-task con-ditions were separated by 17ndash30 mm (average 24 mmEuclidian distances) This indication of spatially differentactivation peaks was statistically confirmed by calculatingthe Euclidian distances for each subject individually Thisanalysis revealed Euclidian distances between 22 and32 mm (average 27 mm) which differed significantlyfrom zero (all prsquos lt 001 one-sample t tests) Furtheranalysis showed that the activation peaks in the single-task conditions were located more anteriorly ( y-axisaccording to Talairach coordinates average 19 mm)and medially (x-axis average 12 mm) than the activa-tion peaks in the DT-fixed condition (all prsquos lt 01two-sample t tests) This indicates that single- and dual-task performances recruited different regions of thelateral prefrontal cortex

Dual-Task Random Order

The second aim of the present study was to testwhether the results revealed by the cognitive subtrac-tion method could be validated by using a parametricmanipulation method For this purpose we introduced

Figure 3 Behavioral data as a function of the different taskconditions Left axis and symbols denote RTs in msec right axis andbars denote error rates including reversal errors For the dual-taskconditions RTs were averaged across the auditory and the visual tasksError bars denote 95 confidence intervals (Loftus amp Masson 1994)

Szameitat et al 1187

Table 1 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activation in theSingle-Task Conditions Compared with the Resting Baseline (AUDndashBASE and VISndashBASE)

AUD VIS

Anatomical Area H BA (x y z) z BA (x y z) z

Frontal

Middle frontal G R 89 28 42 32 730

Middle frontal G L 89 iexcl32 40 32 740 9 iexcl26 32 29 518

Sup frontal G R 8 13 41 42 425

Medial frontal G RL 6 1 5 56 1061 6 iexcl7 iexcl10 56 96

Precentral S R 46 37 iexcl4 52 74

Precentral S R 96 46 4 41 765 69 46 4 41 768

Precentral S L 6 iexcl47 2 32 782 6 iexcl47 2 32 573

Precentral G L 46 iexcl31 iexcl10 53 1089 46 iexcl31 iexcl10 56 1151

Central S R 4 34 iexcl21 59 1099

Central S L 4 iexcl35 iexcl25 56 1261

Lateral Sinsula L 45 iexcl26 23 13 560 4042 iexcl38 iexcl3 15 56

Lateral Sinsula R 44 43 12 6 763 44 43 12 6 521

Parietal

Inf parietal lobe (IPS) R 40 47 iexcl41 39 900 2240 50 iexcl44 27 692

Inf parietal lobe (IPS) L 40 iexcl47 iexcl42 36 974 41 iexcl53 iexcl26 23 83

Sup parietal lobe R 7 11 iexcl64 59 610 5 34 iexcl46 58 564

Sup parietal lobe (precuneus) L 7 iexcl13 iexcl66 53 576 7 iexcl16 iexcl70 60 441

Temporal

Sup temporal Glateral S L 41 iexcl44 iexcl26 11 1325

Sup temporal Glateral S R 4142 53 iexcl30 18 1338

Sup temporal S L 3739 iexcl47 iexcl58 18 776

Other

Calcarine S L 17 iexcl16 iexcl89 4 632 17 iexcl20 iexcl89 5 796

Occipital G R 18 14 iexcl94 8 760 17 iexcl20 iexcl89 4 1085

Globus pallidus L iexcl13 0 5 501

Globus pallidus R 20 iexcl4 10 474

Thalamus R 10 iexcl16 12 585

Thalamus L iexcl11 iexcl17 9 414 iexcl14 iexcl21 16 607

Cerebellum L iexcl14 iexcl55 iexcl9 946

Cerebellum R 26 iexcl67 iexcl9 604 23 iexcl52 iexcl12 874

Activation is thresholded at z gt 33 ( p lt 0005 uncorrected)

Abbreviations H = Hemisphere (L = left R = right) BA = Brodmannrsquos area G = gyrus S = sulcus Sup = superior Inf = inferior Ant = anteriorIPS = intraparietal sulcus IFS = inferior frontal sulcus MFG = middle frontal gyrus

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592

1188 Journal of Cognitive Neuroscience Volume 14 Number 8

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

processing but are more strongly involved in dual-taskprocessing

Parametric Manipulation of Dual-Task-RelatedExecutive Functions

However the cognitive subtraction method is based ona problematic assumption that is the assumption ofpure insertion (Donders 18681969) This assumptionholds that the insertion of an additional process into atask does not change the remaining processes an issuethat has been questioned by a number of researchers(Sartori amp Umilta 2000 Sidtis et al 1999 Braver et al1997 Friston Price et al 1996 Sternberg 1969) Forexample the insertion of an additional task in order tocreate a dual-task situation may not only insert addi-tional executive functions but may also change thegeneral perceptual memory and motor processes ofthe first task as well Consequently dual-task-relatedactivation revealed by DT-fixed cannot unequivocallybe interpreted because it might be caused either byaltered component-task processing or by the involve-ment of additional executive functions in dual-taskperformance One way to circumvent such interpreta-tional problems is to use additional information yieldedby an alternative methodological approach the para-metric manipulation method (Braver et al 1997 Cohenet al 1997) According to this method one experimen-tal factor that affects the operation of a single process isvaried and it is determined whether this manipulationresults in systematic activation changes of certain cor-tical areas If this is the case then these cortical areascan be attributed to the manipulated process

To implement the parametric manipulation method inthe present experiment a second dual-task conditionwas introduced which incorporated generally higherdemands on the dual-task-related executive functionsthan DT-fixed In this dual-task condition the presenta-tion order of the component tasks changed randomlyacross trials (DT-random condition) DeJong (1995) hasshown that coordination demands differ between theconditions because the same scheduling strategy can beimplemented throughout a block in DT-fixed whereasin DT-random the task processes have to be rearrangedwhenever the presentation order changes This resultsin higher demands on the coordination of the process-ing streams in DT-random compared to DT-fixed Inother words the requirement for frequent rearrange-ment makes the coordination of task-order schedulingtask interruption switching and reinstatement moredemanding Accordingly we expected dual-task-relatedareas as identified by the cognitive subtraction methodin DT-fixed to be more strongly activated in DT-randomSuch a result of converging evidence by the cognitivesubtraction and the parametric manipulation methodwould greatly increase the credibility and validity of ourfinal interpretation

RESULTS

Behavioral Data

If not otherwise noted in the following analyses eitheran analysis of variance (ANOVA) with repeated meas-ures or two-sample paired t tests were used Signifi-cance for t tests was calculated two-tailed All erroranalyses were performed with arcsin-transformed rela-tive error frequencies

Analyses of RTs revealed a typical PRP effect in theDT-fixed condition (Figure 2) That is the RTs on the sec-ond task (RT2) increased with decreasing SOA [F(220) =2949 p lt 001] while the RTs on the first task (RT1)remained constant over the range of SOAs [F(220) =036 p gt 05] resulting in an interaction of the factorsResponse (first or second) and SOA [F(220) = 29430p lt 001] According to prior research (eg Pashler1994) the PRP effect points to concurrent drawing ofboth tasks on a bottleneck representing a limited capa-city mechanism in dual-task processing (cf Figure 1)

Further comparisons showed that RTs (Figure 3 sym-bols) increased significantly from VIS to AUD [t(10) =1083 p lt 001] from AUD to DT-fixed [t(10) = 270p lt 05] and from DT-fixed to DT-random [t(10) = 809p lt 001] For these comparisons only RT1 averagedacross the auditory and visual task was used for thedual-task conditions RT2s were further prolonged ascompared to the RT1s in DT-fixed [t(10) = 1007p lt 001] as well as in DT-random [t(10) = 977p lt 001] These results indicated reliable dual-task costsin both dual-task conditions compared to the single-taskconditions An error analysis indicated that dual-task-related RT increases were not due to a speedndashaccuracy

600

700

800

900

1000

reac

tion

time

(mse

c)

50 125 200SOA (msec)

task 1task 2

Figure 2 Behavioral data depicting the PRP effect RTs plotted againstthe SOAs for DT-fixed RTs are averaged across auditory and visual taskThe RTs for the second task significantly increase with decreasing SOAwhile the RTs of the first task are independent of the SOA Error barsdenote 95 confidence intervals (Loftus amp Masson 1994)

1186 Journal of Cognitive Neuroscience Volume 14 Number 8

trade-off Participants made more errors (Figure 3 bars)in DT-random than in any other condition [smallestt(10) = 433 all prsquos lt 01] while they performed equiv-alently in all other conditions [highest t(10) = 202 allprsquos gt 05] It has to be noted that dual-task trials in whichthe correct keys were pressed according to the stimulibut in the wrong order (reversals) were also scored as anerror and consequently they were included in the erroranalysis above However only 759 of the trials in theDT-random condition were reversal errors which indi-cates that the participants were able to process the tasksin the correct order in this condition

fMRI Data

Single Tasks

Activation in the single tasks was determined by com-paring the functional data of these conditions with theresting baseline (AUDndashBASE and VISndashBASE) In thefollowing we report prefrontal activations only Acomplete list of significantly activated areas is given inTable 1 Both conditions led to rather small and circum-scribed activation foci in the lateral prefrontal cortex(Figure 4c) In more detail the analysis of the AUDcondition revealed bilateral activation in the anteriorndashsuperior portion of the middle frontal gyrus (MFGBrodmannrsquos area [BA] 8 9) while the analysis of theVIS condition revealed activation in anterior parts of theleft MFG (BA 9) and right superior frontal gyrus (SFGBA 8)

Dual-Task Fixed Order

The first goal of the present study was to examine dual-task-related activation by using the method of cognitive

subtraction For this purpose we conducted an inter-action analysis composed of the following contrast((DT-fixed ndashAUD) ndash (VIS ndash BASE))2 This comparison(Figure 4b upper row Table 2) revealed a large lefthemispheric prefrontal activation in cortical areas liningthe inferior frontal sulcus (IFS BA 10 45 46) the MFG(BA 8 9 46) and the superior frontal sulcus (SFS BA 6)The activation extended in the anteriorndashposterior axisfrom frontopolar regions to the precentral sulcus (PCS)and in the superiorndash inferior axis from the SFS to theIFS This activation consisted of a chain of six inter-connected (local) peaks along the anteriorndashposterioraxis with the highest peak located in the MFG (seeTable 2 local peaks in italics) In the right hemispherecortical areas were activated in the mid portion of theMFG (BA 46) and in a region along the SFS (BA 6)These results show that performance of the DT-fixedcondition led to extended bilateral prefrontal activationwhich cannot be reduced to the summed effects ofsingle-task performance Further activation related tothe dual-task processing was found along the intra-parietal sulcus (IPS BA 7) bilaterally the left precuneus(BA 7m) the left middle temporal gyrus (BA 37) and theleft cerebellum

In the next step of the analysis we tested whether theprefrontal activation in the DT-fixed condition was dif-ferentially located compared to the single-task condi-tions For this purpose we compared the locations ofthe peak activation (ie Talairach coordinates) in theleft and right MFG in DT-fixed with the correspondinglocations of the peak activation in the single tasks (leftand right MFG in AUD and left MFG and right SFG inVIS3 respectively) With respect to the statistical para-metric maps (SPMs) of the group analysis the activationpeaks in the single-task conditions and dual-task con-ditions were separated by 17ndash30 mm (average 24 mmEuclidian distances) This indication of spatially differentactivation peaks was statistically confirmed by calculatingthe Euclidian distances for each subject individually Thisanalysis revealed Euclidian distances between 22 and32 mm (average 27 mm) which differed significantlyfrom zero (all prsquos lt 001 one-sample t tests) Furtheranalysis showed that the activation peaks in the single-task conditions were located more anteriorly ( y-axisaccording to Talairach coordinates average 19 mm)and medially (x-axis average 12 mm) than the activa-tion peaks in the DT-fixed condition (all prsquos lt 01two-sample t tests) This indicates that single- and dual-task performances recruited different regions of thelateral prefrontal cortex

Dual-Task Random Order

The second aim of the present study was to testwhether the results revealed by the cognitive subtrac-tion method could be validated by using a parametricmanipulation method For this purpose we introduced

Figure 3 Behavioral data as a function of the different taskconditions Left axis and symbols denote RTs in msec right axis andbars denote error rates including reversal errors For the dual-taskconditions RTs were averaged across the auditory and the visual tasksError bars denote 95 confidence intervals (Loftus amp Masson 1994)

Szameitat et al 1187

Table 1 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activation in theSingle-Task Conditions Compared with the Resting Baseline (AUDndashBASE and VISndashBASE)

AUD VIS

Anatomical Area H BA (x y z) z BA (x y z) z

Frontal

Middle frontal G R 89 28 42 32 730

Middle frontal G L 89 iexcl32 40 32 740 9 iexcl26 32 29 518

Sup frontal G R 8 13 41 42 425

Medial frontal G RL 6 1 5 56 1061 6 iexcl7 iexcl10 56 96

Precentral S R 46 37 iexcl4 52 74

Precentral S R 96 46 4 41 765 69 46 4 41 768

Precentral S L 6 iexcl47 2 32 782 6 iexcl47 2 32 573

Precentral G L 46 iexcl31 iexcl10 53 1089 46 iexcl31 iexcl10 56 1151

Central S R 4 34 iexcl21 59 1099

Central S L 4 iexcl35 iexcl25 56 1261

Lateral Sinsula L 45 iexcl26 23 13 560 4042 iexcl38 iexcl3 15 56

Lateral Sinsula R 44 43 12 6 763 44 43 12 6 521

Parietal

Inf parietal lobe (IPS) R 40 47 iexcl41 39 900 2240 50 iexcl44 27 692

Inf parietal lobe (IPS) L 40 iexcl47 iexcl42 36 974 41 iexcl53 iexcl26 23 83

Sup parietal lobe R 7 11 iexcl64 59 610 5 34 iexcl46 58 564

Sup parietal lobe (precuneus) L 7 iexcl13 iexcl66 53 576 7 iexcl16 iexcl70 60 441

Temporal

Sup temporal Glateral S L 41 iexcl44 iexcl26 11 1325

Sup temporal Glateral S R 4142 53 iexcl30 18 1338

Sup temporal S L 3739 iexcl47 iexcl58 18 776

Other

Calcarine S L 17 iexcl16 iexcl89 4 632 17 iexcl20 iexcl89 5 796

Occipital G R 18 14 iexcl94 8 760 17 iexcl20 iexcl89 4 1085

Globus pallidus L iexcl13 0 5 501

Globus pallidus R 20 iexcl4 10 474

Thalamus R 10 iexcl16 12 585

Thalamus L iexcl11 iexcl17 9 414 iexcl14 iexcl21 16 607

Cerebellum L iexcl14 iexcl55 iexcl9 946

Cerebellum R 26 iexcl67 iexcl9 604 23 iexcl52 iexcl12 874

Activation is thresholded at z gt 33 ( p lt 0005 uncorrected)

Abbreviations H = Hemisphere (L = left R = right) BA = Brodmannrsquos area G = gyrus S = sulcus Sup = superior Inf = inferior Ant = anteriorIPS = intraparietal sulcus IFS = inferior frontal sulcus MFG = middle frontal gyrus

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592

1188 Journal of Cognitive Neuroscience Volume 14 Number 8

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

trade-off Participants made more errors (Figure 3 bars)in DT-random than in any other condition [smallestt(10) = 433 all prsquos lt 01] while they performed equiv-alently in all other conditions [highest t(10) = 202 allprsquos gt 05] It has to be noted that dual-task trials in whichthe correct keys were pressed according to the stimulibut in the wrong order (reversals) were also scored as anerror and consequently they were included in the erroranalysis above However only 759 of the trials in theDT-random condition were reversal errors which indi-cates that the participants were able to process the tasksin the correct order in this condition

fMRI Data

Single Tasks

Activation in the single tasks was determined by com-paring the functional data of these conditions with theresting baseline (AUDndashBASE and VISndashBASE) In thefollowing we report prefrontal activations only Acomplete list of significantly activated areas is given inTable 1 Both conditions led to rather small and circum-scribed activation foci in the lateral prefrontal cortex(Figure 4c) In more detail the analysis of the AUDcondition revealed bilateral activation in the anteriorndashsuperior portion of the middle frontal gyrus (MFGBrodmannrsquos area [BA] 8 9) while the analysis of theVIS condition revealed activation in anterior parts of theleft MFG (BA 9) and right superior frontal gyrus (SFGBA 8)

Dual-Task Fixed Order

The first goal of the present study was to examine dual-task-related activation by using the method of cognitive

subtraction For this purpose we conducted an inter-action analysis composed of the following contrast((DT-fixed ndashAUD) ndash (VIS ndash BASE))2 This comparison(Figure 4b upper row Table 2) revealed a large lefthemispheric prefrontal activation in cortical areas liningthe inferior frontal sulcus (IFS BA 10 45 46) the MFG(BA 8 9 46) and the superior frontal sulcus (SFS BA 6)The activation extended in the anteriorndashposterior axisfrom frontopolar regions to the precentral sulcus (PCS)and in the superiorndash inferior axis from the SFS to theIFS This activation consisted of a chain of six inter-connected (local) peaks along the anteriorndashposterioraxis with the highest peak located in the MFG (seeTable 2 local peaks in italics) In the right hemispherecortical areas were activated in the mid portion of theMFG (BA 46) and in a region along the SFS (BA 6)These results show that performance of the DT-fixedcondition led to extended bilateral prefrontal activationwhich cannot be reduced to the summed effects ofsingle-task performance Further activation related tothe dual-task processing was found along the intra-parietal sulcus (IPS BA 7) bilaterally the left precuneus(BA 7m) the left middle temporal gyrus (BA 37) and theleft cerebellum

In the next step of the analysis we tested whether theprefrontal activation in the DT-fixed condition was dif-ferentially located compared to the single-task condi-tions For this purpose we compared the locations ofthe peak activation (ie Talairach coordinates) in theleft and right MFG in DT-fixed with the correspondinglocations of the peak activation in the single tasks (leftand right MFG in AUD and left MFG and right SFG inVIS3 respectively) With respect to the statistical para-metric maps (SPMs) of the group analysis the activationpeaks in the single-task conditions and dual-task con-ditions were separated by 17ndash30 mm (average 24 mmEuclidian distances) This indication of spatially differentactivation peaks was statistically confirmed by calculatingthe Euclidian distances for each subject individually Thisanalysis revealed Euclidian distances between 22 and32 mm (average 27 mm) which differed significantlyfrom zero (all prsquos lt 001 one-sample t tests) Furtheranalysis showed that the activation peaks in the single-task conditions were located more anteriorly ( y-axisaccording to Talairach coordinates average 19 mm)and medially (x-axis average 12 mm) than the activa-tion peaks in the DT-fixed condition (all prsquos lt 01two-sample t tests) This indicates that single- and dual-task performances recruited different regions of thelateral prefrontal cortex

Dual-Task Random Order

The second aim of the present study was to testwhether the results revealed by the cognitive subtrac-tion method could be validated by using a parametricmanipulation method For this purpose we introduced

Figure 3 Behavioral data as a function of the different taskconditions Left axis and symbols denote RTs in msec right axis andbars denote error rates including reversal errors For the dual-taskconditions RTs were averaged across the auditory and the visual tasksError bars denote 95 confidence intervals (Loftus amp Masson 1994)

Szameitat et al 1187

Table 1 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activation in theSingle-Task Conditions Compared with the Resting Baseline (AUDndashBASE and VISndashBASE)

AUD VIS

Anatomical Area H BA (x y z) z BA (x y z) z

Frontal

Middle frontal G R 89 28 42 32 730

Middle frontal G L 89 iexcl32 40 32 740 9 iexcl26 32 29 518

Sup frontal G R 8 13 41 42 425

Medial frontal G RL 6 1 5 56 1061 6 iexcl7 iexcl10 56 96

Precentral S R 46 37 iexcl4 52 74

Precentral S R 96 46 4 41 765 69 46 4 41 768

Precentral S L 6 iexcl47 2 32 782 6 iexcl47 2 32 573

Precentral G L 46 iexcl31 iexcl10 53 1089 46 iexcl31 iexcl10 56 1151

Central S R 4 34 iexcl21 59 1099

Central S L 4 iexcl35 iexcl25 56 1261

Lateral Sinsula L 45 iexcl26 23 13 560 4042 iexcl38 iexcl3 15 56

Lateral Sinsula R 44 43 12 6 763 44 43 12 6 521

Parietal

Inf parietal lobe (IPS) R 40 47 iexcl41 39 900 2240 50 iexcl44 27 692

Inf parietal lobe (IPS) L 40 iexcl47 iexcl42 36 974 41 iexcl53 iexcl26 23 83

Sup parietal lobe R 7 11 iexcl64 59 610 5 34 iexcl46 58 564

Sup parietal lobe (precuneus) L 7 iexcl13 iexcl66 53 576 7 iexcl16 iexcl70 60 441

Temporal

Sup temporal Glateral S L 41 iexcl44 iexcl26 11 1325

Sup temporal Glateral S R 4142 53 iexcl30 18 1338

Sup temporal S L 3739 iexcl47 iexcl58 18 776

Other

Calcarine S L 17 iexcl16 iexcl89 4 632 17 iexcl20 iexcl89 5 796

Occipital G R 18 14 iexcl94 8 760 17 iexcl20 iexcl89 4 1085

Globus pallidus L iexcl13 0 5 501

Globus pallidus R 20 iexcl4 10 474

Thalamus R 10 iexcl16 12 585

Thalamus L iexcl11 iexcl17 9 414 iexcl14 iexcl21 16 607

Cerebellum L iexcl14 iexcl55 iexcl9 946

Cerebellum R 26 iexcl67 iexcl9 604 23 iexcl52 iexcl12 874

Activation is thresholded at z gt 33 ( p lt 0005 uncorrected)

Abbreviations H = Hemisphere (L = left R = right) BA = Brodmannrsquos area G = gyrus S = sulcus Sup = superior Inf = inferior Ant = anteriorIPS = intraparietal sulcus IFS = inferior frontal sulcus MFG = middle frontal gyrus

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592

1188 Journal of Cognitive Neuroscience Volume 14 Number 8

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

Table 1 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activation in theSingle-Task Conditions Compared with the Resting Baseline (AUDndashBASE and VISndashBASE)

AUD VIS

Anatomical Area H BA (x y z) z BA (x y z) z

Frontal

Middle frontal G R 89 28 42 32 730

Middle frontal G L 89 iexcl32 40 32 740 9 iexcl26 32 29 518

Sup frontal G R 8 13 41 42 425

Medial frontal G RL 6 1 5 56 1061 6 iexcl7 iexcl10 56 96

Precentral S R 46 37 iexcl4 52 74

Precentral S R 96 46 4 41 765 69 46 4 41 768

Precentral S L 6 iexcl47 2 32 782 6 iexcl47 2 32 573

Precentral G L 46 iexcl31 iexcl10 53 1089 46 iexcl31 iexcl10 56 1151

Central S R 4 34 iexcl21 59 1099

Central S L 4 iexcl35 iexcl25 56 1261

Lateral Sinsula L 45 iexcl26 23 13 560 4042 iexcl38 iexcl3 15 56

Lateral Sinsula R 44 43 12 6 763 44 43 12 6 521

Parietal

Inf parietal lobe (IPS) R 40 47 iexcl41 39 900 2240 50 iexcl44 27 692

Inf parietal lobe (IPS) L 40 iexcl47 iexcl42 36 974 41 iexcl53 iexcl26 23 83

Sup parietal lobe R 7 11 iexcl64 59 610 5 34 iexcl46 58 564

Sup parietal lobe (precuneus) L 7 iexcl13 iexcl66 53 576 7 iexcl16 iexcl70 60 441

Temporal

Sup temporal Glateral S L 41 iexcl44 iexcl26 11 1325

Sup temporal Glateral S R 4142 53 iexcl30 18 1338

Sup temporal S L 3739 iexcl47 iexcl58 18 776

Other

Calcarine S L 17 iexcl16 iexcl89 4 632 17 iexcl20 iexcl89 5 796

Occipital G R 18 14 iexcl94 8 760 17 iexcl20 iexcl89 4 1085

Globus pallidus L iexcl13 0 5 501

Globus pallidus R 20 iexcl4 10 474

Thalamus R 10 iexcl16 12 585

Thalamus L iexcl11 iexcl17 9 414 iexcl14 iexcl21 16 607

Cerebellum L iexcl14 iexcl55 iexcl9 946

Cerebellum R 26 iexcl67 iexcl9 604 23 iexcl52 iexcl12 874

Activation is thresholded at z gt 33 ( p lt 0005 uncorrected)

Abbreviations H = Hemisphere (L = left R = right) BA = Brodmannrsquos area G = gyrus S = sulcus Sup = superior Inf = inferior Ant = anteriorIPS = intraparietal sulcus IFS = inferior frontal sulcus MFG = middle frontal gyrus

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592

1188 Journal of Cognitive Neuroscience Volume 14 Number 8

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

Table 2 Stereotactic Coordinates (Talairach amp Tournoux 1988) and Anatomical Location of (Local) Peak Activations in theDual-Task Conditions

DT-Fixed DT-Random DT-Random ndash DT-Fixed

Anatomical Area H BA (x y z) z (x y z) z (x y z) zDiff(z)

Frontal

Middle frontal G L 89 iexcl44 16 38 666 iexcl41 14 42 1102 iexcl34 12 33 1038 725

IFSMFG L 1046 iexcl35 41 8 411 iexcl37 40 9 836 iexcl35 45 13 818 511

IFS L 4546 iexcl38 33 16 461 iexcl38 33 16 777 iexcl38 40 17 853 548

MFG L 946 iexcl41 32 24 509 iexcl41 32 24 904 iexcl41 29 25 839 688

MFG L 946 iexcl38 24 32 598 iexcl38 24 32 1070 iexcl38 24 32 837 837

Precentral S L 6 iexcl38 2 42 536 iexcl38 2 42 1051 iexcl41 5 39 1000 915

Sup frontal S L 6 iexcl20 iexcl1 54 671 iexcl20 iexcl1 54 1172 iexcl29 iexcl7 53 1100 836

Middle frontal G R 946 41 30 32 599 40 30 32 1004 35 12 33 1231 718

MFG R 10 31 55 18 593 25 53 20 676

Sup frontal S R 6 20 0 57 442 20 2 56 928 included above 785

Sup frontal G L 8 iexcl4 19 50 873 iexcl2 19 50 1161

Ant insula R 4445 29 19 7 978 29 21 4 1415

Ant insula L 4445 iexcl26 19 7 840 iexcl29 16 11 1128

Cingulate SG L 32 iexcl10 22 37 1172

Parietal

Precuneus L 7 iexcl10 iexcl71 64 607 iexcl8 iexcl71 54 1147 iexcl8 iexcl69 51 1021 913

Sup parietal lobe L 7 iexcl32 iexcl55 48 658 iexcl31 iexcl55 48 1056 included above 639

Sup parietal lobe R 7 28 iexcl66 50 480 28 iexcl54 51 885 20 iexcl68 44 992 540

Inf parietal lobe R 407 41 iexcl47 40 888

Inf parietal lobe L 40 iexcl50 iexcl45 36 829

Temporal

Midinf temporal G L 37 iexcl49 iexcl55 3 594 iexcl49 iexcl55 3 910 iexcl46 iexcl68 0 1052 563

Inf temporal S R 37 46 iexcl62 iexcl1 672 44 iexcl62 iexcl1 918

Sup temporal S R 2221 58 iexcl25 2 671

Other

Cerebellum L iexcl13 iexcl55 iexcl9 422 1 iexcl61 10 871 1 iexcl54 iexcl6 1099 389

Cingulate G L 2331 iexcl11 iexcl44 24 601 iexcl8 iexcl45 21 495

Thalamus L iexcl13 iexcl7 10 690 iexcl8 iexcl8 7 933

Thalamus R 8 iexcl18 19 650 5 iexcl8 7 957

Sup colliculus R 7 iexcl27 3 619 5 iexcl27 3 1003

Sup colliculus L iexcl5 iexcl29 iexcl5 674 iexcl7 iexcl27 3 1045

DT-fixed and DT-random denote the results of the interaction analyses DT-randomndashDT-fixed denotes the results of the direct comparison of bothconditions Diff depicts the z value in the SPM of the comparison DT-randomndashDT-fixed at the location of the peak activation as defined in theinteraction analysis of DT-fixed Local activation peaks of frontal areas are printed in italics

Statistical significance according to Bonferroni adjustment An asterisk denotes activation peaks which proved nonsignificant after Bonferronicorrection for multiple comparisons Significance levels after Bonferroni correction p lt 05 corresponds to z gt 479 p lt 01 to z gt 511 andp lt 0001 to z gt 592 For further information see Table 1

Szameitat et al 1189

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

1190 Journal of Cognitive Neuroscience Volume 14 Number 8

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

the DT-random condition and predicted that dual-task-related areas as defined by the DT-fixed conditionshould reveal higher activation in the DT-random con-dition To ensure that DT-random activated similar areasas DT-fixed we first calculated the same interactioncontrast as for DT-fixed Then we compared the activa-tion in both tasks directly Again lateral prefrontalactivations are reported only and a complete list ofactivated areas is given in Table 2

First the results of the subtraction analysis of theDT-random condition showed activation in similar areasas in the DT-fixed condition In the left hemisphere awidely spread activation covering the IFS (BA 10 4546) the MFG (BA 8 9 46) and the SFS (BA 6) wasfound In the right hemisphere cortical areas in theMFG (BA 10 46) and regions along the SFS (BA 6)revealed significant activation The activation in the MFG(especially in the right hemisphere) extended moreanteriorly in the DT-random condition than in DT-fixedHowever most importantly the (local) activation peakswere nearly identically localized compared to theDT-fixed subtraction analysis This was also true if moreconservative z thresholds were used for the statisticalanalysis (Table 2) To summarize this analysis indicatedalmost identical anatomical structures in the lateralprefrontal cortex to be involved in the processing ofDT-fixed and DT-random

Second we calculated the contrast betweenDT-random and DT-fixed in order to obtain a directcomparison between both dual-task conditions Theresults of this analysis revealed higher activation in theDT-random condition in all cortical areas activated inthe DT-fixed condition (Figure 4b lower row Table 2)Specifically the lateral prefrontal cortex was extensivelyactivated bilaterally (covering the IFS MFG and SFS)

To exclude the possibility that the stronger activationin DT-random is due to the increased error rate in thiscondition we calculated the same contrasts using onlycompletely error-free blocks This analysis (not shown)revealed an almost identical pattern for the relevantactivation foci which indicates that the observed dual-task-related activation was not caused by increasederror processing in the DT-random compared to theDT-fixed condition

Taken together these results show dual-task-relatedactivation in DT-fixed which cannot be reduced to thesummed single-task activation Furthermore althoughprefrontal activation was also observed in the single tasksthis activation was differentially located All dual-task-related areas as defined by DT-fixed especially lateral

prefrontal areas showed higher activation in DT-randomwith striking similarities regarding the location of acti-vation peaks Therefore dual-task-related activation wasshown with two different methodological approachesfirst with an analysis based on the subtraction methodand second with the parametric manipulation method

Regions-of-Interest (ROI) Analyses

To assess the effects of the different task conditions onthe strength of activation in lateral prefrontal areas inmore detail we conducted ROI analyses As described inthe Methods these analyses are based on an analysis ofa cloud of voxels surrounding the voxel with peakactivation in dual-task-related foci of activation As refer-ence points we determined the voxel that was locatedat x = iexcl44 y = 16 z = 38 for the left lateral prefrontalcortex and the voxel at x = 41 y = 30 z = 32 for theright lateral prefrontal cortex (cf Table 2) In the firststep of this analysis we tested whether the percentsignal change (PSC see Methods) in the conditions VISAUD DT-fixed and DT-random differed significantlyfrom that in the condition BASE (Figure 5) While thePSC in the single tasks did not differ significantly fromBASE (lowest p gt 19) it differed in the dual-taskconditions (highest p lt 05) In a next step we testedwhether the PSC in the DT-fixed condition differed fromthe PSC in each single-task condition This analysisrevealed a significantly higher PSC in DT-fixed than ineach single task (all prsquos lt 05) Furthermore DT-randomshowed a higher PSC than DT-fixed (all prsquos lt 01)Altogether these results show that there is only aminor nonsignificant signal change in the single-taskconditions in areas identified as dual-task related bymeans of the SPMs Furthermore the results show thatdual-task performance in DT-fixed significantly enhan-ces the BOLD-signal in these areas and that this en-hancement is more pronounced in the more difficultDT-random condition

Finally we tested whether the PSC observed in thedual-task conditions can be reduced to the sum of thePSC observed in the single-task conditions For thispurpose we compared the PSC in each dual-taskcondition with the summed PSC of both single-taskconditions for each ROI The analyses showed that thePSC increased overadditively in DT-fixed as well as inDT-random as compared to the summed PSC of thesingle-task conditions (all prsquos lt 05) In other words theobserved PSC in the dual-task conditions were notreducible to the summed effects of the single-task

Figure 4 Averaged fMRI data of 11 participants Note the different thresholds of the SPMs which were selected for purpose of illustration (a)Contours and (local) peaks of the activated areas projected onto the surface of a reference brain for the thresholds as given in b and c Colorscorrespond to the condition denotation in b and c Note that contours might change with different thresholds (b) Activation is combined with thesame but white matter segmented reference brain Activation inside white matter is not shown Both cognitive subtraction approach (upper row)and parametric manipulation approach (lower row) revealed dual-task-related activation in lateral prefrontal cortices (c) Activation related tosingle-task performance

Szameitat et al 1191

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

conditions Therefore this analysis confirmed the resultsobserved in the SPMs

DISCUSSION

The concurrent performance of two choice-reactiontasks resulted in additional activation in the dorsolateralprefrontal cortex (DLPFC ie IFSMFG) as compared tothe single-task performance This activation was accom-panied by severe performance decrements in the dual-task compared to the single-task situations Furthermorethe finding of dual-task-related activation was validatedby converging evidence with two methodologically differ-ent neuroimaging approaches the cognitive subtractionmethod and the parametric manipulation method

The results suggest that by using a dual-task paradigmthat ensures overlapping and interfering task processingcortical areas mediating the coordination of task pro-cessing can be localized Thus the identified areas alongthe IFS and in the MFG seem to be involved in the fastadaptation and coordination of actions accordingto current behavioral goals especially in situations ofinterfering information

Dual-Task-Related Brain Areas Assessed by theSubtraction Method

In a first step we identified dual-task-related brain areasin the DLPFC with the cognitive subtraction method Forthis purpose participants performed a dual task accord-ing to the PRP paradigm and the corresponding compo-nent tasks as single tasks The comparison of dual-taskand single-task activation revealed activation not reduci-ble to the summed effects of the activation in both singletasks performed separately Such activation was locatedin lateral prefrontal areas along the IFS and in the MFGThe finding of additional dual-task-related activationindicates that certain processing requirements musthave been additionally present in the dual-task com-pared to the single-task situation

We propose that these additional requirements reflectexecutive functions coordinating interfering processes ofthe component tasks In the present DT-fixed conditionthe occurrence of dual-task interference was indicatedby the PRP effect that is the increase of RTs on thesecond of two consecutively presented tasks (Pashler1994) Recent dual-task theories suggest that the reso-lution of this interference requires executive functionswhich schedule the processing order of the tasksinterrupt and reinstate task processing and switchbetween the processing streams of the tasks (Meyer ampKieras 1997 DeJong 1995) Our data indicate thatthese processes are mediated by lateral prefrontal areasalong the IFS and in the MFG Furthermore they showthat even two easy-to-perform nonmnemonic tasks canimpose substantial coordination demands on theseareas when they are performed concurrently

The present findings converge with those of otherneuroimaging studies investigating related processeswith different paradigms In particular it has beenshown that the inhibition of a task (Konishi et al1999) the switching between tasks (Dove PollmannSchubert Wiggins amp von Cramon 2000 Sohn UrsuAnderson Stenger amp Carter 2000) and the instatementof nondominant tasks (Zysset Muller Lohmann amp vonCramon 2001) also correlate with activation in corticalareas along the IFS and in the MFG Accordingly theseand our results suggest that these cortical areas imple-ment general control over task coordination regardlessof the particular paradigm used as long as interferinginformation has to be processed (see also MacDonaldCohen Stenger amp Carter 2000 Miller 2000)

The present findings are further supported by theresults of Koechlin et al (1999) Goldberg et al (1998)and DrsquoEsposito et al (1995) who also showed thatperformance of a dual task relies on additional neuralprocessing in cortical areas of the MFG as compared tosingle-task performance Moreover a recent study byHerath et al (2001) identified additional activation in avariant of the PRP paradigm too However in this study

Figure 5 Average PSC forROIs in the left and rightmiddle frontal gyrus PSC in thedual tasks was higher than thesummed PSC of both singletasks PSC in DT-random washigher than in DT-fixed

1192 Journal of Cognitive Neuroscience Volume 14 Number 8

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

dual-task-related activation was identified in the rightinferior frontal gyrus (IFG) The reason for this differenceis not completely clear but might result from differencesbetween the employed component tasks Herath et alused simple-reaction tasks (visual and somatosensory)while we used choice-reaction tasks (visual and auditory)Beside the combination of different modalities process-ing differences between simple- and choice-reaction tasks(Schubert 1999 Pashler 1994 Frith amp Done 1986)might have resulted in the involvement of differentcortical areas in the Herath et al and the present study

The results of several other dual task studies are not inaccordance with our findings In particular Adcock et al(2000) Bunge et al (2000) and Klingberg (1998) failed tofind any additional prefrontal areas associated with dual-task performance There may be several reasons for thedifferences between these findings and ours For exam-ple these studies used rather complex paradigms whichdid not allow the occurrence of interference to be con-trolled Due to rather long intervals between the compo-nent tasks the tasks may have been processed withoutsimultaneously drawing on a bottleneck mechanismConsequently interference and executive functionswould have been absent and the brain areas related tothese functions could not be localized This interpreta-tion is supported by the fact that Adcock et al and Bungeet al reported only marginal and even no performancedecrements respectively in the dual-task conditionsThis lack of dual-task decrements would have beencounterintuitive if interference between both tasks hadbeen present in these studies Such caveats were circum-vented in the present study by using the PRP paradigmwith short SOAs between both component tasks

A further reason for the failure to detect dual-task-related activation might be the type of single tasksemployed For instance Adcock et al used a semanticcategorization and a mental rotation task and Bungeet al a sentence evaluation and a memory task while weused easy-to-perform choice-reaction tasks It is conceiv-able that in the former studies even the single tasksalone involved executive functions Accordingly in theabovementioned studies prefrontal activation wasalready present during single-task performance in inferi-or regions of the DLPFC (IFSventral MFG) that is thoseareas that proved to be dual-task related in the presentstudy Thus if dual-task performance relies on thesemore inferior regions of the DLPFC one should expectoveradditively stronger activations in areas already acti-vated by single-task performance rather than newlyactivated areas The results of Adcock et al (2000)Bunge et al (2000) and Klingberg (1998) support thisinterpretation because these studies showed trends ofhigher activation in the MFG during dual-task comparedto single-task performance

In the present study the performance of the singletasks also resulted in activation of lateral prefrontal areasHowever the foci of these activations were located in

anteriorndashsuperior portions of the MFG and SFG andtherefore were differentially located compared to thedual-task-related activation

Anteriorndashsuperior portions of the PFC are assumed toreceive input from higher-order auditory and visualassociation areas (Miller 2000 Petrides amp Pandya1999 Romanski et al 1999) In the present studyactivation strength in these areas remained constantduring all task conditions Consequently it was sub-tracted out in the dual-task interaction contrasts Thiswould be in accordance with the view that activation ofthese prefrontal regions purely reflects the input ofdomain specific sensory information into the prefrontalcortex while the cortical areas identified as dual-taskrelated reflect executive functions acting upon thisinformation to resolve interference This is in disagree-ment with a suggestion made by Adcock et al (2000)and Bunge et al (2000) that additional processing indual-task performance is solely mediated by an increaseof activation in brain areas directly related to single-taskprocessing The ROI analyses of the present datashowed that there was no activation in dual-task-relatedareas during single-task performance Furthermore thevoxels of peak activation in the dual-task and the single-task conditions were clearly spatially separated Theseresults suggest that executive functions resolving inter-ference in dual-task processing can be attributed tospecific cortical areas and that these functions do notsolely originate from altered processing in areas sub-serving component-task processing as proposed byAdcock et al and Bunge et al

In the present study performance of the dual taskactivated not only prefrontal cortices but also parietalareas These activations were located mainly along theIPS and the precuneus These parietal activations are inaccordance with other paradigms that also investigatedexecutive functions (Koechlin et al 1999 Rowe ToniJosephs Frackowiak amp Passingham 2000 Callicottet al 1999 Schubert von Cramon Niendorf Pollmannamp Bublak 1998 Cohen et al 1997) suggesting thatexecutive functions might be mediated by a network ofbrain areas including prefrontal and parietal cortices(Bunge et al 2000 Baddeley amp Della Sala 1996Baddeley 1998) In more detail the presently observedareas along the IPS were previously associated withmore general attentional functions like different typesof visual attention (Wojciulik amp Kanwisher 1999) or theattention to time intervals (Coull amp Nobre 1998) Suchattentional demands specifically arise in the dual-tasksituation where attention has to be switched rapidlybetween modalities in a predefined order

Dual-Task-Related Brain Areas Assessed byParametric Manipulation

As mentioned in the Introduction the approach of cog-nitive subtraction is based on the critical presumption of

Szameitat et al 1193

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

pure insertion (Donders 18681969) that is that pro-cessing of the auditory and visual task is the sameregardless of whether they are performed as single tasksor as component tasks in a dual-task context Further-more in dual-task blocks two task sets have to be heldin memory compared with one task set in single-taskblocks (Kray amp Lindenberger 2000 see also Dove et al2000 Klingberg 1998) The increased memory loadmight interact with task-dependent processes resultingin the overadditive activation observed in the DT-fixedcondition (Rypma Prabhakaran Desmond Glover ampGabrieli 1999)

To circumvent possible pitfalls of pure insertion andmemory load we additionally used the parametricmanipulation method to assess dual-task-related activa-tion For this purpose we manipulated the difficulty ofdual-task-specific executive functions by introducingthe DT-random condition Compared to DT-fixed thissecond dual-task condition imposed higher demands onexecutive functions which coordinate interfering taskprocessing at the stage of the bottleneck (DeJong 1995)The behavioral data showed that DT-random was con-siderably more demanding than DT-fixed The directcomparison of the fMRI data revealed that both con-ditions engaged nearly identical anatomical structureswith remarkable similarities with respect to the locationof the (local) activation peaks Most importantly theseanatomical structures were stronger activated by meansof z value and PSC in the more demanding DT-randomcompared to the DT-fixed condition

These data validate the conclusion suggested by theresults of the cognitive subtraction method that thecoordination of interfering task processes activates addi-tional cortical areas in dual-task compared to single-tasksituations In addition the validity of the parametricmanipulation approach is emphasized by the fact thatcortical areas related to single-task processing likeanteriorndashsuperior portions of the DLPFC or primarysensory and motor cortices showed virtually no increasein activation strength in DT-random compared toDT-fixed (cf Cohen et al 1997) This rules out analternative suggestion that early perceptual or latemotor processes were affected by the parametric manip-ulation of dual-task difficulty and not the proposedcoordination of interfering processing stages at thebottleneck As outlined above this coordination mightinvolve the scheduling of the processing order as well asthe management of task interruption and the switchingbetween different processing streams (Meyer amp Kieras1997 DeJong 1995) It is still an open issue whetherthese aspects are realized by distinct subprocesses orwhether they rely on the same cognitive mechanism Itwould be an interesting question for future studies toelucidate whether a parametric manipulation of thedual-task difficulty different than the one employed inour study would lead to differently localized activationchanges in the observed dual-task-related areas

In addition to similarly localized activation foci therewere also slight differences between the activationpatterns observed in the DT-fixed and DT-random con-ditions for example in the medial SFG Such differ-ences might result from certain cognitive requirementsin the DT-random condition which are related to sideaspects of the parametric manipulation For instancethe observed DT-random activation in the medial SFGmight be due to the additional requirement to perceivethe presentation order of the stimuli or to the require-ment to change rapidly the order of motor programs inDT-random which were present only in a very rudi-mentary form in DT-fixed This would be in accordancewith studies indicating a specific role of this area in theperception of stimulus sequences (Schubotz Friedericiamp von Cramon 2000 Schubotz amp von Cramon 2001) orthe programming of motor sequences (Shima amp Tanji1998 2000)

Conclusion

Taking together the findings gained by the cognitivesubtraction and parametric manipulation method wepresented strong evidence that lateral prefrontal areasalong the IFS and in the MFG are associated with dual-task performance More specifically these areas provedto be related to the coordination of interfering task pro-cessing and not to memory load or altered component-task processing Such coordination requires the rapidadaptation of current processing strategies and involvesthe inhibition and activation of task representations andthe switching between them (cf Baddeley 1986) Whilethis adaptive processing was already required by thebottleneck processing in the DT-fixed condition itplayed an even greater role in the DT-random conditionHere processing strategies for handling the interferencecaused by the bottleneck had to be rapidly changed andadapted on-line during task presentation Interestinglyeven the merging of two rather easy-to-perform tasks canimpose considerable demands on these coordinationprocesses as indicated by the substantial activationsand severe performance decrements in the dual-taskconditions Therefore we conclude that cortical areasalong the IFS and in the MFG mediate the rapid adapta-tion of processing strategies which is a basic prerequisitefor goal-oriented and coherent behavior in situationswhere multiple conflicting actions have to be performedin a defined manner

METHODS

Participants

Twelve participants took part in the fMRI experimenteach having given prior informed consent accordingto the Max-Planck-Institute guidelines The study wasapproved by the local ethics review board at the

1194 Journal of Cognitive Neuroscience Volume 14 Number 8

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

University of Leipzig Germany All participants wereright handed as assessed by the Edinburgh Inventory(Oldfield 1971) Due to technical problems the data ofone participant had to be discarded resulting in 11participants for the analyses The age of the remaining11 participants (5 women) ranged from 21 to 27 yearswith an average of 24 years All participants had normalor corrected to normal vision

Stimuli and Procedure

Participants had to perform an auditory and a visualthree-choice reaction task These tasks were performedblockwise either alone (single-task conditions) ortogether (dual-task conditions) While lying in the fMRIscanner participants viewed a projection screen via amirror They responded on two separate fMRI-suitablekeypads each with four keys The tasks were as follows(see Figure 6)

Single Tasks

Visual single task (VIS) A trial in the VIS conditionstarted with a blank green screen for 150 msec followedby a fixation period of 850 msec During this fixationperiod three black squares (each 168 pound 168) werepresented on a green background and the middlesquare contained a fixation cross (0388 pound 0388) Themiddle square was located at the center of the screenand the two other squares horizontally to the left andright each with a gap of 0448 After the fixation periodone of the three squares (the target) changed its lumi-nance from black to a light gray for 300 msec while theother two black squares remained on the screen Afterthe presentation of the target stimulus the screen wascleared and the participants had to respond during an

interval of 1750 msec Participants had to respond withthe right index finger to the left with the right middlefinger to the middle and with the right ring finger to thepresentation of the right target square After respondingeither a blank screen or a visual error feedback waspresented for 250 msec

Auditory single task (AUD) A trial in the AUD con-dition started with the identical blank screen and fix-ation period as the visual single task After the fixationperiod a tone with a frequency of either 300 600 or1300 Hz was presented for 300 msec while three blacksquares were presented on the screen After the pre-sentation of the tone the screen was cleared Theparticipants had to respond to the low tone with theirleft ring finger to the middle tone with the left middlefinger and to the high tone with the left index fingerThe other characteristics of the procedure were identi-cal to the condition VIS

Dual Tasks

In the dual-task conditions the participants had toperform both tasks together For this purpose bothstimuli (auditory and visual) were presented in rapidsuccession separated by the SOA Participants wereinstructed to respond in the order of task presentationthat is stimulus presentation The order of task presen-tation was balanced so that an equal number of trialsstarted with the auditory and visual stimulus respec-tively There were two dual-task conditions DT-fixedand DT-random

DT-fixed In this condition the order of task presen-tation remained constant within a block Participantswere informed about the upcoming order (AUDndashVIS

Figure 6 Trial design Thetime courses are shown onthe left the stimulusndash responsemappings on the right In thedual task the participants hadto combine both mappingsThe gray bar denotes the pre-sentation time of the auditorystimulus Each trial lasted 3300msec (a) Visual single task (b)Auditory single task (c) Dualtask In this example the parti-cipants first had to respond tothe auditory and then to thevisual stimulus In DT-randomthe SOA was 200 msec

150 msec 850 msec 300 msec 1700 msec1625 msec1550 msec

300 msec

SOA50 ms125 ms200 ms

fixation target(s) responseforeperiod

150 msec 850 msec 300 msec 1750 msec R

150 msec 850 msec 300 msec 1750 msec

1300 Hz

600 Hz

300 Hz

L

a Visual single task (VIS)

b Auditory single task (AUD)

c Dual task (DT)

250 msec

error

feedback on erroror blank screen

250 msec

error

250 msec

error

tone

tone

Szameitat et al 1195

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

or VISndashAUD) by an instruction given immediately beforeeach block The SOA varied randomly between 50 125and 200 msec To ensure that the total trial duration wasthe same in every condition the time available torespond to the second task was decreased by theamount of the used SOA so that the available responsetime for the second task in the SOA 200 msec conditiondecreased to 1550 msec The different SOAs were usedto show the PRP effect as an indicator of concurrent taskprocessing For the analysis of the fMRI data the SOAswere averaged All other characteristics were identical tothe single-task conditions

DT-random In this condition the order of task pre-sentation varied pseudorandomly across trials Partici-pants received no information about the upcomingstimulus order but instead they had to perceive thepresentation order and adapt their response accordinglyWe used only one SOA of 200 msec Prior testing revealedthis SOA as most appropriate for convenient orderdetection by the participants

BASE Additionally we included a resting baseline con-dition (BASE) in which the participants were required tofixate a black cross (0388 pound 0388) presented on a greenbackground at the center of the screen

Design of measurement A block design was usedwith each block consisting of nine trials resulting in aduration of 297 sec per block The blocks whereseparated by an interblock interval of 103 s whichalso served as instruction period for the task in thefollowing block A session consisted of two runs eachbeginning and ending with the condition BASE Single-and dual-task conditions were counterbalanced sothat the probability of transitions between both con-ditions was equal with the only exception that neithersingle-task nor dual-task conditions were presented indirect succession All conditions were presented eighttimes resulting in a total experimental runtime of24 min All participants received an identical stimulusprotocol One to three days before the fMRI measure-ment participants practiced the tasks outside thefMRI scanner

Scanning Procedure

Imaging was carried out with a 3T scanner (Medspec30100 Bruker Ettlingen Germany) equipped with astandard birdcage head coil Participants were supine onthe scanner bed and cushions were used to reducehead motion Fourteen axial slices (192 cm FOV 64 pound64 matrix 5 mm thickness 2 mm spacing) parallel tothe ACndashPC plane and covering the whole brain wereacquired using a single shot gradient recalled EPIsequence (TR 2 s TE 30 msec 908 flip angle) sensitiveto BOLD contrast Two functional runs with 360 volumes

each were administered with each volume sampling all14 slices Prior to the functional runs 16 anatomicalMDEFT slices and 16 EPI-T1 slices were acquired In aseparate session high-resolution whole-brain imageswere acquired from each participant using a T1-weightedthree-dimensional segmented MDEFT sequence Theseimages were linearly rotated and translated but notresized into the stereotactic space of Talairach andTournoux (1988)

Data Analysis

Preprocessing

The fMRI data were analyzed using the software packageLIPSIA (Lohmann et al 2001) First the functional datawere preprocessed For this purpose artifacts at scanborders were removed and a slicewise movement cor-rection in the transverse direction was applied (FristonWilliams Howard Frackowiak amp Turner 1996) A Gaus-sian spatial filter (FWHM 565 mm) was used for smooth-ing The temporal offset between acquisition times ofdifferent slices acquired in one volume were correctedusing a linear interpolation

After preprocessing the functional and anatomicaldata were coregistered First the MDEFT and EPI-T1slices geometrically aligned with the functional sliceswere coregistered with the high-resolution 3-D refer-ence T1 data set of each participant Rotational andtranslational parameters computed for this registrationwere stored in individual transformation matricesSecond each transformation matrix was transformedinto a standard brain size (Talairach amp Tournoux1988) by linear scaling Finally these normalized trans-formation matrices were applied to the individual fMRIdata After anatomical coregistration the functionaldata were spatially rescaled to a resolution of 3 mm3

using trilinear interpolation

Statistics

Statistical analysis was based on a voxelwise leastsquares estimation using the general linear model forserially autocorrelated observations (Friston et al1995) A boxcar function with a response delay of6 sec was used to generate the design matrix Low-frequency signal drifts were controlled by applying atemporal highpass filter with a cutoff frequency of00036 Hz In addition the design matrix and the func-tional data were linearly smoothed with a 4-sec FWHMGaussian kernel The emerging autocorrelation causedby the temporal filtering and the smoothing was takeninto account during statistical evaluation by an adjust-ment of the degrees of freedom (Worsley amp Friston1995) Contrasts between different conditions werecalculated using the t statistics Subsequently t valueswere transformed into z scores As the individual

1196 Journal of Cognitive Neuroscience Volume 14 Number 8

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

functional data sets were aligned to the same stereo-tactic reference space a group analysis of fMRI data wasperformed using a voxelwise one-sample t test (Bosch2000) All resulting SPMs were thresholded at z gt 33( p lt 0005 uncorrected) To account for the possibilityof false positives (Type I errors) given by the numberof multiple tests we further performed a Bonferroniadjustment for an overall false-positive probability of 05(corresponding z gt 479) The results of this adjust-ment are given in Tables 1 and 2

ROI Analyses

As a second approach for analyzing the data we per-formed ROI analyses These analyses were aimed attesting whether the effects observed in the SPMs werealso observed in the PSC of the BOLD response Theanalyses were performed as follows First we deter-mined the activation peaks of dual-task-related activa-tion in lateral prefrontal areas as revealed by the SPM ofthe DT-fixed interaction group analysis Starting fromthese voxels groups of continuously connected voxelsexceeding a threshold of z gt 5 were determined Nextthe signal timecourses of the voxels constituting thesegroups were individually averaged for each conditionPSC of the AUD VIS DT-fixed and DT-random condi-tions was individually calculated by relating their time-courses to the timecourse of the BASE condition Toavoid influences from transient adjustment processes atthe beginning or at the end of a block we discarded thefirst and last four timesteps from each timecourseleaving seven timesteps for analysis Next the PSC ofthe seven timesteps was averaged so that one value wasobtained for each condition and participant Finally weused t tests to compare the PSC values in the differenttask conditions All t tests were calculated one-tailedwith an alpha level of 5 because of directed hypothesesabout the expected effects

Acknowledgments

We thank Anke Mempel Mandy Naumann Anke Pitzmaus andKathrin Wiesner for their assistance in data acquisition andHugh Garavan and an anonymous reviewer for their valuablecomments on an earlier version of the manuscript

Reprint requests should be sent to Andre Szameitat Max-Planck-Institute of Cognitive Neuroscience PO Box 500 355D-04303 Leipzig Germany or via e-mail szameitcnsmpgde

The data reported in this experiment have been deposited inThe fMRI Data Center (httpwwwfmridcorg) The accessionnumber is 2-2002-1134F

Notes

1 There is a recent controversy about whether this bottle-neck is immutable or not (eg Levy amp Pashler 2001Schumacher et al 2001) However this controversy does notconcern the assumption underlying the present study that

executive functions are needed to coordinate task processingat the stage of the bottleneck2 We interpreted our design as 2 pound 2 factorial design withthe factors Auditory Task and Visual Task both incorporat-ing the levels Task Present and Task Absent Both tasksabsent constitutes the resting baseline condition eitherauditory or visual task present the single-task conditions andboth tasks present the dual-task condition This enables totest for interaction between the factors by using the contrast((DT-fixedndashAUD) iexcl ( VISndashBASE))3 Although the activation in the right SFG in the conditionVIS did not reach significance after Bonferroni correction formultiple comparisons ( p = 000011 uncorrected) we consid-ered it for the purpose of this analysis

REFERENCES

Adcock R A Constable R T Gore J C amp Goldman-RakicP S (2000) Functional neuroanatomy of executiveprocesses involved in dual-task performance Proceedings ofthe National Academy of Sciences USA 97 3567 ndash3572

Baddeley A (1986) Working memory Oxford UK ClarendonPress

Baddeley A (1990) Human memory Theory and practiceHove London Erlbaum

Baddeley A (1998) The central executive A concept andsome misconceptions Journal of the InternationalNeuropsychological Society 4 523 ndash526

Baddeley A amp Della Sala S (1996) Working memory andexecutive control Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 3511397 ndash1404

Badgaiyan R D (2000) Executive control willed actionsand nonconscious processing Human Brain Mapping 938ndash41

Bosch V (2000) Statistical analysis of multi-subject fMRI dataThe assessment of focal activations Journal of MagneticResonance 11 61ndash64

Braver T S Cohen J D Nystrom L E Jonides J SmithE E amp Noll D C (1997) A parametric study of prefrontalcortex involvement in human working memoryNeuroimage 5 49ndash62

Broadbent D E (1982) Task combination and the selectiveintake of information Acta Psychologica 50 253 ndash290

Bunge S A Klingberg T Jacobsen R B amp Gabrieli J D E(2000) A resource model of the neural basis of executiveworking memory Proceedings of the National Academy ofSciences USA 97 3573 ndash3578

Callicott J H Mattay V S Bertolino A Finn K Coppola RFrank J A Goldberg T E amp Weinberger D R (1999)Physiological characteristics of capacity constraints inworking memory as revealed by functional MRI CerebralCortex 9 20ndash26

Cohen J D Perlstein W M Braver T S Nystrom L E NollD C Jonides J amp Smith E E (1997) Temporal dynamicsof brain activation during a working memory task Nature386 604 ndash607

Coull J T amp Nobre A C (1998) Where and when to payattention The neural systems for directing attention tospatial locations and to time intervals as revealed by bothPET and fMRI The Journal of Neuroscience 18 7426 ndash7435

Craik K W J (1947) Theory of the human operator in controlsystems I The operator as an engineering system BritishJournal of Psychology 38 56ndash61

DeJong R (1993) Multiple bottlenecks in overlapping taskperformance Journal of Experimental Psychology HumanPerception and Performance 19 965 ndash980

Szameitat et al 1197

DeJong R (1995) The role of preparation in overlapping-taskperformance Quarterly Journal of ExperimentalPsychology 48A 2ndash25

DrsquoEsposito M Detre J A Alsop D C Shin R K Atlas Samp Grossman M (1995) The neural basis of the centralexecutive system of working memory Nature 378 279ndash281

Donders F C (18961969) On the speed of mental processes(W G Koster Trans) In W G Koster (Ed) Attention andperformance II (pp 412ndash431) Amsterdam North-Holland

Dove A Pollmann S Schubert T Wiggins C J amp vonCramon D Y (2000) Prefrontal cortex activation in taskswitching An event-related fMRI study Cognitive BrainResearch 9 103 ndash109

Friston K J Holmes A P Worsley K J Poline J-P FrithC D amp Frackowiak R S J (1995) Statistical parametricmaps in functional imaging A general linear approachHuman Brain Mapping 2 189 ndash210

Friston K J Price C J Fletcher P Moore C FrackowiakS J amp Dolan R J (1996) The trouble with cognitivesubtraction Neuroimage 4 97ndash104

Friston K J Williams S Howard R Frackowiak R S Jamp Turner R (1996) Movement-related effects in fMRItime-series Magnetic Resonance in Medicine 35346 ndash355

Frith C D amp Done D J (1986) Routes to action in reactiontime tasks Psychological Research 48 169 ndash177

Goldberg T E Berman K F Fleming K Ostrem JVanHorn J Esposito G Mattay V S Gold J M ampWeinberger D R (1998) Uncoupling cognitive workloadand prefrontal cortical physiology A PET rCBF studyNeuroimage 7 296 ndash303

Herath P Klingberg T Young J Amunts K amp Roland P(2001) Neural correlates of dual task interference can bedissociated from those of divided attention An fMRI studyCerebral Cortex 11 796ndash805

Klingberg T (1998) Concurrent performance of two workingmemory tasks Potential mechanisms of interferenceCerebral Cortex 8 593 ndash601

Knight R T Grabowecky M F amp Scabini D (1995) Roleof human prefrontal cortex in attention control In H HJasper S Riggio amp P S Goldman-Rakic (Eds) Epilepsyand the functional anatomy of the frontal lobe(pp 1357ndash1371) New York Raven Press

Koechlin E Basso G Pietrini P Panzer S amp Grafman J(1999) The role of the anterior prefrontal cortex in humancognition Nature 399 148 ndash151

Konishi S Nakajima K Uchida I Kikyo H Kameyama Mamp Miyashita Y (1999) Common inhibitory mechanism inhuman inferior prefrontal cortex revealed by event-relatedfunctional MRI Brain 122 981 ndash991

Kray J amp Lindenberger U (2000) Adult age differences intask switching Psychology and Aging 15 126 ndash147

Levy J amp Pashler H (2001) Is dual-task slowing instructiondependent Journal of Experimental Psychology HumanPerception and Performance 27 862ndash869

Loftus G R amp Masson M E J (1994) Using confidenceintervals in within-subject designs Psychonomic Bulletin ampReview 1 476 ndash490

Lohmann G Mueller K Bosch V Mentzel H Hessler SChen L Zysset S amp von Cramon D Y (2001) Lipsiamdasha new software system for the evaluation of functionalmagnetic resonance images of the human brainComputerized Medical Imaging and Graphics 25449 ndash457

MacDonald A W III Cohen J D Stenger V A amp CarterC S (2000) Dissociating the role of the dorsolateralprefrontal and anterior cingulate cortex in cognitive controlScience 288 1835 ndash1838

Meyer D E amp Kieras D E (1997) A computational theoryof executive cognitive processes and multiple-taskperformance Part 1 Basic mechanisms PsychologicalReview 104 3ndash65

Miller E K (2000) The prefrontal cortex and cognitivecontrol Nature Reviews Neuroscience 1 59ndash65

Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia9 97ndash113

Pashler H (1994) Dual-task interference in simple tasks Dataand theory Psychological Bulletin 116 220 ndash244

Petrides M amp Pandya D N (1999) Dorsolateral prefrontalcortex Comparative cytoarchitectonic analysis in thehuman and the macaque brain and corticocorticalconnection patterns European Journal of Neuroscience11 1011 ndash1036

Romanski L M Tian B Fritz J Mishkin M Goldman-RakicP S amp Rauschecker J P (1999) Dual streams of auditoryafferents target multiple domains in the primate prefrontalcortex Nature Neuroscience 2 1131 ndash1135

Rowe J B Toni I Josephs O Frackowiak R S J ampPassingham R E (2000) The prefrontal cortex Responseselection or maintenance within working memory Science288 1656ndash1660

Rypma B Prabhakaran V Desmond J E Glover G Hamp Gabrieli D E (1999) Load-dependent roles of frontalbrain regions in the maintenance of working memoryNeuroimage 9 216 ndash226

Sartori G amp Umilta C (2000) How to avoid the fallacies ofcognitive subtraction in brain imaging Brain and Language74 191 ndash212

Schubert T (1999) Processing differences betweensimple and choice reaction affect bottleneck localizationin overlapping tasks Journal of ExperimentalPsychology Human Perception and Performance 25408 ndash425

Schubert T von Cramon D Y Niendorf T Pollmann Samp Bublak P (1998) Cortical areas and the control ofself-determined finger movements An fMRI studyNeuroReport 9 3171ndash3176

Schubotz R I Friederici A D amp von Cramon D Y (2000)Time perception and motor timing A common corticaland subcortical basis revealed by fMRI Neuroimage 111ndash12

Schubotz R I amp von Cramon D Y (2001) Interval andordinal properties of sequences are associated with distinctpremotor areas Cerebral Cortex 11 210 ndash222

Schumacher E H Seymour T L Glass J M Fencsik D ELauber E J Kieras D E amp Meyer D E (2001) Virtuallyperfect time sharing in dual-task performance Uncorkingthe central cognitive bottleneck Psychological Science 12101 ndash108

Shima K amp Tanji J (1998) Both supplementary andpresupplementary motor areas are crucial for thetemporal organization of multiple movements Journalof Neurophysiology 80 3247 ndash3260

Shima K amp Tanji J (2000) Neuronal activity in thesupplementary and presupplementary motor areas fortemporal organization of multiple movements Journal ofNeurophysiology 84 2148 ndash2160

Sidtis J J Strother C Anderson J R amp Rottenberg D A(1999) Are brain functions really additive Neuroimage 9490 ndash496

Smith E E Geva A Jonides J Miller A Reuter-Lorenz Pamp Koeppe R A (2001) The neural basis of task-switchingin working memory Effects of performance and agingProceedings of the National Academy of Sciences USA 982095 ndash2100

1198 Journal of Cognitive Neuroscience Volume 14 Number 8

Sohn M-H Ursu S Anderson J R Stenger A amp CarterC S (2000) The role of prefrontal cortex and posteriorparietal cortex in task switching Proceedings of theNational Academy of Sciences USA 97 13448ndash13453

Sternberg S (1969) The discovery of processing stagesExtensions of Dondersrsquo method Acta Psychologica 30276 ndash315

Talairach J amp Tournoux P (1988) Co-planar stereotaxicatlas of the human brain New York Thieme

Umilta C Nicoletti R Simion F Tagliabue M E amp Bagnara

S (1992) The cost of a strategy European Journal ofCognitive Psychology 4 21ndash40

Wojciulik E amp Kanwisher N (1999) The generality ofparietal involvement in visual attention Neuron 23747 ndash764

Worsley K J amp Friston K J (1995) Analysis of fMRItime-series revisitedmdashagain Neuroimage 2 173 ndash181

Zysset S Muller K Lohmann G amp von Cramon D Y(2001) Colorndashword matching Stroop task Separatinginterference and response conflict Neuroimage 13 29ndash36

Szameitat et al 1199