Components of Switching Intentional Set

Matthew F. S. Rushworth, R. E. Passingham, and A. C. Nobre Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 Abstract & Despite the intuition that we can shift cognitive set on required shifts of intentional set; subjects selected between instruction, some behavioral studies have suggested that set responses according to one of two conflicting intentional sets. shifting might only be accomplished once we engage in The results demonstrated the existence of more than one performance of the new task. It is possible that set switching constituent process. Some of the processes were linked to the consists of more than one component cognitive process and initiation and reconfiguration of the set prior to actual that the component processes might segregated in time. We performance of the new task. Other processes were time recorded event-related potentials (ERPs) during two set- locked to performance of new task items. Set initiation started switching tasks to test whether different component processes with modulation of medial frontal ERPs and was followed by were responsible for (i) set initiation and reconfiguration modulation over parietal electrodes. Implementation of when presented with the instruction to switch, and (ii) the intentional set was associated with modulation of response- implementation of the new set once subjects engaged in related ERPs. & performing the new task. The response switching (RS) task

INTRODUCTION Styles, and Hsieh (1994) have argued that switching The neural mechanisms for directing selective attention costs that are protracted over several trials of the new to stimuli or locations are increasingly well understood task indicate that active task set reconfiguration did not (Colby & Goldberg, 1999; Corbetta & Shulman, 1999; occur. The debate is important because if task switching Desimone, 1999; Hillyard, Vogel, & Luck, 1999). How- costs are just the consequence of passive dissipation ever, in most situations people must also be able to then the widely held assumption (e.g., Kimberg et al., switch attention from one set of stimuli to another, or to 2000) that task switching is a useful index of executive switch how they respond to stimuli. Such switching is control may be incorrect. usually referred to as task or set switching. The study of The analysis of RT changes across behavioral para- set switching has attracted interest because it is thought digms has demonstrated that set-switching does com- to be a clearly defined, tractable, and central aspect of prise separable component processes (Meiran et al., the executive control of cognitive processes. Indeed, 2000; Monsell, Yeung, & Azuma, 2000; Wylie & some authors have referred to it as an ‘‘operational Allport, 2000; De Jong, 2000; Meiran, 1996; Rogers & measure of executive control’’ (Kimberg, Aguirre, & Monsell, 1995; Allport et al., 1994; Allport & Wylie, D’Esposito, 2000). Since at least the time of Jersild 1999). A prospective process of ‘‘task-set reconfigura- (1927) an important tradition of work has used reaction tion’’ (Rogers & Monsell, 1995), prior to task perform- times (RTs) to study set switching; subjects perform a ance, has been invoked because the cost of switching, task more slowly after previously performing a different as measured by RT increase, is lower when subjects task (‘‘task-switching cost’’). have greater opportunity for advanced preparation. An The constituent components of task switching are alternative account, however, proposes switching costs currently debated (Meiran, Chorev, & Sapir, 2000; Wylie are the consequence of the persisting activation of & Allport, 2000; Rogers & Monsell, 1995). Central to the processes related to the first task ( Wylie & Allport, debate is the degree to which task-switching costs are 2000; Allport et al., 1994; Allport & Wylie, 1999). due to prospective, active reconfiguration for the new According to this theory, advance preparation reduces task prior to its performance or passive dissipation of switching costs because it allows time for interfering the old task set. A related issue concerns the degree to ‘‘task set inertia’’ from the previous task to dissipate. By which switching costs continue beyond the first trial of altering the time when an instructional cue was pre- the new task; Wylie and Allport (2000) and Allport, sented within a preparation interval, Meiran et al. (2000) provided evidence that both passive dissipation of the old task set and active reconfiguration for the University of Oxford new task set occur. Notwithstanding this debate, most

D 2002 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 14:8, pp. 1139–1150

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Despite the success of RT-based analyses of set RS task switching complementary neuroimaging experiments may be informative. For example, Meiran et al. (2000, pp. 250–251) point out that it can be difficult to decide if small switching costs indicate a lack of engagement of left task-switching control processes or, the opposite, very efficient engagement of task-switching processes. In the following experiments we have used event-related

potential (ERP) recording to measure activity first when Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 stay a cue instructed set switches and subsequently when subjects began performing the task in a new way. We have referred to the two epochs as ‘‘set initiation’’ and right ‘‘set implementation’’ periods. If active task set recon- figuration occurs then it should be detected in the first period, the set initiation period. Processes related to left residual switching costs should be recorded in the second period, the set implementation period. Intentional set switching requires subjects to change right the rules by which they select between motor re- sponses while attentional set switching requires sub- jects to change the rules by which they select between sensory stimuli. Although there has been little attempt to distinguish such paradigms, both types of set switch- switch ing are possible (Pollman, 2001; Rushworth, Paus, & Sipila, 2001; Rushworth, Hadland, Paus, & Sipila, 2002; Dove, Pollman, Schubert, Wiggins, & von Cramon, right 2000). The present experiments required subjects to switch between intentional sets and used what we have previously referred to as a response switching (RS) left paradigm (Figure 1). The task was divided into blocks of 8 to 17 trials, and on each trial subjects were presented with either a rectangle- or triangle-shaped stimulus. Subjects started Figure 1. Response switching (RS) paradigm. In the RS paradigm by responding to rectangles and triangles with right- and subjects were presented with a series of task items. The items were left-hand responses, respectively (Figure 1), but on always either rectangles or triangles. Subjects alternated between two other blocks the stimulus-response translation changed; conflicting intentional sets: either triangle stimulus-left hand response and rectangle stimulus-right hand response or triangle stimulus-right hand response and rectangle stimulus-left hand response. Every 8–17 trials a white cue shape instructed subjects either to stay with the current selection rule or to switch to using a selection rule based on stay switch the other stimulus dimension. Stay or switch cues were differentiated cue cue by a ‘‘+’’ or ‘‘Â’’ at their center. The meaning of the ‘‘+’’ and ‘‘Â’’ was counterbalanced across subjects. In the example shown, the ‘‘+’’ and items . . . items . . . the ‘‘Â’’ mean stay and switch, respectively. The figure shows an example where the subject started by selecting left-hand and right-hand responses to the triangle and rectangle, respectively, and then later switches to the selecting right-hand and left-hand responses to the triangle and rectangle, respectively. stay switch block block

authorities agree that ‘‘residual’’ RT costs can be ob- Figure 2. The RS task involved a succession of blocks of 8–17 items served despite advance preparation. The nature of the interspersed with cues instructing subjects to stay with the current residual component of task switching, however, is not selection rule or to switch selection rule. The ERP analyses were based clear (Meiran et al., 2000). It may be a consequence of on two time periods. (1) The cue ERP period was the 1400 msec that a second executive control process that only occurs followed the presentation of stay and switch cues (indicated by gray after onset of the first stimulus in the new task or it rectangles). The ERPs that followed switch cues were compared with those following stay cues. (2) The item ERP period was the 1000-msec may be a reflection of interference from the old task period that followed the presentation of items (indicated by small gray (Meiran et al., 2000; Monsell et al., 2000; Wylie & rectangles). The brain activity after the first two items following switch Allport, 2000). and stay cues was compared.

1140 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 left- and right-hand responses were made to the rec- tangle and triangle, respectively. Blocks of trials were separated by presentation of either ‘‘switch’’ or ‘‘stay’’ cues. Switch cues instructed subjects to switch the intentional set guiding response selection. Stay cues cue period instructed subjects to continue with the current inten- tional set. As explained above, the ERP analysis centered on two 2µV

epochs, the set initiation and the set implementation Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 periods (Figure 2). The set initiation period analysis involved the comparison of the ERPs that followed the switch and stay cues. The set implementation period analysis involved the comparison of ERPs that followed the task items subsequent to either switch or stay cues. -2µV RESULTS 400–440 msec Behavioral Results The behavioral results for the RS task are shown in 4µV Figure 3. Switching intentional set had a clear effect on RTs. RTs recorded after a switch cue were significantly longer than those recorded after a stay cue [main effect of switch: F(1,11) = 5.30, p = .042]. Subjects were significantly slower at the beginnings of task blocks than at their ends [main effect of trial number: F(1.37,15.10) = 12.33, p = .002]. Switching had a greater -2µV effect at the beginning, rather than at the end, of trial 600–640 msec blocks [interaction of switch and trial number: F(2.72,29.93) = 4.817, p = .009]. 4µV

700 switch stay 600 -2µV 700–740 msec

500 1µV RT (msec)

400 -4µV 1300–1340 msec 300 135791113 Figure 4. Set initiation. Scalp topographies of voltage differences Trial Position in Block (switchÀstay) between ERPs recorded after switch or stay cues at various delays. Example 40-msec time bins area shown. The earliest (top) modulation was positive over the medial frontal scalp. At later Figure 3. Behavioral results. The black line shows RTs recorded from times (second and third panels) there was a positive modulation of the switch block trials. The gray line shows RTs recorded from stay block ERPs recorded from posterior midline and adjacent left hemisphere trials. Switching selection rule was associated with an RT cost. electrodes.

Rushworth, Passingham, and Nobre 1141

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 In the RS task it is clear that RTs were slightly slower used a two-dipole model, and identified two medial on the 10th trial. This may have been because just frontal dipoles. One dipole estimate was located in the before the 10th trial was, probabilistically, the most dorsomedial frontal cortex while the other was located likely time at which the ‘‘blink’’ period was presented very anteriorly within the medial frontal cortex (Figure 6, to the subject (see Methods). Subjects may have been example at 440 msec). There was a correlation of .93 slightly slower on the first trial subsequent to the blink between the recorded data and the fitted data (overall period but there was no apparent difference in behav- residual variation throughout period: 12%). ioral performance when the eye-blink period occurred In the second phase (520–1040 msec) switching was

in a switch or stay block. associated with a positive modulation over the posterior Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 midline and left electrodes (Figure 4, example 600- to 640- and 700- to 740-msec bins). This was apparent in ERP Results the significant main effect of switch and the significant Data from two subjects were discarded because of interaction between switch and electrode for the mid- excessive noise levels in the recordings. line, central, lateral, and posterior regions throughout this period (520–840 msec: p < .01; 840–1040 msec: p < .05). A left hemisphere lateralization was apparent Cue ERPs—Set Initiation Period in the significant interaction between switch and hemi- Differential neural activity to cues instructing a set sphere in the posterior region (520–720 msec: p < .05; switch occurred long before subjects had the opportu- 720–840 msec: p < .01, 840–880 msec: p < .05) nity to perform the new task items. The results are between 560 and 880 msec (Figure 4, example 700- to summarized in scalp topographies in Figure 4 and 740-msec bin). Dipole analysis identified a consistent selected electrode recordings in Figure 5. There were left ventromedial occipito-temporal source independent two distinct phases of ERP modulation in the set initia- of starting position (Figure 6, example at 660 msec). tion period. In the first phase (360–520 msec) switching There was a correlation of .95 between the recorded was associated with a positive modulation over frontal data and the fitted data (overall residual variance electrodes (Figure 4, top, example 400- to 440-msec throughout period: 9%). bin). This was apparent as a significant interaction between switch and electrode in the frontal region Task Item ERPs—Set Implementation Period (360–400 msec: p < .05; 400–480 msec: p < .01, 480–520 msec: p < .05). During preliminary dipole The analyses of ERPs to ‘‘items’’ following switch or stay modeling two different sources were identified depend- cues (Figure 2) identified subsequent components of ing on the dipole’s initial starting position. We therefore intentional set switching (Figures 7 and 8). There were

Figure 5. Set initiation. ERP waveforms recorded after stay (dashed) and switch (black) cues.

1142 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 cue period

item period

1µV Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021

-1µV 120–160 msec

440 msec 1µV

-2.5µV 240–280 msec 660 msec

Figure 6. Set initiation. Dipole source localization estimates during 1µV the cue ERP period. The earliest period of modulation (Figure 4, top) over medial frontal electrodes was modeled by ventromedial and dorsomedial dipole source estimates (top and center). The later period (Figure 4, center panels) was associated with dipole source estimates at the ventromedial occipito-temporal boundary (bottom).

two phases of ERP modulation in the set implementa- -2.5µV tion period. A negative ERP modulation was detected 400–440 msec between 200 and 360 msec (significant main effects of switch or interactions of switch and electrode in both posterior and central regions: p < .05). An example time bin (240–280 msec) is shown in Figure 7. The 2µV second phase involved positive ERP modulation (400–920 msec) over central frontal electrodes, partic- ularly the FC electrodes (Figure 7, example 580- to 620-msec bin). This effect was measured as a significant interaction of switch and electrode in the central region (400–480 msec: p < .01; 480–640 msec, p <.05; 840–920 msec, p < .05). We were not able to model -2µV dipole sources successfully during either phase of the 580–620 msec set implementation period.

Figure 7. Set implementation. Scalp topographies of voltage differ- ences (switch À stay) between ERPs recorded after items following DISCUSSION switch or stay cues at various delays. The first significant effect was a negative modulation of an N2-like component (second panel). There The ERPs recorded after cues instructing the switching was a late positive modulation over central and frontal electrodes of set or the maintenance of a current set diverged after (third and bottom panels).

Rushworth, Passingham, and Nobre 1143

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Figure 8. Set implementation. ERP waveforms recorded after the first two tasks items following stay (dashed) and switch (black) cues. Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021

360 msec, more than a second before engagement in the cortex. The prefrontal cortex appears to initiate the subsequent task. This suggests that some reconfigura- change of set prior to selective focusing of attention or tion of task set can occur prior to task performance intention during task performance. The present finding, (Meiran et al., 2000; Rogers & Monsell, 1995). ERP that the determination of intentional set begins with modulation to task items presented in the context of a frontal activity, complements and extends previous dem- set switch indicated a subsequent process of set imple- onstrations that prefrontal lesions disrupt attentional mentation. Both set reconfiguration and set implemen- modulation of the ERP signal (Barcelo, Suwazono, & tation were associated with several distinct phases of Knight, 2000; Gehring & Knight, 2000; Knight, 1997). ERP modulation; this would be expected if the process The ERP voltage distributions and dipole source of set switching consists of several constituent processes estimates suggested that the critical region within the (Meiran, 2000). frontal lobe was the frontal pole and the dorsomedial frontal cortex. Although there is evidence that the lateral parts of the prefrontal cortex are concerned with set Cue ERPs—Set Initiation and Reconfiguration switching (Passingham, 1972; Dias et al., 1996), it may be Intentional set switch initiation, prior to active engage- the case that these regions are most important when set ment in the new task, was associated with an initial switching involves a reassessment of which sensory period of midline frontal electrode modulation (Figure 4, aspects of a stimulus merit attention. There is growing top) that began just 360 msec after the instruction to evidence that it is just such frontal polar and dorsome- switch (360–520 msec). The existence of frontal modu- dial areas (including the presupplementary motor area lation in the earliest stages of set initiation in both tasks [pre-SMA]) that are particularly important when re- is consistent with the importance attributed to the sponse set is manipulated (Rushworth et al., 2002; frontal lobes in set switching (MacDonald, Cohen, Turken & Swick, 1999; Passingham, 1998; Jueptner Stenger, & Carter, 2000; Rogers, Andrews, Grasby, et al., 1997; Paus et al., 1993; Pardo, Pardo, Janer, & Brooks, & Robbins, 2000; Konishi et al., 1998; Rogers Raichle, 1990). et al., 1998; Dias, Robbins, & Roberts, 1996, 1997; Paus, The cue period modulation, prior to active engage- Petrides, Evans, & Meyer, 1993; Owen, Roberts, Polkey, ment in the new task, continued with a modulation Sahakian, & Robbins, 1991; Owen et al., 1992, 1993; over central and posterior electrodes (520–1040 msec). Passingham, 1972). Previous neuroimaging and lesion This effect correlated with a left, ventromedial occipito- studies have highlighted the importance of the frontal temporal dipole source estimate. The ventromedial lobes for set switching but the present study, because of occipito-temporal dipole source estimate implicated in the high temporal resolution of the ERP recording intentional set switching (Figure 6, bottom) is near to a technique, demonstrates that the very first changes, at region implicated in attentional selection in previous the beginning of switching, occur in the prefrontal ERP, PET, and fMRI studies (Hopfinger, Buonocore, &

1144 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Mangun, 2000; Anllo-Vento, Luck, & Hillyard, 1998; 1993; Kok, 1986). Implementing a change of intentional Heinze, Hinrichs, Scholz, Burchert, & Mangun, 1998; set was also associated with an increased late (400–920) Mangun, Buonocore, Girelli, & Jha, 1998; Nobre, Allison, positive modulation over frontal and central electrodes, & McCarthy, 1998; Heinze et al., 1994). The present thought to be sensitive to premotor cortical activity results suggest that not only is activity in this region (Gerloff, Uenishi, & Hallett, 1998). Intentional processes modulated by selective attention but it is also modu- may, therefore, be associated with quite distinct patterns lated by the anticipatory reconfiguration of set in the of modulation to those seen in attentional tasks ( Yama- absence of specific stimuli or responses. The region guchi, Yamagata, & Kobayashi, 2000; Hillyard et al., 1999;

seems to have a similarly important role when either Valdes-Sosa, Bobes, Rodriguez, & Pinilla, 1998; Eimer, Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 attentional or intentional set is reconfigured (M. F. S. 1995; Mangun, 1995). While attention has been shown Rushworth, R. E. Passingham, & A. C. Nobre, unpub- to modulate ERPs related to visual processing, the lished results). Although the region’s role in sensory implementation of an intentional set relies on distinct attention has been emphasized, it probably has a role neural processes that may be associated with the pre- in defining stimuli that will be determinants of action; motor cortex. Intention is known to modulate premotor blood flow changes occur here when subjects learn cell activity (Boussaoud & Wise, 1993). The premotor associations between stimuli and responses (Toni & cortex is critical for the selection of actions according to Passingham, 1999). arbitrary rules (Wise & Murray, 2000); the premotor There has been controversy about whether it is modulation in the RS task may reflect the implementa- possible to reconfigure set at an abstract level prior to tion of a new set of action selection rules. concrete task performance (Monsell et al., 2000; Wylie & An important caveat is whether the task item modu- Allport, 2000; Meiran, 1996, 2000; Meiran et al., 2000; lation reflects a process of set implementation that is Rogers & Monsell, 1995; Allport et al., 1994; Allport & discrete and distinct from the prior ERP modulation Wylie, 1999). Such previous studies question what set carried over from the cue period. The cue period ended switching entails and how it differs from other atten- with a sustained period of negative modulation and the tional processes. It is clear from the present results, first modulation recorded in the item period was also however, that ERP modulation begins prior to any negative. There are several reasons, however, for think- selective focusing of attention on particular task stimuli ing that the cue and item period modulations do not (none are present during the interval after switch or stay constitute one long continuous ERP modulation. First, cue presentation when the ERPs were recorded) or the ERP modulation at the end of the cue period was focusing of intention on particular task stimulus-to- small and not significant. Second, ERP effects during the response transformations (no responses were made item period consisted of two components that dif- during the cue period when the ERPs were recorded). fered—in terms of polarity, time course, and scalp Such ERP modulation may, therefore, indicate that set distribution (Figure 7)—from one another and from switching begins with reconfiguration of set prior to the preceding cue period modulations. It is very difficult active task performance. Some caution, however, to see how the minor and nonsignificant negative frontal is warranted. Another possibility is that differences modulation recorded at the end of the cue period could between stay and switch might be attributed to differ- be considered to have originated from the same ences in encoding effort. It seems possible that the generator as the subsequent posterior negative and subjects subvocalize, or otherwise encode into short- frontal positive modulations recorded some time later term memory, the cues or their identities and do not in the later part of the item period. actually reconfigure for the new task until the target If the item period modulations were related to the item is actually presented. implementation of the new intentional set rather than being somehow a remnant of earlier switch cue modu- lation then it might be expected that similar patterns of Task Item ERPs—Set Implementation modulations would be found when the ERPs following There was a significant difference between the ERPs the items presented later in switch or stay blocks are recorded after the first two trials of a block following compared. If such modulations are found then they a switch cue compared with ERPs recorded after the cannot be interpreted as the continuation of a prior first two trials of a block following a stay cue (Figures 7 period of minor, nonsignificant, negative modulation and 8). In addition to a process of prospective set because the preceding item epochs did not end in any reconfiguration, this result suggests a subsequent such period. Further analyses indeed showed that some process of set implementation. phases of the item modulation were not just observed Implementing an intentional switch modulated ERPs for the first items. They were also observed when the at two time periods. First, it was associated with an 7th and 8th items of switch and stay blocks were increased negative N2-like component (200–360 msec) compared (Figure 9). We compared the patterns of over the central posterior scalp. N2 modulation has intentional set switch modulation observed in trials at previously been related to response suppression (Eimer, the beginning (first and second trials), middle (7th and

Rushworth, Passingham, and Nobre 1145

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 8th trials), and end (12th to 15th trials) of blocks. Because switch or stay cues were presented every 8 to 17 trials, there are less examples of each type of later trial so that an average had to be compiled from 4, as opposed to 2 different trial numbers. ERPs on trials in the middle of blocks were still modulated by whether or not the block started with a switch or stay cue (Figure 9, middle). Although the magnitude of the modulation

was less than at the beginning of the block, there was Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 still a significant positive modulation over frontal and central electrodes (440–600 msec, Switch  Hemi- sphere times; Electrode interaction, p < .05). No sig- nificant switch-related modulation, however, could be observed by the end of the blocks (Figure 9, bottom), indicating that optimal processing of a particular task is eventually reached. In summary, the item period effects cannot be attrib- uted to the continuation of the minor, nonsignificant, negative frontal modulation in the cue period for three reasons. First, the item period modulation is small and nonsignificant. Second, the item period modulations have distinct polarities, time courses, and scalp distribu- tions. Furthermore, some of the item period modula- tions were replicated on subsequent item trials (the seventh and eighth) a long time after the cue period had ended. The identification of distinct neural processes of set implementation is consistent with behavioral studies that have recorded residual RT differences even when subjects are warned to switch in advance and provided with a long preparation interval (Meiran et al., 2000; Monsell et al., 2000; Sohn, Ursu, Anderson, Stenger, & Carter, 2000; Rogers & Monsell, 1995). Recently, fMRI has been used to measure brain activity at the time when subjects first begin to switch set and then subsequently when subjects engage in performance of the new task block (Kimberg et al., 2000; MacDonald et al., 2000). Although these important studies provide an unparal- leled level of spatial information about neural processes, they rely on the imposition of long delays between instruction cue and task engagement in order to sepa- rate the slow fMRI signal associated with each event, which may confound set switching with working mem- ory functions. The present ERP results, therefore, com- plement and extend previous behavioral and Figure 9. Replication of set implementation effects later in the block. neuroimaging studies of task switching. The same pattern of ERP modulation that had occurred to the first The origin of residual switching costs is controversial items following stay or switch cues (Figure 8) could also be seen (Meiran et al., 2000). On the one hand, it has been when later trials from switch and stay blocks were compared. Set argued that they are the consequence of a task set inertia implementation effects recorded on the first items after switch and stay cues are shown at an example electrode FC1 (top). Arrows point that causes interference between the stimulus-response to the two periods of significant difference between switch and stay mappings required by the two different sets (Wylie & ERPs (200–360 msec, 400–920 msec). The effect was replicated Allport, 2000; Allport et al., 1994; Allport & Wylie, 1999). (440–600 msec) when the items that appeared seven or eight trials An alternative view raises the possibility that residual after the switch or stay cues were compared even though the switching costs reflect an additional ‘‘control process, intentional set switch would have been initiated more than six trials (more than 12 sec) previously. The switch/stay difference, however, required to complete task-set reconfiguration, that could disappeared when items 12 to 15, from the end of switch and stay not be carried out until the stimulus onset’’ (Monsell blocks, were compared (bottom). et al., 2000, p. 253). The ERP results address this debate.

1146 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Advocates of control process accounts of residual Behavioral Task costs have tended to emphasize the way in which RT The experiment was conducted with subjects seated in a increases are most notable on the first trial of the new dimly illuminated, electrically shielded room facing a task rather than on the subsequent trials. It has been computer monitor at a distance of 100 cm. Figure 1 argued that it is difficult for theories based on a more summarizes the RS task. On each trial, subjects saw passive process of task set inertia to account for an either a red triangle (5.18 width, 2.78 high) or rectangle abrupt change in RT after the first trial of the new task. (3.78 width, 2.78 high). During the first set of trials, In the current experiment, there was evidence that subjects made a right-hand response to the rectangle

switching set had protracted consequences over several Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 and a left-hand response to the triangle. Subsequently, trials. The analysis of RTs revealed a significant main instruction cues appeared before each set of 8–17 trials. effect of switching even though the effect was greatest Instruction cues were either a vertical (+) or a tilted (Â) on the first trial in the RS task. The slowness of the first cross on a white rectangular background (68 width, trial of stay blocks is probably similar to stop-start effects 58 high) presented for 200 msec. Cues indicated that that have been recorded previously (Gopher, Armony, & the subject should switch intentional set or stay with Greenshpan, 2000). The analyses of ERPs recorded in the current set. There was a 2000-msec interval between the middle of blocks also revealed that ERPs were still the onset of the instructive cue and the onset of the differentially modulated dependent on whether the first item. There were no constraints to the order in block had begun with a stay or switch cue. which switch or stay cues were presented; the presen- On the other hand, not all phases of the item period tation of either type of cue was determined at random. ERP modulations recorded at the beginning of blocks, The meaning assignment (switch, stay) of each cue were observed again in the midblock comparisons. The (Â, +) was counterbalanced across subjects. Subjects earliest item period modulation, of the N2, was only performed a total of 1500 trials; there were five exper- seen when the first trials of stay and switch blocks were imental sessions each of 300 trials, and short pauses compared. This suggests that some cognitive processes were introduced to reduce ocular artefacts. A small may occur on just the first trials immediately after circle (0.98 diameter, 70-msec duration) provided feed- switching while other cognitive processes may still be back to the subjects 300 msec after the response observed in the middle of a switch block. It may, (yellow for correct responses and blue for incorrect therefore, be the case that different ERP modulations responses). Trials with incorrect responses were index either short term control processes or more excluded from the analysis. After the response and protracted interference effects. There is a growing real- feedback times, an interval of 1000 msec followed ization that elements of both theories emphasizing the before the onset of the next trial. The intervals between role of executive control and theories emphasizing item onsets varied with the variable reaction times, and interference will be needed to provide a full and ad- averaged about 1800 msec. equate account of set-switching: ‘‘It is abundantly clear that no simple theoretical dichotomy will be adequate to address the phenomena’’ (Monsell et al., 2000, p. 252, Behavioral Analyses and similarly, Meiran et al., 2000). It may be easiest to account for the diverse ERP modulations by appealing Reaction times of accurate responses were analyzed in a to both control- and interference-based theories and to two-way repeated-measures design. The two factors invoke several and specific control processes that over- were switch, with two levels corresponding to switch ride different aspects of interference. It should be or stay blocks, and trial number, with 14 levels corre- noted that such set implementation control processes, sponding to the item’s position in the block (i.e., first, observed in the item period, are distinct to the prior set second, third, etc.). initiation control processes, observed in the cue period. ERP Recordings

METHODS Electroencephalographic (EEG) activity was recorded continuously from 56 sites using tin electrodes mounted Subjects on an elastic cap (Electro-cap) and positioned according Sixteen subjects participated in the ERP study of the to the 10–20 International system (American Electro- visual switching (RS) task (ages 19–28). All subjects were encephalographic Society [AEEGS], 1991). The montage right handed, according to the Edinburgh handedness included 8 midline sites (FPZ, FZ, FCZ, CZ, CPZ, PZ, inventory (Oldfield, 1971), and had normal or corrected- POZ, PZ) and 21 sites over each hemisphere (FP1/FP2, to-normal visual acuity. The experimental methods were AF3/AF8, F3/F4, F5/F6, F7/F8, FC1/FC2, FC3/FC4, FC5/ noninvasive and had the approval of the Ethics Commit- FC6, FT7/FT8, C3/C4, C5/C6, T7/T8, CP1/CP2, CP3/CP4, tee of the Department of Experimental Psychology, CP5/CP6, TP7/TP8, P3/P4, P7/P8, PO3/PO4, PO7/PO8, University of Oxford. and O1/O2). Additional electrodes were used as ground

Rushworth, Passingham, and Nobre 1147

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 and reference sites, and to monitor the electrooculo- O1/O2. Each region therefore included either five elec- gram (EOG). The ground electrode was placed along the trodes (midline region) or five pairs of electrodes (all midline between FPZ and FZ. ERPs were referenced to other regions). ERP effects were assessed by repeated- the right mastoid, then re-referenced offline to the measures ANOVAs, using the Greenhouse-Geisser epsi- algebraic average of the right and left mastoids. Data lon correction for nonsphericity ( Jennings & Wood, were recorded with a band-pass filter of 0.1–100 Hz, 1976). ERP differences were only considered significant amplified 20,000 times, and digitized at a sampling rate if they persisted for at least two consecutive epochs. of 250 Hz. The EOG signal was recorded bipolarly. Only the corrected probability values are reported. The

Electrodes placed on the left and right of the external analysis compared cue ERPs using ANOVAs that tested Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 canthi measured horizontal eye movements. Electrodes the factors of switch (switch, stay), hemisphere of region placed above and below the left eye measured vertical (left, right), and electrode site (five levels). The hemi- eye movements and blinks. sphere factor was not relevant to the analysis in the ERPs were reconstructed offline, according to stimu- midline region. The analysis of item ERPs was carried out lus type. In the case of the ‘‘switch’’ and ‘‘stay’’ cues, the in a similar way. epoch began 200 msec before presentation and ended 1400 msec afterwards. In the case of the subsequent Dipole Analysis items, the epoch began 200 msec before presentation and ended 1000 msec afterwards. Epochs with excessive An attempt was made to localize the dipole sources of drift, or containing eye movements or blinks were the significant ERP modulations. Source localization was rejected. Trials were automatically eliminated if the carried out with ANT software (Enschede, Netherlands) voltage exceeded ±50 V at midline electrodes or at Advanced Source Analysis (ASA) program version 2.21, either the vertical or horizontal EOG channel at any using a three-shell realistic head model constructed time during the epoch. In addition, the HEOG was from a standardized and normalized ( Talairach & Tour- inspected visually and trials containing saccades were noux, 1988) average brain image. An equivalent current also eliminated. Trials containing incorrect responses in dipole (ECD) approach was used, in which the dipole the RS tasks were also eliminated. position and orientation was automatically iteratively readjusted to minimize the residual variance between the measured EEG data and the simulated field data. We ERP Analyses considered dipole estimates reliable if their locations The physical appearance and frequency of presenta- were independent of the model’s starting positions and tion of stimuli were equated in all ERP comparisons. if there was a greater than .9 correlation between Theanalysiscenteredontwoaspectsofthedata measured EEG data and simulated field data. With one (Figure 3). exception, explained in the results, we have used just 1. Cue ERPs—set initiation period: The ERPs elicited single-dipole models. by cues instructing subjects to ‘‘switch’’ their attentional set or to ‘‘stay’’ with the present set were compared to Acknowledgments isolate mechanisms involved in the in the initiation of intentional set switching. We gratefully acknowledge the assistance of Carlo Miniussi, Anling Rao, Ed Wilding, Ivan Griffin, and Heather Jordan. 2. Item ERPs—set implementation period: The ERPs Supported by the Royal Society and the Wellcome Trust. elicited by the stimulus items following switch and stay cues were compared. ERPs elicited by the first and Reprint requests should be sent to Matthew F. S. Rushworth, Department of Experimental Psychology, University of Oxford, second stimulus items in switch and stay blocks were Oxford, OX1 3UD, UK, or via e-mail: matthew.rushworth@ compared. psy.ox.ac.uk. In order to survey the ERP effects systematically over scalp locations and time periods, we tested for the effects of experimental factors at symmetrical scalp REFERENCES regions over successive 40-msec time bins (see Miniussi, American Electroencephalographic Society (AEEGS). Wilding, Coull, & Nobre, 1999 for a similar approach). (1991). American Electroencephalographic Society Five sets of analyses were performed over midline, guidelines for standard electrode placement position nomenclature. Journal of Clinical Neurophysiology, 8, frontal, central, lateral, and posterior scalp regions. 200–202. The midline analysis included electrodes FPZ, FZ, CZ, Allport, A., & Wylie, G. (1999). Task switching: Positive PZ, and OZ. The frontal analysis included FP1FP/2, and negative priming of task set. In G. W. Humphreys, AF3/AF4, AF7/AF8, F5/F6, and F7/F8. The central analysis J. Duncan, & A. Treisman (Eds.), Attention space and included FC1/FC2, FC3/FC4, C3/C4, CP1/CP2, and action (pp. 273–298). New York: Oxford University Press. Allport, D. A., Styles, E. A., & Hsieh, S. (1994). Shifting CP3/CP4. The lateral analysis included FC5/FC6, intentional set: Exploring the dynamic control of tasks. In C. FT7/FT8, T7/T8, CP5/CP6, and TP7/TP8. The posterior Umilta & M. Moscovitch (Eds.), Attention and performance analysis included P5/P6, P7/P8, PO3/PO4, PO7/PO8, and ( vol. XV, pp. 421–452). Cambridge: MIT Press.

1148 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Anllo-Vento, L., Luck, S. J., & Hillyard, S. A. (1998). Spatio- attention: Electrophysiological and neuroimaging evidence. temporal dynamics of attention to color: Evidence from In G. W. Humphreys, J. Duncan, & A. Triesman (Eds.), human electrophysiology. Human Brain Mapping, 6, Attention, space, and action: Studies in cognitive 216–238. neuroscience. Oxford, UK: OUP. Barcelo, F., Suwazano, S., & Knight, R. T. (2000). Prefrontal Hopfinger, J. B., Buonocore, M. H., & Mangun, G. R. (2000). modulation of visual processing in humans. Nature The neural mechanisms of top-down attentional control. Neuroscience, 3, 399–403. Nature Neuroscience, 3, 284–291. Boussaoud, D., & Wise, S. P. (1993). Primate frontal cortex: Jennings, J. R., & Wood, C. C. (1976). The epsilon-adjustment Effects of stimulus and movement. Experimental Brain procedure for repeated measures analyses of variance Research, 95, 28–40. [letter]. Psychophysiology, 13, 277–278.

Colby, C. L., & Goldberg, M. E. (1999). Space and attention Jersild, A. T. (1927). Mental set and shift. Archives of Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 in parietal cortex. Annual Review of Neuroscience, 22, Psychology, 89(whole). 319–349. Jueptner, M., Stephan, K. M., Frith, C. D., Brooks, D. J., Corbetta, M., & Shulman, G. L. (1999). Human cortical Frackowiak, R. S. J., & Passingham, R. E. (1997). Anatomy of mechanisms of visual attention during orienting and search. motor learning: I. Frontal cortex and attention to action. In G. W. Humphreys, J. Duncan, & A. Triesman (Eds.), Journal of Neurophysiology, 77, 1313–1324. Attention, space, and action: Studies in cognitive Kimberg, D. Y., Aguirre, G. K., & D’Esposito, M. (2000). neuroscience. Oxford, UK: Oxford University Press. Modulation of task-related neural activity in task-switching: De Jong, R. (2000). An intention activation account of residual An fMRI study. Cognitive Brain Research, 10, 189–196. switch costs. In S. Monsell & J. Driver (Eds.), Control of Knight, R. T. (1997). Distributed cortical network for cognitive processes. Attention and performance XVIII. visual attention. Journal of Cognitive Neuroscience, 9, Cambridge: MIT Press. 75–91. Desimone, R. (1999). Visual attention mediated by biased Kok, A. (1986). Effects of degradation of visual stimulation competition in extrastriate visual cortex. In G. W. components of the event-related potential (ERP) in Humphreys, J. Duncan, & A. Treisman (Eds.), Attention space Go/Nogo reaction tasks. Biological Psychology, 23, and action (pp. 273–298). New York: Oxford 21–38. University Press. Konishi, S., Nakajima, K., Uchida, I., Kameyaman, M., Nakahara, Dias, R., Robbins, T. W., & Roberts, A. C. (1996). Dissociation in K., Sekihara, K., & Miyashita, Y. (1998). Transient activation prefrontal cortex of affective and attentional shifts. Nature, of inferior prefrontal cortex during cognitive set shifting. 380, 69–72. Nature Neuroscience, 1, 80–84. Dias, R., Robbins, T. W., & Roberts, A. C. (1997). Dissociable MacDonald, A. W., Cohen, J. D., Stenger, V. A., & Carter, C. S. forms of inhibitory control within prefrontal cortex with an (2000). Dissociating the role of the dorsolateral prefrontal analog of the Wisconsin Card Sort Test: Restriction to novel and anterior cingulate cortex in cognitive control. Science, situations and independence from ‘‘on-line’’ processing. 288, 1835–1838. Journal of Neuroscience, 17, 9285–9297. Mangun, G. R. (1995). Neural mechanisms of visual selective Dove, A., Pollman, S., Schubert, T., Wiggins, C. J., & attention. Psychophysiology, 32, 4–18. von Cramon, D. Y. (2000). Prefrontal cortex activation Mangun, G. R., Buonocore, M. H., Girelli, M., & Jha, A. P. in task switching: An event-related fMRI study. Brain (1998). ERP and fMRI measures of visual spatial selective Research. Cognitive Brain Research, 9, 103–109. attention. Human Brain Mapping, 6, 383–389. Eimer, M. (1993). Effects of attention and stimulus probability Meiran, N. (1996). Reconfiguration of processing mode prior to on ERPs in a Go/Nogo task. Biological Psychology, 35, task performance. Journal of Experimental Psychology: 123–138. Learning, Memory, and Cognition, 22, 1423–1442. Eimer, M. (1995). Event-related potential correlates of transient Meiran, N. (2000). Modelling cognitive control in attention shifts to color and location. Biological Psychology, task-switching. Psychological Research, 63, 234–249. 41, 167–182. Meiran, N., Chorev, Z., & Sapir, A. (2000). Component Gehring, W. J., & Knight, R. T. (2000). Prefrontal-cingulate processes in task switching. Cognitive Psychology, 41, interactions in action monitoring. Nature Neuroscience, 3, 211–253. 516–520. Miniussi, C., Wilding, E. L., Coull, J. T., & Nobre, A. C. (1999). Gerloff, C., Uenishi, N., & Hallett, M. (1998). Cortical Orienting attention in time. Modulation of brain potentials. activations during fast repetitive finger movements in Brain, 122, 1507–1518. humans: Dipole sources of steady-state movement-related Monsell, S., Yeung, N., & Azuma, R. (2000). Reconfiguration cortical potentials. Journal of Clinical Neurophysiology, of task-set: Is it easier to switch to the weaker task? 15, 502–513. Psychological Research, 63, 250–264. Gopher, D., Armony, L., & Greenshpan, Y. (2000). Switching Nobre, A. C., Allison, T., & McCarthy, G. (1998). Modulation of tasks and attention policies. Journal of Experimental human extrastriate visual processing by selective attention to Psychology: General, 129, 308–339. colours and words. Brain, 121, 1357–1368. Heinze, H. J., Hinrichs, H., Scholz, M., Burchert, W., & Mangun, Oldfield, R. C. (1971). The assessment and analysis of G. R. (1998). Neural mechanisms of global and local handedness: The Edinburgh inventory. Neuropsychologia, processing. A combined PET and ERP study. Journal of 9, 97–113. Cognitive Neuroscience, 10, 485–498. Owen, A. M., James, M., Leigh, P. N., Summers, B. A., Marsden, Heinze, H. J., Mangun, G. R., Burchert, W., Hinrichs, H., C. D., Quinn, N. P., Lange, K. W., & Robbins, T. W. (1992). Scholz, M., Munte, T. F., Gos, A., Scherg, M., Johannes, S., Fronto-striatal cognitive deficits at different stages of Hundeshagen, H., Gazzaniga, M. S., & Hillyard, S. A. (1994). Parkinson’s disease. Brain, 115, 1727–1751. Combined spatial and temporal imaging of brain activity Owen, A. M., Roberts, A. C., Hodges, J. R., Summers, B. A., during visual selective attention in humans. Nature, 372, Polkey, C. E., & Robbins, T. W. (1993). Contrasting 543–546. mechanisms of impaired attention: Set-shifting in patients Hillyard, S. A., Vogel, E. K, & Luck, S. J. (1999). Sensory with frontal lobe damage or Parkinson’s disease. Brain, 116, gain control (amplification) as a mechanism of selective 1159–1175.

Rushworth, Passingham, and Nobre 1149

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021 Owen, A. M., Roberts, A. C., Polkey, C. E., Sahakian, B. J., mechanisms of task control following frontal lobe damage & Robbins, T. W. (1991). Extra-dimensional versus and Parkinson’s disease. Brain, 121, 815–842. intra-dimensional set shifting performance following Rushworth, M. F. S., Hadland, K. A., Paus, T., & Sipila, P. K. frontal lobe excisions, temporal lobe excisions, or (2002). The role of the human medial frontal cortex in task amygdalohippocampelectomy in man. Neuropsychologia, switching: A combined fMRI and TMS study. Journal 29, 993–1006. of Neurophysiology, 87, 2577–2592. Pardo, J. V., Pardo, P. J., Janer, K. W., & Raichle, M. E. Rushworth, M. F. S., Paus, T., & Sipila, P. K. (2001). Attention (1990). The anterior cingulate cortex mediates processing systems and the organization of the human parietal cortex. selection in the Stroop attentional conflict paradigm. Journal of Neuroscience, 21, 5262–5271. Proceedings of the National Academy of Sciences, U.S.A., Sohn, M.-H., Ursu, S., Anderson, J. R., Stenger, V. A., & Carter,

87, 256–259. C. S. (2000). The role of prefrontal cortex and posterior Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/14/8/1139/1757675/089892902760807159.pdf by guest on 18 May 2021 Passingham, R. E. (1972). Non-reversal shifts after selective parietal cortex in task switching. Proceedings of the prefrontal ablations in monkeys (Maccaca mulatta). National Academy of Sciences, U.S.A., 97, 13448–13453. Neuropsychologia, 10, 17–26. Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic Passingham, R. E. (1998). Attention to action. In A. C. Roberts, atlas of the human brain. Stuttgart, Germany: Thieme. T. W. Robbins, & L. Weiskrantz (Eds.), The prefrontal cortex: Toni, I., & Passingham, R. E. (1999). Prefrontal-basal ganglia Executive and cognitive functions (pp. 131–143). Oxford, pathways are involved in the learning of visuomotor UK: Oxford University Press. associations: A PET study. Experimental Brain Research, Paus, T., Petrides, M. J., Evans, A. C., & Meyer, E. (1993). 127, 19–32. Role of the human anterior cingulate cortex in the Turken, A. U., & Swick, D. (1999). Response selection in the control of oculomotor, manual, and speech responses: human anterior cingulate cortex. Nature Neuroscience, 2, A positron emission tomography study. Journal of 920–924. Neurophysiology, 70. Valdes-Sosa, M., Bobes, M. A., Rodriguez, V., & Pinilla, T. Pollman, S. (2001). Switching between dimensions, location, (1998). Switching attention without shifting the spotlight: and responses: the role of the left frontopolar cortex. Object based attentional modulation of brain potentials. Neuroimage, 14, S118–S124. Journal of Cognitive Neuroscience, 10, 137–151. Rogers, R. D., Andrews, T. C., Grasby, P. M., Brooks, D. J., & Wise, S. P., & Murray, E. A. (2000). Arbitrary associations Robbins, T. W. (2000). Contrasting cortical and subcortical between antecedents and actions. Trends in Neuroscience, activations produced by attentional-set shifting and reversal 23, 271–276. learning in humans. Journal of Cognitive Neuroscience, 12, Wylie, G., & Allport, A. (2000). Task switching and the 142–162. measurement of ‘‘switch costs.’’ Psychological Research, Rogers, R. D., & Monsell, S. (1995). Costs of predictable switch 63, 212–233. between simple cognitive tasks. Journal of Experimental Yamaguchi, S., Yamagata, S., & Kobayashi, S. (2000). Psychology: General, 124, 207–231. Cerebral asymmetry of the ‘‘top-down’’ allocation of Rogers, R. D., Sahakian, B. J., Hodges, J. R., Polkey, C. E., attention to global and local features. Journal of Kennard, C., & Robbins, T. W. (1998). Dissociating executive Neuroscience, 20, RC72.

1150 Journal of Cognitive Neuroscience Volume 14, Number 8

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/089892902760807159 by guest on 28 September 2021