OUP user menu

A componential analysis of task‐switching deficits associated with lesions of left and right frontal cortex

Adam R. Aron, Stephen Monsell, Barbara J. Sahakian, Trevor W. Robbins
DOI: http://dx.doi.org/10.1093/brain/awh169 1561-1573 First published online: 16 April 2004


Executive functions such as task‐set switching are thought to depend on the frontal cortex. However, more precision is required in identifying which components of such high‐level processes relate to which, if any, subregions of the brain. In a recent study of 19 patients with focal right frontal (RF) lesions and 17 with left frontal (LF) lesions, we found that response inhibition, as measured by the stop‐signal task, was specifically disrupted by damage to the right inferior frontal gyrus (IFG). The present study examined task‐switching performance in this same group of patients and in matched controls on the grounds that inhibitory mechanisms may also be required to switch task‐set. Both RF and LF patients showed significantly larger switch costs (the difference, in reaction time and errors, between changing tasks and repeating the same task) than controls, but apparently for different reasons. For RF patients, a part of the switch deficit could be accounted for by impaired inhibition of inappropriate responses or task‐sets triggered by stimuli, and one measure of the switch cost correlated reliably with damage to the IFG, specifically the pars opercularis (POp). For LF patients, a part of the switch deficit may have arisen from weak top‐down control of task‐set. The degree of top‐down control correlated reliably with the extent of damage to the left middle frontal gyrus (MFG). This study localizes two components of the complex task‐switching process (inhibition of task‐sets and/or responses and top‐down control of task‐set) to the right IFG/POp and the left MFG respectively.

  • Keywords: inhibition; executive function; middle frontal gyrus; inferior frontal gyrus; pars opercularis
  • Abbreviations: C = congruent trial; CCNRP = Cambridge Cognitive Neuroscience Research Panel; exPOp = IFG regions excluding POp; IC = incongruent trial; IFG = inferior frontal gyrus; LF = left frontal; MED = medial; MFG = middle frontal gyrus; N = neutral trial; NART = National Adult Reading Test; NS = non‐switch trial; ORB = orbitofrontal; PFC = prefrontal cortex; POp = pars opercularis; RF = right frontal; ROI = region of interest; RSI = response–stimulus interval; RT = reaction time; SC = switch cost; SFG = superior frontal gyrus; S–R = stimulus–response; SSRT = stop‐signal reaction time; SW = switch trial; ROI = region of interest


Two widely studied executive functions are response inhibition and task‐set reconfiguration. The former can be studied with the stop‐signal paradigm (Logan and Cowan, 1984), in which the stimulus specifies a rapid response which the participant makes on most trials, but must try to suppress if a stop‐signal is concurrently presented. Task‐set reconfiguration has been studied with variants of a task‐switching paradigm (for review see Monsell, 2003) in which the participant responds to a stimulus on each trial according to some stimulus–response (S–R) task rule(s), and on some trials must change to a different task. To perform any such task, a participant must chain together and configure appropriately a task‐set: an appropriate set of processes linking sensory analysis via the categorization or identification of the stimulus to the choice of a response and execution of motor output (Rogers and Monsell, 1995). To change tasks, one or more components of the task‐set must be reconfigured. Behaviourally, the need to reconfigure the task‐set results in a substantial switch cost: longer reaction time (RT) and more errors on task‐switch than on task‐repeat trials (Fig. 1A). A number of authors have suggested that the switch cost arises in part from the need to inhibit competing S–R links specified by the now inappropriate task (Rogers and Monsell, 1995), or to inhibit entire task‐sets (Rogers and Monsell, 1995; Arbuthnott and Frank, 2000; Mayr and Keele, 2000; Mayr, 2002; Schuch and Koch, 2003). A primary motive for this study was to explore the neural correlates of these putative inhibitory processes in task‐switching by assessing behavioural performance in patients with unilateral lesions of the frontal cortex whom we had also tested, in the same sitting, on a specific measure of response inhibition (c.f. Aron et al., 2003).


Fig. 1 The task design and the run‐position effect. (A) Data for the three subject groups from this experiment illustrate the run‐position effect (Rogers and Monsell, 1995). Run‐position 1 (the switch) has elevated RT relative to non‐switch positions 2 and 3. (B) An AAABBB design was employed whereby the relevant task (arrow or word) changed every three trials. The position labels ‘1’, ‘2’ and ‘3’ were not shown to the subject, but are depicted here for explanatory purposes. The position of the thick bar denotes the position of the switch (run position 1). On each trial a cue (‘arrow’ or ‘word’) appeared, followed by the stimulus. On the subsequent trial the cue (and stimulus) appeared at the next, clockwise, position. Stimuli could be congruent (e.g. a left arrow with the word ‘left’, because it specified a left response for both the Word and the Shape tasks), incongruent (e.g. a left arrow with the word ‘right’, because it specified a left response on the shape task but a right response on the other task) or neutral (e.g. where the shape around the word was a rectangle or the letter string within an arrow was ‘xxxx’, so that the irrelevant attribute was associated with no response).

There are several reasons for supposing that the right frontal (RF) cortex may subserve inhibitory processes underlying task‐switching. First, many neuroimaging studies of response inhibition (Konishi et al., 1998; Garavan et al., 1999; Konishi et al., 1999; Rubia et al., 1999; Menon et al., 2001; Bunge et al., 2002; Rubia et al., 2003) and a number of neuroimaging studies of switching task, strategy or dimensions, including reversal‐learning (Nagahama et al., 2001; Cools et al., 2002), the Wisconsin Card Sorting Test (WCST) (Monchi et al., 2001; Nagahama et al., 2001; Nakahara et al., 2002) and task‐set switching (Dove et al., 2000; Sohn et al., 2000; Dreher and Berman, 2002; Brass et al., 2003) have especially, although not exclusively, reported activation, within the right hemisphere, of the inferior frontal gyrus (IFG) (Fig. 2). Secondly, a direct neuroimaging comparison of a form of switching (the WCST) and response inhibition demonstrated a common locus in the right IFG (Konishi et al., 1999). Thirdly, a combined EEG/functional MRI study investigating a Go versus Wait factor and a Switch versus Non‐switch factor suggested that the right IFG locus was related neither to pure switching nor to pure response inhibition, but was instead responsible for ‘switching into a suppression mode’ (Swainson et al., 2003). Although these neuroimaging findings indicate neural substrates activated by these executive functions, human lesion studies can provide more definitive proof that a given brain region is necessary. Recently, we established that the right IFG is indeed necessary for response inhibition by using a region of interest (ROI) MRI‐based method in a sample of patients with excisions of the frontal cortex: the greater the damage to that ROI alone, the longer it took the patients to inhibit their responses (Aron et al., 2003). The present study aimed to extend this approach, in the same group of patients, to task‐set switching in order to test the hypothesis of a common right IFG substrate underlying both task‐set switching and response inhibition executive functions, by assessing whether any switching deficit may plausibly relate to (i) inhibition of an inappropriate task‐set and/or (ii) inhibition of response tendencies activated via an inappropriate S–R rule.


Fig. 2 Talairach coordinates plotted from six neuroimaging studies of switching/sorting/reversing, and boundaries of inferior frontal gyrus (IFG). 3D‐rendered sagittal view showing that several reported task‐set switching, Wisconsin Card Sorting Test and reversal foci fall within the inferior frontal gyrus, defined caudally by the precentral sulcus, ventrally by the sylvian fissure and rostrally by the inferior frontal sulcus. All lateral right frontal coordinates are shown as colour codes: yellow, Cools et al. (2002); blue, Dreher and Berman (2002); purple, Sohn et al. (2000); red, Nakahara et al. (2002); black, Dove et al. (2000); brown, Monchi et al. (2001); green, Brass et al., (2003).

How a task‐set switching experiment may be used to isolate these putative inhibitory factors requires explanation. First, it appears that a stimulus evokes tendencies to perform both a task recently or frequently associated with it and specific responses associated with the stimulus (Rogers and Monsell, 1995; Allport and Wylie, 1999). Switching between tasks may require inhibition at the level of the task‐set and at the level of individual responses. The degree to which such inhibition is effective at both levels may be revealed by comparisons of congruent, incongruent and neutral stimuli in a task‐switching design (Fig. 1B). As congruent and incongruent stimuli are equally associated with the two tasks, any difference in response latencies to them (Stroop‐like interference) must reflect competition due to activation of the irrelevant response. It is also sometimes observed (Rogers and Monsell, 1995) that congruent stimuli (which by definition create no competition at a response level, but are associated with both tasks) are responded to more slowly than neutral stimuli (which are associated with only one task). This can only reflect competition from activation of the irrelevant task‐set. Hence a slower RT for congruent than for neutral trials indexes difficulty in inhibiting the irrelevant task‐set, while a slower RT for incongruent than congruent trials indexes difficulty in inhibiting response tendencies activated via the irrelevant task‐set. As interference from a recently performed task may be observed even in a pure (i.e. non‐switching) task block (Wylie and Allport, 2000), we examined two measures that capture this competition on non‐switch trials alone:



where the subscript NS refers to non‐switch trials and IC, C and N refer to incongruent, congruent and neutral stimuli respectively. Furthermore, as interference effects are typically magnified on switch trials relative to non‐switch trials (Rogers and Monsell, 1995; Meiran, 2000; Meiran et al., 2000) we also examined two measures that capture this more transient competition from the task being switched from:



where the subscripts SW and SC refer to switch trials and the switch cost respectively. The C–N and IC–C measures are not independent, as more activation of the competing task set enables more activation of corresponding response tendencies, and this may facilitate congruent responses. However, a positive SCC–N implies greater activation of the competing task set on switch relative to non‐switch trials. A larger SCIC–C implies greater difficulty in suppressing response activation generated by the irrelevant S–R mapping on switch relative to non‐switch trials. If RF damage affects an inhibitory mechanism generalizable to task‐set switching, RF patients may be expected to show impairments on either or both measures.

We also varied the response–stimulus interval (RSI) between 100 ms (short) and long 1500 ms (long) to assess deficits in voluntary reconfiguration. The standard observation in normal participants is that, as the time available for preparation for the next task (the RSI in the present experiment) increases, switch costs decline but are not eliminated. The reduction has been attributed to endogenous processes that enable the required task‐set and/or suppress the previous task‐set and are carried out in anticipation of the stimulus (Rogers and Monsell, 1995; Meiran, 1996; De Jong, 2000; Rubinstein et al., 2001). Such preparation can be indexed by subtracting the switch cost at long RSI from the switch cost at short RSI (we refer to this difference as SCREDUCT). Interpretation of the residual switch still seen after a long preparation interval (here designated SCRESID and estimated simply as the switch cost at the long RSI) is controversial (see Monsell, 2003). Suggested sources include an exogenous control process that is required to complete task‐set reconfiguration and triggered by the stimulus onset (Rogers and Monsell, 1995; Rubinstein et al., 2001), post‐stimulus completion, on a proportion of trials, of a reconfiguration process that failed to engage endogenously (De Jong, 2000), interference with response selection due to task‐set inertia (Allport et al., 1994; Allport and Wylie, 1999) or, more specifically, persisting task‐set inhibition (Meuter and Allport, 1999; Goschke, 2000; Schuch and Koch, 2003). If switching deficits in patients relate to endogenous control, they would be expected to interact further with RSI, while deficits that are equally apparent at short and long RSI are likely to derive from processes triggered by stimulus onset and/or carry‐over of interference from the pre‐switch trial, which the participant cannot pre‐empt by active preparation.

Another motive for the study was to investigate the consequences of LF damage for task‐set switching and to compare these with the consequences of RF damage. A number of neuropsychological studies suggest that LF cortical damage may affect task‐set switching (Stablum et al., 1994; Rogers et al., 1998; Mecklinger et al., 1999; Keele and Rafal, 2000). Mecklinger et al. (1999) found that patients with speech and language disorders had the greatest switch deficits, suggesting left hemisphere linguistic regions (e.g. the IFG) may be critical. Rogers et al. (1998) found that patients with LF damage had switch deficits only under conditions in which there was interference between the tasks (i.e. where the establishment of task‐set was particularly important). Neuroimaging studies suggest that maintenance and establishment of the task‐set are functions attributable to the left dorsolateral prefrontal cortex (PFC) (especially the middle frontal gyrus, MFG) (MacDonald et al., 2000; Garavan et al., 2002). Keele and Rafal (2000) reported that damage to this same area led to an inability to recover from a task switch with the normal reduction in RT on the second or third trial following the switch in a single, extensively tested, LF patient. In brief, the role of the LF cortex in task‐switching is not particularly clear, perhaps owing to the fact that none of the neuropsychological studies employed particularly sizeable samples, nor could the locus of damage be specified with much precision. We therefore examined the same measures in LF as in RF patients. Specific difficulties inhibiting the inappropriate task‐set would be expected to show up in the C–N and IC–C measures, while more general difficulties imposing a changed task‐set would be expected to result in an abnormally large switch cost or other indices of recovery from a change of tasks. Additionally, the putative importance of the left IFG, MFG or other ROIs might be revealed by reliable correlations between extent of damage to these regions and indices of switching and task control.

Material and methods


Thirty‐six patients were recruited from the Cambridge Cognitive Neuroscience Research Panel (CCNRP) by referral from Addenbrooke’s Hospital, Cambridge. The study was approved by the Cambridge Local Research Ethics Committee and all patients gave informed consent prior to participation. Seventeen patients had a single focal lesion confined to the left PFC and 19 to the right PFC, verified by MRI in 33 of the patients and computer‐assisted tomography (CAT) in three. Lesion aetiology was mostly tumour resection or cerebrovascular haemorrhage (Supplementary Table 1). We excluded patients with current or previous psychiatric diagnosis, colour blindness, or neurological disease other than that determining inclusion in the study. Twenty healthy control volunteers from the East Anglia area were obtained either through advertisement or through the CCNRP and were paid. Controls were matched with patients for age and estimated premorbid verbal IQ as assessed by the National Adult Reading Test (NART) (Table 1). There were no differences between groups in terms of age [F(2,53) < 1, n.s.], NART score [F(2,51) < 1, n.s.] or time since trauma for the two frontal patients groups [F(1,34) < 1, n.s].

View this table:
Table 1

Demographic information on study participants

Sample size201719
Hand (L : R) 19 : 116 : 118 : 1
Sex (M : F)12 : 810 : 77 : 12
Age 52.6 (10.8)51.9 (10)53.6 (10.0)F < 1, n.s.
PV‐IQ115.9 (5.1)116 (5.1)113.2 (8.5)F = 1.1, n.s.
Chronicity 3.1 (2.5)3.2 (3.4)F < 1, n.s.
Lesion size 31.2 (38.1)63.5 (62.3)t = 1.6, n.s.
SFG6.8 (11.9)14.7 (18.0)t = 1.5, n.s.
MFG7.9 (10.3)15.9 (16.6)t = 1.6, n.s.
IFG6.5 (9.4)8.7 (10.7)t < 1, n.s.
ORB4.3 (6.8)5.9 (8.5)t < 1, n.s.
MED1.7 (2.8)3.8 (6.3)t = 1.2, n.s.

PV‐IQ = predicted verbal IQ from the National Adult Reading Test; chronicity = years since frontal trauma; lesion size = total volume of lesion (cc) and average volume of damage to each of the five ROIs (cc). There were no reliable differences between left and right frontal patients in the extent of damage to prefrontal regions of interest.

Neuroradiological assessment

Thirty‐three of the 36 frontal excision patients received MRI scans of the brain, with 3D set acquisition in the coronal plane using a SPGR (spin gradient echo) T1‐weighted sequence and a T2‐weighted axial sequence (using the 1.5 T scanner at the MRIS unit, Addenbrooke’s Hospital, Cambridge). MRI scans were interpreted for the CCNRP by two neurologists blind to the experimental results, and lesions were traced using MRIcro software (www.mricro.com). They were then normalized to a standard template using SPM96 (Wellcome Department of Cognitive Neurology, London) using cost function masking (Brett et al., 2001). For the other three patients, MRI scans could not be obtained and the lesion loci were estimated from CAT scans.

ROI method

The frontal lobes of each hemisphere were divided into five ROIs by Dr P. C. Fletcher (Department of Psychiatry, University of Cambridge), who was blind to experimental results. These ROIs were: superior frontal gyrus (SFG), middle frontal gyrus (MFG), inferior frontal gyrus (IFG), orbitofrontal (ORB) and a medial area (MED) (for figure, see Aron et al., 2003). MRIcro was used to trace these areas onto the standard T1 template used by SPM99 (consisting of averaged scans from 152 healthy subjects). Because it was shown that damage to a subregion of the right IFG, the pars opercularis (POp) [in Aron et al. (2003) we erroneously referred to the pars opercularis as the pars triangularis], was particularly critical for response inhibition (Aron et al., 2003), an additional POp ROI was used from the Automated Anatomical Labelling (AAL) map (Tzourio‐Mazoyer et al., 2002). For control purposes, non‐POp regions of the IFG (i.e. pars triangularis and an orbital region from the AAL map) were combined into an exPOp region (IFG excluding POp). For each hemisphere, the volumes of these ROIs (in cc) were 104.1 (SFG), 68.8 (MFG), 49.2 (IFG), 39.7 (ORB), 56.1 (MED), 12.5 (POp) and 33.1 (exPOp). The normalized lesion for each patient was superimposed onto each of these ROIs in order to compute the volume of damaged grey matter. Although the RF group had more overall damage than the LF group, this was not a significant difference, nor were there significant differences for any of the ROIs (Table 1).

Tasks and procedure

The experiment was run on a PC using ERTS (Frankfurt, Germany), an MS‐DOS program with 0.6 ms timing resolution. Subjects sat 50 cm from a computer screen on which was displayed a framework consisting of three lines radiating 10 cm from the centre at equal angles to form an inverted Y defining three sectors, in one of which the stimulus was displayed about 25 mm from the centre. Stimuli on successive trials were displayed in successive sectors, clockwise (Fig. 1B). Immediately after the previous response, a task cue (the word ‘arrow’ or ‘word’) was displayed about 14 mm above the position in which the next stimulus was then displayed after an RSI of 1500 or 100 ms, varied between blocks (as described below). The cue word, and hence the task, changed every three trials. The position associated with a task switch in that block was additionally indicated by the corresponding limb of the inverted Y being a thicker bar. Hence task switches were predictable (every three trials) and the task was in addition redundantly indicated by the cue word and its location. The position of the switch trial was counterbalanced for each participant (between blocks). Each stimulus remained on the screen until the subject responded. If the participant made an incorrect response, a beep of 200 Hz sounded for 200 ms and the RSI was extended by 2000 ms.

‘Left’ or ‘right’ responses were made with the index and middle fingers of the dominant hand. For the Word task, stimuli were composed of a word (‘left’ or ‘right’) inside a shape (left arrow, right arrow or rectangle). For the Arrow task, stimuli were composed of either a left or right arrow shape surrounding a letter string (‘LEFT’, ‘RIGHT’ or ‘XXX’). For each task, each stimulus was used once in a block in each run position in a random sequence, resulting in three equally frequent congruency conditions: congruent (e.g. a left arrow with the word ‘left’), incongruent (e.g. a left arrow with the word ‘right’) and neutral (the shape around the word was a rectangle or the letter string was ‘XXX’, so that the irrelevant attribute was associated with no left or right response).

Each block contained 36 experimental trials, one for each combination of task (Word, Arrow), run position (1, 2 or 3), response (left, right) and congruency (incongruent, neutral or congruent). In addition, there were one, two or three warm‐up trials at the beginning of each block. Subjects were encouraged to minimize RT while avoiding errors. After each block, the computer displayed a feedback graph of mean RT and error rate so the subject could track performance relative to prior blocks. An instruction screen was then displayed for the next block indicating the identity of the first task, e.g. ‘Word’, and the RSI (short or long), and reminding the subject to respond as quickly and accurately as possible.

Subjects were initially given practice in four single‐task blocks, two for the word task alternating with two for the arrow task, each preceded by an instruction screen explaining the relevant response mappings for that task. In the practice blocks, cues and stimuli were displayed in the successive positions indicated by the inverted Y framework, but without the thick line which later demarcated the switch position. The RSI was 1000 ms. If the subject made an error, a beep was sounded and the RSI was extended by 2000 ms. There followed a demonstration of the switching concept, followed by one block of practice switching tasks for the long (1500 ms) and then the short (100 ms) RSI. The eight blocks of the experiment proper consisted of alternating blocks with long and short RSIs, starting with the long. The thickened limb of the inverted Y was initially upright, and moved one position clockwise after each block.

Analysis of performance

Mean correct RTs and error rates were computed for each cell (task × run position × RSI × congruence) excluding practice trials, warm‐up trials, trials following an error on either of the preceding two trials, and RTs <300 or >4000 ms. Inspection of plots of RT at each run position for each group (Fig. 1A) showed that reduction in RT for positions two and three followed roughly parallel lines for each group [a test of the interaction of run position (for positions 2 and 3) and group was non‐significant; F(2,53) < 1]. Therefore, switch costs for RT and errors were calculated by subtracting the average of trials at run positions two and three from the average of trials at run position one for each combination of RSI and congruency (Table 2). Although patients responded more slowly overall than controls, this was not a significant difference [F(2,53) = 1.7], nor was there a reliable difference at long RSI (when patients appeared particularly slow) [F(2,53) = 2.7, P > 0.05]. Although it thus appeared that patient and control groups could be contrasted for the switching measures, without radical correction for overall slowing, conclusions were checked with respect to proportional measures for crucial effects. Planned pairwise comparisons of controls, RF and LF patients were performed for the relevant measures (significance threshold, P < 0.05, two‐tailed). Reported analyses were no different from those performed with age as a covariate.

View this table:
Table 2

Arithmetic means of RT data and proportion of errors for the task‐set switching experiment

Reaction time
Short RSI
Long RSI
Error rate
Short RSI
Long RSI

NS = non‐switch; SW = switch; SC = switch cost (SW – NS).


Behavioural results

Control subjects

RT and error rates for control subjects were much as expected. Comparing short and long RSI, controls achieved a substantial reduction (from 153 to 39 ms) in the RT switch cost (a 74% reduction in absolute cost, a 57% reduction in cost expressed as a proportion of non‐switch baseline) with preparation (Fig. 3A). Most errors occurred on incongruent stimuli (Table 2), but switching caused only a modest increase in error rate, and this was true at both short and long RSI (Fig. 3B, C).


Fig. 3 Mean (and standard error) of the correct RT and error switch cost variables for controls (light grey), left frontals (darker grey) and right frontals (black). Numbers above bars represent proportional measures in percentages. The top row shows RT indices; the bottom row shows proportion of errors. (A) SCREDUCT (switch cost at short RSI minus switch cost at long RSI). (B) SCshortRSI (switch cost at short RSI); the proportional measure is SCshortRSI/non‐switch_short_RSI. (C) SCRESID (residual switch cost: switch cost at long RSI); the proportional measure is SCshortRSI/non‐switch_long_RSI. (D) SCC–N (switch cost for congruent trials minus switch cost for neutral trials). (E) SCIC–C, shortRSI (switch cost for incongruent trials minus switch cost for congruent trials at short RSI only).

Short RSI

The difference between the LF and the RF groups was apparent largely at the short RSI, when there was almost no time to recover/prepare between each response and the next stimulus (Fig. 3B). Here the LF group, though not particularly slow on non‐switch trials (Table 2) was very slow on switch trials, showing the largest RT cost both for absolute and proportionate switch cost. For absolute cost (SCshort RSI), the LF group had a significant deficit compared with controls [F(1,35) = 7.4, P < 0.01] and a marginally significant deficit compared with the RF group [F(1,34) = 3.0, P = 0.09]. For proportional cost, the LF group had a significant deficit compared with controls [F(1,35) = 6.5, P < 0.05], and a marginally significant deficit compared with the RF group [F(1,34) = 3.4, P = 0.07]. Although abnormally slow on switch trials, the LF group was not particularly error prone on switch trials; the error switch cost was about the same as for the control group (Fig. 3A–E, lower row).

At short RSI, the RF group, though showing somewhat greater (but not reliably so) RT costs than controls [F(1,37) = 1.7, n.s.], exhibited a dramatically elevated error rate on switch trials and hence a large switch cost in error rate (Fig. 3B). This error deficit was reliable compared with both controls [F(1,37) = 6.3, P < 0.05] and the LF group [F(1,34) = 5.2, P < 0.05]. This resulted primarily from incongruent stimuli; the RF group made errors on 30% of incongruent stimuli on switch trails, twice as many as on non‐switch trials (Table 2). By contrast, both controls and the LF group, while making the majority of their errors on incongruent stimuli, showed only a modest (and similar) increase in these errors on switch trials (Table 2). The unusual amplification of the switch cost for incongruent stimuli in the RF group was particularly severe at short RSI, as captured by the contrast SCIC–C(short RSI) (Fig. 3E). For the error measure, this was significantly greater than for controls [F(1,37) = 5.4, P < 0.05] and the contrast with the LF group was marginally reliable [F(1,34) = 3.9, P = 0.06]. This measure interacted further with RSI, so that SCIC–C(short RSI) was significantly greater than SCIC–C(Long RSI) for the RF group compared with controls [F(1,37) = 4.1, P = 0.05]. The amplification of the switch cost at short RSI for incongruent stimuli also occurred for the RT data for the RF group, reliably more than for the LF group [F(1,34) = 4.7, P < 0.05], but not significantly more than for the control group [F(1,37) = 1.0, n.s.]. Again for RT, there was a further interaction with RSI so that SCIC–C(short RSI) was greater than SCIC–C(Long RSI) for the RF group compared with the LF group, and this was marginally reliable [F(1,34) = 2.8, P = 0.10]. Therefore, relative to the LF group, the RF group appeared to have a particular problem suppressing the irrelevant response or the irrelevant task‐set when there was little time between trials. At long RSI, when the RF group did have time to prepare/recover between trials they did not look different from the LF group.


At the long RSI, when the pace of testing was relatively leisurely, both patient groups had about three times the RT switch cost (SCRESID) of the control group (and about 2.5 times the proportional switch cost). For absolute RT the difference in SCRESID was reliable for both LF versus controls [F(1,35) = 10.3, P < 0.01] and RF versus controls [F(1,37) = 4.1, P < 0.05], but for error rate the patient groups did not show reliably greater switch costs than controls [F < 1 for both contrasts] (Fig. 3C). For proportional RT, the difference in SCRESID was reliable for both LF versus controls [F(1,35) = 6.4, P < 0.05] and marginally reliable for RF versus controls [F(1,37) = 3.3, P = 0.08]. For both RT and errors, there was little difference between the LF and RF groups at the long RSI.

Although the deficit in SCRESID for patients was large, its meaning is not immediately clear. One possibility was that the patients simply used the long RSI less effectively to prepare for an upcoming change of task. However, an analysis of the reduction of the switch cost with an increase in RSI (SCREDUCT) provided little support for this (Fig. 3A). In fact, the LF group had at least as large an absolute and proportional reduction in switch cost compared with controls for RT, while the RF group were only mildly (and non‐significantly) different from controls on this measure[F(1,37) < 1, n.s.]. With respect to SCREDUCT for error rate, the LF group was very similar to controls, while the RF group showed a greater reduction, which was marginally significant compared with both the LF group [F(1,34) = 2.9, P = 0.10] and controls [F(1,37) = 3.9, P = 0.07]. Therefore, failure to prepare during the interval was unlikely to explain the residual switch cost deficit of either patient group.

LF group (C–N and IC–C contrasts)

A clue to the greater switch costs at both short and long RSI for the LF group was provided by the C–N and IC–C contrasts. Compared with controls, and for RT, the LF group had a greater (although non‐significant, F < 1) C–N effect for non‐switch trials (Table 2), and the size of this effect was greater on switch trials (Fig. 3D), although not significantly so compared with controls (F < 1). As a neutral stimulus is associated with only one task, while a congruent stimulus is associated with two, a larger C–N effect suggests a greater exogenous task‐cueing effect of the stimulus, possibly induced by weaker ‘endogenous control input’ consequent upon LF damage. This could explain their greater switch costs at both short and long RSI, in the sense that weaker endogenous control of the task‐set would lead to a greater influence of exogenous cueing. If the LF group did have weaker endogenous control, this would mean less efficient biasing of the appropriate S–R links. Consistent with this explanation, the LF group had a notably larger IC–C effect even on non‐switch trials [marginally reliable compared with controls; F(1,35) = 2.4, P = 0.1].

Interim summary

(i) For RT, at short RSI, the RF group had a larger (although non‐significant) switch cost than controls, while at long RSI this was a statistically reliable switch deficit.

(ii) For error rate, the RF group had a significantly greater switch cost for incongruent than congruent trials (SCIC–C(short RSI)) than both controls and the LF group (and a significant effect compared with the LF group for RT). Apparently the RF group had difficulty in reactively suppressing inappropriate responses (or task‐sets), especially at short RSI.

(iii) For errors, the LF group did not have particular difficulty suppressing the inappropriate response on switch trials under speed stress (i.e. short RSIs).

(iv) Instead, for RT, the LF group showed a more general difficulty in imposing the appropriate task set, as indexed by significantly larger switch costs at both short and long RSIs (compared with controls), a marginally significant switch deficit at short RSI compared with the RF group, a larger interference effect (I–C) on all trials, and some suggestion of greater activation of the competing task set (the C–N contrast).

Correlational results

All principal RT and error indices examined above (14 total) were entered into a correlational analysis with the five ROIs [SFG, MFG, IFG, orbitofrontal (ORB) and MED] for RF and LF patients separately. To control for multiple comparisons (140 in total for both hemispheres), an alpha value of P < 0.005 (two‐tailed) was chosen so that the expected number of false positives was less than one (140 × 0.005 < 1). For error measures, there were no reliable correlations at this alpha level (although there were, in some cases, at a weaker level of P < 0.05, and the direction of correlations generally mimicked that for RT). Therefore, correlations are only reported for RT measures (Table 3).

View this table:
Table 3

Correlations between RT variables and volume of damage (cc) to regions of interest

LF group
 SCshort RSI0.290.25–0.02–0.240.01
 SCIC–C(short RSI)–0.04–0.43–0.44–0.14–0.23
RF group
 SCIC–C(short RSI)

Values shown are Pearson’s r. *Significance at P < 0.005 (two‐tailed), a threshold corrected for multiple comparisons.

RF group

The only significant correlations using the strict alpha level were those between IFG damage and SCRESID and between IFG damage and SCREDUCT. Correlations run using the more specific POp ROI were even stronger: POp damage correlated reliably with SCRESID (r = 0.836, P < 0.0001) (Fig. 4B) and with SCREDUCT (r = –0.794, P < 0.0001) (Fig. 4C). As the POp region is more specific and has been shown to be critical for response inhibition (Aron et al., 2003), further analyses were restricted to this rather than the wider IFG.


Fig. 4 The greater the damage to right IFG (specifically the pars opercularis, POp), the greater the disruption of measures of inhibition and switching. There were significant correlations between POp and (A) the stop‐signal reaction time (SSRT) measure of response inhibition reported for these same subjects (r = 0.65) (Aron et al., 2003), (B) SCRESID (residual switch cost) (r = 0.84) and (C) SCREDUCT = SCshortRSI – SClongRSI (r = –0.79). (D) Correlation between SSRT and SCRESID, r = 0.59.

The strength of the correlation between SCRESID and POp was significantly greater than the correlation between SCRESID and any other of the ROIs, with the exception of the MFG [P < 0.05, two‐tailed; test of non‐independent correlations (Williams, 1959)]. However, as damage to POp and MFG was significantly correlated (r = 0.59, P = 0.01), the possibility existed that the significant correlation between MFG and SCRESID was due to conjoint damage to POp. This interpretation was confirmed by using simultaneous multiple regression: damage to MFG did not significantly account for variability in SCRESID (n = 18, t < 1, n.s.) when controlling for damage to POp; but the POp still remained a reliable predictor of SCRESID (t = 4.1, P < 0.001) when controlling for damage to MFG. Similarly, the strength of correlation between SCREDUCT and POp was significantly greater than the correlation between SCREDUCT and any other ROI.

Results were also checked with respect to the exPOp region. The correlation between damage to POp and exPOp was r = 0.49 (P < 0.05). Only the correlation between exPOp and SCRESID was reliable (r = 0.539, P < 0.05). However, the correlation between POp and SCRESID was significantly greater than the correlation between exPOp and SCRESID. Therefore, concomitant damage to exPOp was not as important in producing the switch deficit as POp damage. Additionally, while simultaneously controlling for damage to exPOp, the correlation between POp and SCRESID was still statistically reliable (r = 0.78, P < 0.0001), as was that between POp and SCREDUCT (r = –0.77, P < 0.0001).

Importantly, as reported in a prior study of these same patients (Aron et al., 2003), damage to POp was also correlated with the stop‐signal RT (SSRT) measure of response inhibition (r = 0.65, P < 0.01) (Fig. 4A); greater SSRT implies weaker inhibition. Furthermore, the two behavioural measures, SSRT and SCRESID, were also significantly correlated (r = 0.59, P < 0.005) (Fig. 4D), consistent with a common causal factor—an impaired inhibitory mechanism. To exclude the possibility that the striking correlation between SCRESID and POp damage merely reflected generalized slowing of the patients with greatest damage, a scaled index was created: SCRESIDSCALED [(SWlongRSI – NSlongRSI)/NSlongRSI], where SW and NS are switch and non‐switch respectively. There was still a highly reliable correlation between SCRESIDSCALED and POp damage (r = 0.75, P < 0.0001). Finally, to explore the reliable negative correlation between POp damage and SCREDUCT, a plot was created overlaying SC measures at short and long RSI (Supplementary Fig. 1). It is evident that the five RF patients with greatest damage had SCshortRSI < SClongRSI, while the remainder of RF patients showed the normal pattern of SCshortRSI > SClongRSI. Further inspection of the raw data (Supplementary Table 2) suggested that subjects with the greatest POp damage made little improvement across run positions at the short RSI, but did improve for positions 2 and 3, relative to position 1, at long RSI. This pattern of results is discussed below.

LF group

For this group, there were no reliable correlations for any of the switch cost indices (for RT or errors) for any ROI, just as there were no reliable correlations with SSRT in the study by Aron et al. (2003). The only measures correlated with damage to a ROI were the C–N and (marginally) IC–C contrasts (for non‐switch trials), indicative of task‐set cueing and Stroop‐like interference. However, as scatter plots showed these correlations to derive largely from one patient, we re‐ran correlations for both these indices, at both RSIs, excluding this outlier. There were a number of reliable correlations between these indices and MFG damage at the strict alpha level (Fig. 5). MFG damage correlated reliably with IC–ClongRSI (r = 0.68, P < 0.007) and with C–NshortRSI (r = 0.75, P < 0.002). Therefore, the greater the MFG damage in the LF group, the greater was the tendency for stimuli to activate the competing task and the incongruent response even on non‐switch trials, suggesting weaker endogenous task‐set control.


Fig. 5 The greater the damage to the left middle frontal gyrus (MFG), the weaker the endogenous task control. Correlations are shown for the degree of MFG damage against measures of endogenous task control such as C–N and I–C at short and long RSI. (A) C–NshortRSI (r = 0.68). (B) C–NlongRSI. (C) IC–CshortRSI. (D) IC–ClongRSI (r = 0.75).


To verify that POp damage in the RF group was critical for the behavioural measures, we divided the RF group into those with POp damage (11 patients) and those without (seven patients); one RF patient had no MRI scan. There were significant differences (P < 0.05, one‐tailed) for SCREDUCT, and SCC–N for RT, and SCRESID and SCIC–C(short RSI) for errors. There were marginally significant effects on several other measures (Supplementary Table 3). This confirms that the presence of POp damage is critical for producing behavioural deficits for many of the measures used here.


Thirty‐six patients with unilateral lesions to the right or left PFC were assessed on a test of task‐set switching. A neuroradiological ROI method was used to correlate performance on the switch measures with extent of damage to specific subregions. As a group, RF patients had particular difficulty suppressing Stroop‐like interference from the just‐performed task on switch trials (but not post‐switch trials) at short RSIs. In contrast, as a group, the LF patients showed abnormally slow, but not error‐prone, switching at short RSIs. Both LF and RF groups showed exaggerated time costs of a switch at the long RSI (large residual costs). There may be a number of reasons for this, but there is evidence that the deficit is different for the LF and RF groups. First, although the LF group did not show the RF pattern of high error rates on incongruent stimuli on short‐RSI switch trials, they showed greater IC–C interference overall, which correlated reliably with MFG damage. Secondly, there was an elevated C–N effect for the LF group, which correlated with MFG damage, and this was increased on switch trials. This pattern of data is compatible with the LF group exerting weaker endogenous control consequent upon damage to MFG, resulting in a greater influence on task‐set activation of stimuli associated with the competing task set. This extends the specificity of prior neuropsychological data on left PFC contributions to task‐set reconfiguration. By contrast, the residual switch cost of the RF group was reliably correlated with more specific POp damage, was the measure of response inhibition (SSRT). The residual switch cost and SSRT were themselves reliably correlated. It is plausible therefore that right POp damage disrupts an inhibitory mechanism responsible for suppression of inappropriate responses and/or task sets in both stop‐signal (no‐go) and task‐switching contexts. These results have implications for understanding: (i) the separate contributions of the left and right frontal cortex to executive control, (ii) the specific role of the right POp, (iii) the fractionation of task‐switching into component mechanisms, and (iv) the neuropsychological sequelae of frontal cortical damage.

Components of task‐set

A person’s task‐set at any particular moment results from an interaction of task‐set inertia, exogenous task‐set activation and endogenous control input. Task‐set inertia arises from the persistence of activation/inhibition from the previous trial(s) (Allport and Wylie, 1999; Yeung and Monsell, 2003), and leads to biasing and/or interference on the current trial, especially when it is a switch trial. Exogenous task‐set activation is generated by the stimulus itself because stimuli activate task‐sets associated with them (Lhermitte, 1983; Monsell et al., 2001), and this is especially so on switch trials (Rogers and Monsell, 1995). To overcome both the inertia on a switch trial and the inappropriately activated task‐set, endogenous control is required. Endogenous control consists in top‐down input, which biases a task‐set by directing attention to a particular attribute, selecting a S–R rule, and so forth (Norman and Shallice, 1986; Gilbert and Shallice, 2002; Yeung and Monsell, 2003). However, it cannot be the case that endogenous control completely suppresses the inappropriate task‐set because we usually see IC–C interference due to activation of the inappropriate response tendency. Furthermore, several authors have documented that preparation (endogenous input) can substantially reduce switch costs without reducing IC–C interference (Rogers and Monsell, 1995; Meiran, 2000; Meiran et al., 2000; Monsell et al., 2001), though other studies have seen such a reduction (e.g. Goschke, 2000). Therefore, effective imposition of endogenous control does not prevent some activation of responses afforded by the current stimulus; arguably this activation can only be dealt with reactively, upon detection of conflict (Monsell et al., 2003).

Right frontal cortex and task‐switching

The present data suggest that the right PFC may be crucial to this reactive suppression of inappropriately activated responses, especially when the task set is weakly established (i.e. at short RSI). On this account, damage to the right PFC compromises this function, thus producing a striking switching deficit. For error rate, the RF group had a significantly greater switch cost for incongruent than congruent trials (SCIC–C(short RSI)) than both controls and the LF group (and a significant effect compared with the LF group for RT). For RT, at short RSI, the RF group had a larger (although non‐significant) switch cost than controls, while at long RSI this was a reliable switch deficit.

One explanation for this pattern of results, consistent with our theoretical framework concerning inhibitory mechanisms in task‐switching, is that, at short RSI, when endogenous control is not strongly established, difficulty in reactively suppressing inappropriate responses (or task‐sets) means that the RF group frequently makes the wrong response. This may also explain the fact that those subjects with the greatest POp damage had larger switch costs at long than at short RSI. At short RSI, the patients with greatest right POp damage had inflated RTs at position 1 of the run (the switch position) and they did not recover much at positions 2 and 3 of the run (non‐switch positions), so the switch cost was small. However, at long RSI, we posit that they could recover much better at positions 2 and 3 of the run (relative to position 1) because endogenous control was stronger and could help compensate for weak reactive suppression; hence the switch cost was substantial.

Although this interpretation of the pattern of RF switch‐costs is speculative [especially considering the multiple theoretical frameworks surrounding task‐switching (for review see Monsell, 2003), as well as the multiple cognitive components that doubtless interact to effect a task‐switch], we find it highly plausible in accounting for at least part of the deficit. This is because the RF cortex, and the POp region in particular, has been repeatedly implicated as a focus for inhibitory mechanisms (for review see Aron et al., 2004). With respect to the present study, we note that SCRESID was significantly correlated with damage to POp, as well as with SSRT, a measure of response inhibition. The finding that the amount of damage to POp affected measures from two independent tasks increases confidence that POp implements an inhibitory mechanism required for suppressing responses and/or task sets. The requirement to inhibit a prepotent response upon presentation of a stop‐signal and the requirement to inhibit activation of an irrelevant response by a bivalent‐stimulus in a task‐switching experiment have strong prima facie similarity.

A plausible neural model for the generation of the suppression effect is that once the anterior cingulate cortex (ACC) detects conflict in a task‐setting, the right POp is recruited to inhibit the irrelevant response activation (Gehring and Knight, 2000). That neither the error nor the RT measure of SCI‐C(short RSI) correlated reliably with damage to the right POp (although the correlations were in the right direction) may be because these measures had much less of a range (hence less variability) than SCRESID. Independent evidence from the Eriksen‐Flanker task (effectively an IC–C comparison) indicates significant right IFG/POp activation change in fMRI experiments (Hazeltine et al., 2000; Bunge et al., 2002).

The present results clearly show that RF damage can indeed lead to a switching deficit. It is possible that the failure of prior neuropsychological studies (Rogers et al., 1998; Bedard and Richer, 1999; Mecklinger et al., 1999; Keele and Rafal, 2000) to detect switch deficits in RF patients may have been due to under‐representation within those samples of IFG damage, or damage to other foci critical for switching (Rushworth et al., 2002). It is also noteworthy that the RF deficit in the present study was engendered by specific conditions (particularly the congruency manipulation) which were not always present in prior studies.

Left frontal cortex and task‐switching

In accordance with three prior lesion studies of task‐set switching (Rogers et al., 1998; Mecklinger et al., 1999; Keele and Rafal, 2000) and one using patients with closed head damage (Stablum et al., 1994), we found that LF damage produced a substantial increase in switch costs. Unlike the RF group, the LF group did not have particular difficulty suppressing the inappropriate response on switch trials under speed stress (i.e. short RSIs). Instead, they showed a more general difficulty in imposing the appropriate task set, as indexed by larger switch costs at both short and long RSIs, a larger interference effect (I–C) on all trials, and some suggestion of greater activation of the competing task set (the C–N contrasts). Although the correlations with extent of damage were generally weaker than for the RF patients, there was a reliable correlation between the IC–C contrast and MFG damage (at least at the long RSI) and between the C–N effect and MFG damage (at least at the short RSI). These findings are consistent with weaker endogenous control of task‐set in the LF patients leading to amplification of the influence of exogenous activation of task‐set.

Mecklinger et al. (1999) found that that those LF patients with the greatest switching difficulty also had speech and language disorders, and argued for a specifically linguistic component of endogenous control (Goschke, 2000) associated with left PFC. However, there is also evidence suggestive of a specific role for the left MFG in implementing and maintaining a task‐set. Garavan et al. (2002) showed that the amount of activation in the left MFG (Brodmann area 9) preceding a no‐go trial predicted a correct stop or an incorrect commission error, thus indicating that greater activation (a greater role) for the left PFC meant the subject was better capable of keeping in mind the task‐set [also see MacDonald et al., 2000, who also reported left MFG (Brodmann area 9) activity during task preparation]. Further research by M. Funnell and H. Garavan (submitted for publication) in a split‐brain patient showed that, in a go/no go task and two stop‐signal tasks, the right hemisphere was better able to inhibit motor responses than the left. Conversely, the left hemisphere was better able to stay on task when the task requirements were complicated. Consistent with the present study, these results suggest that the left hemisphere is more important for selection and maintenance of a task‐set, while the right hemisphere is more important for the implementation of inhibitory control.


The results suggest that a subregion of the right PFC, the pars opercularis (POp) of the IFG, plays an inhibitory role related to inhibition of responses and/or task‐sets. The critical importance of this region for both response inhibition and switching revealed in the present cohort confirms two neuroimaging studies directly comparing response inhibition with switching (Konishi et al., 1999; Swainson et al., 2003), both of which found activity in the right IFG common to both tasks. The IFG/POp has been repeatedly activated in a wide range of neuroimaging experiments (for review see Duncan and Owen, 2000) and in studies specifically investigating response inhibition (e.g. Garavan et al., 1999; de Zubicaray et al., 2000; Menon et al., 2001; Bunge et al., 2002; Garavan et al., 2002; Aron et al., 2003; Rubia et al., 2003) and switching/reversing/shifting (e.g. Dove et al., 2000; Monchi et al., 2001; Nagahama et al., 2001; Cools et al., 2002; Dreher and Berman, 2002; Nakahara et al., 2002; Brass et al., 2003) (for review see Aron et al., 2004). The IFG/POp is one of the most heavily connected regions of the PFC (Miller and Cohen, 2001), is one of the last to develop in both ontogeny and phylogeny (Pandya and Barnes, 1987) and is frequently implicated in neuropsychiatric syndromes such as attention deficit/hyperactivity disorder (Rubia et al., 1999). The present characterization of the POp as being critical for suppression of inappropriate responses and/or task sets under conditions of frequent switches between tasks goes some way to clarifying its function (which may extend across multiple task domains). Future research should probe this possibility, as well as the interaction between POp, ACC and other PFC regions (Gehring and Knight, 2000; Garavan et al., 2002). By contrast, the left frontal cortex appears to be required not for the reactive suppression of inappropriate responses, but instead for the implementation of the endogenous control of the task‐set (MacDonald et al., 2000; Garavan et al., 2002). While such top‐down control has been modelled extensively for the task‐switching paradigm in formal/computational terms (Meiran, 2000; Gilbert and Shallice, 2002; Yeung and Monsell, 2003), a neurobiological account is lacking. The current finding that the extent of damage to the left MFG correlates reliably with measures of such top‐down control provides relevant data for such an account.

Task‐switching requires multiple cognitive components. This study indicates the localization of two such components—the reactive suppression of task‐sets and/or responses and top‐down control of task‐set—to the right IFG/POp and left MFG respectively.


The authors wish to thank the managers and subjects of the CCNRP, Cambridge, and C. Rorden for helpful technical advice. The task‐switching paradigm was developed under a Stroke Association Grant to S. M. and I. Robertson. This work was supported by an MRC studentship to A. R. A. and a Program Grant from the Wellcome Trust (019408) and was carried out at the MRC (Cambridge) Centre for Behavioural and Clinical Neuroscience.


View Abstract