We used event-related functional magnetic resonance imaging (fMRI) to measure brain activity when subjects were performing identical tasks in the context of either a task-set switch or a continuation of earlier performance. The context, i.e., switching or staying with the current task, influenced medial frontal cortical activation; the medial frontal cortex is transiently activated at the time that subjects switch from one way of performing a task to another. Two types of task-set-switching paradigms were investigated. In the response-switching (RS) paradigm, subjects switched between different rules for response selection and had to choose between competing responses. In the visual-switching (VS) paradigm, subjects switched between different rules for stimulus selection and had to choose between competing visual stimuli. The type of conflict, sensory (VS) or motor (RS), involved in switching was critical in determining medial frontal activation. Switching in the RS paradigm was associated with clear blood-oxygenation-level-dependent signal increases (“activations”) in three medial frontal areas: the rostral cingulate zone, the caudal cingulate zone, and the presupplementary motor area (pre-SMA). Switching in the VS task was associated with definite activation in just one medial frontal area, a region on the border between the pre-SMA and the SMA. Subsequent to the fMRI session, we used MRI-guided frameless stereotaxic procedures and repetitive transcranial magnetic stimulation (rTMS) to test the importance of the medial frontal activations for task switching. Applying rTMS over the pre-SMA disrupted subsequent RS performance but only when it was applied in the context of a switch. This result shows, first, that the pre-SMA is essential for task switching and second that its essential role is transient and limited to just the time of behavioral switching. The results are consistent with a role for the pre-SMA in selecting between response sets at a superordinate level rather than in selecting individual responses. The effect of the rTMS was not simply due to the tactile and auditory artifacts associated with each pulse; rTMS over several control regions did not selectively disrupt switching. Applying rTMS over the SMA/pre-SMA area activated in the VS paradigm did not disrupt switching. This result, first, confirms the limited importance of the medial frontal cortex for sensory attentional switching. Second, the VS rTMS results suggest that just because an area is activated in two paradigms does not mean that it plays the same essentialrole in both cases.
A number of imaging studies have identified activation within the human medial frontal cortex in tasks involving response conflict and attention to action (Bench et al. 1993; Botvinick et al. 1999; Carter et al. 1995, 1998; Derbyshire et al. 1998;Jueptner et al. 1997a,b; Leung et al. 2000; McDonald et al. 2000; Pardo et al. 1991; Paus et al. 1993; Taylor et al. 1994, 1997). It has, however, proved difficult to interpret the functional role of these activations (Paus 2001) for a number of reasons.
The behavioral tasks used in studies that have activated the medial frontal cortex are often complex, and it is difficult to know exactly which aspect of the task was critical in producing medial frontal activation (Bench et al. 1993; Botvinick et al. 1999; Carter et al. 1995; Derbyshire et al. 1998; Leung et al. 2000; McDonald et al. 2000; Pardo et al. 1991; Sohn et al. 2000). There is longstanding ambiguity about whether conflict occurs at the level of action or sensory selection in many of the paradigms, such as Stroop (1935) or flanker (Eriksen and Eriksen 1974) paradigms, used in imaging studies of attention-to-action and response conflict (McLeod 1991).
There is uncertainty about which medial frontal areas have attentional or task switching functions. Some studies have had only limited spatial resolution and others have been guided by a priori assumptions about the region of interest to be analyzed. There has been some inconsistency in the labeling of human medial frontal activations, and diverse medial frontal regions are likely to have diverse functional roles. A number of studies of response conflict have recorded activations in a relatively dorsal region of the medial frontal cortex defined by −10 < x < 10, 0 <y < 15, 45 < z < 55 (Talairach and Tournoux 1988) in, around, and just posterior to the paracingulate sulcus (Paus et al. 1996a,b). When such activations have been recorded in response conflict studies, they have sometimes been assigned to cingulate cortex, but they may be in a human equivalent of the presupplementary motor area (pre-SMA) (Crosson et al. 1999; Deiber et al. 1999; Luppino et al. 1991;Matsuzaka et al. 1992; Picard and Strick 1996; Sakai et al. 1998–2000).
Although medial frontal activation has been prominent in imaging studies (Paus et al. 1998), there is a paucity of inactivation or lesion data to constrain its interpretation. Just because an activation change is recorded in an area when two tasks are compared does not mean that the area carries out a cognitive operation that is essential for one task but not the other. It is known from animal studies that the activity of single neurons can be modulated during a task even when interference studies indicate that the neurons of the area are not essential for the task's performance. For example, the activity of single cells in the dorsal and ventral premotor cortices is modulated when a monkey reaches, but removal or inactivation of the premotor cortices causes only minor disruption of reaching (Passingham 1988; Rea et al. 1987; Wise et al. 1996). Instead interference studies suggest that the premotor cortex is important for aspects of motor learning and the selection of learned movements (Kurata and Hoffman 1993; Kurata and Hoshi 1999;Passingham 1993; Petrides 1986;Schluter et al. 1998, 1999). Moreover it is now becoming clear that the situation is even more complicated in the case of functional magnetic resonance imaging (fMRI). A combined fMRI, single-unit and field-potential recording study recently suggested that the fMRI signal was more closely correlated with field potential recording than with action potentials recorded from single cells (Logothetis et al. 2001). If this is the case, then fMRI responses may be more reflective of the afferent input to a brain area than the area's output.
Finally, a related issue concerns when an activated brain area makes its critical contribution to the performance of the task. Knowing an area's critical time of operation may help elucidate the nature of the cognitive process it performs in a task. It is not clear if the temporal resolution of human neuroimaging techniques, even event-related fMRI, is sufficient to disentangle the temporal order in which brain activity occurs in different areas. The interposition of large delays within a task (McDonald et al. 2000) may allow temporal separation of task components, but it may also drastically alter the nature of the task by introducing a need for new cognitive processes such as working memory.
To address these issues, we have used event-related fMRI to define medial frontal activation in two simple task-set-switching paradigms. In one paradigm, the response-switching paradigm (RS), switching and conflict only occurred with respect to response selection. Subjects selected between responses on each trial, but the rules for response selection varied between trial blocks. In the second, the visual-switching (VS) paradigm, attentional switching and conflict occurred in relation to the selection of sensory attributes. Subjects selected between stimuli according to their shapes or colors; the critical dimension varied between trial blocks. We have previously described the VS and RS paradigms as manipulating attentional and intentional sets, respectively (Rushworth et al. 2001c). Switching in both paradigms is associated with similar increases in processing demands as indexed by reaction time (RT) increases after switching (M.F.S. Rushworth, R. E. Passingham, and A. C. Nobre, unpublished data). For both VS and RS paradigms, our analysis was performed by comparing increases in event-related blood-oxygenation-level-dependent (BOLD) signals (“activations”) that were time-locked either with the “switch” cue, which instructed subjects to switch from one way of performing the task to the other, or the control “stay” cue, which just instructed subjects to continue performing the paradigm in the same way as before. We recorded prominent dorsomedial frontal activation in both paradigms. The pre-SMA was activated in the RS task.
So that we could assess whether and whendorsomedial frontal activity was essential for task-switching, we used frameless stereotaxy (Paus 1999) to direct repetitive transcranial magnetic stimulation (rTMS) at the activated region while subjects performed the two paradigms. TMS can be used to disrupt, reversibly and transiently, the normal activity of a brain area (Hallet 2000;Jahanshahi and Rothwell 2000; Pascual-Leone et al. 2000; Walsh and Rushworth 1999). Because it is an interference technique, TMS can be used to determine whether a brain area is essential for task performance. Because its disruptive effect is transient, TMS can be used to determine when a brain area plays a critical role (Ashbridge et al. 1997;Schluter et al. 1998, 1999; Terao et al. 1998; Walsh and Cowey 1998; Walsh et al. 1998a,b). When applied over the pre-SMA, TMS disrupted RS performance after a switch cue but not after a stay cue. TMS over a midline control site 4-cm posterior to the pre-SMA site did not disrupt performance in the same way.
To examine the specificity of the effect, we used several procedures during the course of additional TMS experiments (experiments 3 and 4). First we replicated the effect of rTMS in a different group of subjects, using higher frequency TMS (10 Hz rather than 5 Hz). Second, we examined the effect of TMS at a different frontal control site. To control for the effects of having inadvertently stimulated adjacent premotor areas, we tested the effect of TMS over the dorsal premotor cortex area in the vicinity of the superior branch of the superior precentral sulcus (Schluter et al. 1998, 1999).
Third, in addition to examining the spatial specificity of the TMS effect, we also examined its temporal specificity. In addition to applying TMS immediately after switch or stay cues, we also tested its effect on the first trial of a task block after either a switch or stay cue. This is an important control for several reasons. Given the nature of the BOLD signal and its modeling, the results from the fMRI part of the experiment probably reflect cognitive processes related to the performance of the first trials of the block in addition to the preceding switch or stay cues. Moreover, it has been suggested that distinct cognitive processes occur during set switching; switching may begin with a prospective process of set initiation and re-configuration after the switching instruction that can be distinguished from subsequent performance of the new task (Meiran 2000;Meiran et al. 2000; Monsell et al. 2000;Rogers and Monsell 1995; M.F.S. Rushworth, R. E. Passingham, and A. C. Nobre, unpublished results). In this way, we were able to compare the role of the dorsomedial frontal cortex with that of the premotor cortex in prospective set re-configuration and subsequent actual task performance.
The fourth set of control procedures involved testing the effect of TMS on attentional switching in the VS paradigm. The effect of TMS over dorsomedial frontal cortex was compared during two different time periods, either immediately after the switch or stay cues or at the time of the first trials of switch or stay blocks.
In combination, the fMRI experiments and the various TMS experiments demonstrated that a dorsomedial frontal area, probably the pre-SMA, had a role in re-configuring intentional task set in the RS paradigm and that its role could be distinguished from the role of the dorsal premotor cortex in selecting individual task responses. Despite its activation the dorsomedial region did not appear to play the same role in the re-configuration of attentional set in the VS paradigm.
Experiments 1 and 2—fMRI and 5 Hz dorsomedial frontal TMS
In total, 20 right-handed, healthy volunteers participated in the fMRI recording study (ages 19–31 yr). The vision of all subjects was normal or corrected to normal with MRI-compatible glasses. Ten subjects performed the RS paradigm and 10 performed the VS paradigm. The data from two subjects who performed the VS task was lost after mains power failures disrupted data acquisition and storage. Eleven of the 20 subjects participated in the subsequent TMS study, 6 performed the RS paradigm, and 5 performed the VS paradigm. All subjects gave their informed written consent before participation. The procedures were approved by the Research Ethics Committee of the Montreal Neurological Institute and Hospital.
Experiments were conducted both with subjects lying in the coil of the dimly illuminated MRI scanner room or in the dimly illuminated TMS laboratory. Stimulus presentation for fMRI and TMS tests was controlled by essentially identical computer programs. Stimuli were presented on a computer monitor in front of the subjects in the TMS study. In the fMRI study, the stimuli were projected onto a screen using an LCD projector at the head of the scanner tube. Subjects performing the fMRI task used a mirror so that the stimulus appeared directly in front of them.
Figure 1 (left) summarizes the RS paradigm. The RS paradigm concerned intentional set switching and it targeted the mechanisms of task-switching that depend on changing the rules for response selection and response conflict. On each trial, subjects saw either a red triangle (5.1° width, 2.7° high) or rectangle (3.7° width, 2.7°high). During the first set of trials, subjects made a right-hand response to the rectangle and a left-hand response to the triangle. A small circle (0.9° diam, 70-ms duration) provided feedback to the subjects 100 ms after the response (yellow for correct responses and blue for incorrect responses). An interval of 800 ms followed before the onset of the next trial. The intervals between trial onsets varied according to the variable reaction times, and averaged approximately 1,500 ms.
Each experimental session was broken down into blocks of 9–11 trials. The rules by which responses were selected varied between blocks; on some blocks, subjects responded with a left-hand response to rectangles and a right-hand response to triangles. Each block was preceded by an instruction cue. Instruction cues were either a vertical (+) or a tilted (×) cross appearing in a white rectangular background (6° width, 5° high) presented for 200 ms. Cues indicated that the subject should either switch rules for response selection orstay with the current response selection rules. There was a 1,000-ms interval between the onset of the instructive cue and the onset of the first pair of items. The meaning assignment (switch, stay) of each cue (×, +) was counterbalanced across subjects.
In the fMRI, study trials were presented in four sessions each of 5-min duration. The event-related analysis was centered on a comparison of BOLD signal after the switch and stay cues. In each of the four sessions, there were approximately 10 switch and 10 stay cues (approximately 40 switch and 40 stay cues in total). Each cue was separated from the subsequent and preceding cues by a variable interval of 11–14 s. The timing of both types of cue onset was recorded with respect to the onset of acquisition of each frame of fMRI data (see following text).
The TMS study was conducted on a separate subsequent day. In the TMS study, each session consisted of 400 trials. As in the fMRI experiment the switch and stay cues were the focus of the investigation. Half of both types of cues were followed by a 5-Hz 4-pulse rTMS train applied over the pre-SMA area activated in the fMRI study (Fig.2). It should be emphasized that the rTMS train was completed before the initiation of the first trial. The reaction times (RTs) for responses made on the first trial after stay or switch cues, either after rTMS or no rTMS, were recorded. Median RTs (average of 8–10 trials) for each category of trial (stay with no rTMS, stay with rTMS, switch with no rTMS, switch with rTMS) for each subject were then calculated. RTs from both correct and incorrect trials were included; there was a slight, but nonsignificant, tendency for more errors to be made on trials preceded with rTMS. Two dorsomedial frontal sites were tested: the pre-SMA (activated in the switch-stay comparison) and a control site, 4 cm posterior to the pre-SMA (a region not activated in the switch-stay comparison). The RTs from each site were tested with two-way repeated-measures ANOVAs. The first factor was switch, with two levels corresponding to stay and switch trials. The second factor was TMS, with two levels TMS and non-TMS control.
Figure 1 (right) summarizes the VS task. The VS paradigm complemented the RS paradigm and shared most aspects of its formal design. It was designed to study the mechanisms involved in set switching that depend on switching attention between different sensory dimensions of stimuli. Two visual stimulus items were presented simultaneously (70-ms duration) to either side (1.7° eccentricity) of a white central fixation cross (1.3° width, 1.1° high) on a black background on a PC monitor. The two items always consisted of one rectangle shape (1.7° width × 2° high) and one triangular shape (2.6° width [with base up], 2° high). One of the items was always green and one of the items was always red. Either shape could be combined with either color. Subjects used either a particular shape (e.g., rectangle) or color (e.g., red) to direct their attention to the relevant item to detect occasional embedded targets (see following text). There was a variable 1,200- to 1,500-ms interval between trials.
As in the RS paradigm, each experimental session was broken down into shorter blocks of 9–11 trials. At the beginning of an experimental block, during the first set of trials, subjects were told to attend to one particular stimulus feature (e.g., red color) and identify targets that appeared within the relevant (red) item. Subsequently, instruction cues appeared before each set of 8–17 trials. Instruction cues were either a vertical (+) or a tilted (×) cross appearing in a white rectangular background (6° width, 5° high) presented for 200 ms. Cues indicated that the subject should switch the current visual rule for selection or stay with the current visual rule. There was a 1,000-ms interval between the onset of the instructive cue and the onset of the first pair of items.
The visual selection rule was switched between particular predefined features in different dimensions (e.g., red and rectangle). For example, starting with the relevant feature “red,” the switch cue (e.g., ×) would inform the subject that the relevant feature became “rectangle.” The next switch cue instructed the subject that the relevant feature returned to being red. The appearance of the stay cue (+) instructed subjects to continue selecting items based on their current visual rule. The meaning assignment (switch, stay) of each cue (×, +) was counterbalanced across subjects. The specific features for each dimension relevant for selection (red/rectangle, red/triangle, green/rectangle, green/triangle) were also counterbalanced across subjects.
The counterbalancing of cue assignment and selection features ensured that behavioral measures were un-confounded with artifacts due to different physical appearances of the stimuli. Five levels of matched red and green luminosities were used randomly for item colors throughout the experiment. Differences in the physical intensity of stimuli therefore were unlikely to contribute systematically to attentional effects.
To ensure feature-guided sensory attention, subjects were asked to discriminate small target stimuli embedded within the items. A small (0.7° long and 0.06° high) horizontal or angled line was presented in the middle of each item. The embedded stimulus appeared only briefly (15 ms) at the end of each item presentation (55 ms after item onset) to maximize the advantage of orienting toward the relevant item. On most trials (80%), embedded nontargets were presented; the nontarget was either a horizontal line or a line angled upward (approximating a “v”) to different degrees (0.06–2.9°). On rare (20%) target trials, the line was deviated downward (into a “w,” always by 2.9°). Subjects responded on the detection of the rare target (w) with a single key-press. Targets (w) only ever appeared in the relevant visual dimension to which subjects were attending.
As for RS, trials in the fMRI study of VS were presented in four sessions each of 5-min duration. The event related analysis was centered on a comparison of BOLD signal after the switch and stay cues. In the fMRI version of the paradigm, the low probability of target presentation (20%) meant that the event-related fMRI analysis of the switching of sensory attention would be largely uncontaminated by response-related brain activity.
The TMS study version of the paradigm, RT on the trials after switch and stay cues was the measured behavioral index of attention switching. In the TMS version, there was a very high probability of target presentation (80%) on the first trial following both stay and switch cues. This ensured sufficient data for analysis. Subjects were not told of variations in target presentation probability.
The TMS study was conducted on a separate, subsequent day in a session of 600 trials. The switch and stay cues were the focus of the investigation. As in the RS paradigm, half of both types of cues were followed by a 5-Hz 4-pulse rTMS train applied over the dorsomedial frontal cortex area activated in the fMRI study (Fig. 2). It should be emphasized that the rTMS train was completed before the initiation of the first trial. The RTs for responses made on the first trial after stay or switch cues, either after rTMS or no rTMS, were recorded. Median RTs (average of 8–10 trials) for each category of trial (stay with no rTMS, stay with rTMS, switch with no rTMS, switch with rTMS) for each subject were then calculated. The RTs were tested with a two-way repeated-measures ANOVA. The first factor was Switch, with two levels corresponding to stay and switch trials. The second factor was TMS, with two levels TMS and non-TMS control.
Scanning was performed on a 1.5 T Siemens Vision magnet. The scanning procedure began with the acquisition of a T1 structural anatomical scan (80 slices at a thickness of 2 mm, 256 × 256 matrix size,T R = 22 ms,T E = 10 ms, flip angle = 30°, voxel size = 1 × 1 × 2 mm3). This was immediately followed by acquisition of four series of 120 gradient-echo images (20 slices of 5-mm thickness in the same orientation as the Sylvian fissure starting above the most dorsal cortex, 64 × 64 matrix size, TR = 2.441 ms, T E = 50 ms, flip angle =90°, voxel size = 5 × 5 × 5 mm3) of BOLD signal while subjects performed the behavioral tasks.
EVENT-RELATED fMRI DATA ANALYSIS.
All images were transformed into standardized stereotaxic space. This was accomplished by using an automatic image-registration method (Collins et al. 1994) based on multi-scale three-dimensional (3-D) cross-correlation with an average (n = 305) MR image aligned with the Talairach stereotaxic space (Talairach and Tournoux 1988). The transformation is linear, yielding three scaling factors for the width (x axis), length (y axis), and height (z axis) of the brain and effectively removing inter-individual differences in brain size. BOLD signal images were smoothed with a 3-D 6-mm (full-width half-maximum) Gaussian kernel, corrected for head motion artifact and transformed into the same standard stereotaxic space. The statistical analysis was carried out with adapted in-house software (Worsely et al. 2000) using a method based on a linear model with correlated errors and a random-effects analysis. Task-related brain activity was measured by examining the BOLD signal following the switch and stay cues in the VS and RS paradigms; the BOLD signal was convolved with a hemodynamic response function that was modeled as a gamma-density function with a mean lag of 7 s and a SD of 3 s (Zarahn et al. 1997) timed to coincide with the onset of switch or stay cues. Drift was removed by adding third-order polynomial covariates in the volume acquisition times in the design matrix (which were not convolved with the hemodynamic response function). Random effects T-statistical maps of significant difference between cue related BOLD signals were constructed by using a spatially smoothed (150mm full width half-maximum Gaussian kernel) estimate of the random effects variance. The t-statistical maps were then thresholded (t > 4.75, P < 0.01;t > 5.19, P < 0.001) in accordance with the Bonferonni correction for multiple comparisons (for the entire 20-slice brain-volume scanned) and nonisotropic random field theory (Worsely et al. 1996, 1999).
A Cadwell high-speed magnetic stimulator and a 5-cm-diam Cadwell (Kennewick) figure-8 “cone” coil were used to administer rTMS. Each rTMS train was delivered as a 5-Hz sequences of four pulses. Intensity of stimulation was set to be 5% above the threshold for eliciting a visible twitch of the foot when the TMS was applied during mild dorsiflexion of the ankle in all subjects (subjects were instructed to dorsiflect at 10% of full force). TMS intensity was therefore set to be between 85 and 90% of the Cadwell stimulator's maximum output. The rTMS trains used in the experiment began 200 ms after the onset of the switch or stay cue.
Coil placement, in both VS and RS experiments, was guided by the position of dorsomedial frontal activation in each individual subject. Because it was soon apparent that rTMS over the pre-SMA significantly disrupted RS task performance, we also tested the effects of rTMS over a control site, 4 cm posterior to the pre-SMA. The control site stimulation was approximately over the site of the SMA (Fink et al. 1997). No significant switch-stay BOLD signal differences were recorded at the control site.
The coordinates of the dorsomedial activation peak were determined individually for each subject. This target position was marked on the subjects anatomical MRI scan using Brainsight (Rogue Research, Montreal, Canada) software. The subject's anatomical MRI scan was then co-registered with the subject's head using frameless stereotaxy (Paus 1999). The subject's head position was assessed by using the Polaris (Northern Digital, Waterloo, Canada) infra-red tracking system to measure the position of scalp land marks (nasion, nose-tip, intra-trageal notch of left and right ears) also visible on the subject's anatomical MRI. Once the subject's head and MRI scan were co-registered, infra-red tracking was used to monitor the position of the TMS coil with respect to the subject's brain. The TMS coil was then placed over the target brain area.
Experiment 3: 10-Hz dorsomedial frontal TMS at cue and item periods
In this experiment, TMS was again applied over the dorsomedial frontal cortex while subjects performed both VS and RS paradigms. Although the experiment resembled the preceding TMS experiment, there were three important differences.
First, a higher rate of TMS, 10 Hz, rather than 5 Hz, was used in this experiment. It is possible that the failure of TMS trains to impair switching in the VS paradigm was due to the relatively slow rate of TMS pulse presentation in that experiment. In experiment 3, the rate of TMS pulse presentation was doubled. We (Hadland et al. 2001) and others (Harmer et al. 2001) have shown that 10-Hz TMS, at or just above the motor threshold for visible movement of the foot during dorsiflexion, is sufficient to elicit behavioral effects when applied over the dorsomedial frontal cortex.
Second, we employed longer testing sessions in experiment 3 so that it was possible to gather more data for each subject performing each condition. It was therefore possible to exclude incorrect responses from the RT analysis of the TMS effects in experiment 3.
Third, the effects of TMS were tested at two different time periods. In experiment 2, TMS was applied after some switch and stay cues (Fig. 2). In experiment 3, we again tested the effect of applying TMS after stay and switch cues, but we also tested the effect of applying TMS on presentation of the first task item after either a switch or a stay cue (Fig. 3). We have therefore referred to the two different times of TMS application as the cue (Fig. 2) and item (Fig. 3) periods.
The effect of TMS during the cue period of the RS paradigm was studied in six subjects. The effect of the TMS during the item period of the RS paradigm was studied in six different subjects. The effect of TMS during the cue period of the VS paradigm was studied in six subjects. The effect of TMS in during the item period of the VS paradigm was studied in six different subjects. All subjects gave their informed consent before participation and the procedures were approved by the Central Oxfordshire Research Ethics Committee (reference No. C99.178).
The same RS and VS paradigms were used as in experiments 1 and 2. The sessions were longer than in experiment 2; subjects usually performed about 800 trials, although, if too many mistakes were made or if the randomized delivery of TMS occurred on too few occasions, then the number of trials was sometimes increased to 900 or 1,200 trials. The results for the RS and VS paradigms were analyzed separately using a between-subjects ANOVA approach similar to that used in experiment 1; within-subject factors of switch and TMS were used as before, and in addition, a between subject factor of period (with 2 levels corresponding to the cue or item periods) was used.
A Magstim (Whitland, Wales, UK) rapid high-speed magnetic stimulator and a Magstim double-cone coil were used to administer rTMS. The cue period rTMS consisted of a 1-s train at 10 Hz. Item period rTMS consisted of 0.5-s train at 10 Hz (some subjects began to respond within 0.5 s of task item presentation). The rTMS trains used in the experiment began with the onset of the switch or stay cue in the cue-period experiment or, in the case of the item-period experiment, 30 ms prior to the onset of the first task item after the switch or stay cue (in pilot experiments, we found that TMS trains that started co-incidentally with the item period stimulus onset caused some subjects to sometimes blink during presentation of the brief 15-ms target used in the VS experiment). Intensity of stimulation was set to 5% above the threshold for eliciting a visible twitch of the foot when the TMS was applied during mild dorsiflexion of the ankle in all subjects (subjects were instructed to dorsiflect at 10% of full force). TMS intensity was therefore set to be between 45 and 70% of the Magstim stimulator's maximum output.
Coil placement, in both VS and RS experiments, was determined as the position 5 cm anterior to the maximally excitable leg representation in the motor cortex (Hadland et al. 2001). We have previously shown that this leads to placement of the center of the coil over the pre-SMA region (Hadland et al. 2001), and this was confirmed in seven of the subjects using the MRI-guided frameless stereotaxic procedures described in the preceding text. The frameless stereotaxic procedure confirmed that the coil was placed similarly in both experiments.
Experiment 4: 10-Hz dorsal premotor TMS at cue and item periods
Experiment 4 was conducted in a similar way to experiment 3. The main difference was that TMS was directed over the dorsal premotor cortex.
Five subjects were tested on two occasions in the RS paradigm. On each occasion TMS was either delivered in the cue period or in the item period. All subjects gave their informed consent before participation and the procedures were approved by the Central Oxfordshire Research Ethics Committee (reference No. C99.178).
The RS paradigm was used in the same way as in experiment 3. The analyses performed were the same as in experiment 3 except that the factor period was now a within-subject factor because the same subjects had been tested with both cue and item period TMS.
A Mastim Rapid was used to apply 10-Hz TMS trains in either the cue or item period as in experiment 3. Instead of a cone coil, however, a flat 70-mm Magstim figure-8 coil was used. Stimulation intensity was no longer set with respect to the threshold for stimulating the leg area on the medial wall because now the stimulation site was on the lateral surface adjacent to the motor cortex representation of the hand area. Instead stimulation intensity was set at 5% above the threshold for eliciting a visible thumb twitch when the coil was placed over the maximally excitable hand representation in the motor cortex. As before the intensity was set to be appropriate for each individual subject (normally between 50 and 70% of stimulator maximum output). To place the coil over the dorsal premotor cortex, it was moved 2 cm anterior and 2 cm medial to the motor cortex using procedures similar to those previously described and those that we and others have shown leads to placement of the coil in the vicinity of the superior branch of the superior precentral sulcus (Praamstra et al. 1999;Schluter et al. 1998, 1999), and this was confirmed in five of the subjects using the MRI-guided frameless stereotaxic procedures described in the preceding text. The coil was held tangential to the skull with the handle pointing backwards approximately parallel to the mid-sagittal axis.
Experiment 1: fMRI
In the scanner, switching was associated with a behavioral cost measurable in reaction time (RT). Nine of the 10 RS subjects responded more slowly on the first trial of a switch block (mean, 605 ms) than they had on the first trial of a stay block (mean, 505 ms). The difference was significant (Wilcoxon T = 0,n = 10, P = 0.008). There were significant increases in BOLD signal on switching (switch-stay comparison) in four medial frontal regions (Table1, Fig.4). All four activations were in the left hemisphere. The most prominent activation had a peak in or just posterior to the paracingulate sulcus (x = −10, y = 9, z = 53) and extended dorsally to cover the adjacent medial aspect of the superior frontal gyrus. We have therefore labeled this activation as pre-SMA. Two activations had peaks in the cingulate sulcus, approximately 2 cm anterior and posterior the vertical plane at the anterior commissure (VCA plane), and were labeled as rostral and caudal cingulate zones (RCZ and CCZ). The fourth medial frontal activation was considerably more anterior, extended beyond the medial surface and was labeled as frontal pole.
There were also areas of significant decrease in BOLD signal on switching (stay-switch comparison) in the medial frontal cortex (Table2, Fig. 5). The peaks were situated in both very anterior (y > 42) and subcallosal cingulate cortex.
In the scanner, switching was associated with a behavioral cost measurable in RT. All eight VS subjects responded more slowly on the first trial of a switch block (mean, 665 ms) than they had on the first trial of a stay block (mean, 579 ms). The difference was significant (Wilcoxon T = 0, n = 8,P = 0.012). Fewer medial frontal increases in BOLD signal on switching (switch-stay comparison) were recorded in the VS paradigm (Table 1, Fig. 4). There was a significant difference at just two voxels in the cingulate sulcus (x = −2,y = 21, z = 37) in the RCZ region. More prominent was a more dorsomedial activation at the posterior end of the paracingulate sulcus (x = −8, y = 3,z = 60), adjacent to that labeled pre-SMA in the RS experiment. The peak activation in the VS task was more than 9 mm (direct distance in 3-D space) from that recorded in the RS task. The more dorsal and caudal (just anterior to the VCA plane) position of the VS activation meant that it was not clear if it should be ascribed to the SMA or pre-SMA. It was therefore described as “SMA/pre-SMA.”
There were also areas of significant decrease in BOLD signal on switching (stay-switch comparison) in the medial frontal cortex (Table2, Fig. 5). As in the RS task, the peaks were situated in very anterior (y > 49) and subcallosal cingulate cortex.
Experiment 2: 5-Hz dorsomedial frontal rTMS
Figure 6 shows the target sites in the subjects taking part in the rTMS experiment. Figure7 shows the co-registration of the TMS coil with the pre-SMA target area, using the frameless stereotaxic procedure in one subject.
There was a significant main effect of applying rTMS over the pre-SMA site on subjects subsequent RTs (F = 13.253, df = 1, 4, P = 0.022). Subjects' RTs on the first trials after switch cues were a mean of 265 ms slower when rTMS had been given (Fig. 8 A); this difference was significant (t = 2.668, df = 4, P= 0.028). The RTs of all five subjects were slowed by rTMS on switch trials. Subjects' RTs on the first trials after stay cues were a mean of 3 ms slower when rTMS had been given (Fig. 8 A); this difference was not significant (t = 0.62, df = 4,P > 0.05). In summary, rTMS over pre-SMA disrupted RS performance but only on switch trials.
The effect of applying rTMS over the more posterior control site was quite distinct (Fig. 8 B); none of the three subjects' RTs were slowed on switch trials. None of the effects of applying rTMS over the control site were significant.
Figure 6 shows the targets sites for the subjects taking part in the TMS experiment. The application of rTMS over the SMA/pre-SMA activation did not disrupt performance on either stay or switch trials (Fig.8 C). There was a general trend for subjects to perform slightly faster after rTMS although the effect did not approach significance.
Experiment 3: 10-Hz dorsomedial frontal TMS at cue and item periods
Figure 9 shows the TMS target sites in a group of seven subjects as measured with the frameless stereotaxic procedure. The average point of intersection of the coil trajectory with cortex was at −7, 8, 64) (Talairach and Tournoux 1988). Figure 9 also shows the co-registration of the TMS coil with the pre-SMA target area in one example subject.
There was a significant main effect of applying TMS over the pre-SMA site on subjects' subsequent RTs (F = 6.025, df = 1, 10, P = 0.034) and a significant main effect of Switch (F = 21.364, df = 1, 10, P= 0.001). The effect of TMS depended on whether it was applied after a switch or stay cue; there was a significant interaction between TMS and Switch factors (F = 6.054, df = 1, 10,P = 0.034). In addition the effect of TMS depended on whether it was applied in the earlier cue period or the later item period; there was a significant three-way interaction among the factors of TMS, Switch, and period (F = 5.701, df = 1, 10,P = 0.038). From Fig.10 (A and B) it is clear that the statistical interactions were due to TMS having its most disruptive effect when it was applied in the cue period after a switch cue (compare • and □ in Fig. 10 A, right). Subjects' RTs on the first trials after switch cues were a mean of 295 ms slower when rTMS had been given; this difference was significant (1-tailed t = 2.499, df = 5, P = 0.028). RTs were slowed for all six subjects.
As in the RS paradigm, in the VS paradigm there was also a significant main effect of Switch (F = 8.431, df = 1, 10,P = 0.016). There was, however, no significant main effect of TMS nor interaction between TMS and Switch factors. TMS and period factors did interact (F = 5.554, df = 1, 10, P = 0.040). From Fig. 10 (C andD) it is apparent that this is due to the fact that TMS tended to speed RTs when it was applied in the cue period (Fig.10 C), and it tended to slow RTs when it was applied in the item period (Fig. 10 D), regardless of whether or not subjects were switching between sets. The TMS induced slowing in the item period was seen in half of individual subjects' data and was not significant. The TMS-induced facilitation in the cue period, which was similar to that observed in experiment 2, again varied between subjects and did not reach significance.
Experiment 4: 10-Hz dorsal premotor TMS at cue and item periods
Figure 11 shows the TMS target sites in a group of five subjects as measured with the frameless stereotaxic procedure. The average point of intersection of the coil trajectory with cortex was at −36, 0, 64 (Talairach and Tournoux 1988). Figure 11 also shows the co-registration of the TMS coil with the dorsal premotor target area in one example subject.
As in the case of dorsomedial TMS during the RS task, TMS also had a significant effect when it was delivered over the dorsal premotor cortex (F = 15.2044, df = 1, 4, P= 0.018). In other respects, however, the results were different to the dorsomedial TMS results in experiments 2 and 3. First, although there was a significant main effect of Switch (F = 9.599, df = 1, 4, P = 0.036), it clearly did not interact with TMS (F = 0.005, df = 1, 4, P= 0.945) nor was there any suggestion of a three-way interaction of TMS, Switch, and period (F = 0.007, df = 1, 4, P = 0.935). On the other hand, again unlike dorsomedial TMS, there was a significant interaction between TMS and period (F =16.090, df = 1, 4, P = 0.016). From Fig. 12 (A andB) it is clear that dorsal premotor TMS slowed subjects performance when it was applied in the item period (Fig.12 B), regardless of whether or not subjects had just switched sets. All five subjects showed the same pattern of RT slowing when TMS was delivered in the item period, and the slowing was significant, both in the context of switching set (t = 3.869, df = 4, P = 0.018) or staying with the same set as previously (t = 4.002, df = 4,P = 0.0016). There was a slight and nonsignificant speeding of RT when TMS was delivered during the cue period.
In the present experiments, we used event-related fMRI to measure brain activity when subjects were performing identical tasks either in the context of a behavioral switch or in the context of a continuation of earlier performance. The context, switching-task set or staying with the current-task set, influenced medial frontal cortical activation; the medial frontal cortex is transiently activated at the time that subjects switch from one way of performing a task to another. The medial frontal activation was more extensive in the RS paradigm, which required intentional set switching and involved changing the rule for response selection and response conflict. One area of activation was probably in the pre-SMA. The application of rTMS over the pre-SMA disrupted RS performance but only on switch trials. The effect was most clear when the TMS was applied during the cue period when subjects engage in a prospective process of set re-configuration prior to actual performance of the new task. TMS over adjacent premotor regions did not have the same effect. TMS over the dorsal premotor cortex disrupted the selection of individual task responses, but it did not affect wholesale task set reconfiguration. The results suggest a transient but essentialrole for the pre-SMA in intentional set switching that can be dissociated from the role of the dorsal premotor cortex in selecting individual, specific responses. An area at the SMA/pre-SMA border was the only medial frontal area activated in the VS paradigm, which entailed attentional set switching between rules for stimulus selection and stimulus conflict. The application of rTMS over the SMA/pre-SMA border, at the same time (after the switch cue), did not disrupt performance of VS. There was some equivocal evidence for an effect of medial frontal TMS at the later time period (the item period) when subjects were actively performing the task, but the effects were not specific to switching trials and did not reach statistical significance. The medial frontal cortex does not appear to play the same role in re-configuring attentional set in the VS task as it does in the RS task.
Location of switch-related activations in the medial frontal cortex
Switching in the RS task was associated with activation changes in several medial frontal regions (Fig. 4). The locations of the activations suggest that the attentional/task switching role of the medial frontal cortex is closely tied to its motor role; three of the activated regions, in the paracingulate and cingulate sulci, probably correspond to medial premotor areas (Deiber et al. 1999;Fink et al. 1997; Paus 2001; Paus et al. 1993; Picard and Strick 1996). There were two activations in the cingulate sulcus, approximately 2 cm anterior and 2 cm caudal to the VCA plane. These activations fall into premotor regions that have been described as RCZ and CCZ (Deiber et al. 1999; Picard and Strick 1996). Task switching was associated with BOLD signal decreases in both RS and VS in anterior and in ventral subcallosal cingulate areas (Fig. 5). Such decreases are consistent with models of cingulate cortex that emphasize its functional heterogeneity (Devinsky et al. 1995;Koski and Paus 2000; Paus et al. 1998). The proposed cognitive functions of the cingulate cortex seem closely related to its motor functions and depend on a relatively restricted supracallosal region extending only a limited distance anterior to the VCA plane in humans as is the case in monkeys (Rushworth et al. 2000). The more anterior and ventral cingulate cortex in both species may be more concerned with social and emotional processes (Bush et al. 2000; Devinsky et al. 1995; Rushworth et al. 2000).
The more dorsal medial frontal activation recorded in RS was in or just posterior to the paracingulate sulcus. Activations in this region have been recorded in a number of response switching or response conflict paradigms (Pardo et al. 1991; Paus et al. 1993; Taylor et al. 1994) although they have not been labeled consistently. Crosson et al. (1999) have discussed the difficulty of deciding whether activations in this region are in the pre-SMA or the cingulate cortex. In the present study, although its peak was in a sulcus, the activation appeared to extend dorsally into the medial aspect of the superior frontal gyrus. Because of this, and in accordance with previous studies of this region (Crosson et al. 1999; Deiber et al. 1999;Picard and Strick 1996; Sakai et al. 1998,1999a,b), the activation has been labeled as pre-SMA.Disbrow et al. (2000) have compared the position of fMRI-recorded BOLD signals with microelectrode recordings of activity in the same task and found that the BOLD signal may be biased toward the position of local blood vessels. In the same way, the current fMRI-based estimate of the pre-SMA's position may be biased ventrally toward the vessels in and around the paracingulate sulcus.
VS was associated with definite activation in just one medial frontal area in or just posterior and dorsal to the paracingulate sulcus (Fig.4). The peak of this activation was 9 mm distant from the pre-SMA activation recorded in RS. We have referred to this peak as SMA/pre-SMA because of its position at the proposed boundary between the SMA and the pre-SMA (Passingham 1995; Picard and Strick 1996). Stephan et al. (1995) suggested that even within the SMA proper there is a division or transition between a more anterior region, near the VCA plane, and a more posterior region. It is tempting to identify the VS activation with anterior SMA (see also following text). The certain conclusion that VS and RS activated distinct areas in the anterior SMA and the pre-SMA respectively awaits an intra-individual direct comparison of the two paradigms with a higher resolution scanning method. It should be emphasized that the lack of medial frontal activation in VS cannot be attributed to VS just being easier. Behavioral data gathered during scanning and in previous experiments showed a similar RT cost for switching in both VS and RS (Rushworth et al. 2001; M.F.S. Rushworth, R. E. Passingham, and A. C. Nobre, unpublished data); moreover, the VS task clearly activated other frontal areas, such as those in vicinity of superior precentral sulcus (−35, −1, 69; 18, −13, 71; −26, −5, 61; 25, −2, 56; an example can be seen in the top left quadrant of Fig.4).
Attention to action vs. sensory attention
All three medial premotor areas activated in RS, RCZ, CCZ, and pre-SMA, have previously been associated with response conflict and attention to action and its consequences (Bench et al. 1993; Botvinick et al. 1999; Carter et al. 1995, 1999; Jueptner et al. 1997;Leung et al. 2000; MacDonald et al. 2000; Pardo et al. 1991; Passingham 1998; Paus et al. 1993, 1998;Posner and DiGirolamo 1998; Taylor et al. 1994,1997; Turken and Swick 1999). Because of the complex nature of the tasks used in some of these studies; however, it is not always clear that conflict is occurring at the response level, as opposed to an earlier sensory level; there is ambiguity about the locus of conflict in Stroop (Stroop 1935) and flanker (Eriksen and Eriksen 1974) paradigms (MacLeod 1991). Confirmation of the importance of response rule switching and conflict for activation of RCZ and CCZ came from the fMRI results in the VS paradigm; there was no activation in CCZ, and just two voxels of significant activation in the RCZ region, associated with switching.
The importance of response conflict for RCZ and CCZ activation is consistent with a recent metanalysis of positron emission tomography (PET) studies recording cingulate activation; Paus et al. (1998) found that blood flow changes in this region were associated with experiments involving fast manual responding.Turken and Swick (1999) reported that a patient with a restricted cingulate lesion was only impaired on the Stroop task when responses were manual rather than vocal. The human RCZ and CCZ are thought to be homologous with rostral and caudal cingulate motor areas (CMAs) in the macaque monkey brain (Dum and Strick 1993;Shima et al. 1991). All the CMAs have direct connections with the spinal cord and the motor cortex (Dum and Strick 1991,1996; He et al. 1995; Lu et al. 1994; Luppino et al. 1991).
Exactly which aspect of the response demands of a task are critical for activating the cingulate sulcal regions remains to be elucidated. It has been suggested that the cingulate may play a critical role in monitoring responses, perhaps for errors (Bush et al. 2000; Carter et al. 1998;Dehaene et al. 1994; Luu et al. 2000;MacDonald et al. 2000). Single-cell, field recording, and muscimol inactivation studies in the monkey have provided evidence consistent with this hypothesis (Gemba et al. 1986;Shima and Tanji 1998). It is possible that subjects in the present experiments might have monitored their own responses for errors in the context of a response set switch. It should be noted, however, that the cingulate BOLD increases cannot reflect the actual commission of errors; very few errors were made in the fMRI scanner (only 1–4% of trials were error trials across all subjects), and these were not more frequent on switch block trials.
The pre-SMA was also activated on switching in the RS paradigm. Identifying the pre-SMA with intentional switching and attention to action, as opposed to sensory attention switching, however, was more difficult; an adjacent area on the SMA/pre-SMA border was activated in VS. There are anatomical reasons for thinking that the more caudal and dorsal VS activation is a distinct region to the pre-SMA area activated in RS (see preceding text). The current rTMS results, however, do show that the pre-SMA plays an essential role in intentional set switching in RS but that the SMA/pre-SMA region, although activated, isnot essential for attentional set switching in VS. The delivery of TMS, in the cue period (Fig. 2) in the RS task slowed subsequent performance when subjects were switching set but not when they were staying with the current set (experiments 2 and 3, Figs.8 A and 10 A). The delivery of TMS, in the cue period in the VS task did not slow subsequent performance regardless of whether or not subjects were switching set (experiments 2 and 3, Figs.8 C and 10 C). There was a slight slowing when TMS was delivered after the first item in a task block (Fig. 3), in both the VS and RS paradigm, in the context of both switching and maintaining set (experiment 3, Fig. 10, B and D). These item period effects varied between subjects and did not approach significance.
Essential and nonessential activations
To establish the causal importance of a brain area for a cognitive function, it is necessary to use interference techniques in addition to techniques for measuring the activity of single cells or populations of cells. For example, the activity of premotor neurons change when a monkey makes a reaching movement, but lesion or inactivation of the premotor cortex has a minimal effect on reaching movements under normal circumstances (Passingham 1988; Rea et al. 1987; Wise et al. 1996). Instead, interference studies, conducted with permanent lesions (Passingham 1993; Petrides 1986), temporary muscimol inactivation (Hoshi and Kurata 1999; Kurata and Hoffman 1993), and TMS (Schluter et al. 1998,1999), suggest that the premotor cortex is essential for aspects of motor learning and the selection of learned movements. Although the importance of fMRI imaging for the understanding of human brain function cannot be overestimated, it can be instructive to combine the technique with others. In the case of fMRI, is possible that the BOLD signal may reflect the afferent input to an area rather than the output of the area; a combined fMRI, single-unit and field-potential recording study recently suggested that the fMRI signal was more closely correlated with field-potential recording than with action potentials recorded from single cells (Logothetis et al. 2001).
The pre-SMA region is critical for switching in RS. The application of rTMS over this region, after a switch cue, disrupted subsequent performance (experiments 2 and 3, Figs. 8 A and10 A). The effect was unlike that seen when TMS was delivered over other areas. Dorsal premotor cortex TMS (experiment 4) did not disrupt task performance when it was delivered during the cue period (Fig. 2, Fig. 12 A), although it did interfere with the selection of individual task responses when it was delivered in conjunction with task items (Figs. 3 and 12 B).
The SMA/pre-SMA was activated in VS. We did not, however, obtain evidence that the medial frontal cortex had the same essential role in reconfiguring attentional set in the VS paradigm as was the case for the RS paradigm; rTMS, in the cue period (after the switch cue was presented and before subjects re-engaged in task performance) did not disrupt VS performance, either on switch or nonswitch trials (Figs.8 C and 10 C). One explanation for the absence of a disruptive effect in VS might be that switching was not as difficult in VS as it was in RS. This possibility seemed unlikely from our previous behavioral experiments when longer RTs were recorded in the VS task than the RS task, suggesting that, if anything, the VS task was the more difficult (Rushworth et al. 2001; M.F.S. Rushworth, A. C. Nobre, and R. E. Passingham, unpublished results, Fig.2). Unlike our previous experiments, however, the subjects studied with rTMS in the present experiments demonstrated smaller RT costs on switching that the subjects taking part in the RS rTMS experiment. Nevertheless there was clear evidence of an effect of the switching manipulation in both the experiments that tested the effect of dorsomedial frontal TMS directed over the SMA/pre-SMA region (experiments 2 and 3). All five subjects studied in the VS part of experiment 2 were slower to respond after a switch cue than a stay cue, demonstrating that there was still a behavioral cost of switching. Not all subjects were slower on switching trials in the VS part of experiment 3, but the manipulation was still effective enough to produce a statistically significant slowing effect on the group as a whole. Moreover, it is not the case that the VS paradigm is somehow less susceptible to TMS; it is clear that the VS paradigm issusceptible to rTMS interference but only if it is applied over brain areas other than the SMA/pre-SMA; specifically posterior parietal rTMS disrupts VS performance (Hadland and Rushworth 2000). TMS over the posterior parietal cortex has been shown to affect several visual-attention tasks (Ashbridge et al. 1997;Göbel et al. 2001; Rushworth et al. 2001; Walsh and Cowey 1998, 2000;Walsh et al. 1998, 1999) and we have recently presented evidence that, not surprisingly, TMS at the same site affects VS performance (Hadland and Rushworth 2000).
Although there was no evidence that medial frontal TMS during the cue period (immediately after the presentation of the switching cue and prior to actual task performance) disrupted switching, there was some equivocal evidence of a TMS effect when it was applied at the item period when subjects were actively engaged in the task. This aspect of the results must be interpreted with caution. First, it should be noted that because these effects varied between subjects, they did not approach statistical significance. Because these effects are not specific to just switching trials and because a similar effect was not observed when TMS was applied during the anticipatory interval of the cue period, it is difficult to associate them specifically with switching set rather than some other aspect of component task performance.
In summary, the interpretation of the SMA/pre-SMA activation recorded during the VS paradigm is not clear. Although it may reflect a role for the medial frontal cortex in switching attentional set in the VS paradigm, it is not clear that its role here is similar to its role in switching intentional set in the RS paradigm. An alternative hypothesis is that it might be due to modulation of subjects' sustained preparation to respond to the rare targets. Cortex in the SMA/pre-SMA region has been associated with imagining or preparing responses (Lee et al. 1999; Stephan et al. 1995). We have recorded a greater bereitshafts potential (BP) after a switch as opposed to a stay cue in the VS paradigm (M.F.S. Rushworth, A. C. Nobre, and R. E. Passingham, unpublished observations), and the BP has been associated with preparation (Ikeda et al. 1992). Such preparatory activity, however, is not confined to the pre-SMA and/or SMA but is also prominent in some parietal and premotor areas (Deiber et al. 1996; Krams et al. 1998; Rushworth et al. 2001; Wise 1985). If preparatory activity is so widespread, then that seen in the pre-SMA may not be essential for response preparation.
Role of the pre-SMA in task switching
The present results demonstrate that the pre-SMA plays a role in task set switching. The role that it is played by the pre-SMA can now be distinguished from that of a number of other medial and lateral premotor areas.
First, the role of the pre-SMA can be distinguished from that of the cingulate cortices. Cingulate areas were also activated during intentional set switching in the RS task. A series of recent fMRI studies (Botvinick et al. 1999; Carter et al. 2000; MacDonald et al. 2000), however, have suggested that cingulate activation is more closely tied to the subsequent period when response conflict is greatest rather than the prior period of top-down control of switching. We have also shown that lesions in the macaque cingulate cortex do not affect the top-down control of set switching (Rushworth et al. 2000). The BOLD signal modeling in the present experiments did not just capture set re-configuration related activity, but it is also likely to have reflected activity associated with the first trials of the task blocks when response conflict is greatest. The cingulate area activation (experiment 1) may, therefore, reflect such processes. The TMS results, however, suggest a very different role for the pre-SMA. The delivery of TMS over this region (experiments 2 and 3) was most effective during the cue period (Figs. 2 and 3) when subjects only had the opportunity to re-configure task set prospectively and beforethe subjects had any opportunity to select specific responses to particular task items or before subsequent response conflict monitoring could occur.
The role of the pre-SMA can also be distinguished from that of the dorsal premotor cortex. BOLD signal increases on switching in the RS paradigm were also recorded in the vicinity of the superior precentral sulcus and adjacent gyrus (−30, −7, 73; −26, 14, 62; −16, 8, 62; −16, 8, 62; examples can be seen in the bottom left quadrant of Fig. 4). The timing of the disruptive effects of dorsal premotor and pre-SMA TMS, however, are quite distinct and suggest that the two regions have distinct roles. There was an interaction between the main effects of TMS and set switching, and a three-way interaction between the main effects of TMS, set switching and stimulation period (cue or item period) in the pre-SMA results (experiment 3, Fig. 10, A and B) that showed that the region was most important when subjects were switching set, not otherwise, and that the region was most important when set switching was initiated rather than when subjects were selecting specific responses. In the case of the dorsal premotor cortex (experiment 4), however, there was no interaction between the effect of TMS and set switching, but there was an interaction between the main effects of TMS and stimulation period (Fig. 12, A and B). This results suggests that the premotor cortex has a role in selecting between specific responses rather than the in the wholesale switching of response set. This is consistent with other single-unit recording, lesion, temporary inactivation, neuroimaging, and TMS data that confirms that the dorsal premotor cortex (PMd) has a role in selecting individual responses that are arbitrarily associated with stimuli (Kurata and Hoffman 1994; Passingham 1993; Petrides 1986; Schluter et al. 1998; Toni et al. 1999; Wise et al. 1996). We suggest that the pre-SMA may have a role in selecting between sets of such response selection rules. This hypothesis not only accommodates the current finding that the pre-SMA is involved in task switching in the RS paradigm but accommodates demonstrations of the pre-SMA's involvement in motor sequences (Gerloff et al. 1998;Hikosaka et al. 1996; Jueptner et al. 1997; Nakamura et al. 1998, 1999; Sakai et al. 1998, 1999; Shima et al. 1996). The pre-SMA concern with motor sequences may be just a particular instance of its more general role in selecting between sets of responses.
Consistent with this view is the finding that pre-SMA activity is most prominent during the learning of a new sequence (Hikosaka et al. 1996; Jueptner et al. 1997b; Nakamura et al. 1998, 1999; Sakai et al. 1998; Shima et al. 1996). Hikosaka and colleagues have developed the “2 × 5 task” (Hikosaka et al. 1995) for testing the learning and retention of motor sequences by monkeys. In this task, the monkey has to press two illuminated targets in the correct order. Five sets of two targets are presented in a fixed order “hyperset.” Pre-SMA neurons are active during the performance of the 2 × 5 task, but it is clear that their activity is often restricted to just the first of the two movements of a set or sometimes even just for the first movement of a hyperset (Nakamura et al. 1998). The cell recording results can be interpreted in the same way as the present fMRI and rTMS results; the pre-SMA is concerned with the selection of a superordinate set of responses rather than each individual response.
Cell recording studies also confirm the anticipatory role of the pre-SMA in reconfiguring the response selection prior to execution.Matsuzaka and Tanji (1996) taught monkeys to make one of two movements on hearing a 1-kHz go signal. On “stay” trials, the monkey simply made the same movement as on the previous trial. On “switch” trials, however, a 50-Hz sound indicated that the monkey should make the opposite movement on hearing the next 1-kHz, imperative go signal. Matsuzaka and Tanji found that pre-SMA neurons began firing when the monkey heard the switch cue in advance of the imperative go signal. In a similar vein,Gerloff et al. (1997) found that TMS over the pre-SMA disrupted movements made some time later in a movement sequence; pre-SMA TMS-induced disruption occurred 1 s after motor cortex TMS-induced disruption.
In summary, the pre-SMA appears to have an anticipatory or prospective role in the wholesale re-configuring intentional or response set. This superordinate role in selecting between sets of responses can be distinguished from the subordinate role of the PMd in selecting between individual responses. The pre-SMA is more concerned with the switching of intentional set than it is with the switching of attentional set.
We gratefully acknowledge the advice of K. Worsley, C. Liao, V. Petre, B. Pike, and H. Johansen-Berg and the assistance of P. Hobden.
This work was supported by the Royal Society, Medical Research Council of Great Britain, the Canadian Institutes of Health Research, and the Canadian Foundation for Innovation.
↵207 Deceased 22 May 2001.
Address for reprint requests: M.F.S. Rushworth, Dept. of Experimental Psychology, University of Oxford, South Parks Rd., Oxford OX1 3UD, UK (E-mail:).
- Copyright © 2002 The American Physiological Society