Converging lines of evidence show that volitional movement prevention depends on the right prefrontal cortex (PFC), especially the right inferior frontal gyrus (IFG). Selective movement prevention refers to the rapid prevention of some, but not all, movement. It is unknown whether the IFG, or other prefrontal areas, are engaged when movement must be selectively prevented, and whether additional cortical areas are recruited. We used rapid event-related fMRI to investigate selective and nonselective movement prevention during performance of a temporally demanding anticipatory task. Most trials involved simultaneous index and middle finger extension. Randomly interspersed trials required the prevention of one, or both, finger movements. Regions of the right hemisphere, including the IFG, were active for selective and nonselective movement prevention, with an overlap in the inferior parietal cortex and the middle frontal gyrus. Selective movement prevention caused a significant delay in movement initiation of the other digit. These trials were associated with activation of the medial frontal cortex. The results provide support for a right-hemisphere network that temporarily “brakes” all movement preparation. When movement is selectively prevented, the supplementary motor cortex (SMA/pre-SMA) may participate in conflict resolution and subsequent reshaping of excitatory drive to the motor cortex.
Cortical networks enable the initiation, termination, and prevention of voluntary movements. The cortical network recruited to rapidly prevent movement has been investigated across a number of studies (for a review, see Aron, Durston, et al., 2007). However, what of the situation where certain aspects of movement must be suppressed, while others are executed at the same time? Here, we examine the interaction between movement execution and prevention within the brain.
The right prefrontal cortex (PFC) plays a fundamental role when stopping all movement (nonselective inhibition), as the extent of right, but not left, PFC lesions correlates with time required to prevent a prepared response (Aron, Fletcher, Bullmore, Sahakian, & Robbins, 2003). Furthermore, repetitive transcranial magnetic stimulation (TMS) of the right lateral PFC diminishes inhibitory control (Chambers et al., 2006; Knoch, Gianotti, et al., 2006; Knoch, Pascual-Leone, Meyer, Treyer, & Fehr, 2006). Functional magnetic resonance imaging (fMRI) studies of response inhibition provide complementary evidence, with the consistent finding of right middle and inferior frontal gyri (MFG and IFG) activation during successful movement prevention (Leung & Cai, 2007; Aron & Poldrack, 2005, 2006; Rubia, Smith, Brammer, & Taylor, 2003; Garavan, Ross, & Stein, 1999; Konishi, Nakajima, Uchida, Sekihara, & Miyashita, 1998). Concomitant activation is also observed across a broad network of cortical areas, including the bilateral superior temporal gyrus, the inferior parietal lobe, and midline areas including the precuneus, the anterior cingulate cortex, and the presupplementary motor area [pre-SMA] (Aron, Behrens, Smith, Frank, & Poldrack, 2007; Leung & Cai, 2007; Aron & Poldrack, 2006; Garavan, Hester, Murphy, Fassbender, & Kelly, 2006; Pliszka et al., 2006; Garavan et al., 1999). Parietal and temporal cortex activation may relate to sensory processing of the stop cue, or to the inhibitory control process. Presumably, this information is conveyed to the frontal cortex via association pathways.
Band and van Boxtel (1999) proposed a cortical stopping network whereby the PFC suppresses motor output via circuitry within the basal ganglia. There is evidence that response inhibition is impaired in patients with basal ganglia lesions (Rieger, Gauggel, & Burmeister, 2003) and that activity in the subthalamic nucleus is increased during volitional inhibition (Aron & Poldrack, 2006; Kuhn et al., 2004). The right lateral PFC may “brake” the basal ganglia motor loop via “hyperdirect” projections to the subthalamic nucleus (Aron, Behrens, et al., 2007; Aron & Poldrack, 2006; Frank, 2006; Nambu, Tokuno, & Takada, 2002) or via input to the striatum that engages the “indirect” basal ganglia pathway (Mink, 1996). In either case, the ensuing reduction of thalamo-cortical output may explain why primary motor cortex (M1) excitability is suppressed during volitional inhibition tasks (Coxon, Stinear, & Byblow, 2006).
The assertion that the right IFG is critical for response inhibition has been based on studies employing imperative (visual/auditory) “go” and “stop” cues (Aron, Behrens, et al., 2007; Leung & Cai, 2007; Aron & Poldrack, 2006; Chambers et al., 2006; Aron et al., 2003). In the present study, we examined the neural basis of selectively preventing movement using an anticipatory paradigm (Coxon, Stinear, & Byblow, 2007; Slater-Hammel, 1960) with many similarities to the stop-signal task, including roughly equivalent stop-signal reaction time [SSRT] (Logan, Cowan, & Davis, 1984). The blood oxygen level-dependent (BOLD) signal was measured with rapid event-related fMRI to investigate both nonselective and selective movement prevention. We hypothesized that both forms of movement prevention share common activation in the right lateral PFC. Specifically, we expected right IFG activation, consistent with fMRI investigations of the stop-signal task.
Selective movement prevention is the ability to prevent part, but not all, of a prepared movement. In a behavioral study, we found that selective movement prevention was associated with substantial delays of up to 100 msec for movements made concurrently (Coxon et al., 2007). This was the case for movements prepared both within and between hemispheres, suggesting that suppression of prepared movement is global, reflecting a limit of the brain's functional architecture. Furthermore, movement preparation and prevention mechanisms appear to interact upstream from the primary motor cortex. Presently, little is known about the neural basis of stopping and going at the same time.
It is possible that selective movement prevention recruits additional cortical areas, such as regions of the medial frontal cortex, which have been ascribed functions like error detection/monitoring and conflict resolution (for reviews, see Rushworth, Buckley, Behrens, Walton, & Bannerman, 2007; Ridderinkhof, Ullsperger, Crone, & Nieuwenhuis, 2004). Primate research provides convincing evidence that the pre-SMA is involved in switching actions by first suppressing the unwanted action and subsequently facilitating desired action (Isoda & Hikosaka, 2007). The timing of pre-SMA activity is suggestive of a causal role in the (delayed) movement execution when stimulus–response associations change (Isoda & Hikosaka, 2007). Selective movement prevention and task switching may share some cognitive–motor processes. We hypothesized that selective movement prevention would additionally activate the medial frontal cortex, consistent with the requirement to resolve pre-response conflict.
Thirteen subjects with no neurological impairments participated in the experiment (aged 22–40 years; 8 men). Twelve subjects were right-handed (mean laterality quotient = 0.93, SD = 0.08) and one was left-handed (laterality quotient = −0.83), determined by the Edinburgh Handedness Inventory (Oldfield, 1971). The right hand was tested for all subjects. Subjects gave written informed consent prior to participation and the local ethics committee approved the procedure.
Subjects lay supine on the scanner bed with the right index and middle fingers positioned over the switches of a modified computer mouse. The state of each switch was transmitted to a control room PC. Custom software generated a visual display that was projected onto a screen within the scan room and viewed via a 45° mirror attached to the head coil. A trigger from the scanner initiated the software.
Movement Prevention Task
The display consisted of two vertical indicators (Figure 1). Both indicators moved at an equal rate from the bottom upward, filling in 1 sec, and crossing a horizontal line 800 msec from onset. Subjects were instructed that their primary task was to stop both indicators at this target line by extending the index and middle fingers. The extension movement released a switch under each finger (referred to as go trials). The software recorded the time (in milliseconds) each indicator stopped relative to the target. It was emphasized that the go task was to be performed as accurately as possible. The subject was informed that the indicators may stop prior to reaching the target, and that for these trials they were to prevent the index/middle finger extension movement. These are referred to as stop trials, and there were three types: stop all—both indicators stopped at the same time prior to the target; stop middle—the right indicator stopped prior to the target; and stop index—the left indicator stopped prior to the target. Index and middle refer to the digit to be prevented from moving. For this task, the closer in time the indicator stops relative to the target, the greater the probability of responding (Coxon et al., 2006, 2007). This relationship is equivalent to that of increasing stimulus onset asynchrony of the stop cue relative to the go cue for the stop-signal task (longer stimulus onset asynchrony, greater probability of responding). The time that the indicator stopped prior to the target was set so that movement prevention would be successful on approximately 80% of trials. To ensure comparable performance, these times were 200 and 260 msec prior to the target for stop all and stop selective (stop middle and stop index), respectively. These stop times were determined from the results of a previous experiment (Coxon et al., 2007).
The experiment consisted of four scanning runs, each 381 sec in duration. An algorithm (Wager & Nichols, 2003) was used to create pseudorandomized event sequences (www.columbia.edu/cu/psychology/tor/software.htm). The algorithm optimized design efficiency, maximizing detection power, while maintaining the unpredictability of the task. Within each scanning run, there were 68 go trials (71%), 12 stop all trials (12.5%), 12 stop middle trials (12.5%), and 4 stop index trials (4%). Stop index trials were included to maintain task difficulty. There were 384 trials in total. The intertrial interval was 3 sec for 80% of trials. Null time was included by randomly altering the intertrial interval to 6 sec (15% of trials) and 9 sec (5% of trials). For each scanning run, the relationship between the presentation of events and time to repetition (TR) for echo-planar imaging (EPI) was offset by 0.25 × TR to increase effective sampling resolution of the hemodynamic response function (HRF).
A go trial consisted of the following sequence (Figure 1A). A black rectangle appeared in the subject's peripheral vision to warn that trial onset was imminent, and after 500 msec both indicators started filling. The indicators stopped when the subject extended both fingers, and the software recorded the lift time (LT) relative to the target. One second after trial onset, the indicators were reset to empty and the peripheral rectangle disappeared. The empty indicators and the target line remained visible in the null period between trials. Stop trials were identical to go trials except that either one, or both, indicators stopped automatically before reaching the target, signaling that movement be prevented (Figure 1B and C). In the case where only one indicator stopped (stop selective trials), the other indicator continued until the subject extended the corresponding finger and released the switch.
Participants completed 192 trials outside the scanner to ensure that their performance was stable. They were instructed that their primary task was to perform the go task as accurately as possible, but if either/both indicators stopped automatically prior to reaching the target, they were to prevent the corresponding extension movement.
A Siemens 1.5-T Magnetom Avanto MRI scanner (Siemens, Erlangen, Germany) was used for image acquisition. For all subjects, high-resolution T1-weighted structural images were acquired using a FLASH sequence (TR = 11 msec, echo time [TE] = 4.94 msec, 1 mm isotropic voxels, field of view [FOV]: 208 × 256 × 160 mm, sagittal slices). Functional data (fMRI) were acquired using a descending gradient EPI pulse sequence for T2*-weighted images (TR = 3000 msec, TE = 40 msec, 3 mm isotropic voxels, 36 oblique axial slices, 0.5 mm gap, flip angle = 90°, 64 × 64 matrix). Two “dummy” scans at the beginning of each run were discarded from the fMRI analysis.
Image preprocessing was conducted using MRIcro (University of Nottingham, UK, www.mricro.com) and SPM2 (Wellcome Department of Imaging Neuroscience, University College, London) within Matlab 7.1 (Mathworks, Sherborn, MA, USA). The T1 was “skull stripped” using a Brain Extraction Toolbox (Smith, 2002). EPI image volumes were corrected for differences in slice acquisition time (reference = 10th slice) then spatially realigned to the first volume in the time series to correct for rotation and translation of the head during scanning. Head movement across all scanning runs was <2 mm. Realigned images were processed using the “Unwarp” toolbox to remove variance due to the Susceptibility by Movement interaction (Andersson, Hutton, Ashburner, Turner, & Friston, 2001). The mean EPI image was coregistered to the T1 and the same transformation applied to all EPI images. Images were normalized to the smoothed (skull stripped) MNI template brain then spatially smoothed with an isotropic Gaussian kernel of 8 mm FWHM.
Subject-level Modeling of the Hemodynamic Response
Events were specified at the time of indicator onset and modeled as delta functions convolved with the canonical HRF and its temporal derivative within the general linear model. Within each run there were four types of events: go, stop all, stop selective, errors (unsuccessful stop trials). Remaining null time constituted an implicit baseline. Data were filtered in the temporal domain using a high-pass cutoff of 128 sec, and global differences in BOLD signal were removed by scaling to the grand mean. Using the method outlined by Calhoun, Stevens, Pearlson, and Kiehl (2004), the temporal derivative was incorporated into the HRF beta images to obtain a “latency-independent” amplitude estimate.
For each subject, the following contrast images were generated: (a) go > null, (b) go > stop all, (c) stop all > go, (d) stop selective > go, (e) stop respond > go, (f) stopping > go, and (g) stop selective > stop all. The algorithm was optimized for the detection of the contrasts stop all > go, and stop selective > stop all. Contrast (a) detects visuomotor activity related to performing the go task (in the presence of the possibility of a stop trial). Contrast (b) controls for visual tracking, thus active areas reflect the go process more specifically. Contrasts (c) and (d) weight the stop conditions against go separately to determine their individual contributions. Contrast (e) determines activation associated with unsuccessful stopping, such as the detection of errors. Contrast (f) pools stop all and stop selective trials to look for “stopping”-related activity. Finally, contrast (g) detects activation when selectively preventing movement that is greater than when preventing all movement.
Group-level Random Effects Analysis
Second-level analysis was performed using SPM5 (Wellcome Department of Imaging Neuroscience, University College, London). The subject-level contrast images were input to a random effects model, under the assumption that first-level intersubject variance is homogeneous. The height threshold for group-level one-sample t tests was set at t > 3.05 and significant clusters were determined using the Gaussian Random Field Theory (p < .05, corrected). Where possible, localization of functional activation was assigned using probabilistic cytoarchitectonic maps from the SPM Anatomy toolbox (Eickhoff, Heim, Zilles, & Amunts, 2006; Eickhoff et al., 2005). Conjunction analysis was used to determine common activation across contrasts. Using the method outlined by Nichols, Brett, Andersson, Wager, and Poline (2005), common stop activation [(c) and (d)] was determined. In addition, a conjunction was performed for activation common to go and stop selective [(b) and (g)].
Trial Evoked Response Analysis
The MarsBar region-of-interest (ROI) toolbox (Brett, Anton, Valabregue, & Poline, 2002) was used to define 4 mm radius spheres centered on group-activated voxels. The time course of the raw signal within the ROI was extracted for each subject. For each scanning run, the signal at each of 4 TRs from event onset were averaged for each condition and expressed in terms of percent signal change from the mean. This trial evoked response (TER) was then averaged over scanning runs. Because ROIs were determined using significantly active coordinates from the whole-brain analysis, a statistical comparison between conditions is inappropriate (Poldrack, 2007). Thus, the TER analysis is exploratory.
Subjects performed the task to a comparable level within the scanner as outside the scanning environment (Coxon et al., 2007). On go trials, LT for index and middle was 13.5 ± 2.9 and 14.2 ± 2.6 msec after the target, respectively (mean ± SE). Lift variability (one standard deviation of LT distribution) was 29.1 ± 1.5 msec for index and 31.1 ± 1.6 msec for middle. The probability of responding on stop trials was .2 ± .05 for stop all and .17 ± .02 for stop selective (two-tailed paired t test p = .5). SSRT was estimated as 188.4 ± 5.2 msec for stop all and 248.5 ± 3.5 msec for stop selective, by rank ordering the LT averaged across fingers and finding the nth LT that corresponded to the probability of responding (Floden & Stuss, 2006; Logan et al., 1984). The eventual LT of the noncued digit on stop selective trials was 108.2 ± 10.4 msec for index (on correct stop middle trials) and 86.3 ± 8.1 msec for middle (on correct stop index trials). As demonstrated in a previous study (Coxon et al., 2007), LT was significantly delayed relative to go trial performance for each digit (one-tailed paired t tests, both p < .001).
Preventing a Prepared Movement
Significant activity for the contrast stopping > go is shown in Figure 2A. This contrast combined both types of stop trial to examine activity related to stopping in general (Supplementary on-line material, Table 1). There was bilateral activation of the inferior parietal lobe including the probabilistically defined intraparietal areas hIP1 and hIP2. The left-hemisphere cluster was largely confined to the lateral aspect of the intraparietal sulcus. Activation in the right hemisphere extended into the angular, supramarginal, and middle temporal gyri and the precuneus. Frontal lobe activation was located predominantly within the right MFG, extending into the dorsal IFG (pars opercularis and pars triangularis). Of this 241-voxel cluster, 9.8% was within area 44 (pars opercularis) and 5.8% was in area 45 (pars triangularis). In total, 9.1% of probabilistically defined area 44 was active and 4.4% of probabilistically defined area 45 was active. A significant cluster of activation was present within the MFG of the left hemisphere extending into the middle orbital gyrus, situated anterior to the location of activation in the right hemisphere.
Activation for selective and nonselective stop trials was also determined separately. The contrast stop selective > go is shown in Figure 2B and stop all > go in Figure 2C. For selective stopping, there was an 89-voxel cluster, centered on the right MFG and extending into the dorsal IFG (29% of the cluster in area 44, 3% of the cluster in area 45, and 10% of probabilistically defined area 44 active). For nonselective stopping (stop all > go), there were two clusters of activation in the right lateral PFC. The largest (96 voxels) was centered on the right MFG (<1% of cluster in area 44). The second activation was located ventrally including the right insular and temporal pole (<1% assigned to each of probabilistic areas 44 and 45).
Unsuccessful stopping was examined with the contrast stop respond > go (Figure 2D). There was activation within the right ventral IFG (pars opercularis, extending into the pars orbitalis) and within the anterior medial frontal cortex. For the 143-voxel IFG cluster, 31.6% of the cluster was located in area 44, and 12% of the cluster was located in area 45. In total, 16.7% of probabilistically defined area 44 and 5.2% of probabilistically defined area 45 were active. The 69-voxel cluster in the anterior region of the medial wall included the rostral cingulate zone (RCZa) with MNI coordinates ranging from x = −9 to 13 mm, y = 19 to 47 mm, and z = 21 to 52 mm.
To specifically examine activation common to both nonselective and selective stopping, we calculated the conjunction of stop all > go and stop selective > go (Figure 3A). This revealed activation extending laterally from the intraparietal sulcus in both hemispheres. The extent of parietal activation was greater in the right hemisphere, including the angular and supramarginal gyri and the temporo-parietal junction. In the frontal cortex, activation common to both nonselective and selective stopping was observed solely in the MFG of the right hemisphere (centered on MNI coordinate x, y, z = 49, 19, 41).
Selective Movement Prevention
The contrast stop selective > stop all identified areas with greater activation for selective compared with nonselective stopping. As selective stopping involved movement of one digit, activity in the go network (contralateral sensorimotor cortex, ipsilateral cerebellum) was observed (Supplementary on-line material, Table 2). Stopping selectively led to significantly greater activity within the postcentral sulcus bilaterally relative to stopping all movement. There were no voxels in the prefrontal cortex of either hemisphere that were more active for stop selective than stop all. Importantly, a region of the medial wall exhibited greater activity for selective compared with nonselective stopping. There was activation in the supplementary motor area (SMA) region for the conjunction of stop selective > stop all and go > stop all. However, the extent of activation was greater and extended anteriorly into the pre-SMA for selective stopping indicating that medial wall activation on stop-selective trials was not entirely related to extending one digit (Figure 3B). Additional analysis revealed that SMA/pre-SMA activation was present for the contrast stop selective > null, but not for stop all > null (Figure 3D). Plotting the TER confirmed that activation in the left sensorimotor cortex and the SMA/pre-SMA was due to enhanced BOLD during stop selective trials, rather than negative BOLD during stop all trials (Figure 3E).
Go Trial Performance
The go > null contrast revealed significant clusters of activation in regions expected for visuomotor task performance (Supplementary on-line material, Table 3). Controlling for the visual tracking element of the task, the contrast of go > stop all revealed common activation in the left sensorimotor cortex and right cerebellum, lobule VI, consistent with extension movements of the right index and middle fingers (Figure 3C). Compared with go > stop all, activation for go > null extended into more anterior regions of the frontal cortex bilaterally, located within the caudal aspect of the superior frontal gyrus. This coincides with the location of the functionally defined human frontal eye fields (FEF and SEF) (Gagnon, Paus, Grosbras, Pike, & O'Driscoll, 2006; Rosano et al., 2002; Grosbras, Lobel, Van de Moortele, LeBihan, & Berthoz, 1999; Paus, 1996). Activation associated with rapid movement prevention (as described above) is clearly distinct from the FEF/SEF and unlikely to include activity associated with eye movements.
Temporal Modulation of the Hemodynamic Response in the Motor Cortex
Movement preparation continues in motor cortical regions well beyond the time that stopping processes are initiated. Thus, one might expect that the BOLD signal will reflect such movement preparation. The time course of hemodynamic response within M1 differed across conditions, as shown in Figure 3E. When all movement was prevented (stop all), the BOLD signal within M1 initially increased, even though there was no overt movement. This is consistent with research using TMS that demonstrate M1 excitability increases until 140 msec after presentation of the stop signal (Coxon et al., 2006). For selective stop trials, the TER profile was similar to go trials, even though part of the movement was prevented. This likely reflects spatial and temporal limitations of the technique. The BOLD signal reflects the input to an area, and local processing by excitatory and inhibitory interneurons (Viswanathan & Freeman, 2007; Logothetis, 2003). Thus, on stop selective trials, the hemodynamic response likely reflects both excitatory and inhibitory neural processing which cannot be distinguished.
In accordance with previous fMRI studies, movement prevention engaged a right-hemisphere network, including the inferior parietal cortex, the MFG, and the IFG (Figure 2). A novel finding of this study is that medial frontal cortex activation was greater for selective than for nonselective movement prevention (Figure 3). This suggests involvement of the SMA/pre-SMA in selective movement prevention.
Movement Prevention Involves a Predominantly Right-hemisphere Network
Conjunction analysis revealed activity common to both selective and nonselective movement prevention in the right MFG (Figure 3). This overlap supports the hypothesis that selective movement prevention engages the same mechanism as when preventing movement altogether. In our paradigm, the SSRT is comparable to the stop-signal task (Coxon et al., 2006, 2007). However, there are fundamental task differences that may account for the dorsolateral locus of PFC activation in our study. Our findings raise the possibility that lateral PFC regions other than the IFG may initiate a stopping process, depending on the task context.
Very little research has considered the neural mechanism underlying inhibitory control based on visual motion. A notable exception is the work of Sakagami, Pan, and Uttl (2006), Sakagami et al. (2001), and Sakagami and Tsutsui (1999), who trained macaque monkeys to perform a go/no-go task whereby the salient cue was color in some blocks and motion in others. Interestingly, ventrolateral regions of the PFC were responsive for color but not motion (Sakagami et al., 2001). In contrast, dorsolateral regions of the PFC were tuned to visual motion (Sakagami & Tsutsui, 1999). However, this dissociation remains to be demonstrated for the human PFC. It is also possible that the complexity of the stop task impacted on PFC activity. Interestingly, conditional stop-signal task performance activates the right MFG to a greater extent than standard stop-signal task performance, for the same group of participants (Aron, Behrens, et al., 2007; Aron & Poldrack, 2006). Thus, the MFG may play a greater role when the stop task also involves discrimination (e.g., stop if you hear the tone, but only for a left-hand response). In the present task, discrimination was also necessary (a stop trial could be either selective or nonselective).
Another possible contribution to the observed activation in the PFC may relate to the concomitant activation in the posterior cortex. Visual processing is traditionally considered to occur via two main routes, the dorsal stream processing spatial information (“where”) and the ventral stream processing object/perceptual information (“what”) (Goodale & Westwood, 2004; Ungerleider & Haxby, 1994; Goodale & Milner, 1992). A third visual stream that is neither dorsal nor ventral has evolved for the human cortex (Husain & Nachev, 2007). The angular and supramarginal gyri of the right inferior parietal lobe are thought to process temporal information (“when”), based on evidence that this region processes visual event timing for both visual fields (Battelli, Pascual-Leone, & Cavanagh, 2007; Martinez-Trujillo, Cheyne, Gaetz, Simine, & Tsotsos, 2007). We found activation within the right angular and supramarginal gyri when the indicators stopped unexpectedly, which may reflect violation of the anticipated temporal order of events. These regions project via the superior longitudinal fasciculus to dorsal regions of the lateral PFC (Schmahmann et al., 2007; Croxson et al., 2005; Makris et al., 2005). Furthermore, the prominent inferior parietal cortex activity may reflect early inhibitory processing. The lateral parietal cortex encodes information regarding the behavioral relevance, or meaning, of stimulus motion (Freedman, Riesenhuber, Poggio, & Miller, 2001). For our task, the parietal cortex may convey the meaning associated with indicator motion cessation on stop trials. However, the parietal cortex does not appear to activate differentially for successful and unsuccessful movement prevention.
Selective Movement Prevention Activates the Medial Frontal Cortex
In addition to the right lateral PFC, some studies implicate the pre-SMA in (nonselective) response inhibition (Simmonds, Pekar, & Mostofsky, 2008, Picton et al., 2007, Floden & Stuss, 2006; Mostofsky et al., 2003). In a review of this literature, Mostofsky and Simmonds (2008) propose that response inhibition and selection are two sides of the same coin, with the pre-SMA being critical for both processes. However, this conclusion is based primarily on studies of the go/no-go task, which tests the ability to decide, following the imperative cue, that a movement should not be prepared. Therefore, this task places high demands on stimulus discrimination and response selection as indicated by go trial reaction time being substantially longer than simple reaction time. For example, Mostofsky et al. (2003) report go trial reaction times of 340 msec. Simple reaction time to a visual stimulus is typically around 200 msec. Brain activation during the go/no-go task is therefore likely to reflect the resolution of pre-response conflict. Here we show that the pre-SMA is active when part of a movement must be prevented (selective inhibition), but pre-SMA activity is not required to override all prepared movement (nonselective inhibition). This finding is in accordance with primate research indicating that the pre-SMA is essential for successfully resolving pre-response conflict so that a desired action can be selected (Isoda & Hikosaka, 2007).
Medial frontal cortex activation for go trials was observed posterior to a vertical line through the anterior commissure (VCA). This area demarcates the boundary between the SMA and the pre-SMA (Johansen-Berg et al., 2004; Picard & Strick, 1996, 2001). In contrast, activation was more profuse for selective movement prevention trials, extending anterior to the VCA (Figure 3B). The TER analysis confirmed that BOLD was greatest for selective movement prevention in both the SMA proper and the pre-SMA and corresponds with results of imaging studies which investigated paradigms of pre-response conflict (for a review, see Ridderinkhof et al., 2004). Because selective movement prevention delayed initiation of the responding digit by more than 75 msec, this clearly indicates the presence of pre-response conflict in this condition. The observed activation of the pre-SMA/SMA cannot be attributed to performance errors because only correct task performance was analyzed.
Activation in the pre-SMA is observed in situations of increased stop-signal task complexity (Aron, Behrens, et al., 2007; Nachev, Wydell, O'Neill, Husain, & Kennard, 2007). Aron, Behrens, et al. (2007) and Aron, Durston, et al. (2007) report that the pre-SMA is active during conditional stop-signal task performance. Our task differs from the conditional stop-signal paradigm in one important respect—an updated action plan is required during selective movement prevention. This may explain why we observed activation extending into more posterior medial frontal cortex territory in the present study. Connections from the pre-SMA to the SMA exist for the arm representation (Luppino, Matelli, Camarda, & Rizzolatti, 1993). In contrast to the pre-SMA, the SMA has strong projections to M1 digit representations (Tanji, 1996) and is somatotopically organized (Picard & Strick, 1996). Sumner et al. (2007) have provided evidence that the SMA mediates the effector-specific suppression of motor plans when condition–action associations change. Further, Nachev et al. (2007) demonstrate that a focal lesion of the pre-SMA impairs stop–change performance to a greater extent than simple stopping. We propose that during selective movement prevention, the pre-SMA resolves pre-response conflict, and via its connections with the SMA, initiates an updated action plan. Examining both nonselective and selective inhibition in patients with lesions of the pre-SMA, or transient disruption of this region with TMS, may be an interesting direction for future research.
One potential limitation of this study was that the stop times did not vary between trials. To examine the difference between selective and nonselective stopping, it was necessary to match task difficulty across conditions. Without a dynamic tracking algorithm to adjust stop times, constant stop times were determined based on our previous studies (Coxon et al., 2007). Participants required more time for stop selective (260 msec) conditions than for stop all conditions (200 msec) in order to match the probability of responding at 20%. It is possible that the constant stop time under each condition allowed participants to preactivate a stopping process on go trials. This seems unlikely, however, given the imperfect performance on stop trials and that movement initiation was not delayed on go trials. In fact, the probability of responding on stop trials and measures of go trial performance (LT and variability) correspond closely to the results of a previous study with unpredictable stop times (Coxon et al., 2007). Therefore, it does not appear that participants were able to predict when a stop trial would be presented, or what type it would be, although we cannot completely discount this possibility.
A second potential limitation was that our analysis was primarily based on the contrasts for which the experiment design was optimized. We compared selective and nonselective stopping with go trials. It is possible that this introduced oddball effects, given the relative frequency of each event (Braver, Barch, Gray, Molfese, & Snyder, 2001). However, Chikazoe et al. (in press) demonstrate that the right lateral PFC (MFG and IFG) activation persists after controlling for the frequency of go trials. Importantly, in our study, the contrast of stop selective > stop all was not confounded by differences in the frequency of events, Therefore, oddball effects cannot explain the increased activity in the pre-SMA and the SMA during selective movement prevention.
Rapid movement prevention engages a right-hemisphere network including the right inferior parietal lobe and the right lateral PFC. This network is thought to instigate a stopping process that acts to “brake” movement preparation. Stopping may involve more dorsal regions of the lateral PFC when the task is anticipatory in nature. Selectively preventing part of a movement requires more time than preventing all movement, causes a delay in the initiation of concurrently prepared movement, and is associated with activation of the pre-SMA and the SMA. This likely reflects neural activity to resolve pre-response conflict and the subsequent reshaping of excitatory drive to the motor cortex to initiate the updated action plan.
We thank the technical assistance of Fung Yang, Timothy Verstynen, Peter Smale, Brett Cowan, and the staff at the Centre for Advanced MRI.
Reprint requests should be sent to Winston D. Byblow, Movement Neuroscience Laboratory, Department of Sport & Exercise Science, University of Auckland, Private Bag 92019, Auckland, New Zealand, or via e-mail: email@example.com.