When response inhibition is followed by response reengagement: An event‐related fMRI study

Abstract

In the course of daily living, changing environmental demands often make our actions, once initiated, unnecessary or even inappropriate. Under such circumstances, the ability to inhibit the obsolete action and to update behavior can be of vital importance. Previous lesion and neuroimaging studies have shown that the right prefrontal cortex and the basal ganglia seem to play an important role in the inhibition of already initiated motor responses. The present study was designed to investigate whether the neural activity of inhibitory motor control was altered if the inhibition process was succeeded by an additional process, namely the reengagement into an alternative action. Therefore, cerebral blood oxygenation during performance of a stop‐change paradigm was registered in 15 male participants using event‐related functional magnetic resonance imaging. Data analysis showed, that during successful and failed stopping and changing (response inhibition and subsequent response reengagement) of initiated motor responses a very similar network was activated including primarily the right inferior frontal cortex (IFC). Besides, stopping‐related activation in right IFC was significantly greater for fast inhibitors than for slow ones. Results of the present study thus further underline the important role of right IFC in response inhibition and suggest that the inhibition process functions similarly regardless whether changing task demands require the complete suppression of an already initiated motor response or its suppression and a subsequent response reengagement into an alternative action. Hum Brain Mapp, 2010. © 2010 Wiley‐Liss, Inc.

INTRODUCTION

As everyday life is characterized by constantly changing environmental demands, the dynamic control of behavior is a crucial ability to flexibly adapt one's actions to altered circumstances. Deficits or even the inability to exert behavioral control can lead to difficulties relating to the flexible interaction with the environment as can be seen in patients with a variety of mental and neurological disorders (e.g., attention deficit hyperactivity disorder, obsessive‐compulsive disorder, trichotillomania, and Parkinson's disease [Aron et al., 2003a; Chamberlain et al., 2006; Gauggel et al., 2004; Konrad et al., 2000a, b; Rieger et al., 2003; Van den Wildenberg, 2006]). Because of its enormous significance for the understanding of the neuropathology of the aforementioned disorders as well as for the better understanding of behavioral correction and adaptation in general, much research was devoted to the investigation of the neural implementation of behavioral control in recent years. In doing so, the main focus of previous research was concentrated on response inhibition [Verbruggen and Logan, 2008].

This focus is in so far justified as the first step to behavioral adjustment consists of the suppression of an already planned or even initiated action. However, everyday life rarely calls for the complete suppression of actions without subsequent behavioral adjustments, but altering environmental conditions normally also require an update of behavior. To our knowledge little attention has been paid to this aspect of behavioral control in previous lesion or neuroimaging studies. Therefore, the purpose of the present study was to investigate whether the neural activity of inhibitory motor control was altered by having the inhibition process succeeded by an additional process, namely the reengagement into an alternative action. One of the paradigms used to investigate inhibitory control seems exceptionally efficient and suited to follow this research question: the stop‐signal paradigm [Lappin and Eriksen, 1966; Logan, 1994; Verbruggen and Logan, 2008]. First, in recent years the stop‐signal paradigm has been increasingly applied in lesion and neuroimaging studies to reveal the neural source of inhibitory motor control [e.g., Aron and Poldrack, 2006; Aron et al., 2003b, 2007a; Chevrier et al., 2007; Gauggel et al., 2004; Li et al., 2006a, b; Plizka et al., 2006; Ramautar et al., 2006; Rieger et al., 2003; Rubia et al., 1999, 2000, 2001, 2003; Van den Wildenberg et al., 2006; Vink et al., 2005]. And—meanwhile even first models concerning the neural implementation of stop‐signal inhibition have been proposed [Aron and Poldrack, 2006; Band and van Boxtel, 1999]. Second, there also exists an extended version of the stop‐signal paradigm, the so‐called stop‐change paradigm [Logan, 1994] which allows the investigation of response reengagement besides the exploration of response inhibition. However, up to now the stop‐change paradigm has been almost exclusively used in behavioral studies (e.g., [Bekker et al., 2005; Logan, 1994; Logan and Burkell, 1986; Schachar et al., 1995].

As to the neural implementation of inhibitory motor control, Band and van Boxtel [1999] proposed a theoretical model, according to which response inhibition in the stop‐signal task is conjointly accomplished by cortical and subcortical structures. In doing so, the prefrontal cortex seems likely to be in charge because it is supposed to be capable of modulating subcortical input to the motor cortex by gating the thalamic transmission of associated activity from the basal ganglia and the cerebellum. The role of the prefrontal, especially the right prefrontal cortex as well as of the basal ganglia as important structures within the inhibitory control network has been confirmed by lesion, animal, and fMRI studies [e.g., Eagle and Robbins, 2003; Gauggel et al., 2004; Rieger et al., 2003; Rubia et al., 1999, 2000, 2001]. Recently, Aron and Poldrack [2006] specified the neural model of stop‐signal inhibition. In doing so, they referred to the horse‐race model which had been proposed to explain inhibitory control on the behavioral level [e.g., Logan, 1994], and tried to find the neural correlates of the two postulated model processes, namely the go‐ and the stop‐process. Aron and Poldrack suggested that inhibition (i.e. the stop‐process) in the stop‐signal paradigm is implemented via the “hyperdirect” fronto‐subthalamic pathway [e.g., Nambu et al., 2002]. Accordingly, they proposed that during stop‐trials the right inferior frontal cortex (IFC) sends excitatory output to the subthalamic nucleus (STN), a subcortical region in the basal ganglia, which then excites the globus pallidus pars interna and so suppresses basal ganglia thalamocortical output [Mink, 1996]. Thus the execution of the go‐response is blocked.

Several findings support the critical role of the right IFC and STN as key regions for the inhibition of already initiated responses. IFC contribution to the canceling of motor responses has been found in various functional magnetic resonance imaging (fMRI) studies [Aron and Poldrack, 2006; Aron et al., 2007a; Chevrier et al., 2007; Li et al., 2006a, b; Rubia et al., 2003]. Moreover, Aron et al. [2003b] found lesion volume in the right IFC to be highly correlated with inhibition performance. In addition, Chambers et al. [2006] could show in a recent transcranial magnetic stimulation (TMS) study that temporary deactivation of the pars opercularis in the right IFC selectively impairs the ability to stop an initiated action. STN involvement in response inhibition has been evidenced by fMRI [Aron and Poldrack, 2006; Aron et al., 2007a], by deep brain stimulation in patients with Parkinson's disease [Van den Wildenberg et al., 2006] and by a lesion study of Eagle et al. [2008] indicating that damage to a midbrain region including the STN worsened inhibition performance in rats. More evidence that stop‐signal inhibition might be implemented via the “hyperdirect” pathway was found by Aron et al. [2007a] who could show by diffusion‐weighted imaging (DWI) tractography that the right IFC is indeed directly connected to the ipsilateral STN region in humans.

Despite the growing knowledge concerning the functional architecture of inhibitory motor control, there are still several open questions. Thus, it is still not closingly solved whether by and large the same network underlies the inhibition process in successful and failed inhibition of initiated actions. The horse‐race model behaviorally describes stop‐signal inhibition as a race between the two stochastically independent go‐and stop‐processes, which are triggered by the go‐ or the stop signal, respectively [e.g., Logan, 1994]. Whether the inhibition process is successful or not, therefore depends on the speed of the go‐response, the speed of the inhibitory process and the delay between the onset of the go‐ and the stop‐processes. Following the horse‐race model, one could expect a similar activation pattern during successful and failed inhibition—with the stop‐process being triggered in both cases, but being too slow during unsuccessful inhibition trials. However, the functional findings have so far been inconsistent. Rubia et al. [2003] and Li et al. [2006a] found for example clearly different activation patterns, when successful and failed stopping were directly contrasted. Successful inhibition was associated with right IFC in the Rubia et al. study and among other regions with bilateral IFC in Li et al. study. On the other hand, Aron and Poldrack [2006] found a similar network—including the right IFC—to be activated during successful and failed stopping, with activation differences above all within bilateral putamen. These differences in putamen activation were discussed by the authors as being a result of speed differences of the go‐processes associated with successful and failed inhibition trials and not as a genuine difference in inhibition processes. Exclusive bilateral striatal activity was also found by Vink et al. [2005] when they contrasted successful and failed stopping.

Besides the problem of the neural implementation of successful and failed stopping there is another open question of importance: In how far is the inhibition process itself altered when it is followed by another process—i.e., the reengagement into an alternative action? Previous behavioral studies using the so‐called change paradigm suggest that the go‐ and the stop‐process function pretty much the same in the stop‐signal and the change paradigm even if stopping seems to be more difficult in change‐trials [e.g., Logan, 1994]. The higher difficulty level of inhibitory control within change‐trials can be derived from the longer so‐called stop signal reaction time (SSRT) in change‐trials, a behavioral mark for the latency of the stopping process. On the neural level there exists to our knowledge only one study using functional near‐infrared spectroscopy (fNIRS) to investigate cerebral blood oxygenation in bilateral dorsolateral prefrontal cortices (DLPFC) during the performance of the stop‐change paradigm [Boecker et al., 2007]. In this study we were able to show bilateral prefrontal activation during both—successful and failed stopping, with activation being more pronounced in the right DLPFC. However, results concerning the similarity of the inhibition process in‐between stop‐ and change‐trials were ambiguous as similar dorsolateral prefrontal activation patterns for stopping and changing were only found for half of the participants.

Therefore, the main goal of our study was to further investigate these two just outlined open questions—this time by using rapid event‐related fMRI. The stop‐change paradigm as it was used in the present study was well suited to investigate both questions: Beside the frequent go‐trials it was composed of pure inhibition trials (stop‐trials) and of trials in which participants had to inhibit an already initiated motor response and subsequently execute an alternative action (change‐trials). A staircase tracking procedure ensured successful stopping and changing rates of about 50%, allowing a comparison of failed and successful inhibitory control.

To reveal the neural correlates of successful and failed stopping, we adopted a twofold strategy: We not only contrasted activation during successful or failed inhibition, respectively with go‐process activation (StopInhibit‐Go, StopRespond‐Go), but also compared neural activity associated with successful and failed inhibition directly.

Further, we focused on the neural correlates of changing and wanted to clarify whether the subsequent reengagement into an alternative action altered the neural implementation of the stopping process. Therefore, two contrasts were of particular interest, namely the contrast between activation during successful change‐trials and go‐trials (StopChange‐Go) and activation differences between successful change‐trials and successful stop‐trials (StopChange‐StopInhibit). Following the terminology of the horse‐race model, activation during successful changing should reflect coactivation of an already initiated go‐process, of the later started stop‐process and of the subsequent response reengagement. Subtracting the go‐condition from successful changing should therefore reflect activity due to successful stopping and response reengagement and the contrast StopChange–StopInhibit should reveal activity associated above all with response reengagement. If the inhibition process functioned similarly in stop‐ and in change‐trials, the contrasts StopInhibit‐Go and StopChange‐Go should yield a similar activation pattern, with the latter showing additional activation due to the additional motor response.

MATERIALS AND METHODS

Participants

A total of 17 healthy male subjects participated in this study. According to the Edinburgh Inventory [Oldfield, 1971] all of them were right‐handed. All participants had normal or corrected‐to‐normal vision and none of them had a history of neurological, major medical, or mental disorders. At the time of measurement no participant was taking medication. Written informed consent was obtained from all participants after complete description of the study before the session. Participants were paid for study participation. The project was approved by the local Ethics Committee of the Medical Faculty RWTH Aachen University and was in accordance with the latest version of the Declaration of Helsinki. Two participants had to be excluded from the analysis because of technical problems with the response button interface. Thus, all analysis were based on the data of 15 participants (mean age 24.8 ± 4.6).

Stop‐Change Paradigm

The stop‐change task

The stop‐change paradigm was composed of four different trial types: the go‐trials, stop‐trials, change‐trials, and null events (see Fig. 1). The first three trial types will be referred to as stimulus trials. The duration of the stimulus trials was 3,350 ms whereas the null events either lasted 2,800 or 3,350 ms. During null events a blank screen was presented. They were included into the experiment according to the randomized event‐related design proposed by Burock et al. [1998].

Figure 1.

Trial types used in the event‐related stop‐change paradigm and graphical illustration of the horse‐race model and its most important parameters. Horse‐race model: The horse‐race model describes stop‐signal inhibition as a race between the two stochastically independent go‐ and stop‐processes, which are triggered by the go‐ or the stop signal respectively [e.g. Logan, 1994]. In doing so, the independence assumption premises that the distribution of go‐processes on stop‐ and change‐trials is the same as the observed distribution of go‐responses during go‐trials. Whether the inhibition process is successful or not, depends according to the model on the speed of the go‐response, the speed of the inhibitory process and the delay (SOA) between the onset of the go‐ and the stop‐processes.

The go‐trials which made up 70% of the stimulus trials were part of a simple choice reaction time (RT) task in which participants had to discriminate a black circle and a black triangle. Depending on which of the two go‐stimuli was presented, participants had to respond either with a left or right key press using index and ring fingers of their right hand. On stop‐trials (15% of stimulus trials) the choice RT task was followed by an auditory stop signal (a 1,000‐Hz tone of 500‐ms duration) after a variable delay (stimulus onset asynchrony, SOA). It signaled that the participants should inhibit their response to the choice RT task. In another 15% of the stimulus trials the change signal (a 400‐Hz tone of 500‐ms duration) was presented after the presentation of the choice RT task. The change signal indicated that the participants should inhibit their response to the choice RT task and press instead the middle of the three response buttons with their right middle finger.

The stop signal delay (stop‐SOA) and the change signal delay (change‐SOA) were set by a staircase‐tracking algorithm [Kaernbach, 1991] which adapted to the response rate. The stop‐SOA and the change‐SOA were adjusted independently in such a way that participants reached an inhibition rate of ∼50% (stop condition) or a change rate of likewise ∼50%, respectively (change condition). The staircase‐tracking algorithm worked in the following way: initially the stop‐SOA and the change‐SOA were set at 250 ms. If participants successfully inhibited their response on stop‐trials or successfully changed their response on change‐trials, the SOA was increased by 50 ms in the next stop‐trial or change‐trial, respectively, so that it was made more difficult for participants to inhibit or change their response. If, however, participants failed to inhibit or change their response, the SOA was reduced by 50 ms in the respective next stop‐ or change‐trial, enhancing the chance of successful inhibition or changing. This tracking procedure provides a measure of inhibition performance, the so‐called SSRT, which is relatively robust against any violations of independence between go‐ and stop‐processes. Moreover, the individual adjustment of the stop‐ and change‐SOA allows keeping the difficulty level of inhibition high and at the same time homogenous across subjects by making the participants work at the edge of their inhibitory capacity.

Procedure

Participants first performed four practice blocks outside the scanner at 44 trials each, in which the different trial types were successively introduced. They were instructed to respond as fast as possible to the choice RT task while maintaining a high level of accuracy. They were told not to delay their responses in anticipation of the stop or change signal, but to make an effort to withdraw the response if they detected the stop signal or to change the response if they detected the change signal. Participants were informed that they would not always succeed in withdrawing the response to the choice RT task and that the computer would adjust to their efficiency, yielding ∼50% success rate.

The scanning session itself consisted of two subsessions with 308 trials each (196 go‐trials, 42 stop‐trials, 42 change‐trials, and 28 null events). The visual stimuli were presented via a head‐mounted video display (http://www.mrivideo.com) designed to meet MR requirements. Vision was corrected to normal if necessary by inserting adapting lenses in the goggles. The auditory stimuli were played through headphones. The volume was individually set at a level both sufficient to exceed the noise of the tomograph and comfortable to the participant.

Data Acquisition by fMRI

MR images were acquired at the University Hospital of the RWTH Aachen University, using a 1.5 Tesla Philips Gyroscan NT with standard head coil and foam padding to restrict movements. Axial multislice T2*‐weighted images were obtained with a gradient‐echo planar imaging sequence (TE = 50 ms; TR = 2800 ms; 64 × 64 matrix; flip angle = 90°; 30 slices, 3.4375 × 3.4375 mm in‐plane resolution; slice thickness 3.75 mm; no gap), covering the entire brain. The investigation in the tomograph consisted of two subsessions. During each subsession 380 volumes were acquired of which the first 10 were applied to allow tissue magnetization to reach its steady state. These first volumes were discarded from further data analysis.

It is inherent to the stop‐change paradigm that the experimenter does not know in advance on which of the stop‐ and change‐trials the participant will successfully inhibit or change the initiated response and on which he will fail. To ensure nevertheless an optimized scanning of the hemodynamic response in all brain slices for the four critical stop‐ and change‐conditions, namely StopInhibit, StopRespond, StopChange, and ChangeRespond, the experiment was programmed the following way: The TR‐time was subdivided into four equally sized so‐called TR‐classes (TR1: 0–700 ms, TR2: 700–1,400, TR3: 1,400–2,100, TR4: 2,100–2,800). An adaptive TR‐class tracking algorithm, which was used from the beginning of the second subsession, ensured that approximately the same proportion of StopInhibits, StopResponds, StopChanges, and ChangeResponds fell into each of the TR‐classes. This was achieved by taking the probable answer of the participant into account (on the basis of whether the stop‐ or change‐SOA was reduced or enhanced in the next stop‐ or change‐trial) and by altering the original onset of the respective trial.

Data Analysis

Behavioral data

On the behavioral level the standard parameters of the stop‐change paradigm were determined: mean go‐RT for correctly answered go‐trials; mean change‐RT for correctly answered change‐trials; mean StopRespond‐RT, i.e., the response speed on stop‐trials for which the participants did not stop; mean ChangeRespond‐RT, i.e., the response speed on change‐trials for which the participants did not change; mean stop‐SOA; mean‐change‐SOA; SSRT, i.e., the time a person needs to inhibit an already initiated motor response on stop‐trials; change signal reaction time (CSRT), i.e., the time a person needs to inhibit an already initiated motor response on change‐trials; percentage of correctly answered go‐trials; percentage of correct inhibition (StopInhibit) and percentage of correct changing (StopChange). The SSRT was estimated by calculating the difference between the mean go‐RT and the mean stop‐SOA. The CSRT was calculated analogously. In addition to these parameters, two further reaction times were estimated: According to the horse‐race model participants should be able to inhibit an already initiated response if the underlying go‐process in the respective stop‐ or change‐trial belongs to the slower part of the go‐speed distribution and should fail if the go‐process is rather fast. To investigate whether there were any activation differences between fast and slow go‐trials, which should be considered when setting up the contrasts with the stop‐ and change‐conditions, the go‐trials were divided into slow‐ and fast go‐trials by means of a median split. Therefore, on the behavioral level the mean RTs of fast go‐trials and of slow go‐trials are also reported.

Before further calculations assumptions for statistical analysis were proved according to the recommendations of Tabachnick and Fidell [1996]. All data were screened for deviation from normality, outliers, and homogeneity of variance. A one‐tailed paired Student's t‐test was conducted to compare the SSRT and CSRT expecting a significant faster SSRT. Additionally, effect sizes were calculated as effect sizes are sample size independent and allow an estimation of the magnitude of the difference between the investigated conditions. Following the recommendations of Dunlap et al. [1996] the effect size was calculated for independent variables instead of dependent variables as effect sizes for dependent variables often overestimate the actual size of effect. Accordingly, the effect size was calculated as the difference of the means of the SSRT and CSRT divided by the pooled standard deviation: d = (m ₁ − m ₂)/[(SD₁ + SD₂)/2].

A repeated measure ANOVA was performed to compare mean go‐RT, StopRespond‐ and ChangeRespond‐RT expecting no difference between the latter two, but in accordance with the race model a significant slower mean go‐RT. To test these specific hypotheses a Helmert‐contrast was defined, contrasting the go‐RT with StopRespond‐RT and ChangeRespond‐RT and the StopRespond‐RT with ChangeRespond‐RT. In addition, the correlations between the go‐RT and the SSRT and CSRT were determined. Consistent with the race model assumption of independence of the go‐ and the stop‐process, no significant correlations were expected. However, the correlation between SSRT and CSRT, both indicators for the time participants need to inhibit their already initiated responses, were expected to be significant.

fMRI data

Functional images were analyzed with statistical parametric mapping software (SPM2; http://www.fil.ion.ucl.ac.uk./spm/spm2.html). Image preprocessing included motion correction, realignment of the images to each participant's first image, and normalization into standard stereotactic space. Spatial smoothing was performed on the functional images using a Gaussian kernel of 8‐mm FWHM. No participant had to be discarded from further analysis because of movement artifacts exceeding a limit of one voxel size per axis.

For each participant the following events were modeled with a canonical hemodynamic response function: Go, fastGo, slowGo, StopInhibit, StopRespond, StopChange, and ChangeRespond. For the go‐conditions events were modeled at the time of the go‐stimulus. In view of the high intraindividual and interindividual variability for the stop‐SOA and the change‐SOA, events of the stop‐ and change‐conditions were modeled at the time of the auditory stimulus to optimally cover activation related to the inhibition of initiated responses. Afterward the contrast images of interest were computed for each subject across the two subsessions and subsequently submitted to a random effects group analysis. To test for consistency of participants' activation, one‐sample t tests on constraint estimates were conducted. Results are reported in accordance with the questions raised in the introduction. On the whole‐brain level the threshold for the activation patterns elicited by the contrasts of interest was set to P < 0.0001 (uncorrected on voxel level) combined with a spatial extent threshold of 10 adjacent voxels. However, only activations significant on cluster level (P < 0.01) will be reported. For some selected contrasts of interest additional regions of interest (ROI) analyses in subcortical regions were performed. In doing so images were thresholded at P < 0.05 using a correction for multiple comparisons [False Discovery Rate (FDR), Genovese et al., 2002]. For the StopInhibit‐Go and StopRespond‐Go contrasts the coordinates of the subcortical ROIs reported by Aron and Poldrack [2006] were chosen to see whether we would find a similar activation pattern: (1) STN, centered at Talairach coordinates 9.9, −14.7, −3.5 with a box of size 10 × 10 × 10 mm³, and (2) globus pallidus (GP) from the automated anatomical labeling (AAL) atlas [Tzourio‐Mazoyer et al., 2002]. The same ROI analyses were computed for the two contrasts within the change conditions, namely the StopChange‐Go and ChangeRespond‐Go contrasts.

RESULTS

Behavioral Results

Mean go‐RT, mean change‐RT, SSRT, and CSRT were in the typical range for young adults (see Table I). Expectedly, participants made hardly discrimination errors on go‐trials and the inhibition as well as the change rate was close to the expected 50%.

Table I.

Behavioral data in the stop‐change paradigm

Behavioral measure	M (SD)
go‐RT (ms)	546 (171)
fast go‐RT (ms)	463 (135)
slow go‐RT (ms)	642 (211)
change‐RT (ms)	662 (124)
StopRespond‐RT (ms)	492 (139)
ChangeRespond‐RT (ms)	501 (135)
SSRT (ms)	214 (42)
CSRT (ms)	238 (39)
P% correct go	96.7 (2.8)
P% correct inhibition	51.2 (3.1)
P% correct changing	49.6 (4.2)

Note. M: mean; SD: standard deviation; P%: percentage; go‐RT: speed of responding on correct go‐trials; fast (slow) go‐RT: refers to the speed of responding on the 50% faster (slower) go‐trials after median split (see text for details); change‐RT: speed of the change response; StopRespond‐RT: speed on stop‐trials for which the participants did not stop; ChangeRespond‐RT: speed on change‐trials for which the participants did not change; SSRT: inhibition speed on stop‐trials; CSRT: inhibition speed on change‐trials.

The one‐tailed paired Student's t‐test revealed that the SSRT was significantly faster than the CSRT (T = 3.11, df = 14, P < 0.005) and the effect size of 0.57 indicated a RT‐difference of medium size. The repeated measure ANOVA for mean go‐RT, StopRespond‐RT, and ChangeRespond‐RT was highly significant (F = 15.1, df = 2, 13, P < 0.005, Greenhouse‐Geisser corrected: ε = 0.64), with the Helmert‐contrast confirming the expected RT‐difference between the go‐RT on the one side and the StopRespond‐ and ChangeRespond‐RT on the other side (F = 16.9, df = 1, 14, P < 0.005). The RT‐difference between the StopRespond‐ and ChangeRespond‐RT were not significant (F = 2.9, df = 2, 13, P = 0.11). Correlation analyses showed that go‐RT was neither significantly correlated with SSRT (r = 0.36, n.s.) nor with CSRT (r = 0.10, n.s.), whereas SSRT and CSRT correlated highly (r = 0.75, P < 0.005). Finally, as expected, the go‐RT of fast go‐trials was significantly faster than the go‐RT of slow go‐trials (T = 8.25, df = 14, P < 0.000) and the effect size of 1.01 indicated a large RT‐difference.

fMRI Results

No activation differences between slow and fast go‐trials

Contrasting activation during slow go‐trials against activation during fast go‐trials and vice versa did not reveal any activation differences between these two go‐conditions. Consequently in the following, activation during all correctly answered go‐trials was considered for calculation when contrasts with go‐process activation were of interest.

StopInhibit: When the stop‐process wins the race

As activation during StopInhibit‐trials reflects coactivation of an already initiated go‐process and of the later started stop‐process, the go‐condition was subtracted from the StopInhibit‐trials to isolate the neural correlates specific to successful stopping. In doing so, successful inhibition was associated with a network of brain regions in bilateral IFC (BA 47), DLPFC (BA 46 and BA 9), and medial frontal cortex (BA 6 and BA 8)—including presupplementary motor area (preSMA). In addition other regions such as bilateral auditory cortex (BA 41; the stop signal was presented binaurally), inferior parietal lobes (BA 40), frontopolar cortex (BA 10), and right insula (BA 13 and BA 47) were recruited during successful inhibition (see Fig. 2A; Supporting Information Table I, available at www.mrw.interscience.wiley.com/suppmat/1065‐9471/suppmat/). Even though the prefrontal activations were bilaterally distributed, the focus of prefrontal activations was clearly right‐lateralized. ROI analyses did not reveal any significant signal change in right or left STN, whereas bilateral activity was observed in the GP (Talairach coordinates: x = 15, y = 3, z = 3, Z = 3.52, FDR corrected, P < 0.005 and x = −12, y = 3, z = 2, Z = 4.30, FDR corrected, P < 0.01).

Figure 2.

Brain areas associated with A. successful response inhibition; B. failed response inhibition; C. successful changing (successful response inhibition and subsequent response reengagement); D. failed changing; and E. response reengagement. All SPM(Z)s were uncorrected on voxel level (P < 0.0001, spatial extent threshold of 10 adjacent voxels), but corrected on cluster level (P < 0.01).

StopRespond: When the stop‐process loses the race

Although inhibition failed on StopRespond‐trials, StopRespond‐Go activated a similar network as it was found for the contrast StopInhibit‐Go (see Fig. 2B; Supporting Information Table II, available at www.mrw.interscience.wiley.com/suppmat/1065‐9471/suppmat/). Besides the bilateral superior temporal (BA 41 and BA 42) and inferior parietal (BA 40) activations, bilateral activations in prefrontal brain areas were found in the IFC (BA 47 and BA 45) and in medial frontal cortex (BA 6 and BA 8)—including preSMA. Additional right prefrontal activation was found in DLPFC (BA 9 and BA 46), the insula (BA 13 and BA 47), and frontopolar cortex (BA 10). The ROI did not show significantly increased activity change—neither in the STN, nor in the GP.

No activation differences between StopInhibit and StopRespond

The finding that StopRespond‐Go activated a similar network as StopInhibit‐Go, suggests that on StopRespond‐trials the stop‐process was at least triggered on the neural level, although it failed on the behavioral one. To further investigate whether there were any activation differences between successful and failed response inhibition, StopInhibit‐trials were contrasted with StopRespond‐trials and vice versa. Neither of the two contrasts did show any activation differences.

Fast stoppers show greater IFG activation than slow stoppers

Li et al. [2006a] and Aron and Poldrack [2006] investigated whether brain activation in subjects with fast SSRTs differed from subjects with slow SSRTs assuming that a more efficient stop‐process would result from greater brain activation of regions that mediated the inhibition of initiated responses. The two research groups chose different contrasts for the comparison of “fast” and “slow stoppers.” Whereas Aron and Poldrack [2006] used the StopInhibit‐Go contrast, Li et al. [2006a] reported their findings for the StopInhibit–StopRespond contrast. For better comparison, in the present study both contrasts were calculated and for each of the contrasts a between‐samples t test was used to compare brain activation of fast and slow stoppers. Whole brain analyses were again thresholded at P < 0.0001 (uncorrected on voxel level), but were corrected on cluster level (P < 0.01). ROI analyses were performed using the ROIs described in Aron and Poldrack [2006]: (1) IFC, consisting of combined pars opercularis and pars triangularis from the AAL atlas; (2) STN; (3) preSMA, using the SMA region from the AAL atlas (with y > 0) and (4) GP from the AAL atlas. In the study of Li et al. [2006a] the left preSMA had been more active in fast inhibitors than in slow ones. For ROI analyses images were again thresholded at P < 0.05 using the FDR‐correction. To divide the 15 participants into fast and slow stoppers a median split of the SSRT was used. On the behavioral level both groups did neither differ in go‐RT (fast SSRT‐group: 519 ms ± 132; slow SSRT‐group: 575 ms ± 214; T = −0.63, df = 13, P = 0.54; effect size: −0.32) nor in inhibition rate (fast SSRT‐group: 50.9% ± 1.4; slow SSRT‐group: 51.5% ± 4.4; T = −0.4, df = 13, P = 0.73; effect size: = −0.19) whereas the difference in SSRT was as expected significant (fast SSRT‐group: 185 ms ± 36; slow SSRT‐group: 247 ms ± 17; T = −4.1, df = 13, P < 0.005; effect size: −2.13).

Whole brain analysis did not reveal any activation differences between fast and slow stoppers for both contrasts and the ROI analysis for the StopInhibit–StopRespond contrast did not show any activation differences either. For the StopInhibit‐Go contrast ROI analysis detected a stronger activation within the right IFC for the fast stoppers, which just failed to be significant (Talairach coordinates: x = 42, y = 41, z = −5, Z = 3.42, FDR corrected, P = 0.051). As the contrast just failed to reveal significant activation differences and as the power to detect significant differences was limited due to the smaller sample size of this subgroup comparison, this result seems to suggest that the fast stoppers activate the right IFG more than the slow stoppers do.

StopChange: When successful stopping is followed by response reengagement

The crucial question of the present study was whether the subsequent reengagement into an alternative action substantially altered the neural activity of the stopping process. Previous behavioral studies using the change paradigm suggested that the go and the stop‐process functioned pretty much the same in stop and change‐trials. As stated in the introduction, activation during StopChange‐trials should reflect coactivation of an already initiated go‐process, of the later started stop‐process and of the subsequent response reengagement. Subtracting the go‐condition from StopChange‐trials should therefore reflect activity due to successful stopping and response reengagement and the contrast StopChange‐StopInhibit should reveal activity associated above all with response reengagement.

For the StopChange‐Go contrast a similar network as in the StopInhibit‐Go contrast was found (see Fig. 2C; Supporting Information Table III, available at www.mrw.interscience.wiley.com/suppmat/1065‐9471/suppmat/)—apart of additional mainly left hemispheric activations in left motor cortex (BA 4), premotor cortex (BA 6) and left postcentral gyrus (BA 1‐3). This network consisted again of the bilateral auditory cortex (BA 41 and BA 42) and inferior parietal lobes (BA 40), as well as of prefrontal activations in bilateral IFC (BA 47, BA 44, and BA 45), DLPFC (BA 9 and BA 46), medial frontal cortex (BA 6 and BA 8)—including preSMA, frontopolar cortex (BA 10) and right insula (BA 13 and BA 47). ROI analyses of STN and GP revealed right STN activation, (Talairach coordinates: x = 11.9, y = −14.8, z = −4.3, Z = 3.11, FDR corrected, P < 0.05) and bilateral GP activation (Talairach coordinates: x = 12, y = 3, z = 0, Z = 3.39, FDR corrected, P < 0.05; x = −12, y = 6, z = 0, Z = 2.88, FDR corrected, P < 0.05; x = −21, y = −9, z = −2, Z = 2.44, FDR corrected, P < 0.05 and x = −24, y = −17, z = 4, Z = 2.44, FDR corrected, P < 0.05).

StopChange–StopInhibit significantly activated left motor cortex (BA 4), premotor cortex (BA 6) as well as left postcentral gyrus (BA 1‐3) and inferior parietal lobe (BA 40) (see Fig. 2E). The opposite contrast, namely StopInhibit–StopChange, did not reveal any activation differences.

ChangeRespond: When the stop‐process loses the race—Part II

ChangeRespond‐trials are very similar to StopRespond‐trials insofar as—spoken on the behavioral level—in both trial types the stop‐process is defeated by the go‐process and the go‐response comes to execution. On the neural level the ChangeRespond‐Go contrast revealed similar activations as the other contrasts with the go‐condition, though less marked and the prefrontal activations being exclusively in the right hemisphere (see Fig. 2D; Supporting Information Table IV, available at www.mrw.interscience.wiley.com/suppmat/1065‐9471/suppmat/). Thus, ChangeRespond‐Go significantly activated bilateral inferior parietal lobes (BA 40), bilateral auditory cortex (BA 41) as well as right IFC (BA 47), DLPFC (BA 46), preSMA (BA 6), frontopolar cortex (BA 10), and the insula (BA 13). In the ROI analysis a significant activation of right GP was found (Talairach: x = 12, y = 3, z = 0, Z = 3.62, FDR corrected, P < 0.05), whereas no activation differences were found for the left GP and STN.

Contrasting ChangeRespond with StopRespond did not reveal any activation differences, whereas during StopRespond‐trials the left auditory cortex (BA 41) was significantly more activated than during ChangeRespond‐trials. To investigate whether there were any activation differences between successful and failed change‐trials, ChangeRespond‐trials were contrasted with StopChange‐trials and vice versa. The former contrast did not reveal any activation differences. However, during StopChange‐trials left fusiform gyrus (BA 37), left parietal lobe and precuneus (both BA 7), left postcentral gyrus as well as bilateral paracentral lobe were significantly stronger activated than during ChangeRespond‐trials.

DISCUSSION

The aim of the present study was to gain more insight into the neural mechanisms underlying successful and failed inhibitory motor control of already initiated actions using the stop‐change paradigm and rapid event‐related fMRI. Furthermore, we wanted to extend previous research by investigating in how far cerebral blood oxygenation due to response inhibition was altered if the stopping process was followed by an additional process, namely response reengagement.

The behavioral data found in the present study, namely the mean go‐RT and the time needed to inhibit the initiated responses (SSRT and CSRT), were in the typical range for young adults. As predicted by the horse‐race model, the go‐ and stop‐process were independent as could be seen by means of the uncorrelated go‐RT with SSRT and CSRT. The go‐process racing against the inhibition process in stop‐ and change‐trials seems to have functioned similarly in both trial types shown by almost identical StopRespond‐ and ChangeRespond‐RTs, both being as expected significantly faster than the mean RT in pure go‐trials.

Imaging data revealed a pattern of activation being associated with successful inhibition in stop‐trials which was consistent with results obtained in previous studies using the stop‐signal paradigm in block design as well as event‐related fMRI studies [e.g., Aron and Poldrack, 2006; Leung and Cai, 2007; Rubia et al., 1999, 2000, 2001, 2003; Xue et al., 2008]. Thus, successful inhibition when contrasted with activation during go‐trial performance, activated a network of brain regions consisting of bilateral IFC, DLPFC and medial frontal cortex including preSMA as well as frontopolar cortex, right insula, and bilateral auditory cortex (the stop signal was presented binaurally) and inferior parietal lobes. Even though being bilaterally distributed, the focus of prefrontal activations was clearly right‐lateralized. Failed inhibition when contrasted with activation during go‐trials activated a network being very similar to the network which was found for successful inhibition. This and the fact that no activation differences between successful and failed stopping were observed when both were directly compared, suggest that the stop‐process was at least initiated on the neural level during StopRespond‐trials. This result fits well the horse‐race model [Logan, 1994] assuming that during all stop‐trials, the go‐process and the stop‐process are initiated, but that on StopRespond‐trials the stop‐process starts too late or is too slow and therefore is defeated by the go‐process. This finding is also in line with the results of Aron and Poldrack [2006] who also found a similar network being associated with failed and successful inhibition.

Moreover, Aron and Poldrack [2006] could show that when considering the time course of the estimated blood oxygenation level‐dependent (BOLD) response in left motor cortex, StopRespond‐trials peaked at the same time and amplitude as go‐trials and then showed a strong undershoot as if the stop‐process was too slow and had later effects. On the other hand, StopInhibit‐trials had a delayed response of similar amplitude as the two former trial types, as if the stop‐process was fast enough to prevent activation in left motor cortex from reaching a critical threshold at a critical time. However, our results are at variance with the findings from Rubia et al. [2003] and Li et al. [2006a] who found right or bilateral IFC activation, respectively, when successful and failed inhibition were directly compared. But as both studies did not report the contrast StopInhibit‐Go it remains unclear how weighty the found differences between the studies are. Aron and Poldrack [2006] not only found no activation differences between StopInhibit and StopRespond in the IFC, but they could even show that right IFC activation did not discriminate between early and late inhibition when the SOA was used as parametric regressor.

It is important to note that the contrasts with the go‐condition most probably not only reveal activation associated with response inhibition but also the coactivation of brain regions due to other uncontrolled noninhibitory cognitive functions such as e.g., response selection, response competition, and response monitoring [Chevrier et al., 2007; Li et al., 2006a; Rubia et al., 2003]. Therefore, we adopted the strategy already used by Li et al. [2006a, b] and Aron and Poldrack [2006] and compared fast and slow stoppers, thereby assuming that a more efficient stop‐process would result from greater brain activation of regions that mediated the inhibition of initiated responses. We could not replicate the findings of Li et al. [2006a] that the left preSMA was significantly more activated in participants with a fast SSRT. However, as in the study of Aron and Poldrack [2006] we found a more efficient inhibition process being associated with right IFC activation. This finding once more underlines the crucial role of the right IFC in response inhibition—a finding not only supported by imaging studies, but also by lesion [Aron et al., 2003b] and TMS studies [Chambers et al., 2006, 2007].

The question in how far brain activation was altered in change‐trials as compared to stop‐trials was of particular interest as we could not conclusively settle it in a previous study using fNIRS to investigate cerebral blood oxygenation in bilateral DLPFC during the performance of the stop‐change paradigm [Boecker et al., 2007]. Results of the present study strongly suggest that the inhibition process functions similarly regardless whether the complete suppression of actions was required as it was the case during stop‐trials, or whether the withdrawing from an initiated response should be followed by the engagement into an alternative response as it was the case during change‐trials. Thus, the contrasts with the go‐condition revealed a network of brain regions being associated with the change‐trials that was very similar to the network found to be associated with the stop‐trials including again the IFC. However, as expected, successful change‐trials yielded some additional activations—especially in left premotor and motor cortex and in the left inferior parietal lobe. The greater activation in the left motor and premotor cortex, which was also found in the contrast StopChange–StopInhibit, provides validating evidence for the current results, since an actual motor response was only required in change‐trials and all participants responded with their right hand.

The fact that we did not find motor cortex activity when contrasting StopRespond and StopInhibit might be explained by the fact, that the motor cortex could also be activated during StopInhibit‐trials due to the initiated go‐process but that in StopInhibit‐trials activation peaks too late in the motor cortex for an open response to be executed [Aron and Poldrack, 2006]. During StopChange‐trials however, two possible response options were activated—first the go‐process and with some delay the reengagement into the alternative response, which probably led to the increased activation during StopChange‐trials. The greater activation found in left inferior parietal lobe (BA 40) during StopChange‐trials might be associated with the need to maintain more representations of possible responses than during go‐ but also than during stop‐trials [e.g., Bunge et al., 2002].

Considering the contrasts of all stop‐ and change‐conditions with the go‐condition, we did not find consistent results concerning basal ganglia activity, which should be expected according to all current neural models of response inhibition. Thus, we did not find any basal ganglia activity in whole brain analysis. However, ROI analysis revealed bilateral GP activation for the StopInbit‐Go and StopChange‐Go contrasts, right GP activation in the ChangeRespond‐Go contrast and right STN activation for the StopChange‐Go contrast. The fact that we could not show consistent GP and STN activation in all contrasts might be grounded on the fact that the activity was simply not found at the chosen threshold level or caused by insufficient resolution. However, basal ganglia activity is often not detected in neuroimaging studies. Thus, Aron and Poldrack [2005] summarized the results of several neuroimaging studies which investigated response inhibition and found that only two out of eleven studies reported basal ganglia activity.

A very interesting result of the present study is the consistent activation of the right preSMA in all stop‐ and change‐conditions. This finding once more raises the question whether the right preSMA plays a direct role in the withdrawing of already initiated responses [Aron et al., 2007a; Chevrier et al., 2007; Floden and Stuss, 2006]. Thus, Floden and Stuss [2006] found that inhibition performance in patients with right dorsomedial frontal lesions was significantly worse than in controls. Aron et al. [2007a] found a striking overlap between the network underlying the inhibition of initiated responses and the network underlying conflict induced slowing—both consisting of right preSMA, IFC and STN—so that the authors raised the question whether the same system might be responsible for response inhibition and for “putting on the brakes” on responses (i.e., to temporarily withhold them) in conflict situations.

Chevrier et al. [2007] who tried to disentangle go‐phase activities in StopInhibit‐trials from the actual inhibition process rather suggest that right superior and middle frontal activities which are often found in neuroimaging studies of inhibition are related to response withholding instead of response withdrawal whereas the right IFG is associated with actual response withdrawal.

Contrasting activation during stop‐ and change‐trials with go‐trial activity yielded a consistent activation of right DLPFC, which also had been found in previous neuroimaging studies using the stop‐signal or go/no‐go paradigm to investigate the neural correlates of response inhibition [e.g., Boecker et al., 2007; Garavan et al., 2002; Leung and Cai, 2007; Pliska et al., 2006]. This raises the question whether the found DLPFC activation reflects inherent inhibitory control or the coactivation due to another cognitive function such as response selection [e.g., Bunge et al., 2002; Kelly et al., 2004]. However, a direct contribution to inhibitory control cannot be excluded. Thus, Staedtgen et al. (submitted) found in a recent single‐pulse TMS study that stimulation of the right DLPFC 50 and 100 ms after the stop signal led to an acceleration of the inhibition process whereas it did not influence the execution of the go‐process.

As already noted above, not all brain regions found activated during stop‐ and change‐trials in the present study might be necessarily needed for the accomplishment of the inhibition process as some of the found activations were certainly due to uncontrolled noninhibitory cognitive functions such as e.g., response selection and competition. Therefore, future research should further combine the different available techniques such as imaging, lesion, neurophysiological, TMS, and pharmacological studies to solve the “puzzle” of the neural implementation of the inhibition process and to better understand the interaction of the different cognitive functions. In doing so, the effects of complications of the stopping process as they were investigated in the present study in which during change‐trials the stop‐process was directly followed by response reengagement should be also more strongly considered. One interesting line of future fMRI studies might be to scrutinize more closely the time course of the go‐ and stop‐processes (and reengagement) using time‐resolved fMRI [e.g., Sack et al., 2008] and to directly test the existing models of the neural implementation of response inhibition [e.g., Aron and Poldrack, 2006; Aron et al., 2007b] and response reengagement.

However, there seems to be a major problem, if one takes the horse‐race model as basis for the description of what is going on during response inhibition. The horse‐race model is based on an independence assumption of the stop‐ and the go‐process as it was originally constructed to account for the description of behavioral data. However, research investigating the neural implementation of inhibitory motor control, strongly suggests that response inhibition on the behavioral level depends on neural inhibition on the neural level [e.g., Hanes et al., 1998] and therefore the stop‐ and go‐process have to interact at a given time point. A recent computational modeling study addressed this apparent paradox [Boucher et al., 2007] and could show that the original horse‐race model still provides a good description of the underlying mechanisms of response inhibition and that the SSRT still provides a valid marker for the time needed to inhibit an already initiated response. Thus, Boucher et al. found that both—the behavioral and neurophysiological data—were best fitted by an interactive model if the stop‐ and go‐processes were modeled as independent for most of their durations and only interacted strongly for a brief period in their final stages. In doing so, the go‐process could take the following pathway: It could be generated by premotor areas and then take the direct pathway of the basal‐ganglia (through striatum, GP, and thalamus) to excite primary motor cortex [e.g., Aron and Poldrack, 2006]. As described in the introduction, the stop‐process could be generated by the right IFC and take the hyperdirect pathway via the STN trying to intercept the go‐process in the GP. However, scrutinizing the time course of these possible pathways more closely, one will have to consider that there are go/no‐go studies and a stop‐signal study suggesting that response inhibition could be governed by different mechanisms, depending on how urgent the inhibition is [e.g., Aron et al., 2007a; Kelly et al., 2004] and that different strategies might be used to solve the task [Li et al., 2006b]. The whole matter is even more complicated by the fact that our brain seems to be able to reallocate relatively quickly inhibitory mechanisms to other brain regions if there is the need for it as Chambers et al. [2006] could show in their TMS study.

CONCLUSION

In the present study it could be shown that a similar network of brain regions was activated during successful and failed inhibition of initiated motor responses. This rather right‐localized network included the right IFC, which has been already shown to be a probable key structure for response inhibition in previous studies. Furthermore, the results of the present study strongly suggest that the inhibition process functions similarly regardless whether participants are required to completely suppress an initiated action or to suppress the action and subsequently engage into an alternative action.

Supporting information

Additional Supporting Information may be found in the online version of this article.

Supporting Information Materials.