People are able to adapt their behavior to changing environmental contingencies by rapidly inhibiting or modifying their actions. Response inhibition is often studied in the stop-signal paradigm that requires the suppression of an already prepared motor response. Less is known about situations calling for a change of motor plans such that the prepared response has to be withheld but another has to be executed instead. In the present study, we investigated whether electrophysiological data can provide evidence for distinct inhibitory mechanisms when stopping or changing a response. Participants were instructed to perform in a choice RT task with two classes of embedded critical trials: Stop signals called for the inhibition of any response, whereas change signals required participants to inhibit the prepared response and execute another one instead. Under both conditions, we observed differences in go-stimulus processing, suggesting a faster response preparation in failed compared with successful inhibitions. In contrast to stop-signal trials, changing a response did not elicit the inhibition-related frontal N2 and did not modulate the parietal mu power decrease. The results suggest that compared with changing a response, additional frontal and parietal regions are engaged when having to inhibit a response.
Humans are able to rapidly withhold or modify ongoing actions when unanticipated events or changes occur. The classic paradigm to study the inhibition of already prepared actions is the stop-signal paradigm, in which go stimuli are occasionally followed by a stop-signal, indicating the need to stop the already prepared response (Verbruggen & Logan, 2008c; Logan, Cowan, & Davis, 1984). Alternatively, one might think of situations calling for a change of motor plans such that the prepared response has to be withheld but another has to be executed instead (De Jong, Coles, & Logan, 1995). The question is whether the same brain mechanisms are underlying the inhibitory action control required in both of these situations. The present study sought to address this question by means of ERPs and task-related oscillations extracted by wavelet analysis.
In the stop-signal paradigm, participants are performing in a primary choice–reaction time task with infrequent stop trials embedded (Verbruggen & Logan, 2008c; Logan et al., 1984). In stop trials, the go signal is followed after a short delay by an auditory or visual stop signal, indicating to inhibit the already prepared response. An influential model of the processes at work in this task is the horse-race model of response inhibition, which assumes the go and the stop processes to proceed independently with the winner of this “race” determining whether a response is given or not (Band, van der Molen, & Logan, 2003; Logan et al., 1984). This model allows one to estimate the latency of the stop-process, that is, the stop-signal RT (SSRT) (Band et al., 2003; Logan et al., 1984).
Although the horse-race model of response inhibition can successfully predict behavioral variables in the stop-signal task, it remains silent with respect to the underlying neural processes. fMRI studies with the stop-signal task consistently show stop-related activations in right inferior frontal cortex, dorsolateral pFC, inferior parietal sulcus, and pre-SMA as well as subcortical structures such as the subthalamic nucleus (Marco-Pallares, Camara, Münte, & Rodriguez-Fornells, 2008; Aron, Behrens, Smith, Frank, & Poldrack, 2007; Aron & Poldrack, 2006; Li, Huang, Constable, & Sinha, 2006; Rubia, Smith, Brammer, & Taylor, 2003). However, the relative contribution of these areas to the inhibitory process is still under debate. ERP studies on the stop-signal task have mainly focused on the frontal N2 and P3 components (Ramautar, Kok, & Ridderinkhof, 2004, 2006; Schmajuk, Liotti, Busse, & Woldorff, 2006; Kok, Ramautar, De Ruiter, Band, & Ridderinkof, 2004). The N2 refers to a large negative deflection around 200 to 300 msec after onset of the stop signal, which is consistently seen in visual stop-signal tasks (Ramautar et al., 2004, 2006; Schmajuk et al., 2006; Pliszka, Liotti, & Woldorff, 2000). Interestingly, Schmajuk et al. (2006) could show a right frontally enhanced N2 in successful compared with failed inhibitions, proposed to reflect increased inhibitory activity in right frontal regions. The stop-related P3 on the other hand has been linked to the evaluation of the inhibitory process rather than to the response inhibition per se because it occurs after the time of failed inhibitions and the SSRT (Kok et al., 2004).
Although considerable research has been done on response inhibition in the stop-signal task, less is known about neural underpinnings of more selective types of inhibitory motor control such that one response has to be inhibited but another response has to be given instead. Whereas performance in the stop-signal task likely recruits a global inhibitory signal suppressing any motor preparation, this might not be the case for more selective task demands. One electrophysiological study compared the stop-signal and change-signal tasks (De Jong et al., 1995) and showed a prolonged latency of the inhibitory process (change-signal RT [CSRT]) and reduced lateralized readiness potentials (LRPs) in the latter. Supportive evidence for a dissociation between these types of inhibitory action control comes from a recent lesion single case study (Nachev, Wydell, O'Neill, Husain, & Kennard, 2007). The patient, a woman with a rare lesion in the right pre-SMA, showed reduced inhibition rates in the change-signal task when she had to inhibit a left-hand (i.e., contralesional) response and change to a right-hand response, whereas no impairments were seen for ipsilesional to contralesional changes. Interestingly, the patient did not show any difficulties when she had to inhibit left-hand responses in the stop-signal task. The authors argued that the pre-SMA seems to be particularly involved in response inhibition in the context of a response conflict, whereas response inhibition in the stop-signal task might be instigated by right prefrontal regions (Nachev et al., 2007). An important role of the pre-SMA in particular for switching from one response to another is indicated by electrophysiological recordings in monkeys while they performed in a saccade task (Isoda & Hikosaka, 2007). A subset of neurons in the pre-SMA selectively fired when monkeys switched from automatic behavior to a controlled alternative action. The authors concluded that the pre-SMA might be relevant to suppress unwanted actions and facilitate the alternative response (Isoda & Hikosaka, 2007). On the basis of these findings, we asked whether electrophysiological data and more specifically the N2 as correlate of response inhibition can provide evidence for distinct processes underlying these types of response control.
In addition to ERPs, we assessed the time course of event-related oscillatory changes. Preparation and execution of manual responses are known to be associated with distinct spatio-temporal patterns in the mu and beta bands (Hari, 2006; Neuper, Wortz, & Pfurtscheller, 2006; Pfurtscheller & Lopes da Silva, 1999). Both oscillations in the mu band (typically around 10–13 Hz) and in the beta band (around 20 Hz) show a prominent event-related desynchronization (ERD) during the preparation and execution of a movement, followed by an event-related synchronization after the movement. There is some evidence that the mu and beta ERD reflect activation in primary somatosensory and motor cortex, respectively (Ritter, Moosmann, & Villringer, 2009; Rau, Plewnia, Hummel, & Gerloff, 2003; Leocani, Toro, Zhuang, Gerloff, & Hallett, 2001; Pfurtscheller & Lopes da Silva, 1999). Mu and beta event-related synchronization on the other hand have been suggested to reflect a rebound to a “resting state” or even an active intracortical inhibition in motor areas (Neuper et al., 2006; Rau et al., 2003; Leocani et al., 2001).
Given their strong association with motor activity, the mu and the beta rhythms are particularly interesting with respect to response inhibition. Yet, only a few EEG studies on the stop-signal task have examined task-related oscillations (Alegre, Alvarez-Gerriko, Valencia, Iriarte, & Artieda, 2008; Marco-Pallares et al., 2008). Marco-Pallares et al. (2008) compared successful inhibitions with both failed inhibitions and go trials in an Eriksen flanker task with stop trials embedded. They reported a relative increase in the beta band in successfully inhibited trials compared with both failed inhibitions and go trials during the time of the responses (Marco-Pallares et al., 2008). In a go/no-go study, Alegre et al. (2004) showed a power decrease over central electrodes in the alpha and beta bands in go trials, which was absent in no-go trials. On the contrary, no-go trials even demonstrated an increase of beta power over frontal areas, supposedly reflecting frontal inhibitory processes (Alegre et al., 2004). Swann et al. (2009) demonstrated distinct effects in the beta band over inferior frontal and motor areas on the basis of intracranial recordings. Beta power over inferior frontal areas increased relative to baseline in inhibited stop trials but not during failed inhibitions. Beta power over motor areas showed in all trials a decrease relative to the baseline, and this was stronger in failed compared with successful stop trials. In the current study, we focused on motor-related oscillatory changes in the beta and mu bands and on the basis of previous results predicted a reduced beta and mu ERD in inhibited stop trials, which we sought to compare with the time course of motor activation and inhibition in change-signal trials. Summarizing, we wanted to compare selective and nonselective response inhibition by investigating the N2 as index of frontal inhibitory motor control and LRPs as well as mu and beta band activity as correlates of response preparation and execution.
Twenty students (14 women; mean age = 25.0 years, SD = 2.3 years) of the University of Magdeburg took part in the experiment after giving informed consent. All participants were right-handed and free of any neurological or psychiatric disorder (self-report). Data of five participants were excluded from further analysis because of extensive artifacts (n = 4) or technical problems (n = 1). The study was performed in agreement with the Declaration of Helsinki and approved by the ethics committee of the University of Magdeburg. Participants received money or course credit.
Task and Procedure
We applied a modified variant of the Eriksen flanker task (Eriksen & Eriksen, 1974) that required the participants to respond to the central arrow in an array of five arrows (with the right hand following a right-directed arrow and with the left hand following a left-directed arrow). The four surrounding arrows were either compatible or incompatible to the central arrow. We presented 37.5% of compatible and 37.5% of incompatible trials. In 12.5% of trials, we included stop-signal trials in which the central green arrow changed to red after a variable delay, indicating participants to inhibit the response in these trials. Finally, the remaining 12.5% of trials featured a change signal as the central arrow changed after a variable delay its direction indicating the participant to press the other response button (Figure 1A). Thus, if the arrow pointed first right and then left, the right-hand response had to be inhibited and a left-hand response was required. The delay was adapted on a trial-by-trial basis by means of a staircase tracking algorithm separately for stop-signal and change-signal trials (Band & van Boxtel, 1999). The signal delay was set to 140 msec initially. After a successful inhibition, the signal delay was increased by 10 msec (making the inhibition harder), after a failed inhibition the signal delay was reduced by 10 msec. This procedure was applied to yield a rate of 50% of successful inhibitions. The order of go, stop, and change trials was randomized. The SSRT was calculated by subtracting the participant's average stop-signal delay (with an inhibition rate of 50%) from the median RT of correct go responses (Band et al., 2003). The CSRT is calculated, respectively, by subtracting the participant's average change-signal delay from the median RT of correct go responses.
Each stimulus array was presented in the middle of the screen. Stimulus duration was 300 msec, and the SOA was between 900 and 1100 msec. Participants received 30 training trials to get acquainted with the task. The experiment was divided in forty blocks, each comprising 96 trials, resulting in a total of 3840 trials. The duration of the experiment was about 1.5 hours. Participants were informed that the experiment was programmed in a way that made it impossible for them to perform correctly every stop and change trial. This was explained to prevent them from slowing down in the primary task. They were instructed to perform as fast and accurately as possible.
The EEG was recorded from 29 tin electrodes mounted in an elastic cap (Easycap; positions: Fp1/2, F3/4, C3/4, P3/4, O1/2, F7/8, T7/8, CP1/2, P7/8, FC1/2, FC5/6, CP5/6, PO3/4, Fz, Cz, and Pz) with reference electrodes placed on the right and left mastoid. During recording, all scalp electrodes were referenced against the right mastoid and off-line rereferenced against the algebraic mean of the activity at the two mastoid processes. Electrode impedances were kept below 5 kΩ. To monitor horizontal eye movements, electrodes were placed on the outer canthus of the right and left eye. Vertical eye movements and blinks were monitored by electrodes placed below and above the right eye. EEG and EOG were recorded continuously with a band-pass filter of 0.01 to 70 Hz and digitized with a sampling rate of 250 Hz. Eye and muscle artifacts were rejected off-line automatically. For two participants with extensive blinks, eye movements were corrected with the technique of second-order blind identification (Joyce, Gorodnitsky, & Kutas, 2004).
Visual inspection of the ERPs and behavioral data suggested slight differences between compatible and incompatible stop-signal and change-signal trials. To avoid confusion of the additional factor “compatibility,” we restricted all behavioral and EEG analyses on compatible trials unless otherwise stated. ERPs and LRPs for incongruent trials are presented as supplementary on-line material (Figure 1). For the change-signal trials, that means that the go-signal was compatible, changing to an incompatible stimulus configuration.
RTs and error rates were computed for both the compatible and the incompatible go trials and statistically tested for differences with paired sample t tests. We also assessed RTs of responses in the change and stop condition to test for differences compared with correct go trials. Previous studies have found slower responses in trials after choice errors compared with trials after correct responses (Rabbitt, 1966), a phenomenon dubbed post-error slowing. Other studies have reported slower responses after failed inhibitions in stop-signal trials, partly also after correctly inhibited stop trials (Boehler et al., 2009; Verbruggen & Logan, 2008a; Krämer et al., 2007; Rieger & Gauggel, 1999). To assess possible slowing after stop and change trials, we computed the average RT in correct go trials after stop and change trials (separately for inhibited and error trials) and compared it with the average RT of correct go trials preceding stop and change trials. Note that the correct go trial preceding stop or change trials were themselves again following another correct go trial (yielding thus sequences of go–go–stop/change–go). To increase the number of analyzed sequences, these analyses were done for both compatible and incompatible change and stop trials together. In contrast to the common procedure to compare post-error trials with post-go trials in general, we were thereby able to also compare whether go trials before successful and failed inhibitions did significantly differ. To test for statistical significance, we subjected mean RTs for pre- and posttrials for the different conditions to a repeated measures ANOVA with the within-subject factors Stop (stop-all vs. stop-change), Inhibition (successful vs. failed), and Order (pre vs. post).
We examined the ERPs in stop and change trials in three different ways on the basis of previous results in the stop-signal task (Ramautar et al., 2004, 2006). First, we computed the ERPs stimulus locked to the go stimulus to assess possible differences in the processing of the go signal in trials of successful and failed inhibitions (see Ramautar et al., 2004, 2006). Second, the ERPs stimulus locked to the stop or change stimulus were computed, again separately for successful and failed inhibitions. Successful and failed inhibitions systematically differ in the delay of the stop/change signal with successful inhibitions being associated with on average shorter delays. These conditions differ in the overlap of ERPs to the go and the stop/change stimulus. To account for this, we matched trials in these two conditions in terms of the delay of the second stimulus. This was done by finding for every successfully inhibited trial an error trial with a corresponding stop- or change-signal delay (SD = 1 time point, i.e., 4 msec). This matching procedure resulted in on average 93 (SD = 7) stop trials and 90 (SD = 7) change trials per subject. Thirdly, to assess differences in the stop-signal-processing irrespective of differences in the go-stimulus processing, we computed difference waveforms separately for successful and failed inhibitions. Specifically, we shifted go-ERPs from the respective RT distribution (fast reactions for errors and slow reactions for successful inhibitions) across the range of individuals' stop- or change-signal delays weighted by the actual occurrence of that delay and averaged them (for visualization, see Supplementary Figure 2). These “virtual” go-ERPs were then subtracted from successful and failed inhibitions. Note that this procedure is similar to the ADJAR technique (Woldorff, 1993), except that go-ERPs are computed separately for fast versus slow reactions and then shifted and averaged for further comparison with failed and successful inhibitions, respectively (Ramautar et al., 2004, 2006).
LRPs were computed by a double subtraction using the electrodes C3 and C4 with the following formula: LRP = left hand (C4 − C3) − right hand (C4 − C3). The left hand and the right hand refer to the response hand indicated by the go stimulus. For change trials, a negative LRP results from an increased negativity contralateral to the wrong response hand (indicated by the go stimulus), a positive deflection reflects an increased negativity contralateral to the correct response hand (indicated by the change stimulus).
To assess statistical differences between the conditions, average amplitudes in the time windows of interest were subjected to repeated measures ANOVAs. The time windows of interest were based on previous studies (Ramautar et al., 2006; Schmajuk et al., 2006) and visual inspection of the peak latency of the components of interest. As visual inspection of the ERPs suggested centrally focused effects and following previous studies (Krämer et al., 2007; Ramautar et al., 2004, 2006), only midline electrodes were included in statistical analyses. The ANOVAs comprised the within-subject factors Stop (stop-all vs. stop-change), Inhibition (successful vs. failed), and Electrode (Fz, Cz, Pz). To study the time course of LRPs, we performed repeated measures ANOVAs for consecutive 50-msec time windows from go-stimulus onset until the average RT of stop and change errors, that is, 400 msec. The ANOVAs comprised the factors Stop (stop-all vs. stop-change) and Inhibition (successful vs. failed). For visualization purpose only, ERPs are depicted with a 15-Hz low-pass filter in the figures, but statistical analyses were performed on unfiltered data, unless otherwise stated.
To study the time course of mu and beta band changes with respect to response inhibition, we performed repeated measures ANOVAs for consecutive 50-msec time windows from go-stimulus onset until the average RT of stop and change errors. We thus analyzed eight time windows from 0 to 400 msec (relative to go-stimulus onset). To account for multiple comparisons, we considered relevant only effects that yielded significance in three consecutive time windows. The ANOVA comprised the factors Stop (stop-all vs. stop-change), Inhibition (successful vs. failed), Anterior-Posterior (F3/4, C3/4, P3/4), and Hemisphere (contralateral vs. ipsilateral).
For all statistical effects involving more than one degree of freedom in the numerator, the Huynh–Feldt correction was applied to correct for possible violations of the sphericity assumption (Huynh & Feldt, 1976). We report the uncorrected degrees of freedom and the corrected probabilities.
Participants reacted faster and more accurately in compatible (mean RT = 401 msec, SD = 51 msec; mean error rate 2.4, SD = 2.1) than incompatible go trials (mean RT = 424 msec, SD = 51 msec; mean error rate = 5.1, SD = 4.5), in line with usual findings in the flanker task; RT difference, t(14) = 9.2, p < .001; error rate difference t(14) = 4.2, p < .001.
RTs of errors in both stop trials (mean = 364 msec, SD = 39 msec) and change trials (mean = 361 msec, SD = 38 msec) were faster than correct go responses (both t > 7.2, p < .001) but did not differ from each other (p > .2). RTs of correctly changed responses were 582 msec (SD = 87, relative to the go-signal; 417 msec relative to the change signal). Participants inhibited on average 49.0% (SD = 1.8) of stop-signal trials and correctly changed to the new response in 48.7% (SD = 2.2) of the change-signal trials. Participants pressed the wrong response button (i.e., the button not indicated by the arrow) in a small percentage of failed stop trials (mean = 2.3%, SD = 2.4%). The variability (SD) of the stop- and change-signal delay within each participant was in average 30 msec for change trials and 33 msec for stop trials and thus slightly higher in stop trials, t(14) = 2.2, p = .04. Interestingly, the SSRT (mean = 246 msec, SD = 31 msec) was significantly longer than the CSRT (mean = 224 msec, SD = 27), t(14) = −7.7, p < .001 (Figure 1B), in contrast to the results of De Jong et al. (1995). The average stop-signal delay in inhibited trials was accordingly shorter (mean = 143 msec, SD = 56 msec) than the average change-signal delay in inhibited trials (mean = 165 msec, SD = 50 msec).
The results regarding post-stop and post-change slowing are shown in Figure 1B. The repeated measures ANOVA with the within-subject factors Stop (stop-all vs. stop-change), Inhibition (successful vs. failed inhibition), and Order (pre vs. post) yielded a significant interaction of all three factors, F(1, 14) = 21.7, p < .001. Subsequent paired-sample t tests confirmed order effects for all four trial types (paired-sample t tests, all p values < .01). However, as can be assessed from the figure, participants got faster after inhibited stop trials but slower after both error trials and correct change trials. Note that the RT in correct change trials is considerably slower (∼200 msec) than both errors and correct go trials, which likely accounts for the slower RT in the next trial. In addition to the differences between trials preceding and following stop or change trials, it is also noteworthy that trials preceding errors were significantly faster compared with those preceding inhibited trials: stop trials, t(14) = 3.98, p = .001; change trials, t(14) = 3.97, p = .001.
The ERPs for stop- and change-signal trials are shown in Figures 2, 3 and 4. The go-stimulus–locked data (Figure 2) revealed an enhanced P3 in failed compared with successful inhibitions for both stop and change trials. In stop trials, the typical stop-N2 and P3 were observed in the stop-stimulus–locked ERPs (Figures 3 and 4). However, the stop N2 was absent in change-signal trials (Figures 3 and 4). After failed inhibitions, we observed an additional later frontal negativity (error-related negativity [ERN]), supposedly related to error processing (Figure 4). In the following, we will first present the results for the go-stimulus–locked data and then for the stop-/change-stimulus–locked data.
Note that ERPs of successful and failed inhibitions were matched with respect to the stop-/change-signal delay, and there is no difference in the overlap of go- and stop-ERPs between these conditions. Go-stimulus–locked ERPs showed an increased P3 in trials of failed versus successful inhibitions with a maximum at posterior electrodes (Figure 2; P3 is highlighted). This effect was found for both stop- and change trials. To test this, we submitted the average amplitude in the time window around the P3 (300–400 msec) to a repeated measures ANOVA with the within-subject factors Stop, Inhibition, and Electrode (Fz, Cz, Pz). A significant main effect of Inhibition, F(1, 14) = 18.40, p = .001, confirmed the increased positivity for errors compared with inhibited trials with the typical maximum at Pz, Inhibition × Electrode interaction, F(2, 28) = 5.54, p = .017, FPz > FCz > FFz. No main effect or interaction with the factor Stop was observed (all p values > .1). A comparable P3 difference was obvious when contrasting go trials of the respective RT distribution, that is, fast and slow reactions. Trials with fast reactions were associated with an enhanced P3 amplitude compared with slower reactions, paired t test, t(14) = 3.4, p = .004 (Figure 2C, and for all three midline electrodes, see Supplementary Figure 3).
As can be assessed from Figure 2, the P3 differences developed after the average presentation time of the stop and change signals. To verify that the P3 differences reflected differential processing of the go and not of the stop/change stimuli, we reexamined this difference in the stop- and change-stimulus–locked data (Figure 3). Here, the difference corresponded to the time window 100 to 200 msec. Again, successful and failed inhibitions were matched with respect to the stop-/change-signal delay, and there is no difference in the overlap of go- and stop-ERPs. If the increased positivity was related to the go stimulus as opposed to the stop/change stimulus, the effect should be smaller in the stop-/change-stimulus–locked data. Indeed, the main effect inhibition did not yield significance in the respective ANOVA, F(1, 14) = 2.31, p = .15, but the interaction Inhibition × Electrode did, F(1, 14) = 9.65, p = .003. We compared the two effects in an ANOVA with the additional factor Reference Event, which refers to go- versus stop-/change-stimulus–locked ERPs. A significant interaction was found for Reference Event × Inhibition, F(1, 14) = 14.29, p = .002. The difference in the positivity at Pz was clearly smaller in stop-/change-stimulus ERPs (0.79 μV) compared with go-stimulus ERPs (1.54 μV).
The ERP stimulus locked to the stop and change stimuli is depicted in Figures 3 and 4. As addressed in the previous paragraph, the stop-/change-stimulus–locked ERPs showed early on a larger posterior positivity for failed inhibitions. These differences were present despite a matching in terms of the stop-stimulus delay in these conditions and indicated differences in the processing of the go stimulus. To assess stop-stimulus–related ERPs independently of go-related differences, we computed difference waveforms (subtracting aligned go-ERPs; Figure 4) as described in the Methods section (see also Supplementary Figure 2).
The stop-stimulus–locked ERPs for both inhibited and noninhibited trials (Figure 4A and B) showed a clear N2, peaking around 240 msec (posterior earlier than anterior), which was more extended in failed inhibitions. No N2 could be detected in successful change trials, but a later negativity around 300 msec was present in the change-error trials (Figure 4B). The N2 and the later negativity are marked in Figure 4, with the N2 being marked with light gray boxes. As this later negativity was seen in error trials only, we assumed it to reflect error processing (ERN) rather than inhibitory processes. To assess the early difference in the N2 time range, we subjected the mean amplitude of this time window (220–280 msec) to a repeated measures ANOVA with the factors Stop (stop-all vs. stop-change), Inhibition (failed vs. successful), and Electrode (Fz, Cz, Pz). A significant main effect of Stop, F(1, 14) = 34.79, p < .001, and Stop × Electrode interaction, F(2, 28) = 8.55, p = .005, reflected the increased negativity in stop compared with change trials with a maximum at frontocentral electrodes (Figure 4A). Post hoc comparisons showed significant effects of the factor Stop for all three electrodes (all p values < .01), but higher F values for Fz and Cz. In addition, failed inhibitions demonstrated an increased negativity relative to successful inhibitions; Inhibition × Electrode, F(2, 28) = 3.752, p = .046; all three electrodes, p < .05 and FCz > FPz > FFz. These two effects did not interact (both Stop × Inhibition and Stop × Inhibition × Electrode: p > .1).
To evaluate differences in the later time window (280–340 msec), we subjected the average amplitude to a second repeated measures ANOVA with the same factors as above. The results were clearly different: We observed a large main effect of Inhibition, F(1, 14) = 15.98, p = .001, and an interaction of Inhibition and Electrode, F(2, 28) = 11.01, p = .001, reflecting the increased negativity in failed compared with successful inhibitions over frontocentral electrodes (Fz and Cz: p < .003; Pz: ns; Figure 4B). This effect did not interact with the factor Stop (F < 1). Neither the main effect of Stop, F(1, 14) = 3.17, p = .097, nor the Stop × Electrode interaction (F < 1) was significant.
If the negativity in change trials was related to the erroneous response rather than to inhibition, it should be clearly detectable in the response-locked ERPs. We directly compared the response-locked ERPs for change and stop errors. We computed the difference waves of the response-locked ERPs in error trials relative to correct reactions in go trials with a pre-response baseline of 400 msec. To best assess the ERN, we band-pass filtered the ERPs with 2 to 8 Hz (for similar analyses, see Krämer et al., 2007). The results are shown in Figure 4D. We detected a clear ERN after both stop and change errors, which did not differ from each other. This was verified with a paired t test comparing the average amplitude in a 20- to 120-msec time window (p > .05). This strengthens the argument that erroneous stop and change trials did not differ in terms of error-related activity, but only in the earlier N2.
Lateralized Readiness Potentials
LRPs were clearly detectable in noninhibited as well as in inhibited trials both in stop and in change trials (Figure 4E). The statistical analyses confirmed increased LRP amplitudes in errors compared with inhibited trials from 150 msec after go-stimulus onset on (main effect Inhibition in all time windows 150 to 400 msec: p < .05). To further test whether LRPs for inhibited trials did significantly differ from 0, we ran one-sample t tests separately for inhibited stop and change trials. Under both conditions, LRPs were different from 0 between 250 and 350 msec after go-stimulus onset (all p values < .05).
Time-frequency Data: Mu Band (10–13 Hz)
The analysis of the mu band revealed differences between stop-all and stop-change trials (Figure 5A and B), such that (i) inhibited change trials were associated with a stronger power decrease over ipsilateral electrodes compared with inhibited stop trials and (ii) change errors showed a stronger power decrease over contralateral electrodes relative to stop errors (200–400 msec; Figure 5C). This was reflected in a significant threefold Stop × Inhibition × Hemisphere interaction between 200 and 400 msec; all time windows, F(1, 14) > 11.7, p < .01. The subsequent ANOVA for error trials revealed a significant main effect of Stop over the contralateral hemisphere from 200 to 400 msec; all time windows, F(1, 14) > 6.4, p < .03. As can be assessed from Figure 5C, the mu decrease was stronger for change compared with stop errors over frontocentral electrodes. The subsequent ANOVA for inhibited trials yielded a significant main effect of Stop over the ipsilateral hemisphere from 250 to 400 msec; all time windows, F(1, 14) > 5.6, p < .04. Note that for change trials, ipsilateral (to the first target's direction) corresponds to contralateral to the new target's direction. As depicted in Figure 5C, the mu decrease was stronger for inhibited change compared with stop trials over centroparietal electrodes.
Time-frequency Data: Beta Band (15–25 Hz)
Inhibited trials demonstrated a relative power increase compared with failed inhibitions around the time of the erroneous responses (∼350 msec; Figure 5D and E). This was confirmed by an Inhibition × Anterior-Posterior interaction in the time windows between 250 and 400 msec; 250–300 msec, F(2, 28) = 5.0, p = .02; 300–400 msec, F(2, 28) > 9.1, p < .002. The beta difference was focused over central electrodes (main effect of Inhibition in all time windows: C3/C4: p < .02; F3/F4 and P3/P4: p > .2; Figure 5F). In addition, error trials showed a stronger beta power decrease from around 50 msec after the go-stimulus onset. This was reflected in a significant main effect of Inhibition between 50 and 200 msec; all time windows, F(1, 14) > 4.8, p < .05 (Figure 5F). This effect did not interact with the stop condition or electrode location (all respective effects: p > .05).
We investigated whether electrophysiological data can provide evidence for distinct mechanisms subserving inhibitory action control when stopping or changing a planned response. Under both the stop and the change conditions, we observed go-stimulus–related differences in a parietal positivity (“P3”) between successful and failed inhibitions, indicating that differences in the processing of the go-signal determined whether responses could be inhibited or not. We also observed a beta decrease in error trials already shortly after the go signal, suggesting a faster response preparation in these trials. We detected clear differences in inhibition-related effects between the stop and the change conditions. In contrast to stop-signal trials, we did not observe the typical inhibition-related frontal negativity (“N2”) in change-signal trials. Moreover, inhibited stop trials were associated with a reduced parietal mu power decrease compared with inhibited change trials. The results confirm previous data suggestive of a dissociation between situations calling for a change to another response and the inhibition of any response.
Response Preparation in Trials of Failed and Successful Inhibitions
We observed differences in response preparation between inhibited and noninhibited trials on the behavioral level as well as in terms of ERPs and oscillatory effects. Behaviorally, RTs in trials preceding an error were shorter compared with trials preceding a successful inhibition. This indicates that a bias for fast responding to the primary go task drove the behavior in the stop and change task. Moreover, these RT differences were reversed after stop and change trials, such that reactions after inhibited stop trials were faster and those after errors (as well as after correct change trials) got slower. This replicates previous studies showing reactive behavioral adaptations after stop trials (Verbruggen & Logan, 2008a; Krämer et al., 2007; Rieger & Gauggel, 1999). The evidence is less clear, however, when comparing behavioral changes after successful and failed inhibitions because some studies reported slower reactions after any stop trial (Boehler et al., 2009; Rieger & Gauggel, 1999) and others only after stop errors (Verbruggen & Logan, 2008a, 2008b). The latter would be reminiscent of the so-called “post-error slowing,” which is reliably found after errors in choice RT tasks (Rabbitt, 1966). The present results support flexible trial-by-trial adaptations manifested as a trade-off between responding fast to the go signal and accurately to the stop and change demands.
Both ERP and oscillatory correlates of go-stimulus processing and response preparation differed between trials of failed and successful inhibitions. Specifically, the parietally focused P3 to the go stimulus was larger in error trials compared with inhibited trials for both stop and change conditions. This strengthens earlier go-P3 findings in the stop task, reporting an increased P3 amplitude in erroneous trials (Ramautar et al., 2004). Despite being the most-studied ERP component, there is no clear consensus on what specific neural process(es) the P3 reflects (Polich & Kok, 1995; Picton, 1992). The most influential account of the P3 is the “context updating” hypothesis (Donchin & Coles, 1988; Donchin, 1981), interpreting it as a working memory update in response to task-relevant, unexpected events, and there is extensive literature on how the P3 is affected by probability, uncertainty, and resource allocation (Polich & Kok, 1995; Picton, 1992; Johnson, 1986). There is also evidence that the P3 is not only related to stimulus processing per se but in fact reflects the link between perceptual processing and response preparation (Verleger, Jaskowski, & Wascher, 2005). Others have argued that distinct subcomponents of this late positivity are associated with stimulus evaluation on the one hand and the cognitive response selection process on the other hand (Falkenstein, Hohnsbein, & Hoormann, 1994). An increased P3 amplitude in trials of failed inhibitions might thus indicate a stronger bias toward the primary task stimuli and toward fast responses (Ramautar et al., 2004).
This interpretation is supported by the fact that the motor-related beta power decrease was stronger in errors compared with inhibited trials already about 50 msec after go-signal onset, that is, before presentation of the stop or change stimulus. These early beta power differences were found to be less spatially focused compared with the beta effect around the time of the motor response (see discussion below), which suggests that it might not be limited to activity in motor areas. In fact, a recent article (Marco-Pallares et al., 2008) reported a frontal beta increase related to post-error slowing in a Flanker task. The authors interpreted this effect as a correlate of inhibitory cognitive control kicking in after committing an error. Using intracranical recordings, Swann et al. (2009) observed a right inferior frontal beta increase in successful stop trials suggestive of an inhibitory control process. Also in a go/no-go paradigm, Alegre et al. (2004) observed a frontal beta increase in no-go trials, supposedly reflecting motor inhibition. It is tempting to speculate that the early relative increase of beta power, which we observed in inhibited trials, reflects reduced excitability in motor areas and/or stronger prefrontal activity.
On the basis of our observation of trial-by-trial adaptations in RTs, we suggest that these differences in response preparation do not merely reflect random fluctuations but indicate flexible adjustments in the preparatory set to account for the go and stop demands. This fits with imaging results showing that the preparation to inhibit complements response inhibition in a stop-signal task (Chikazoe et al., 2009). Our results are also in line with recent evidence from an MEG study with the stop-signal paradigm (Boehler et al., 2009). The authors demonstrated that increased early visual responses to the go or stop signal predicted a failure or success in inhibition, respectively. Moreover, the visual N1 to the go-stimuli in trials after stop trials was shown to be reduced. This strengthened the author's argument that the N1 differences indicated a flexible allocation of attentional resources to the go or stop demands (Boehler et al., 2009). Note that we did not observe these N1 modulations in the present EEG experiment (data not shown).
Finally, our data have methodological implications for studies with the stop-signal paradigm and specifically the comparison of failed and successful inhibitions because they show that the performance in stop trials partially depends on the go process and not only on the stop process (Li et al., 2006). This is in agreement with the horse-race model of response inhibition, which posits that the slow but not the fast portion of the go RT distribution can be inhibited (Logan et al., 1984). For the comparison with go trials, this accordingly means that failed and successful inhibitions should be compared with go trials from the corresponding fast and slow fraction of the RT distribution, respectively (Ramautar et al., 2004, 2006).
Neural Mechanisms of Inhibitory Action Control
Interestingly, the CSRT was found to be faster than the SSRT, which is in contrast to the findings of De Jong et al. (1995), who reported an increased latency of the stop process in the stop-change condition. This may be because of task differences. In the present study, the change condition involved a change of the response hand only, whereas in the study of De Jong et al., participants were asked to give manual responses in the primary go task and respond with the foot in the stop-change trials. Notably, however, the behavioral differences between stop-all and stop-change trials in the present study point already at differences in the underlying mechanisms.
Stop signals elicited an increased frontal N2 compared with change signals, whereas no differences between these conditions were detected in error-related activity. There is ongoing debate on what the N2 in the stop-signal paradigm and in other cognitive control paradigms reflects (Folstein & Van Petten, 2008). In the stop-signal literature, the N2 is often viewed as a correlate of prefrontal inhibitory control, supposedly stemming from activity in right inferior frontal regions (Schmajuk et al., 2006) or medial frontal regions as the ACC (Huster, Westerhausen, Pantev, & Konrad, 2010; Ramautar et al., 2004, 2006; van Boxtel, van der Molen, Jennings, & Brunia, 2001). There is evidence supporting this notion. Schmajuk et al. (2006) demonstrated a right prefrontally enhanced N2 for successful compared with failed inhibitions, and attention deficit/hyperactivity disorder children showed a reduction of this N2 effect together with a slower SSRT compared with healthy children (Pliszka et al., 2000). Participants with fast compared with slow SSRTs have an increased stop N2 over frontomedial regions (van Boxtel et al., 2001), and the N2 has been shown to peak earlier in successful compared with failed inhibitions (Kok et al., 2004; Ramautar et al., 2004). There is also some evidence that the N2 is independent of the stop-signal modality (Ramautar et al., 2006). Note, however, that the direct comparison of successful and failed inhibitions is problematic because our results indicate an overlap of the N2 and ERN in failed inhibitions, which precludes a clear assessment of N2 amplitude and latency in these trials. In any case, the present data suggest that the frontal inhibitory processes indexed by the N2 are specific to the nonselective inhibition required in stop-all tasks. One might argue though that the N2 differences are reflecting an overlap with processes of movement replanning or conflict resolution in the change but not in the stop trials. We cannot resolve this question with the present data. Future studies would be needed to investigate whether the N2 differences between stop and change trials depend on movement replanning in the change task by manipulating the response selection process or response conflict.
The time course of the mu power decrease differentiated between stop and change trials, whereas a reduced beta decrease over central electrodes was found for both inhibited stop-all and stop-change trials. As previously noted, mu and beta power decrease before and during movements has been attributed to activity in the somatosensory cortex and the primary motor cortex, respectively (Ritter et al., 2009; Hari, 2006; Pfurtscheller & Lopes da Silva, 1999). The reduced beta decrease in inhibited trials corroborates previous findings of changes in beta power related to response inhibition in the stop-signal task (Swann et al., 2009; Marco-Pallares et al., 2008) and go/no-go tasks (Alegre et al., 2004; Leocani et al., 2001). It also converges with results from TMS studies showing reduced corticospinal excitability after stop signals, presumably caused by intracortical inhibitory networks in M1 (van den Wildenberg et al., 2010; Stinear, Coxon, & Byblow, 2009; Coxon, Stinear, & Byblow, 2006). We found a similarly reduced beta decrease in inhibited change trials, which indicates that in these trials the motor excitability was reduced. Recent behavioral studies have investigated more selective types of response inhibition where one out of two prepared responses had to be withheld while the other response had to be given. These studies found delayed RTs for the responding finger in these trials (Coxon, Stinear, & Byblow, 2007). This was taken as evidence for a global inhibitory process that transiently shuts down any motor output via the subthalamic nucleus, before the movement of the responding finger is initiated, supposedly involving activity in (pre-) SMA and premotor regions.
We also observed LRP differences between inhibited trials and errors but no differences between the stop and change condition. This is in contrast to the study of De Jong et al. (1995), who reported reduced LRPs for the change compared with the stop condition, which again might be caused by task differences. LRPs, the late part of the so-called Bereitschaftspotential (Kornhuber & Deecke, 1965), reflect response preparation in the contralateral primary motor cortex and lateral premotor cortex, a view supported by various source localization techniques (for a review, see Shibasaki & Hallett, 2006). Several studies demonstrated different temporospatial dynamics of LRPs and beta desynchronization though, suggesting that these reflect distinct neuronal mechanisms underlying response preparation and execution (Gladwin, 't Hart, & de Jong, 2008; de Jong, Gladwin, & 't Hart, 2006; Shibasaki & Hallett, 2006). Two points should be noted about the present LRP results. First, LRPs were also detectable in inhibited trials indicating response preparation up to premotor and primary motor areas even in these trials. Second and more important, stop and change trials did not differ in terms of LRPs, which indicates that these different task demands similarly modulated activity in premotor and primary motor cortex.
In contrast to the beta band effects, the temporal dynamics in the mu band differentiated between stop-all and stop-change conditions. Specifically, change errors demonstrated a stronger mu decrease compared with stop errors over frontocentral regions contralateral to the inhibited response hand. The topography was suggestive of differences not only in central presumably motor-related areas but additionally in frontal areas (Figure 5). Moreover, inhibited change trials were associated with a stronger mu power decrease relative to inhibited stop trials over centroparietal electrodes ipsilateral to the inhibited hand. Note that in change trials, ipsilateral to the inhibited hand is equivalent with contralateral to the side of the executed response. The results indicate differences in response preparation in parietal, possibly sensorimotor areas between stop-all and stop-change conditions. Notably, this effect was already apparent shortly after presentation of the stop and change signals and more than 200 msec before the execution of the alternative response in change trials. Previous EEG studies on the stop-signal task have not addressed oscillatory effects in the mu band, but research with a go/no-go-paradigm showed reduced mu desynchronization in no-go trials over sensorimotor regions (Alegre et al., 2004). Importantly, our study suggests that the more selective response control in the stop-change paradigm modulated activity in primary motor cortex mainly as indicated by the LRP and beta results, whereas the inhibitory control required in stop-all trials seems to affect parietal regions.
Because of the limited spatial resolution of EEG, claims about the neural sources of the current neurophysiological effects remain speculative. As the present data were based on low-density recordings (29 scalp electrodes), we did not perform source analyses, which might be a useful approach with higher density recordings (Huster et al., 2010; Pastötter, Hanslmayr, & Bäuml, 2008; Guderian & Düzel, 2005). Evidence from a lesion study points to the pre-SMA as especially important for the inhibition of a response when having to execute another one instead (Nachev et al., 2007). Interestingly, a recent fMRI study compared selective and nonselective response inhibition by asking participants to execute two responses simultaneously and inhibit either one or both responses in a subset of trials (Coxon, Stinear, & Byblow, 2009). The authors reported activity in pre-SMA in trials of selective inhibition but not in nonselective stop trials. Although the task requirements and temporal dynamics of the response execution in their task were different to the stop-change task, the involved processes are likely similar, as in both cases a specific response needs to be withheld while an alternative response has to be executed. Electrophysiological data from primates confirm that the pre-SMA is involved in this kind of behavioral switching, supposedly by suppressing the unwanted action first and then facilitating the alternative response (Isoda & Hikosaka, 2007). In the present study, we observed increased frontal activity for the nonselective inhibition in stop-all trials, possibly stemming from frontomedial or frontolateral areas.
The present results address two key issues in research on inhibitory motor control, namely, whether there are different neural mechanisms of response inhibition and how variability in the go process contributes to the performance in stop trials (Verbruggen & Logan, 2008c). Increased frontal activity was found when any response had to be withheld compared with situations that required to change to another response. Moreover, analyses of event-related oscillatory changes and LRPs suggested an effect of stop-change trials on activity in primary motor cortex mainly, whereas nonselective response inhibition showed additional modulations over parietal, presumably sensorimotor areas. Conversely, go-stimulus processing and response preparation differed between successful and failed inhibitions, as verified by modulations of the go-P3 and the beta band power decrease. Together with observations of behavioral adaptations on a trial-by-trial basis, these effects point to a flexible allocation of resources between go and stop demands. This provides neurophysiological evidence for the horse-race model of response inhibition because we observed systematic differences in go-related processes between inhibited and noninhibited trials.
The authors thank Kathrin Ibert for help during data acquisition. U. M. K. was supported by a fellowship from the DFG (KR 3691/1-1). T. F. M. was supported by various grants of the DFG.
Reprint requests should be sent to Ulrike M. Krämer, Helen Wills Neuroscience Institute, University of California, 132 Barker Hall, Berkeley, CA 94720, or via e-mail: firstname.lastname@example.org.