Establishing a Right Frontal Beta Signature for Stopping Action in Scalp EEG: Implications for Testing Inhibitory Control in Other Task Contexts

Johanna Wagner1, Jan R. Wessel2,3, Ayda Ghahremani4,5, and Adam R. Aron1

Abstract ■ Many studies have examined the rapid stopping of action pants in total) performing versions of the standard stop signal task. as a proxy of human self-control. Several methods have shown We identified a spectral power increase in the band 13–20 Hz that a critical focus for stopping is the right inferior frontal cor- that occurs after the stop signal, but before the time of stopping tex. Moreover, studies have shown beta elapses, with a right frontal topography in the EEG. This right band power increases in the right inferior frontal cortex and frontal beta band increase was significantly larger for success- in the BG for successful versus failed stop trials, before the time ful compared with failed stops in two of the three studies. We of stopping elapses, perhaps underpinning a prefrontal–BG also tested the hypothesis that unexpected events recruit the network for inhibitory control. Here, we tested whether the same frontal system for stopping. Indeed, we show that the same signature might be visible in scalp electroencephalogra- stopping-related right-lateralized frontal beta signature was also phy (EEG)—which would open important avenues for using active after unexpected events (and we accordingly provide this signature in studies of the recruitment and timing of pre- data and scripts for the method). These results validate a right frontal inhibitory control. We used independent component frontal beta signature in the EEG as a temporally precise and analysis and time–frequency approaches to analyze EEG from functionally significant neural marker of the response inhibition three different cohorts of healthy young volunteers (48 partici- process. ■

INTRODUCTION (see reviews by Schmidt & Berke, 2017; Aron et al., One component of executive function is top–down inhib- 2016; Jahanshahi, Obeso, Rothwell, & Obeso, 2015; itory control. The domain in which this is best explored is Kenemans, 2015; Schall & Godlove, 2012; Chambers, the stopping of action. In the lab, this is often operation- Garavan, & Bellgrove, 2009). Here, we focus on the alized using the stop signal task (SST; Logan & Cowan, cortical aspect. Lesion and stimulation approaches in 1984). On each trial, participants initiate a go response, humans have pointed to the critical importance of the which they sometimes have to try to stop when a stop right inferior frontal cortex (rIFC; reviewed by Aron, signal is subsequently presented. The latency of the stop Robbins, & Poldrack, 2014), and this region is also acti- process, referred to as stop signal RT (SSRT), can be es- vated in many fMRI studies (for meta-analysis, see, e.g., timated with models that take into account the stop sig- Cai, Ryali, Chen, Li, & Menon, 2014; Swick, Ashley, & nal delay (SSD), the probability of stopping, and the RT Turken, 2011). Moreover, electrocorticography (ECoG) on go trials (Logan & Cowan, 1984). Understanding the studies, which record (EEG) sig- neural architecture of the putative inhibitory process in nals directly from the surface of the , have shown stopping action is important for several reasons. One of that one electrophysiological signature of stopping is a these is that if a clear brain signature of inhibitory control power increase in the rIFC for successful versus failed can be established for the stopping of action, it could then stop trials, before the time of stopping elapses (i.e., SSRT; be asked if the same signature occurs in other behavioral Wessel, Conner, Aron, & Tandon, 2013; Swann et al., 2009; and cognitive contexts. but see Fonken et al., 2016). These power increases oc- There have been many studies of the neural correlates curred in the beta band, as do power increases in the BG of action stopping, specifically for the stop signal (and during stopping (e.g., Wessel, Ghahremani, et al., 2016; related) tasks, across species, and using many methods Ray et al., 2012), perhaps underpinning a prefrontal–BG network for inhibitory control (reviewed by Aron, Herz, Brown, Forstmann, & Zaghloul, 2016; Zavala, Zaghloul, 1University of California, San Diego, 2University of Iowa, 3Univer- & Brown, 2015). The cortical electrophysiology results sity of Iowa Hospitals and Clinics, 4Krembil Research Institute, are especially important because they show both the fre- Toronto, Canada, 5University of Toronto quency and the timing of activation of a region known

© 2017 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 30:1, pp. 107–118 doi:10.1162/jocn_a_01183 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 to be critical for the stopping process. Here, we test METHODS whether a similar signature is detectable in scalp EEG. If Participants so, this would open avenues for testing the generalizabil- ity of inhibitory control in other task contexts, moreover We analyzed data from participants in three different in healthy individuals (rather than surgical patients). It studies, all using slightly different versions of the stop would also open up more ecologically relevant scenarios signal paradigm. Other results from these data sets were because scalp EEG is much more portable than ECoG reported in Wessel and Aron (2013, 2014, 2015). Impor- or fMRI. tantly, one of the studies also included an unexpected We used independent component analysis (ICA) and perceptual events task (Wessel & Aron, 2013). The number time–frequency approaches to analyze scalp EEG data of participants was as follows: Study 1: 11, Study 2: 13, from three different cohorts of healthy young volunteers and Study 3: 24 (overall mean age = 21.47 years, SEM = performing versions of the standard SST. First, based on 0.77, 34 women, 3 left-handed). All participants provided the results from ECoG (above), we expected to detect a written informed consent according to the local ethics spectral power increase in the band 13–20 Hz that occurs committee at the University of California, San Diego. after the stop signal, but before SSRT, with a right frontal topography in the scalp EEG. We also predicted that this right frontal beta band increase would be larger for Stop Signal Versions successful compared with failed stops. Second, we The SST requires participants to make a quick motor re- aimed to establish the reliability of this signature by test- sponse to a go signal. On a minority of trials, a stop signal ing for it in the three independent data sets. Third, we is presented, which requires participants to try to stop aimed to test if this same signature of inhibitory control their response. The delay at which the stop signal occurs occurs in a different behavioral context—the processing after the go signal (SSD) was initially set to 200 msec in all of unexpected events. We could test this because, for studies and then dynamically adjusted during the task one of the data sets, each participant had performed according to the participant’s performance; that is, SSD an unexpected events task and then an SST. For the was increased by 50 msec when participants successfully unexpected events task, on each trial, the participant pre- stopped and decreased by 50 msec when they failed to pares to respond—and just before the imperative stim- stop their response. The staircase method typically yields ulus, a sound occurs; on a minority of trials, the sound a probability of successful stopping of around .5. See is a deviant (we refer to these as “novels”; cf. Vachon, Figure 1 for an overview of the different versions of the Hughes, & Jones, 2012; Wessel, Danielmeier, Morton, & SST. Ullsperger, 2012; Gentsch, Ullsperger, & Ullsperger, 2009; Parmentier, Elford, Escera, Andrés, & San Miguel, 2008; Barcelo, Escera, Corral, & Periáñez, 2006). Whereas an earlier study (Wessel & Aron, 2013) showed that un- expected events and stopping recruit a common central midline component in the EEG—a negative deflection at200 msec (N2) and a positive deflection at 300 msec (P3)—here we aimed to test if there was a common right frontal beta increase for unexpected events and stopping. As above, this was motivated by the putative specificity of the right frontal beta as a signature of rIFC (based on ECoG). We suppose this reflects a different part of the stop process compared with the N2/P3 complex in the EEG. We tested the idea of a common process as follows. In each participant, we concatenated the EEG data from the unexpected events task and the SST. We then used ICA to decompose the concatenated EEG data into tem- porally independent source signals (Onton, Westerfield, Townsend, & Makeig, 2006). The goal was to identify a component that has the characteristic frontal beta signature of stopping and then look at whether this component is active after unexpected events (Wessel, 2016). If there is a common inhibitory process, then the same independent component (IC) with a right frontal topography identified in the SST should show a spectral power increase in the beta band after unexpected Figure 1. Three SSTs for the three data sets. Note the higher events. percentage of stop trials in Study 3.

108 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Study 1: Manual Responses, Visual Stop Signals, (which should be faster than go RT according to the n = 11 (Wessel & Aron, 2014) race model; Logan & Cowan, 1984), the probability of stopping (which should be in the range of .4 and .6, con- Go signals were white arrows pointing left or right at the firming the efficacy of the SSD staircase algorithm and center of the screen. Participants pressed a left or right the prospect of accurately estimating SSRT), and SSRT key on a standard computer keyboard. The stop signal itself. SSRT was computed using the mean method, that was a red exclamation mark over the white arrow. The is, subtracting the observed mean SSD on stop trials from probability of a stop signal was 25%. Participants per- the observed mean RT on go trials. formed 240 trials split across six blocks. Trial timing was as follows: 500 msec fixation, 1500 msec response dead- line, 500 msec intertrial interval (ITI), overall trial duration: EEG Recording 2500 msec. EEG data were recorded using a BioSemi system (Bio- Semi Instrumentation, Amsterdam, The Netherlands). Study 2: Verbal Responses, Visual Stop Signals, In Studies 2 and 3, 64 scalp electrode sites in the extended n = 13 (Wessel & Aron, 2013) 5% international 10/20 electrode system (Oostenveld Go signals were white letters (K or T) at the center of & Praamstra, 2001) were recorded at a 512-Hz sampling the screen. Participants spoke the letter into a micro- rate. In Study 1, 128 scalp electrode sites were recorded phone. The stop signal was the white letter turning at a 1024-Hz sampling rate. The 128-channel montage red. The probability of a stop signal was 25%. Partici- was in accordance with the BioSemi-designed equiradial pants performed 400 trials split across 10 blocks. Trial system. In each data set, we also recorded six EOG elec- timing was as follows: 1000 msec fixation, 1000 msec re- trodes placed above and below each eye and at the outer sponse deadline, 500 msec ITI, overall trial duration: canthi of the eyes. 2500 msec.

EEG Preprocessing Study 3: Manual Responses, Visual Stop Signals, n = 24 (Wessel & Aron, 2015) EEG data analysis was performed using custom scripts in MATLAB2015b (The MathWorks, Natick, MA) incor- Go signals were white arrows pointing left or right in the porating EEGLAB 14.0b functions (sccn.ucsd.edu/ center of the screen. Participants pressed a left or right ; Delorme & Makeig, 2004). For Study 3, the EEG key on a standard computer keyboard. The stop signal data were down-sampled to 512 Hz. Preprocessing then was the white arrow turning red. The probability of a stop proceeded as follows. First, the EEG data were high- signal was 33%. Participants performed 300 trials split pass filtered at 2 Hz (to minimize slow drifts; zerophase across six blocks. Trial timing was as follows: 500 msec FIR filter order 3381) and low-pass filtered at 200 Hz fixation, 1000 msec response deadline, 1000 msec ITI, (zerophase FIR filter order 565). Second, EEG channels overall trial duration: 2500 msec. with prominent artifacts were identified by visual in- spection and removed. Subsequently, channels that Unexpected Events Task (Wessel & Aron, 2013) were decorrelated (r < .4) from neighboring channels for more than 1% of the time were automatically re- On each trial, a sound was played, followed by a letter. moved. On average, 122 of 134 channels for Study 1, Participants spoke the letter into a microphone. On 80% 64 of 70 channels for Study 2, and 68 of 70 channels of trials (standards), the sound was a 600-Hz sine wave; on for Study 3, per participant (Study 1: SD =3.1;Study2: 20% of trials (novels), a unique birdsong segment was SD =2.3;Study3:SD = 1.7) remained in the analy- presented. Trial timing was as follows: 1000 msec fixa- sis. Third, EEG data were then rereferenced to a com- tion plus a variable jitter (100–500 msec), 200 msec mon average reference. Fourth, the continuous EEG sound, 300 msec delay before letter appeared, 1000 msec data were then visually inspected for artifacts, and response deadline (for full details, see Wessel & Aron, affected segments were removed from further analysis. 2013). Participants performed 450 trials equally divided Fifth, to perform automatic rejection of affected seg- into six blocks of 75 trials (15 of which were novel trials ments, the data were partitioned into epochs of 0.5 sec. and 60 of which were standard trials). Only participants Epochs containing values exceeding the average of the in Study 2 did this task—before the SST (i.e., the order probability distribution of values across the data seg- was always unexpected events task followed by SST). ments by 5 SDs were rejected. On average, the number of stop and go trials and the percentage of EEG data that remained in the analysis were as follows: Study 1: 131 go, Behavioral Analysis 64 stop and 81% of EEG data; Study 2: 224 go, 94 stop and For the SST, for each participant, we estimated go RT, 80% of EEG data; Study 3: 175 go, 87 stop and 87% of number of erroneous or missed go trials, failed stop RT EEG data.

Wagner et al. 109 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 ICA Approach follows. First, we selected the relevant parts of each Stop Signal Data Sets feature. For example, for the ERP we selected the mean over trials of 0–600 msec after the stop signal, which re- The preprocessed EEG data were decomposed using sulted in approximately 307 data points (given the sam- adaptive mixture ICA (AMICA; Palmer, Makeig, Kreutz- pling rate of 512 Hz). We then reduced these 307 points Delgado, & Rao, 2008; Palmer, Kreutz-Delgado, & Makeig, to 10 dimensions using the PCA. This method finds orthog- 2006). AMICA is a generalization of the Infomax ICA onal subspaces that explain maximal variance of the data, (Makeig, Bell, Jung, & Sejnowski, 1996; Bell & Sejnowski, with the first principal component explaining the largest 1995) and multiple-mixture (Lewicki & Sejnowski, 2000; partofthevarianceofthedata.Weappliedthesamepro- Lee, Lewicki, Girolami, & Sejnowski, 1999) ICA approaches. cedure to the spectra, ERSP, and scalp maps. Dipoles have AMICA performed blind source separation of all pre- inherently only three dimensions (the Tailarach coordi- processed data for each participant individually based on nates x, y, z). We thus ended up with five feature vectors, the assumed temporal near-independence of the effective four each with 10 dimensions and one with three dimen- EEG sources (Makeig, Debener, Onton, & Delorme, 2004; sions (related to dipoles) for each IC. Makeig et al., 2002). ICA results in as many ICs for each data set as there are Weighting of feature vectors. The dimensionally re- channels recorded; thus, Studies 2 and 3 resulted in an duced feature vectors were then weighted for sub- average of respectively 64 and 68 ICs per participant, sequent clustering (i.e., dipole locations: weight 12; scalp whereas Study 1 resulted in an average of 122 ICs per projection: weight 4; power spectra: weight 3; ERPs: weight participant (i.e., some channels were removed for each 3; ERSPs: weight 10). These weights were chosen based participant). We then calculated a best-fitting single ECD on our hypothesis: We expected a beta band increase matched to the scalp projection of each IC source using a (13–20 Hz) with a right frontal scalp distribution and a standardized three-shell boundary element head model timing from 0 to 600 msec after the stop signal. We also implemented in the DIPFIT toolbox (Delorme & Makeig, know that an ERP occurs at around 300 msec following 2004; Oostenveld & Oostendorp, 2002). The IC scalp pro- the stop signal. The relatively large weight for the ERSPs jection or scalp map represents the relative weights with made clustering similarly sensitive to IC ERSP differences “ ” which the source projects to each of the scalp channels. as for dipole location. The theoretical basis of ICA assumes brain source loca- tions and projection maps to be spatially fixed, whereas Concatenation and clustering. The five feature vectors “ ” source activations represent their activity time courses were then concatenated for each IC and further reduced (Onton et al., 2006). to 10 principal components using PCA. We then ran The standard electrode locations corresponding to the k-means clustering (k =11). extended 10–20 system were aligned with a standard brain model (Montreal Neurological Institute). The Bio- Outlier clusters. ICs were identified as outliers if their Semi equiradial system montage used in Study 1 and locations in the clustering vector space were 4 SDs from the 5% international 10/20 system partly overlap, allowing the obtained cluster centers. Only clusters including ICs us to align equiradial system electrode positions to the from more than half of the participants were considered standard 10/20 system and in the following to a standard for further analysis. ICs were separated into 11 clusters brain model (Montreal Neurological Institute) using elec- representing either muscle or cortical activity. Cortical trode landmarks such as Cz. clusters were then screened for remaining artifact ICs We excluded ICs from further analysis for which the by visually inspecting single IC spectra and ERSP images equivalent dipole model was located outside the brain for broadband activity from 20 to 100 Hz. It is possible and explained less than 85% of variance of the cor- that some ICs contain cortical activity mixed with EMG responding IC scalp map. or the clustering algorithm did not correctly assign all Below, we explain the process by which automatic ICs containing muscle activity to a separate cluster. Those clustering of ICs is done across participants. ICs were moved to the outlier cluster.

Overview. Clustering used feature vectors derived from (1) IC dipole locations, (2) scalp projection, (3) power Unexpected Events (Study 2 Only) spectra in the range of 3–200 Hz, (4) ERPs in the time The approach was slightly different. Now, in each partic- window from 0 to 600 msec following the stop signal, ipant, the EEG data from both the unexpected events and (5) event-related spectral perturbations (ERSPs) in task and the SST were preprocessed as above and the frequency range from 3 to 20 Hz and time window concatenated. Then ICA was run on the combined data. from 0 to 600 msec following the stop signal. The process of clustering ICs was automated, as above (using the same features). We then identified a cluster Preclustering dimensionality reduction. Before clus- that had a right-lateralized frontal scalp distribution (as tering, we reduced the dimensionality of the features as above for SST only) and showed a beta band increase

110 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 (13–20 Hz) relative to baseline in the stop trials of SST we then averaged a band of ±2 Hz around this selected that elapses following the stop signal but before SSRT. beta band, but now in the unexpected events data (the This cluster was defined as the stop cluster (the “proto- “candidate process”), and averaged over participants. type process”; see Wessel, 2016), and each component We also checked for a single trial correlation on novel therein (most participants contributed just one compo- trials between beta band power and RT for each partici- nent) was used to compute time–frequency decomoposi- pant. We selected the frequency band as described above tions for the unexpected events task (the “candidate and averaged beta power in single trials in the interval of process”). 150–300 msec after the unexpected event. We then cor- related those values to single trial RT. We did this analysis on the first half of the novel trials, because the effect of Hypothesis Testing Using the ERSP unexpected events in this study wore off over time. Stop Signal Task The data were segmented into time epochs relative to Effect Size onsets of the go cue (i.e., from −1 to 2 sec around the We computed effect size as a guide to powering future go cue). ERSPs (Makeig, 1993) were computed for each studies. This was for the right frontal beta power increase IC in the cluster of interest. We used Morlet wavelets for successful stop trials versus baseline. We selected a (Morlet, Arens, Fourgeau, & Glard, 1982) for time– time window of −150 to +50 msec around mean SSRT frequency analyses (4–30 Hz). Starting with three cycles for each study and averaged power values at each fre- at 4 Hz, we linearly increased the number of cycles used quency over time. We then averaged power over fre- by 0.5 as frequency increased. To generate ERSPs, single- quencies in the beta band from 13 to 20 Hz, resulting trial spectrograms were computed and time-warped to in one value for beta power increase and one value for the median RT latency for go trials and to the median baseline power per participant. We then computed one SSD latency for stop trials (across participants within overall value for all three studies for Cohen’s d. each study) using linear interpolation. This procedure aligned time points of RTs and stop signals over trials following each go cue. Relative changes in spectral power RESULTS were obtained by computing the mean difference between Behavior each single-trial log spectrogram and the mean baseline spectrum (the average log spectrum between −1and Stop Signal Task 0 sec preceding the go cue). If participants had more than Behavioral performance was fairly typical for healthy one IC in a cluster, the respective IC ERSP matrices were young participants (Table 1), with quite fast go RT, p(stop) averaged for that participant. Significant deviations from about 50%, few omissions or errors on go trials (error the baseline were detected using a nonparametric boot- rates at 1%, SEM = 0.3), and failed stop RT faster than go strap approach (Delorme & Makeig, 2004) and corrected RT, consistent with the race model. Go RT was longer for false discovery rate (Benjamini & Yekutieli, 2001) with a significance level set a priori at .05. Table 1. RTs and SSDs for Each Study

Unexpected Events Task Error Go RT FS RT SSRT SSD p(Stop) Rates The data were segmented into epochs relative to novel and standard tones (i.e., from −1to2secaroundthe Study 1 (Manual Responses, Visual Stop Signals) tone). Again, relative changes in spectral power were Mean 488 415 247 241 .52 .01 obtained by computing the mean difference between SEM 45.0 31.3 20.3 13.3 .02 .02 each single-trial log spectrogram and the mean baseline spectrum—in this case, the average log spectrum between −1 and 0 sec preceding the tone. Across-participant ERSP Study 2 (Verbal Responses, Visual Stop Signals) maps were then generated for novel trials, standard trials, Mean 557 495 257 300 .52 .001 and novel–standard trials. A follow-on analysis specifically tested whether un- SEM 14.8 13.9 6.4 13.3 .005 .0004 expected events recruited the beta band. To do this, we selected the beta frequency band for each participant individually using combined successful and failed stop Study 3 (Manual Responses, Visual Stop Signals) trials from SST (the “prototype process”). For each par- Mean 520 446 204 315 .52 .01 ticipant, we selected the band in the range of 13–20 Hz SEM 12.4 11.0 6.2 14.6 .005 .004 that was maximally modulated over time in the window 300–800 msec relative to the go signal. For each participant, Go RT = go trial RT; FS RT = failed stop trial RT.

Wagner et al. 111 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 than RT on failed stop trials in all participants, indicating beta band power increase was present for successful stop the validity of the race model. Mean SSDs were 241, 300, andfailedstoptrialsbutabsent for go trials. Further- and 315 msec in Studies 1, 2, and 3, respectively, and more, for Studies 1 and 2, the beta power increase was mean SSRTs were 247, 257, and 204 msec for Studies 1, significantly different from baseline only for the success- 2, and 3, respectively. These are typical values. Notably, ful stop trials, not for failed stop trials, and in Studies 1 SSRT was longest for Study 2, which involved stopping and 2, the beta band increase was significantly greater for of speech. For each of the three studies, p(stop) was successful versus failed stop trials. In Study 3, the beta approximately .5, indicating the efficacy of the SSD stair- power increase was significantly different from baseline case algorithm. for both successful and failed stop trials, but now there was no difference between successful and failed trials. The Cohen’s effect size for right frontal beta power in- Unexpected Events Task crease in successful stop trials above baseline was 0.74 As reported in Wessel and Aron (2013) over all blocks, RT for Study 1, 0.96 for Study 2, and 0.77 for Study 3. The was numerically slower for novel compared with standard mean effect size was 0.82. Note that the features that went trials. Although the main effect of Trial type was not sig- into the clustering method were (1) scalp distribution, (2) nificant, F(1, 12) = 1.6, p = .23, there was a significant dipoles, (3) spectra, (4) ERP after stop signal, and (5) ERSP – Block × Trial type interaction, F(5, 60) = 7.61, p <.0001, after stop signal for 600 msec in the range of 3 20 Hz. indicating that the difference between novels and stan- These features do not guarantee a stopping-related beta dards wore off over time. There was a significant novelty- increase with the specific timing of interest (just before induced slowing for Block 1, t(12) = 5.37, p <.001,d = SSRT), nor a successful versus failed difference, or a right 0.75, but not for the other blocks (all ps > .25). frontal topography. We also observed low-frequency power increases in the RF cluster after both successful and failed stop trials. This Cortical IC Source Clusters Related to Stopping increase is probably related to perceptual processing of the stop signal and not to the actual stopping process be- Our main analysis focuses on cluster RF with a right frontal cause it occurs for both failed and successful stop trials. scalp distribution (possibly related to the ECoG rIFC signa- ture; Wessel et al., 2013; Swann et al., 2009) as explained above. In addition, we report analysis from two other Cluster SML putative stopping-related components, cluster MC with a midline central scalp distribution (cf. Wessel & Aron, For all three studies, cluster SML (putative M1) showed a 2013, 2015) and cluster SML with a left “sensorimotor” power decrease in the mu (7–12 Hz) and beta band (13– topography (contralateral to the hand that participants 30 Hz) significant from baseline relative to the go signal. were responding with in Studies 1 and 3). Although in This mu and beta band power decrease was present for Studies 1 and 3 the cluster RF appeared to have a more go trials and successful stop and failed stop trials (Fig- frontocentral than right-lateralized scalp distribution, the ure 4). The decrease was most pronounced for failed stop respective distribution of IC dipoles (not shown here) trials > go > successful stop (in that order). This is a nice was skewed to the right hemisphere. Number of contribut- validation of earlier results with ECoG from the hand area ing participants and ICs to each cluster are listed in Table 2. of M1 in an SST (Swann et al., 2009) and also findings from motor physiology (Stinear, Coxon, & Byblow, 2009), as will be discussed below. Event-related Spectral Perturbations Cluster RF ERPs As predicted, we found that, in all studies, the cluster RF showed a stopping-related power increase in the beta Cluster MC band (13–20 Hz) significant from baseline starting as This cluster was similar to that reported in earlier studies early as 150 msec after the stop signal (see Figure 2). This of stopping (Wessel & Aron, 2013, 2014, 2015). Analysis

Table 2. Number of Contributing Participants and ICs to Each Cluster

Cluster RF (Right Frontal) Cluster MC (Midline Central) Cluster SML (Left “Sensorimotor”) Study 1 10 participants, 13 ICs 9 participants, 9 ICs 10 participants, 12 ICs Study 2 10 participants, 10 ICs 10 participants, 11 ICs 8 participants, 8 ICs Study 3 19 participants, 22 ICs 20 participants, 23 ICs 20 participants, 21 ICs

112 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Figure 2. ERSP images for right frontal (RF) clusters in (A) Study 1, (B) Study 2, and (C) Study 3. From left to right in each row: Mean cluster scalp projection; cluster mean ERSP images time-locked to the go signal for go trials, successful stop trials, failed stop trials, and the successful–failed difference. Nonsignificant differences are masked with transparency. In all three studies, these RF clusters show a significant increase in beta band power after the stop signal that elapses before SSRT for successful stop trials. In Studies 1 and 2, this beta band increase is significantly larger for successful than failed stop trials.

of ERPs in this cluster confirmed prior results by showing, 2013), we did observe an N2/P3 complex in the RF cluster, for all three studies, P3 onset latencies earlier for success- albeit much smaller in amplitude than in central midline ful stop compared with failed stop trials before SSRT (see areas (MC cluster). This N2 in the RF cluster could have Figure 3). Note that this cluster is different in spatial arisen because not all sources were completely sepa- location from the RF cluster (Figure 4). rated, leading to a “merged source” in some participants, that is, one containing a right frontal beta increase and the N2/P3 complex. Indeed in some participants, there Cluster RF was only one “frontal source,” and this source was local- Although this and other studies have suggested that the ized in between midline and right frontal topographies N2/P3 has a dorsomedial topography relative to the stop sig- and combined features from both clusters. Another pos- nal (Wessel, 2017; Wessel & Aron, 2013, 2014, 2015; Huster, sibility is that the N2 is a distributed phenomenon like Enriquez-Geppert, Lavallee, Falkenstein, & Herrmann, the P3a and —with the P3a occurring earlier and

Figure 3. ERPs and mean cluster scalp projections for midcentral (MC) clusters in (A) Study 1, (B) Study 2, and (C) Study 3. ERP plots are time-locked to the stop signal and show a positive deflection in the EEG occurring around 300 msec after the stop signal () previously shown in EEG studies (Wessel & Aron, 2015). For all three studies, P3 latencies occurred earlier for successful stop compared with failed stop trials relative to the stop signal.

Wagner et al. 113 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Figure 4. ERSP images for sensorimotor left (SML) clusters in (A) Study 1, (B) Study 2, and (C) Study 3. From left to right in each row: Mean cluster scalp projection; cluster mean ERSP images time-locked to the go signal for go trials, successful stop trials, failed stop trials, and the successful–failed difference. Nonsignificant differences are masked with transparency. All three SML clusters show a significant decrease in mu and beta bands after the go signal for go trails and successful and failed stop trials. This decrease was most pronounced for failed stop trials > go > successful stop (in that order).

over medial frontal areas and the P3b occurring later and beta band, but now in the unexpected events data (the over more posterior medial areas. In our study, the N2 “candidate process”). In this way, we confirmed that occurring over dorsomedial areas (MC cluster) had a the RF cluster displays a significant beta band power shorter latency than the N2 occurring over right frontal (13–20 Hz) increase for novels versus baseline starting areas (RF cluster). We note that Schmajuk et al. (2006) around 100 msec after the tone and lasting for around observed an N2 relative for stopping with a right frontal 200 msec. This beta band power increase was absent voltage distribution at 200-msec latency. It would be after standard tones, and the novel versus standard dif- interesting to look at the voltage distributions Schmajuk ference was significant. However, the correlation be- et al. (2006) observed at fine-grained time points between tween single trial beta power and RT on novel trials was 150 and 250 msec after the stop signal to see how scalp not reliable. distributions change over these latencies. DISCUSSION Unexpected Events Task We reanalyzed scalp EEG from three stop signal data sets. We ran ICA over the concatenated EEG for the SST and In each case, we identified a brain signature for action unexpected events task. We identified a cluster RF with stopping with a beta power increase following the stop a right frontal scalp distribution similar in spatial loca- signal that occurred before the time of stopping and with tion and activity to the cluster that we found for the a right frontal topography. In two of the three data sets, SST in Studies 1–3. When testing the ERSP in this clus- beta band power was stronger for successful than failed ter for the unexpected events task, we observed a power stop trials. We interpret this signature as the scalp EEG increase for novel versus standard trials in the range of counterpart to the stop-related beta power increase in 10–16Hzinthetimefromaround100to500msec rIFC shown with intracranial ECoG indexing prefrontally after the novel (see Figure 5). To more specifically test mediated inhibitory control. Furthermore, we show that the beta band, we first selected the frequency band in the same characteristic brain signature of stopping is ac- the range of 13–20 Hz in each participant that was max- tive after unexpected events. This provides important imally modulated over time in the window 0–500 msec validation of a scalp EEG signature of inhibitory control after the stop signal in SST (“the prototype process”). that could be leveraged by other investigators in other task We then averaged a band of ±2Hz around this selected contexts.

114 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Figure 5. ERSP images for the right frontal (RF) cluster in Study 2 for the unexpected events task. (A) From left to right: Mean cluster scalp projection; cluster mean ERSP images time-locked to the sound for novel and standard sounds and the difference between the two conditions. (B) Task schematics for the unexpected events task. (C) Mean beta band power changes (solid line) for novel (red) and standard (blue) sound trials, and one standard error confidence interval (rose and turquois areas).

Right Frontal Beta Band Increase as a Putative approach and classification methods, identified 15 ICs Index for Action Stopping and 45 time–frequency features in the EEG contributing to the stopping process. Interestingly, an IC with a scalp Using ICA, we identified a cluster of ICs over right frontal map that had a somewhat right frontal distribution and areas in each of the three studies. These clusters show a showed a frequency cluster with an increase in the beta transient beta power (13–20 Hz) increase that occurs band around 150 msec after the stop signal was 1 of the after the stop signal and before the time of stopping 10 best predictors for differences between successful and is absent on go trials. This beta band power increase stop and go trials. That study did not, however, differenti- was significantly larger for successful compared with ate between successful and failed stop trials as we did here. failed stop trials in Studies 1 and 2 (see Figure 2). Our In addition to the IC with the right-lateralized beta study was motivated by ECoG studies showing a beta effect, ICA also revealed, for each of the three data sets, – band (13 20 Hz) power increase in the rIFC for success- another stopping-related signature—a dorsomedial frontal ful versus failed stop trials before SSRT (Wessel et al., topography with a negative deflection around 200 msec 2013; Swann et al., 2009). Here we show something and a positive deflection in the EEG occurring around strikingly similar with scalp EEG in healthy young par- 300 msec after the stop signal. The P3 latency occurred ticipants. Although only limited anatomical claims can earlier for successful compared with failed stop trials. The be made from EEG results alone (without an individual N2/P3 signature has now been well established as relat- MRI-based head model), the similarity of these features ing to the stopping process in several EEG studies (Wessel, (right-lateralized, frontal and beta power) in the current 2017; Wessel & Aron, 2013, 2014, 2015; Huster et al., 2013). scalp EEG results to the abovementioned ECoG results Yet this has different features from the right-lateralized beta strongly suggests that EEG reflects the same process. effect and could putatively reflect the pre-supplementary One discrepancy in our results was that Studies 1 and 2 motorarearatherthanrIFC—although this remains to showed increased beta power for successful versus failed be established. The pre-supplementary motor area, which stop trials, but Study 3 did not (notably, however, there is dorsomedial, has also been implicated in stopping by a was a strong beta increase in that study for successful variety of brain stimulation, lesion, imaging, ECoG, and stop trials). This lack of difference between conditions connectivity studies (see Jahanshahi et al., 2015; Aron could be due to the higher percentage of stop trials in et al., 2014; Herz et al., 2014). Study 3 (33%, compared with 25% for Studies 1 and 2). Although we also observed an N2/P3 complex with a Also, in Study 3, participants performed the simple stop smaller amplitude over right frontal brain areas similar task that we analyzed after they had already performed, to Schmajuk et al. (2006), it is unclear how these two in the same session, a different kind of stop task that phenomena are related. Increases in beta power relative induced more preparation to stop and effectively had a to the stop signal have been shown in ECoG in the rIFC 50% stopping rate (unpublished). Thus, they might have (Wessel et al., 2013; Swann et al., 2009), and the rIFC is been biased to use a different strategy (indeed RT was thought to be part of a prefrontal–BG network for inhib- much longer than the other studies). itory control (reviewed by Aron et al., 2016; Zavala et al., A recent study by Huster, Schneider, Lavallee, Enriquez- 2015). Furthermore, the role of beta oscillations in motor Geppert, and Herrmann (2017), using a group level ICA control is well defined (Pogosyan, Gaynor, Eusebio, &

Wagner et al. 115 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Brown, 2009; Gilbertson et al., 2005). Thus, a beta power but it leaves open the question about what is the critical increase with right frontal topography is functionally element that drives this (Parmentier, 2014, 2016). Future more specific than the N2/P3 complex, which can occur studies should better control the type of unexpectedness. in a wide range of tasks. We expect that further studies will look into the functional relationship between the N2 Sensorimotor Cortical Areas and the right frontal beta power increase. We note here a limitation of our article, which is that A useful validation of our ICA method is that we also all three stop signal studies were visual-go/visual-stop identified an IC cluster for each study located over left paradigms, whereas often stop signal studies are done sensorimotor areas and contralateral to the responding with visual-go/auditory-stop. For this latter kind of para- hand in Studies 1 and 3. In all three studies, we found digm, Kenemans (2015), reviewing ERP findings, observed event-related desynchronization (see Pfurtscheller & that there is little sign of right frontal ERP signature of Da Silva, 1999) in the mu and beta band starting after successful versus failed stop. Although our results here the go cue for stop and go trials. This desynchronization mainly concern a right frontal time–frequency signature, was more pronounced for go and failed stop trials than it would be important to validate, in future studies or anal- for successful stop trials, and it was more pronounced yses, that this does also apply in the case of a visual-go/ for failed than go trials (consistent with a faster/stronger auditory-stop paradigm. movement on those trials). This is a nice validation of earlier results with ECoG from the hand area of M1 in an SST (Swann et al., 2009). The fact that failed stop Unexpected Events showed a greater mu/beta power decrease (desynchroni- For Study 3, where each participant did both an unexpected zation) than go trials is consistent with failed stop trials events task and an SST, we used ICA to decompose the having faster RT (an earlier/more energetic movement). concatenated EEG data. We thenusedtheautomated Moreover, relatively less desynchronization on successful clustering method to identify a cluster localized over right stop trials is consistent with both a slower RT and also a frontal scalp areas that showed a transient beta power (13– stopping-related synchronization in M1—for example, 20Hz)increaseelapsingafterthestopsignalandbeforethe stopping increases GABA in M1 (Stinear et al., 2009), time of stopping (the “prototype process”). When testing and increased GABA in M1 has been related to increased this same component for the candidate process (i.e., unex- power in mu/beta (Jensen et al., 2005). pected events), we identified a transient beta band power In Study 2 with verbal responses, in addition to the increase (13–20 Hz) that was significantly greater for unex- left SML discussed above, we also identified a right pected events versus standards. This increase started about SML cluster with the same properties. This fits with 150–200 msec after the unexpected event, consistent with bilateral representations of lip and tongue in the motor the timing of a putative suppression process indicated by cortex—which should be active in both hemispheres other methods such as TMS and subthalamic nucleus local during verbalization. field potentials (see Wessel, Jenkinson, et al., 2016; Wessel & Aron, 2013). Although this striking finding supports the idea Methodological Considerations that unexpected events activate the same (pre)frontal sys- tem as stopping in the stop signal paradigm, there are some Our results help define methodological guidelines for caveats. First, we did not observe a reliable relation between EEG studies that want to employ a frontal beta band sig- single trial beta power and RT on unexpected events trials. nature of inhibitory control. (1) Number of participants: However, we note that post-novel slowing was quite small in We estimated that, with an overall Cohen’s effect size of this particular study. Also, our theory of unexpected events 0.82 for all three studies, we would need 18 participants is that they recruit the stopping system within, for example, to get 95% power to detect a significant beta band power 150 msec and disrupt action (motor suppression) and task effect with an alpha of .05, one-tailed. (2) Total number set. Yet this particular task was so simple, with the partic- of stop trials: In Study 1, we only had, on average, 60 stop ipant saying “K” or “T” to an imperative stimulus about trials for the analysis (only about half of these were 1 sec later, that there might have been little task set to successful). Nevertheless, even with the small number of interrupt. Future studies will need to more carefully test trials, we detected a significant difference between suc- whether this right frontal beta signature of unexpected cessful and failed trials in the beta band. We would there- events relates to behavioral and cognitive interruptions. fore suggest an absolute minimum with these numbers, A second caveat is that the unexpected event in this study but preferably many more trials. (3) Percentage of stop was a 200-msec birdsong segment (20%), whereas the stan- trials: Studies 1 and 2 revealed a difference in the beta dard was a 200-msec sine wave (80%). This means that the band between successful and failed stop trials whereas birdsongs were unexpected at the informational level and at Study 3 did not. Notably, the first two studies had a 25% the stimulus characteristic level too (saliency, pitch). This stop rate, whereas Study 3 had a 33% rate. Several lines does not speak against the finding that such unexpected of evidence suggest that the way in which stopping is events activated the same brain signature as the stop signals, done changes with increased probability of stop signals

116 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 (Wessel, 2017; Chikazoe et al., 2009; Verbruggen & Logan, Cai, W., Ryali, S., Chen, T., Li, C. S. R., & Menon, V. (2014). 2009), and we therefore recommend the use of a 25% stop Dissociable roles of right inferior frontal cortex and anterior insula in inhibitory control: Evidence from intrinsic and signal rate. task-related functional parcellation, connectivity, and response profile analyses across multiple datasets. Journal of Neuroscience, 34, 14652–14667. Conclusion Chambers, C. D., Garavan, H., & Bellgrove, M. A. (2009). Insights into the neural basis of response inhibition from cognitive We show with scalp EEG in healthy participants that we and clinical neuroscience. Neuroscience & Biobehavioral can derive a signature of stopping action that manifests as Reviews, 33, 631–646. a beta power increase for successful versus failed stop Chikazoe, J., Jimura, K., Hirose, S., Yamashita, K., Miyashita, trials, before the time of stopping, and with a right frontal Y., & Konishi, S. (2009). Preparation to inhibit a response complements response inhibition during performance of topography. We interpret this signature as the scalp EEG a stop-signal task. Journal of Neuroscience, 29, 15870–15877. counterpart to the stop-related beta power increase in Delorme, A., & Makeig, S. (2004). EEGLAB: An open source rIFC in ECoG. We further show that the same frontal toolbox for analysis of single-trial EEG dynamics including right-lateralized beta signature of stopping is active after independent component analysis. Journal of Neuroscience – unexpected events—which strikingly supports the argu- Methods, 134, 9 21. Fonken, Y. M., Rieger, J. W., Tzvi, E., Crone, N. E., Chang, E., ment that unexpected events leverage the same brain Parvizi, J., et al. (2016). Frontal and motor cortex contributions system as stopping in the stop signal paradigm (Wessel to response inhibition: Evidence from electrocorticography. & Aron, 2017). We anticipate that other researchers can Journal of Neurophysiology, 115, 2224–2236. deploy this same signature for stopping in scalp EEG to Gentsch, A., Ullsperger, P., & Ullsperger, M. (2009). Dissociable test the recruitment and timing of inhibitory control in medial frontal negativities from a common monitoring system for self- and externally caused failure of goal achievement. other task contexts (cf. Wagner, Makeig, Gola, Neuper, Neuroimage, 47, 2023–2030. & Müller-Putz, 2016; van Gaal, Ridderinkhof, Fahrenfort, Gilbertson, T., Lalo, E., Doyle, L., Di Lazzaro, V., Cioni, B., & Scholte, & Lamme, 2008). Accordingly, we provide com- Brown, P. (2005). Existing motor state is favored at the panion material, which shares the data and analysis expense of new movement during 13–35 Hz oscillatory stream. synchrony in the human corticospinal system. Journal of Neuroscience, 25, 7771–7779. Herz, D. M., Christensen, M. S., Bruggemann, N., Hulme, O. J., Ridderinkhof, K. R., Madsen, K. H., et al. (2014). Motivational Acknowledgments tuning of fronto-subthalamic connectivity facilitates control – We thank NIH DA026452 and James S McDonnell Scholar of action impulses. Journal of Neuroscience, 34, 3210 3217. Award 20133896 for funding. Huster, R. J., Enriquez-Geppert, S., Lavallee, C. F., Falkenstein, M., & Herrmann, C. S. (2013). Electroencephalography of Data Sharing: Two stop signal data sets standing alone and one response inhibition tasks: Functional networks and cognitive stop signal plus unexpected events data set were analyzed. We contributions. International Journal of Psychophysiology, provide access to the EEG and behavioral data for the latter 87, 217–233. (combined) data set along with detailed instructions and scripts Huster, R. J., Schneider, S., Lavallee, C. F., Enriquez-Geppert, S., to generate the results shown in the article: https://osf.io/ukvmh/. & Herrmann, C. S. (2017). Filling the void—Enriching the feature space of successful stopping. Human Brain Mapping, Reprint requests should be sent to Johanna Wagner, Psychology 38, 1333–1346. Department, University of California, San Diego, 3133 McGill Hall, Jahanshahi, M., Obeso, I., Rothwell, J. C., & Obeso, J. A. (2015). 9500 Gilman Drive, La Jolla, CA 92103, or via e-mail: j9wagner@ A fronto-striato-subthalamic-pallidal network for goal-directed ucsd.edu. and habitual inhibition. Nature Reviews Neuroscience, 16, 719–732. Jensen, O., Goel, P., Kopell, N., Pohja, M., Hari, R., & REFERENCES Ermentrout, B. (2005). On the human sensorimotor-cortex beta rhythm: Sources and modeling. Neuroimage, 26, Aron, A. R., Herz, D. M., Brown, P., Forstmann, B. U., & 347–355. Zaghloul, K. (2016). Frontosubthalamic circuits for control Kenemans, J. L. (2015). Specific proactive and generic reactive of action and cognition. Journal of Neuroscience, 36, inhibition. Neuroscience & Biobehavioral Reviews, 56, 11489–11495. 115–126. Aron, A. R., Robbins, T. W., & Poldrack, R. A. (2014). Inhibition Lee, T. W., Lewicki, M. S., Girolami, M., & Sejnowski, T. J. and the right inferior frontal cortex: One decade on. Trends (1999). Blind source separation of more sources than in Cognitive Sciences, 18, 177–185. mixtures using overcomplete representations. IEEE Signal Barcelo, F., Escera, C., Corral, M. J., & Periáñez, J. A. (2006). Processing Letters, 6, 87–90. Task switching and novelty processing activate a common Lewicki, M. S., & Sejnowski, T. J. (2000). Learning overcomplete neural network for cognitive control. Journal of Cognitive representations. Neural Computation, 12, 337–365. Neuroscience, 18, 1734–1748. Logan, G. D., & Cowan, W. B. (1984). On the ability to inhibit Bell, A. J., & Sejnowski, T. J. (1995). An information-maximization thought and action: A theory of an act of control. Psychological approach to blind separation and blind deconvolution. Review, 91, 295. Neural Computation, 7, 1129–1159. Makeig, S. (1993). Auditory event-related dynamics of the Benjamini, Y., & Yekutieli, D. (2001). The control of the false EEG spectrum and effects of exposure to tones. Advances in discovery rate in multiple testing under dependency. Annals Neural Information Processing Systems, 86, 283–293. of Statistics, 29, 1165–1188.

Wagner et al. 117 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021 Makeig, S., Bell, A. J., Jung, T. P., & Sejnowski, T. J. (1996). meets go. Neuroscience & Biobehavioral Reviews, 33, Independent component analysis of electroencephalographic 662–673. data. In D. Touretsky, M. Mozer, & M. Hasselmo (Eds.), Swann, N., Tandon, N., Canolty, R., Ellmore, T. M., McEvoy, Advances in Neural Information Processing Systems L. K., Dreyer, S., et al. (2009). Intracranial EEG reveals a (pp. 45–151). Cambridge, MA: MIT Press. time- and frequency-specific role for the right inferior Makeig, S., Debener, S., Onton, J., & Delorme, A. (2004). frontal gyrus and primary motor cortex in stopping initiated Mining event-related brain dynamics. Trends in Cognitive responses. Journal of Neuroscience, 29, 12675–12685. Sciences, 8, 204–210. Swick, D., Ashley, V., & Turken, U. (2011). Are the neural Makeig, S., Westerfield, M., Jung, T. P., Enghoff, S., Townsend, correlates of stopping and not going identical? Quantitative J., Courchesne, E., et al. (2002). Dynamic brain sources of meta-analysis of two response inhibition tasks. Neuroimage, visual evoked responses. Science, 295, 690–694. 56, 1655–1665. Morlet, J., Arens, G., Fourgeau, E., & Glard, D. (1982). Wave Vachon, F., Hughes, R. W., & Jones, D. M. (2012). Broken propagation and sampling theory—Part I: Complex signal expectations: Violation of expectancies, not novelty, captures and scattering in multilayered media. Geophysics, 47, 203–221. auditory . Journal of Experimental Psychology: Onton, J., Westerfield, M., Townsend, J., & Makeig, S. (2006). Learning, , and Cognition, 38, 164–177. Imaging human EEG dynamics using independent van Gaal, S., Ridderinkhof, K. R., Fahrenfort, J. J., Scholte, H. S., component analysis. Neuroscience & Biobehavioral Reviews, & Lamme, V. A. (2008). Frontal cortex mediates unconsciously 30, 808–822. triggered inhibitory control. Journal of Neuroscience, 28, Oostenveld, R., & Oostendorp, T. F. (2002). Validating the 8053–8062. boundary element method for forward and inverse EEG Verbruggen, F., & Logan, G. D. (2009). Proactive adjustments computations in the presence of a hole in the skull. Human of response strategies in the stop-signal paradigm. Journal Brain Mapping, 17, 179–192. of Experimental Psychology: Human Perception and Oostenveld, R., & Praamstra, P. (2001). The five percent Performance, 35, 835–854. electrode system for high-resolution EEG and ERP Wagner, J., Makeig, S., Gola, M., Neuper, C., & Müller-Putz, measurements. Clinical Neurophysiology, 112, 713–719. G. (2016). Distinct β band oscillatory networks subserving Palmer, J. A., Kreutz-Delgado, K., & Makeig, S. (2006). Super- motor and cognitive control during gait adaptation. Journal Gaussian mixture source model for ICA. In J. Rosca, of Neuroscience, 36, 2212–2226. D. Erdogmus, J. C. Principe, S. Haykin (Eds.), Lecture Wessel, J. R. (2016). Testing multiple psychological processes notes in computer science (pp. 854–861). Berlin: Springer. for common neural mechanisms using EEG and independent Palmer, J. A., Makeig, S., Kreutz-Delgado, K., & Rao, B. D. (2008). component analysis. Brain Topography, doi:10.1007/s10548- Newton method for the ICA mixture model. IEEE 016-0483-5. International Conference on Acoustics and Signal Wessel, J. R. (2017). Prepotent motor activity and inhibitory Processing (pp. 1805–1807). Las Vegas. control demands in different variants of the go/no-go paradigm. Parmentier, F. B. (2014). The cognitive determinants of Psychophysiology, doi:10.1111/psyp.12871. behavioral distraction by deviant auditory stimuli: A review. Wessel, J. R., & Aron, A. R. (2013). Unexpected events induce Psychological Research, 78, 321–338. motor slowing via a brain mechanism for action-stopping Parmentier, F. B. (2016). Deviant sounds yield distraction with global suppressive effects. Journal of Neuroscience, 33, irrespective of the sounds’ informational value. Journal of 18481–18491. Experimental Psychology: Human Perception and Wessel, J. R., & Aron, A. R. (2014). Inhibitory motor control Performance, 42, 837–846. based on complex stopping goals relies on the same brain Parmentier, F. B., Elford, G., Escera, C., Andrés, P., & San Miguel, network as simple stopping. Neuroimage, 103, 225–234. I. (2008). The cognitive locus of distraction by acoustic novelty Wessel, J. R., & Aron, A. R. (2015). It’s not too late: The onset of in the cross-modal oddball task. Cognition, 106, 408–432. the frontocentral P3 indexes successful response inhibition Pfurtscheller, G., & Da Silva, F. L. (1999). Event-related EEG/MEG in the stop-signal paradigm. Psychophysiology, 52, 472–480. synchronization and desynchronization: Basic principles. Wessel, J. R., & Aron, A. R. (2017). On the globality of motor Clinical Neurophysiology, 110, 1842–1857. suppression: Unexpected events and their influence on Pogosyan, A., Gaynor, L. D., Eusebio, A., & Brown, P. (2009). behavior and cognition. Neuron, 93, 259–280. Boosting cortical activity at beta-band frequencies slows Wessel, J. R., Conner, C. R., Aron, A. R., & Tandon, N. (2013). movement in humans. Current Biology, 19, 1637–1641. Chronometric electrical stimulation of right inferior frontal Ray, N. J., Brittain, J. S., Holland, P., Joundi, R. A., Stein, J. F., cortex increases motor braking. Journal of Neuroscience, Aziz, T. Z., et al. (2012). The role of the subthalamic nucleus 33, 19611–19619. in response inhibition: Evidence from local field potential Wessel, J. R., Danielmeier, C., Morton, J. B., & Ullsperger, M. recordings in the human subthalamic nucleus. Neuroimage, (2012). Surprise and error: Common neuronal architecture 60, 271–278. for the processing of errors and novelty. Journal of Schall, J. D., & Godlove, D. C. (2012). Current advances and Neuroscience, 32, 7528–7537. pressing problems in studies of stopping. Current Opinion Wessel, J. R., Ghahremani, A., Udupa, K., Saha, U., Kalia, S. K., in Neurobiology, 22, 1012–1021. Hodaie, M., et al. (2016). Stop-related subthalamic beta Schmajuk, M., Liotti, M., Busse, L., & Woldorff, M. G. (2006). activity indexes global motor suppression in Parkinson’s Electrophysiological activity underlying inhibitory control disease. Movement Disorders, 31, 1846–1853. processes in normal adults. Neuropsychologia, 44, 384–395. Wessel, J. R., Jenkinson, N., Brittain, J. S., Voets, S. H., Aziz, Schmidt, R., & Berke, J. D. (2017). A pause-then-cancel model T. Z., & Aron, A. R. (2016). Surprise disrupts cognition via of stopping: Evidence from basal ganglia neurophysiology. a fronto-basal ganglia suppressive mechanism. Nature Philosophical Transactions of the Royal Society of London, Communications, 7, 11195. Series B, Biological Sciences, 372. Zavala, B., Zaghloul, K., & Brown, P. (2015). The subthalamic Stinear, C. M., Coxon, J. P., & Byblow, W. D. (2009). Primary nucleus, oscillations, and conflict. Movement Disorders, 30, motor cortex and movement prevention: Where stop 328–338.

118 Journal of Cognitive Neuroscience Volume 30, Number 1 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_01183 by guest on 01 October 2021