Response inhibition rapidly increases single-neuron responses in the subthalamic nucleus of patients with Parkinson's disease

Article

Reference

Response inhibition rapidly increases single-neuron responses in the subthalamic nucleus of patients with Parkinson's disease

BENIS, Damien, et al.

BENIS, Damien, et al . Response inhibition rapidly increases single-neuron responses in the subthalamic nucleus of patients with Parkinson's disease. Cortex , 2016, vol. 84, p. 111-123

DOI : 10.1016/j.cortex.2016.09.006 PMID : 27745848

Available at:

http://archive-ouverte.unige.ch/unige:95962

Disclaimer: layout of this document may differ from the published version.

1 / 1

Research report

Response inhibition rapidly increases

single-neuron responses in the subthalamic nucleus of patients with Parkinson's disease

Damien Benis

, Olivier David

, Brigitte Piallat

, Astrid Kibleur

, Laurent Goetz

, Manik Bhattacharjee

, Val erie Fraix

,

Eric Seigneuret

, Paul Krack

, St ephan Chabard es

and Julien Bastin

aUniversite Grenoble Alpes, Grenoble Institut des Neurosciences, GIN, Grenoble Cedex 9, France

bInserm, U1216, Grenoble Cedex 9, France

cCHU Grenoble Alpes, Department of Neurology, Grenoble Cedex 9, France

dCHU Grenoble Alpes, Department of Neurosurgery, Grenoble Cedex 9, France

a r t i c l e i n f o

Article history:

Received 16 March 2016 Reviewed 30 May 2016 Revised 12 July 2016 Accepted 12 September 2016 Action editor Lesley Fellows Published online 21 September 2016

Keywords:

Subthalamic nucleus Motor inhibition Electrophysiology Stop-signal task

a b s t r a c t

The subthalamic nucleus (STN) plays a critical role during action inhibition, perhaps by acting like a fast brake on the motor system when inappropriate responses have to be rapidly suppressed. However, the mechanisms involving the STN during motor inhibition are still unclear, particularly because of a relative lack of single-cell responses reported in this structure in humans. In this study, we used extracellular microelectrode recordings during deep brain stimulation surgery in patients with Parkinson's disease (PD) to study STN neurophysiological correlates of inhibitory control during a stop signal task. We found two neuronal subpopulations responding either during motor execution (GO units) or during motor inhibition (STOP units). GO units fired selectively before patients'motor re- sponses whereas STOP units fired selectively when patients successfully withheld their move at a latency preceding the duration of the inhibition process. These results provide electrophysiological evidence for the hypothesized role of the STN in current models of response inhibition.

©2016 Published by Elsevier Ltd.

1. Introduction

Neuroimaging studies have shown that response-stopping processes activate a frontal-subcortical network (Aron, 2006;

Chikazoe et al., 2009; Li, Yan, Sinha, & Lee, 2008). The dy- namics of this network has been conceptualized in several computational models of response inhibition (Frank, 2006;

Ratcliff&Frank, 2012; Wiecki&Frank, 2013) in which rapid

*Corresponding author. Institut des Neurosciences de Grenoble, B^atiment Edmond J. Safra des Neurosciences, Chemin Fortune Ferrini, Universite Joseph Fourier, Site Sante La Tronche, BP 170, 38042 Grenoble Cedex 9, France.

E-mail address:[email protected](J. Bastin).

Available online atwww.sciencedirect.com

ScienceDirect

Journal homepage:www.elsevier.com/locate/cortex

http://dx.doi.org/10.1016/j.cortex.2016.09.006 0010-9452/©2016 Published by Elsevier Ltd.

suppression of a prepared move involves a fast stopping signal neurally implemented via a putative hyperdirect pathway (Monakow, Akert,&Ku¨nzle, 1978; Nambu, Takada, Inase,& Tokuno, 1996). This hyperdirect pathway assumes projections of the frontal cortex (pre-SMA, inferior frontal gyrus and anterior cingular cortices) onto the subthalamic nucleus (Haynes&Haber, 2013). This hypothesis is attractive, but it is currently only supported by a few intracerebral elec- trophysiological recordings in humans-during the stop signal task (SST)-showing that beta (15e35 Hz) oscillations of local field potentials (LFP) increased during stopping in the inferior frontal gyrus (Jha et al., 2015; Swann et al., 2009; Wessel, Conner, Aron,&Tandon, 2013) and in the STN (Benis et al., 2014) at latencies that precede the time needed to cancel movements (stop signal reaction time, SSRT). Indirect evi- dence also comes from reports that showed that STN lesions cause ballistic and involuntary movement in non-human primates and humans (Crossman, Sambrook, & Jackson, 1984; Nishioka, Taguchi, Nanri, & Ikeda, 2008) as well as inaccurate and premature responding in reaction time tasks performed by rats (Baunez, Nieoullon,&Amalric, 1995; Eagle et al., 2008) and humans (Obeso et al., 2014). Finally, effects of bilateral high frequency stimulation of the STN in patients with Parkinson's disease (PD) remain to be firmly established, as both improvement (Mirabella et al., 2012; Swann et al., 2011;

van den Wildenberg et al., 2006) or worsening of inhibitory performance (Obeso, Wilkinson, Rodrı´guez-Oroz, Obeso, &

Jahanshahi, 2013; Ray et al., 2009) have been reported.

The STN is thus modeled as an important region within the response inhibition brain-network, but the functional mech- anisms involved at the cellular level during response inhibi- tion remain unclear despite recent advances in monkeys and rats (Isoda & Hikosaka, 2008; Schmidt, Leventhal, Mallet, Chen, & Berke, 2013). Thus, even if a consensus seems to emerge from many models that hypothesize that a neuronal population within the STN should increase its firing rate during successful stopping at a latency that should precede the SSRT (Boucher, Palmeri, Logan,&Schall, 2007; Frank, 2006;

Logan&Cowan, 1984; Ratcliff&Frank, 2012; Schall, Stuphorn,

&Brown, 2002; Wiecki&Frank, 2013), the electrophysiological

evidence supporting this idea remains weak. We recently identified such neurons in the putative associative-limbic area of the STN of patients suffering from obsessive-compulsive disorders (OCD) (Bastin et al., 2014). But whether such cells can be observed in the putative sensorimotor territory of the STN remains unclear. To test this hypothesis, we used STN micro-recordings during deep brain stimulation (DBS) surgery in patients with PD and analyzed single unit responses during a stop signal paradigm. We found that STN neurons respon- ded selectively during successful response inhibition, at la- tencies that preceded the SSRT.

2. Materials and methods

Intraoperative recordings were obtained from 21 PD patients (13 male and 8 female; mean±SEM age: 58.4±1.5 y.o.; 19 right-handed; no psychiatric comorbidity; additional details in

Table 1). They were selected for their ability and readiness of collaborating in a demanding cognitive task while undergoing therapeutic bilateral STN implantation of DBS electrodes (Benabid, Chabardes, Mitrofanis, & Pollak, 2009). They had been suffering from idiopathic PD for 10.6± .8 years. They responded well to Levodopa (Unified Parkinson Disease Rating Scale, UPDRS ON Levodopa: 15.9 ± 2.1; OFF Levodopa:

42.17 ± 2.5) and had disabling motor fluctuations and/or Levodopa-induced dyskinesia refractory to the adjustment of anti-parkinsonian medication. They presented no relevant deterioration in overall cognitive evaluation (Table 1, average Mattis Dementia Score was 138.6±1, on a scale from 0 to 144, with higher scores indicating preserved cognition, Mattis, 1976) and only relatively mild dysexecutive syndrome (fron- tal score: 41.29±1.4, on a scale from 0 to 50, with a higher score indicating preserved executive functions,Pillon, Dubois, Lhermitte, & Agid, 1986). The electrophysiological signals were recorded while patients were in an OFF state since medication was stopped on the night before surgery (12 h preoperatively). All patients gave their written informed con- sent to participate to this study that was approved by our local ethics committee (Grenoble University Hospital, Comite de Protection des Personnes Sud-Est I, protocol number: 2011- A00083-38).

2.2. Electrode implantation and neuronal recordings

As in routine DBS procedure of Grenoble University Hospital (Benabid et al., 2009; Thobois et al., 2010), electrophysiological signals recorded from five tungsten microelectrodes (2 mm apart, tip diameter<10mm; impedance of .2e6 MUat 1 KHz, FHC microelectrodes, Bowdoinham, USA) were used to opti- mize STN targeting based on the spiking properties of STN cells. Raw neuronal activity was amplified (10), band-pass filtered between 300 and 6000 Hz and sampled at 48 kHz.

The STN was preoperatively targeted using stereotactic mag- netic resonance imaging (MRI). In patients with PD, the following coordinates are used to target the putative senso- rimotor STN: 6/12 of the anterior-posterior commissural (AC- PC) length posterior to the AC, 12 mm lateral to the midline, and 3 mm below the AC-PC line. This targeting was further refined by the pattern of electrophysiological activity typically observed in the STN area (Benabid et al., 2009), i.e., the pres- ence of asymmetrical spikes at high frequency with bursting patters and/or proprioceptive responses to passive move- ments. Importantly, we previously reported task-related electrophysiological modulations from the putative non- motor part of the STN recorded in OCD patients. Note that in patients with OCD, the non-motor part of STN is targeted 2 mm anterior and 1 mm medial to the PD target (Chabardes et al., 2012). When STN cells were detected the behavioral task was proposed to patients. Patients were laying horizon- tally, their head maintained in the stereotaxic frame. Correct vision of the monitor by the patient was ensured before starting the task.

2.3. Behavioral task

Patients performed 0e4 intra-operative SST sessions of 100e200 trials per STN side, depending on clinical constraints

(for example, 13 out of 21 patients performed the task only during the first STN side surgery whereas both STN sides could be successively recorded in the 8 other patients). In the task, two types of trials were presented in a randomized order (Fig. 1). During GO trials (70% of trials), an imperative GO cue prompted patients to press a button quickly with the right index after a variable fixation period (the range was from 800 to 1200 msec). GO cue vanished after button press or after 1000 msec. During STOP trials (30% of trials), a STOP signal unpredictably followed the imperative GO cue after a variable stop signal delay (SSD). The STOP cue vanished after button press or after 500 msec. The ability to stop a response is related to the SSD value: the longer the SSD is, the more difficult it is to stop (Logan&Cowan, 1984). The SSD varied from trial to trial to adjust task difficulty using a procedure composed of a single staircase (50 msec steps) to obtain suc- cessful withholding of button presses in approximately 50% of STOP trials. The initial SSD chosen for each patient was esti- mated from results obtained during a training session pre- ceding the electrophysiological study and the range varied from one patient to another (SSD values ranged from 150 msec to 668 msec across patients). This staircase procedure pro- vided a simple estimation of the duration of the inhibitory process (SSRT), which was computed by subtracting the average SSD to the average GO reaction time for each patient.

We chose to use the mean difference method instead of the integration method to estimate SSRT because of the low number of trials recorded during each session, even if this procedure tends to overestimate SSRT (Verbruggen&Logan, 2008; Verbruggen, Chambers, & Logan, 2012). Note that to avoid another possible bias in SSRT estimation, while the response window during GO trials was limited to 1s to obtain a Table 1ePatients demographics and clinical details.

Patient Gender Age at surgery

(years)

Disease duration

(years)

UPDRS III^a Mattis Dementia

Score

Frontal score

L DOPA daily dose (DOPA equivalent)

Hand- laterality On^a Off^a

1 F 69 18 13 35,5 144 50 1030 R

2 F 68 13 9 41 138 44,9 625 R

3 M 56 14 12 46 132 44 600 R

4 M 59 7 8 32 130 27,3 525 R

5 F 61 12 11 39 137 47,4 1235 L

6 M 70 6 31 47 140 32,8 320 R

7 M 53 7 8 25 142 ND 1250 R

8 M 49 9 21 79 127 29,6 1650 R

9 M 54 11 10 49,5 143 41,8 2750 R

10 M 47 14 11 48 138 42,25 320 R

11 F 46 10 6,5 31 142 48,5 300 L

12 F 59 11 6 39 144 47 785 R

13 M 49 18 17 39 141 43 1800 R

14 F 59 10 9 40 138 38,42 1225 R

15 M 64 7 8 20 135 31,6 1225 R

16 F 52 6 9 33,5 144 50 1450 R

17 M 61 13 18 35,5 143 44,7 1000 R

18 M 57 4 40,5 51,5 134 42,7 1420 R

19 F 63 13 37 50 139 45 1017 R

20 M 61 9 23 56,5 139 30,75 800 R

21 M 70 11 26 47,5 140 44 1128 R

R: Right; L: Left.

a LDOPA treatment. ND: not done. UPDRS: Unified Parkinson Disease Rating Scale.

Fig. 1eStop signal task and behavioral results. A. After a fixation period (Fix, 1500 msec maximum), patients were instructed to press a button as fast as they could after the GO cue (GO trials, 70% of trials), and to withhold their movement when a STOP signal occurs (STOP trials, 30% of trials). Task difficulty during STOP trials was adjusted by shortening or lengthening the delay between GO and STOP cues (SSD) following a staircase procedure. At the end of each trial a positive or negative feedback was presented (FB, see methods). B. Average across patient's success rate and reaction times (RT) measured during GO and STOP (ST) trials. Stars indicate significant differences (Pairedt-test;

***p<.0001). Error bars represent±SEM.

positive feedback (in order to encourage fast and automatic motor responses), we used the full reaction time distribution (including RT longer than one second) to estimate SSRT. To maintain patient's attention and motivation at a high level, despite the repetitiveness of the SST, a feedback was dis- played during 1500 msec after each trial (the onset of the feedback was 1000 msec after button presses or 500 msec after STOP cue onset). Successful GO or STOP trials increased the score by 1 or 3 points, respectively, whereas failed trial feed- back consisted of losing 1 or 3 points, respectively. Button presses had to occur before 1000 msec after GO cue to be considered as successful. During feedback, the score was updated and displayed. Visual stimuli were delivered on a 15.4-inch monitor with a refresh rate of 60 Hz with Presenta- tion 14.1 software (Neurobehavioral Systems, Albany, CA).

Patients responded to the task using right (20 patients) or left (2 patients) index finger button presses.

2.4. Data analysis

We computed the percentage of successful strop trials, the average GO reaction time, the average SSD and the SSRT. To exclude recordings during which patients did not perform the task accurately, blocks of trials whose success rate during GO trials was below 75% and/or whose success rate during STOP trials was outside a [30e70%] range were excluded.

Extracellular recordings were processed using SigTOOL (Lidierth, 2009). Spike detection and spike sorting was applied to continuous data using Waveclus clustering algorithm (Quiroga, Nadasdy,&Ben-Shaul, 2004). Subsequently, clusters were classified into single or multi-units according to a crite- rion based on the spike shape and variance, the signal-to- noise ratio and the refractory period observed in the inter- spike interval histograms (Tankus, Yeshurun,&Fried, 2009).

To quantify for each identified cell the spike activity during the task, peri-stimulus spike histograms (PSTH) (75 msec bins;

900 msec time window in the [125,775 msec] time interval after the STOP signal for STOP trials, and in the [425,475 msec] time interval around button presses) and continuous neuronal activation functions (spike density functions) were generated. Spike density functions were ob- tained by convolving each spike with a Gaussian kernel (SD¼75 msec).

To estimate whether each neuronal unit significantly responded to the task, we used a bootstrap randomization process (resampling test) that was applied 10,000 times by shuffling the spike timing within each trial type (Successful STOP, Unsuccessful STOP, GO). The resampling method maintained the number of observed spikes during each trial and allowed the computation of a surrogate PSTH. This pro- cedure tests the null hypothesis that spikes are randomly distributed across time around the task-event. Hence, if spiking activity increase from the original PSTH exceeded the 95% confidence interval of the permuted set during at least one 75 msec bin, the cell was considered to be significantly task-modulated (p<.05).

For each task responsive cell, the firing rate for each single trial was then normalized (Z-scores were computed using a 1000 msec pre-stimulus baseline). Normalized data from all trials of interest were entered in a general linear mixed model

(GLM) with the following parameters: normalized firing rate was used as dependent variable whereas time bin (12 levels in the [125,775] msec time interval) and trial type [6 levels:

successful STOP trials that were either time locked to STOP cues or to a virtual response, unsuccessful STOP, button press during GO trials, button press during unsuccessful STOP trials and latency matched GO (LMGO) trials] were entered as fixed effects. In addition, each unit anatomical location (depth in trajectory), frontal score, UPDRS and disease duration were entered as continuous variables, and patient, neuron discharge pattern (3 levels: Irregular, Bursted, Oscillatory Burst), unit type (2 levels, single unit and multi-unit), and subject handedness (2 levels: Left and Right) were used as random variables in the GLM analysis. This statistical approach was chosen because it accounts for variance con- tributions by clinical and behavioral factors (such as disease severity and peroperative fatigue) and differences in neuronal firing pattern, and thus provides an accurate estimation of the contribution of task-related factors over confounding vari- ables. If the GLM analysis displayed a significant interaction between trial type and time, we ran two a priori defined contrast analyses. First, to test if task responsive STN cells responded during motor execution, we ran a contrast analysis to compare for each time bin the firing rate measured during correct button presses (GO trials) to the firing rate observed during successfully stopped movement centered to the STOP cue (Successful STOP trials) and during successfully stopped movements centered to a virtual button press (Virtual Response centered Successful STOP, vr-SS). A second contrast was used to examine if STN cells were responding to behav- ioral stopping: we compared STN firing rate during Successful STOP trials and during LMGO trials. Note that to avoid statis- tical biases inherent to multiple comparison,p-values for all contrast analyses were corrected by controlling the false dis- covery rate (FDR,Genovese, Lazar,&Nichols, 2002; Hochberg, 1995).

LMGO trials were selected as follows: each successful STOP trial was paired to a GO trial randomly picked among GO trials presenting a reaction time longer than SSD þ SSRT of the current STOP trial. Note that the SSD of each particular suc- cessful STOP trial was used, to take into account SSD vari- ability. An additional constraint was that a GO trial could never be used twice (once a given GO trial was selected using this latency matching procedure, it was excluded from the GO trials distribution for subsequent matching). Neuronal activity in LMGO trials was then aligned to latency of STOP signal appearance in the corresponding successful STOP trials, so that GO (LMGO trials) and STOP (Successful STOP) processes could be directly compared. Thus, if a STOP signal had occurred during a LMGO trial, the patient would probably have succeeded to stop his/her response. This procedure was preferred to the possibility to align neural activity on GO cues because preliminary analyses showed that this kind of pro- cedure was likely to suppress neural responses observed during Successful STOP trials, reflecting the variability of the SSDs in this study.

Virtual Response Centered Successful STOP trials (vr-SS trials) were implemented in the analyses as follows: for each successful STOP trial, we attributed a virtual button press that was randomly picked from the center part of the GO trials

reaction time distribution so that the average virtual STOP reaction time corresponded to the mean GO reaction time. In a control analysis, instead of using the center part of the GO trials reaction time distribution, we used the latter half of the distribution because slower responses are more likely to be suppressed than faster responses. In both cases, an additional constraint was that a reaction time from a GO trial could never be used twice to generate a vr-SS trial (once a given reaction time trial was selected from the center of the distribution, it was excluded from the GO trials distribution for subsequent matching). Neuronal activity during vr-SS trials was therefore time-locked to the latency of a virtual button presses onset corresponding to lLMGO trials, so that activity during response to GO trials and vr-SS trials could also be directly compared when the activity was time-locked to motor-control processes.

Finally, instead of quantifying STN neuronal activity using the entire pool of task-responsive neurons, we ran a more specific analysis that classified individual task-responsive neurons according to the pattern of activity that was observed during the task, using a previously validated approach (Bastin et al., 2014; Isoda and Hikosaka, 2008;

Schmidt et al., 2013): units were classified into two func- tional clusters: STOP units and motor (GO) units. STOP units were defined according to the following criteria: STOP unit's firing rate had to increase significantly (i.e., the firing rate had to exceed the statistical threshold estimated from the resampling method) and selectively during Successful STOP trials (i.e., at least one 75 msec bin in the peristimulus histo- gram aligned on the STOP cue within a time window of 775 msec following the occurrence of the STOP cue). Note that an additional constraint was that the activity of STOP units during correct button presses (GO trials) had to remain below the statistical threshold estimated from the resampling method (non-significant activity during these conditions).

Conversely, GO units were defined according to the following criteria: 1. Their firing rate had to increase significantly around motor responses during GO trials (i.e., at least one 75 msec bin in the peristimulus histogram within 900 msec around response to GO trials had to exceed the statistical threshold estimated from the resampling method) while the activity of these cells had to remain non-significant during successful STOP trials.

Modulations of the normalized firing rate for each func- tional class of neuron were then studied using slightly modi- fied GLM approaches. The normalized activity of STOP units was statistically tested for the 12 time bins3 trial types (SS, US and LMGO trials were compared) and the same random variables was used (see the full GLM description above).

Contrast analyses were then performed to compare for each time bin z-normalized firing rate between successful STOP, unsuccessful STOP and latency matched GO trials. Note that the procedure employed regarding the use of a virtual STOP to time-lock STN activity during latency matched GO could potentially blur modulations of STN activity related to the processing of GO cues. To test this possibility, we also exam- ined STN activity during latency matched GO activity time- locked to the mean SSD obtained from SSD distribution observed during STOP trials. A general linear model similar to the above-mentioned analysis was then applied to Successful

STOP; Unsuccessful STOP and mean SSD centered LMGO tri- als, to ensure consistency of our results across the two methods.

To compute a temporally refined estimation of the latency at which a significant difference emerged (compared to the analyses above that were based on PSTH with a time resolu- tion of 75 msec per time bin) between Successful STOP and LMGO trials (i.e., for STOP units), we looked at the first time point for which the difference between spike density func- tions measured during STOP and during LMGO trials exceeded the mean differenceþ1.96 standard deviation of the differ- ence measured during a 200 msec baseline immediately pre- ceding cue onset, with the additional constraint that the significant effect on the difference had to last at least 50 msec so that significant differences corresponded to sustained re- sponses and did not reflect a transient (false positive) modu- lation of activity.

In GO units, z normalized firing rate was compared be- tween successful STOP, response to GO and Unsuccessful trial. Effect of trial type and trial time interaction on z normalized firing rate was then completed using a linear mixed model with similar parameters as used for STOP cells analysis. Also, to better estimate the onset latency of motor execution processes, we identified the first time point for which the spike density function for response to GO trials exceeded meanþ1.96 SD of baseline firing rate (200 msec on the middle of the interval before cue onset), with the addi- tional constraint that the significant effect on the difference had to last at least 50msec.

To determine how STN neurons mediated the two different functional responses across patients, we computed the grand- average activity separately for each group of cells, i.e., units that responded significantly to inhibition (Successful STOP trials) and motor responses (GO or Unsuccessful STOP trials).

3. Results

In total, 50 SST sessions were performed by 21 PD patients during the surgery (2.08±.21 SST sessions per STN side per patient). After excluding the sessions during which patients were unable to perform accurately the task (see methods), we further analyzed data obtained from 42 SST sessions from 21 PD patients (23 recording depths in the left STN and 19 in the right STN). Neuronal activity was recorded from 167 micro- electrodes in total (3.98 ± .15 microelectrodes per session).

Each SST session was composed of 123.21±5.45 SST trials on average (89.83±4.24 GO and 33.38±1.24 STOP trials).

3.1. Behavior

Fig. 1B shows the behavioral performances across all experi- mental sessions (n¼42). During GO trials, patients responded accurately on average (93.19±.96% correct responses) during GO trials whereas during STOP trials, accuracy was close to 50% (52.26± 1.355%), demonstrating that the staircase pro- cedure used to adjust the SSD as a function of patients'per- formance was effective in this study (see methods). On average, the SSRT was 312±17 msec (after STOP signal) and the SSD occurred 384±23 msec after GO signal. GO reaction

times and SSRT were not correlated [r(41)¼.1919,p¼.2234]

and reaction times during unsuccessful STOP (Unsuccessful STOP) trials were significantly shorter than reaction times observed during GO trials [pairedt-test:t(41)¼10.54,p<.0001].

To further investigate the relationship between PD patho- physiology and behavioral performance in the stop-signal task, we looked at the correlation between SSRT and pa- tients'demographic and clinical ratings. We found a signifi- cant positive correlation between SSRT and patients' age (Pearson correlation coefficient,r¼.32,n¼42,p¼.039), in line with a previous study demonstrating a significant effect of aging on SSRT in healthy subjects (Andres, Guerrini, Phillips,&

Perfect, 2008). However, no other significant correlation was identified between SSRT and clinical ratings (disease duration, UPDRS, Mattis and frontal score), in agreement with results from previous studies (Obeso, Wilkinson,&Jahanshahi, 2011;

Obeso et al., 2013).

3.2. Electrophysiological results

From the 167 available recordings, we found that 87 micro- recordings presented a neuronal spiking activity. These 87 recordings corresponded to the neural activity obtained from 42 SST sessions performed during the surgery by 21 patients with PD (on average, 2.02±.87 electrodes per SST experiment).

After spike sorting, we identified 98 STN units (1.14±.04 STN unit per electrode) that corresponded to 34 multi-units (MU) and 64 single-units (SU) (see methods,Tankus et al., 2009). 39 out of 98 recorded units showed a significant task-related response (p<.05; resampling test).

If STN cells are involved during response stopping, STN spiking activity should increase just after the STOP cue and before SSRT. To test this hypothesis using an unbiased esti- mation of STN neural activity, we first used a linear mixed model analysis on the activity of the entire population of task- responsive cells (n¼39 cells; see methods). We found a sig- nificant interaction between factors time trial type [F(55,70194)¼ 3.76,p<.00001] and contrast analysis revealed that STN firing rate was significantly higher during successful STOP than during latency matched GO trials. The effect reached significance at three (75 msec duration) time bin after STOP signal presentation (t ¼ 287.5; 437.5 and 512.5 msec;

p<.05 FDR corrected). Thus, the onset latency of the effect of stopping on STN neural activity at the population level occurred slightly later than the estimated SSRT obtained across the 39 task-responsive cells (SSRT¼281.6±14.78 msec, mean±SEM). However, note that there was a strong statistical tendency for the entire population of STN units to discharge before SSRT; indeed, the uncorrected effect reached signifi- cance before SSRT (t¼67.5 msec; uncorrectedp<.05;p¼.053 FDR corrected).

The GLM also revealed that task responsive cells were significantly modulated by motor execution processes: STN activity was higher during response to GO trials compared to successful STOP trials and virtual response centered STOP trials. Contrast analysis reached significance slightly before motor execution (at t ¼ 237.5 msec before button press compared to SS trials andt¼312.5 msec before button press for vr-SS trials; p <.05 FDR corrected). These results were replicated when centering vr-SS trials on virtual button

presses taken from the latter half of the GO reaction time distribution instead of the center part (see methods).

To further characterize the functional role of STN cells during the task and following previous studies in human (Bastin et al., 2014) and in animals (Isoda and Hikosaka, 2008;

Schmidt et al., 2013), we performed a second set of analyses after classifying each individual cell as a motor cell (GO cell) or a stopping cell (STOP cell, see methods).

Fig. 2a shows an example of single STOP cell that selec- tively increased its firing rate during successful STOP trials, at a latency that preceded the SSRT by 144 msec, thereby sug- gesting that this neuron could mediate a fast stopping signal.

We identified 11 STOP units in the STN (11.11% of recorded units, using our classification criteria, see methods). Linear mixed model analysis applied to STOP cells revealed a sig- nificant interaction between factors time trial type [F(22,6698)¼2.62,p<.00001] that was expected given the clas- sification criteria. Critically, contrast analysis showed that activity of STOP cells was significantly higher during suc- cessful STOP than unsuccessful STOP and latency matched GO trials as early as 185 msec before the average SSRT (p<.05 FDR corrected). To test whether STOP cells are modulated by the processing of GO cues, we examined STN activity during LMGO trials time-locked to the mean SSD observed during successful STOP trials (see methods). This analysis yielded identical results: linear mixed model analysis applied to STOP cells revealed a significant interaction between factors timetrial type [F(22,6736)¼2.26,p<.001], and contrast anal- ysis showed that activity of STOP cells was significantly higher during successful STOP than unsuccessful STOP and latency matched GO trials time-locked to the mean SSD as early as 185 msec before the average SSRT (p<.05 FDR corrected). This clearly shows that STOP cells are not modulated by the pro- cessing of GO cues.

To further test whether the increase of activity observed during successful STOP trials was compatible with response stopping (i.e., activity should increase at a latency that pre- cedes SSRT, Logan& Cowan, 1984), we computed for each STOP cell the latency at which the difference between suc- cessful STOP and latency matched GO activity reached sig- nificance (latency onset) and the peak of this difference (see methods). This analysis confirmed that the stopping effect (the difference between successful STOP and latency matched GO neural activity) became significant at a latency that significantly preceded the SSRT by 187.7±58.93 msec [paired t-test:t(10)¼3.187,p¼.0097] and peaked significantly earlier than the SSRT [130.1±50 msec before SSRT (mean±SEM), pairedt-test:t(10)¼2.612,p¼.0260]. To control a possible bias inherent to the use of a Gaussian kernel to compute the spike density function, we ran a control analysis using a causal kernel (Hanes, Patterson,&Schall, 1998) to estimate the la- tency at which the stopping effect emerged and peaked rela- tive to the SSRT. We found that the latencies (of both onsets and peaks) of the stopping effect still preceded SSRT using a causal kernel [onsets: 147.4 ± 47 msec before SSRT:

t(10)¼3.125,p¼.0108; peaks: 126.5±48 msec before SSRT;

t(10)¼2.641,p¼.0247].

We also identified 16 GO units in the STN (16.16% of the recorded units, using our classification criteria, see methods).

Fig. 3a shows a single GO cell that increased its firing rate

selectively during button presses (p<.05, resampling test).

Linear mixed model analysis revealed a significant interaction between factors timetrial type [F(33,22428)¼4.36,p<.0001].

Contrast analysis showed that the firing rate of GO cells was significantly higher during GO than during SS trials and vr-SS trials before motor execution (att¼ 162.5 msec before button presses;p<.01, FDR corrected). Results were similar when vr- SS trials were centered on button presses taken from the latter half of the GO-RT distribution (contrast reached significance slightly earlier, at t ¼ 237.5 msec before button presses;

p<.01, FDR corrected). We then computed the mean onset

latency of the firing rate increase during GO button press using the normalized spike density function (see methods). The la- tency onset of this firing rate increase occurred 316±67 msec before the response [one samplet-test,t(15)¼4.74,p¼.0003]

whereas the latency of the peak occurred 25.75 ± 23 msec before the response [one samplet-test:t(15)¼1.160;p¼.26].

This effect was not related to presentation of the GO cues because the increase in firing rate in a 400 msec time window after GO cue was significantly lower than the activity around button press presentation [pairedt-test,t(15)¼6.582,p<.0001].

We did not find any evidence of functional lateralization since Fig. 2eSelective response of STOP cells during motor inhibition. A. Activity of a single representative STOP cell activity during Successful STOP (left,n¼8 trials), latency matched GO trials (middle,n¼8 trials) and, Unsuccessful STOP (right, n¼8 trials) shown as rastergrams (black vertical lines, top), spike density functions (continuous lines, bottom) and peristimulus histograms. In rastergrams, blue dots indicate response time and green dots represent GO signal occurrence.

For illustrative purpose, the same number of trials is displayed in the rastergrams for the 3 trial types presented. The horizontal dashed lines on the peristimulus histogram represent the significance threshold of the permutation procedure (p<.05, see methods). The red vertical dashed line indicates the SSRT. The blue vertical dashed line indicates the reaction time. B. Grand average Z normalized spike activity of STOP cells (n¼11 STOP units) during Successful STOP (red), latency matched GO trials (green) and Unsuccessful STOP (purple). Note that neuronal activity is time-locked to the STOP signal (during Successful and Unsuccessful STOP trials) or to a virtual STOP cue (during LMGO trials, see methods). The red vertical dashed line indicates the SSRT. The black vertical dashed line indicates the STOP signal occurrence. Error bars represent 95% confidence intervals of the GLM. *Successful STOP versus latency matched GO and Unsuccessful STOP trialsp<.05 FDR corrected. ***Successful STOP versus latency matched GO and Unsuccessful STOP trialsp<.001 FDR corrected, #Successful STOP versus Unsuccessful STOPp<.05 FDR corrected.

6 out of 11 STOP neurons and 6 out of 16 GO neurons were found in the STN ipsilateral to the response hand.

4. Discussion

We identified two neuronal populations in the STN of PD pa- tients presenting a selective increase of firing rate either during motor execution (GO cells) or during response inhibi- tion (STOP cells). STOP cells fired rapidly, i.e., before the SSRT, suggesting that this increase could mediate a fast stopping signal predicted by several computational models (Boucher et al., 2007; Wiecki& Frank, 2013). This result suggests the existence of functionally antagonistic STN neuronal pop- ulations: a population would promote motor execution whereas another population may contribute to inhibitory control.

Before further developing the theoretical impact of our findings, let us consider some of the limitations inherent to this study. Because patients with PD have been associated with impaired reactive inhibitory (Gauggel, Rieger,&Feghoff, 2004; Obeso, Wilkinson, Casabona, et al., 2011), whether the results of our study can be generalized to healthy subjects remains controversial. Specifically, we cannot rule out the possibility that the increase of STN neuronal activity found before the SSRT might be induced by the longer SSRT that is characteristic of PD. Note that alternatively, it is also possible that STN neurons fire actually more slowly in PD than in healthy subjects during the stop signal task.

4.1. STN neuronal activity during stopping

We hypothesize that the fast increase of firing rate observed in the STN during successful stopping may reflect a fast global Fig. 3eSelective response of GO cells during motor preparation and execution. A. Activity of a single representative motor cell activity during Successful STOP trials (column 1,n¼7 trials), correct motor responses to GO trials (column 2,n¼35 trials) and incorrect motor responses to Unsuccessful STOP trials button presses (US-BP) (column 3,n¼8 trials), shown as rastergrams (black vertical lines, top), spike density functions (continuous lines, bottom) and peristimulus histograms. In rastergrams, green dots represent GO signal occurrence. For illustrative purpose, the same number of trials (7 trials) is displayed in the rastergrams for the 3 trial types presented. The horizontal dashed lines on the peristimulus histogram represent the significance threshold of the permutation procedure (p<.05, see methods). B. Grand average Z normalized spike activity of motor cells (n¼16 motor cells) during Successful STOP (red), response to GO trials (blue) and response to Unsuccessful STOP trials (purple). Note that neuronal activity is time-locked to the STOP signal (during Successful STOP trials) or to correct/incorrect button press (during GO or unsuccessful STOP trials). The black pointed line indicates the STOP signal occurrence. The black dashed line indicates the response occurrence. Error bars represent 95% confidence intervals of the GLM. ***GO button press versus successful STOP trialsp<.001 FDR corrected.

stopping signal (Aron, 2011) that would act as a brake on the motor system. This type of signal might also be useful to adjust the speed-accuracy trade-off in the context of decision- making tasks, in particular to favor accurate (but slow) de- cisions (Frank, 2006; van Maanen et al., 2011). In line with this hypothesis, a neuronal population was shown to increase its activity during difficult decisions in the STN (Zaghloul et al., 2012). Previous functional fMRI studies have indicated that such signal could be mediated by the STN (Aron, 2006; Li et al., 2008) and might underlie STN's critical involvement during motor inhibition (Baunez et al., 1995; Crossman et al., 1984;

Obeso et al., 2014). Furthermore, a relative increase of the amplitude of STN local field potentials in the beta band occurs during inhibition (Alegre et al., 2013; Ku¨hn et al., 2004; Ray et al., 2012), at latencies preceding the SSRT, in both PD (Benis et al., 2014) and OCD (Bastin et al., 2014) patients.

The identification of STOP cells is consistent with these studies and provides a simple single-cell mechanism that fits with both models and experimental/animal data. This finding also confirms previous extracellular recordings of STN activity during cognitive tasks that tapped on similar executive con- trol processes. In monkeys, a population of STN neurons was shown to increase its firing rate as early as 175 msec after a NOGO cue during an action reprograming task (Isoda &

Hikosaka, 2008). However, the design of the GO/NOGO task does not allow a clear estimation of the precise duration of inhibitory processes and therefore lacks a measure such as SSRT to interpret this latency. Conversely, another study suggested that in the STN of rats, neurons increase their ac- tivity at very fast latencies, around 15 msec post STOP cue, during both correct and incorrect STOP trials (Schmidt et al., 2013). In contrast, the human STN seems to distinguish cor- rect from incorrect STOP trials, at least in patients with PD or OCD (Bastin et al., 2014). These inter-species differences might reflect distinct electrophysiological recording context (e.g., the spatial sampling of STN during extracellular recordings is generally weak in humans), inter-species differences regarding STN connectivity with the prefrontal cortex (Astafiev et al., 2003) or a different implementation of inhibi- tory control signals in the human prefrontal cortex that would impact basal ganglia activities (Wessel& Aron, 2015). On a related note, thalamic and pedunculopontine nucleus pro- jections to the STN could be the substrate for the very quick STN response observed in rat (Schmidt et al., 2013), while STN activity in humans could be driven by a slower prefrontal- subthalamic pathway (originating in the pre-SMA or in the inferior frontal gyrus, seeRae, Hughes, Anderson, Rowe,&

Rowe, 2015).

Our findings also fit well with current models of action inhibition. The STN has been postulated to implement a fast stopping process, a“hold your horse”signal to quickly stop an action when required (Frank, 2006; Wiecki & Frank, 2013).

According to the horse race model of the SST, this process must take place before the SSRT (Logan&Cowan, 1984). Our results support this hypothesis by demonstrating that STOP neurons discharge as early as 187 msec before the SSRT. This latency is consistent with a direct activation of STN neurons by the cortex via the fast conducting hyperdirect pathway to inhibit a prepotent response (Nambu, Tokuno, & Takada, 2002). Recent recordings within the basal ganglia network of

rats also suggested a two-step process of reactive inhibition: a fast transient movement“pause” process through the acti- vation of the STN-SNr pathway via the fast hyperdirect connection with the cortex and a late movement “cancel” process via the feedback loop between the striatum and the GPe (Mallet et al., 2016; Schmidt et al., 2013). Our results showing an early pre-SSRT and a late post-SSRT response of STOP neurons in the STN fit well with this theoretical frame- work and suggest that, in humans, the STN might also be involved in the late cancellation of ongoing movement, perhaps through the reciprocal STN-GPe connectivity.

According to our results, neuronal populations encoding facilitation and quick inhibition of movement coexist in the STN. This raises an interesting question: how can two neuronal subpopulations in the same structure mimic the two independent GO and STOP processes of the horse race model?

In a previous study, a potential mechanism has been sug- gested to explain this phenomenon through differential downstream connections of both populations (Isoda &

Hikosaka, 2008). Accordingly, while STOP neurons could pre- sent a direct excitatory connectivity with the substantia nigra pars reticulata (SNr) and internal Globus Pallidus (GPi) to inhibit the thalamus, GO neurons could be indirectly con- nected with the SNr through the external Globus Pallidus (GPe), so that GO cells would have an opposite excitatory effect on the thalamus (Isoda&Hikosaka, 2008). An alternative hy- pothesis was raised by a recent model, the interactive horse race model, assuming inhibitory connections between GO and STOP units (Boucher et al., 2007). While the two units would be acting independently for most of the STOP trials duration, the STOP unit would start to interact with the GO unit towards the time of the SSRT to inhibit GO units activity. Finally, the role of STN units during GO and STOP trials might actually reflect action initiation processes. More specifically, a classical model postulates that STN activity contributes to the suppression of alternative motor pattern to ensure the execution of the planned movement (Mink, 1996; Nambu et al., 2002). In this framework, activity of GO neurons would correspond to this process of selective inhibition of competing motor plans just before motor execution whereas activity of STOP neurons would reflect a more global stopping process when all actions have to be inhibited. However, this hypothesis predicts an increase of activity of STOP units above baseline level after the GO cue (i.e., when movement initiation begins) and an in- crease of activity above baseline level of GO units during STOP trials. We were not able to provide strong support for this model since no STOP cell increased its activity after the GO cue relative to baseline levels and no GO units increased their activity on STOP trials (p>.05; bootstrapping test). Conversely, results from this study are consistent with the idea according to which STN neurons could potentially represent a “wide- spread motor inhibitory process” that would modulate the gain in the motor system according to the context (Greenhouse, Sias, Labruna,&Ivry, 2015).

As a number of studies showed that STN activity was related to the processing of behaviorally relevant cues (Sarma et al., 2012; Sauleau et al., 2009), STN neuronal activity observed in this study could also reflect the processing of salient stimuli, rather than motor or inhibitory processes per se. A number of elements go against this alternative

explanation of our results. Indeed, STOP cells displayed a significantly higher activity during successful versus unsuc- cessful STOP trials whereas a salient event (i.e., the STOP signal) was present in both cases; furthermore, STOP cells did not present a significant increase in firing rate during GO tri- als, whereas GO cues are also clearly behaviorally relevant stimuli. Similarly, no significant increase in firing rate was observed for GO cells during successful STOP trials, despite the fact that the STOP cue is a highly salient event.

4.2. Neuronal activity in the STN of PD versus OCD patients

Here, we found that a majority of task responsive neurons recorded in PD patients responded to motor preparation/

execution events (n¼16 GO cells) whereas a minority of units responded to executive control (n¼11 STOP cells). Interest- ingly, we previously reported an opposite pattern of results in the STN of OCD patients, where a majority of task responsive neurons responded to executive processes (n ¼ 12 error- cellsþn¼10 STOP cells) and a minority to motor processes (n¼10 GO cells). Note that both studies included human STN cells recorded during an identical task and experimental setting so that we are in an interesting position to compare directly functional responses of STN neurons across these studies. First, note that the difference of proportion between motor versus non-motor cells across PD versus OCD patients was significant (c2 test,p¼.0309). This significant difference mainly comes from the absence of units firing consistently after behavioral errors in PD patients. Several competing explanations could account for this important difference between studies.

A first possibility might be related to the anatomical loca- tion of the recordings that are likely to differ between studies:

in OCD patients, the clinical target is located in the associative and limbic territory whereas in PD patients, the clinical target is located in the sensorimotor STN (Chabardes et al., 2012), and previous anatomical reconstruction did demonstrate that STN neurons recorded in OCD are more anterior than in PD (Piallat et al., 2011; Welter et al., 2011). Therefore, we speculate that the absence of error-monitoring cells in PD might be related to a lower sampling of the STN subregions where these cells might be clustered. This idea would support a tripartite model of STN functional subterritories, in coherence with previous neuroimaging (Haynes&Haber, 2013; Karachi et al., 2005; Parent, 1990) and stimulation studies (Greenhouse et al., 2011; Lambert et al., 2011; Mallet et al., 2007;

Sudhyadhom et al., 2007).

Conversely, we also note that the presence of two identical functional neuronal subpopulations (GO and STOP cells) in the STN of both OCD and PD patients is also compatible with models postulating that the STN may actually integrate motor, associative and limbic information via modulations of firing rate in different neuronal populations (Burbaud et al., 2013; Peron, Fru¨hholz, Verin, & Grandjean, 2013; Temel, Blokland, Steinbusch, & Visser-Vandewalle, 2005). Note, however, that the low number of task-responsive cells found in both studies combined with the absence of high resolution clinical anatomical images and tractography data did not allow us to provide clearer empirical support for hypotheses related to STN functional territories.

Difference between the cognitive profiles of PD and OCD patients might also explain a significant part of the intriguing difference found between the two studies: for example, it re- mains possible that while deficits in inhibitory control in the STOP signal task have been described in Parkinsonian (Gauggel et al., 2004) and OCD patients (Lipszyc&Schachar, 2010), error-monitoring processes might be differentially affected. Hence, deficits in post-error behavioral adjustments have been reported in PD patients compared to healthy con- trols (Farooqui et al., 2011) whereas converging evidences suggest that performance monitoring processes are over- active in OCD patients (Endrass, Klawohn, Schuster, &

Kathmann, 2008; Ursu, Stenger, Shear, Jones,&Carter, 2003).

Note that in our studies, PD patients displayed a more serious inhibitory deficit than the OCD population in our previous study (SSRT of 242 ± 64 msec in OCD patients against 312±17 msec for PD patients). Finally, the absence of error- monitoring units in PD patients could also reflect a sampling bias even if this seems unlikely since the number of recorded neurons is even larger in the PD study (n¼98 units;n¼21 patients) than in the OCD study (n¼75 units, 7 patients).

5. Conclusion

In conclusion, our results demonstrate the existence of two functionally opposite neuronal populations in the STN.

Whereas GO units promote motor execution, STOP units inhibit the motor system in situations during which a conflict has to be rapidly resolved to select context-dependent re- sponses. Our study provides the electrophysiological corre- lates of such braking role of some STN neurons on the motor system. Furthermore, the identification of GO units during motor execution suggests that a variety of cognitive responses could be represented in the human STN by distinct neuronal populations.

Acknowledgements

The authors would like to thank the patients for their partic- ipation as well as the clinical team of the department of neurosurgery and neurology of the Grenoble University Hos- pital. Funding was provided by the Grenoble University Hos- pital (project DRCI 2010 IP-ESCP) by“Agence Nationale pour la Recherche”, grant ANR-14-CE13-0030-01“Physiobs”. The au- thors declare no competing financial interests. The neuro- physiology facility was partly funded by the French program

“Investissement d’Avenir”run by the“Agence Nationale pour la Recherche”: grant“Infrastructure d'avenir en Biologie Sante e ANR-11-INBS-0006”. Astrid Kibleur was partly funded by Fondation pour la Recherche Medicale (FDT20150532565).

r e f e r e n c e s

Alegre, M., Lopez-Azcarate, J., Obeso, I., Wilkinson, L., Rodriguez- Oroz, M. C., Valencia, M., et al. (2013). The subthalamic nucleus is involved in successful inhibition in the stop-signal

task: A local field potential study in Parkinson's disease.

Experimental Neurology, 239, 1e12.

Andres, P., Guerrini, C., Phillips, L. H., & Perfect, T. J. (2008).

Differential effects of aging on executive and automatic inhibition.Developmental Neuropsychology, 33(2), 101e123.

Aron, A. R. (2006). Cortical and subcortical contributions to stop signal response inhibition: Role of the subthalamic nucleus.

Journal of Neuroscience, 26(9), 2424e2433.

Aron, A. R. (2011). From reactive to proactive and selective control: Developing a richer model for stopping inappropriate responses.Biological Psychiatry, 69(12), e55e68.

Astafiev, S. V., Shulman, G. L., Stanley, C. M., Snyder, A. Z., Van Essen, D. C., & Corbetta, M. (2003). Functional organization of human intraparietal and frontal cortex for attending, looking, and pointing.The Journal of Neuroscience, 23(11), 4689e4699.

Bastin, J., Polosan, M., Benis, D., Goetz, L., Bhattacharjee, M., Piallat, B., et al. (2014). Inhibitory control and error monitoring by human subthalamic neurons.Translational Psychiatry, 4, e439.

Baunez, C., Nieoullon, A., & Amalric, M. (1995). In a rat model of parkinsonism, lesions of the subthalamic nucleus reverse increases of reaction time but induce a dramatic premature responding deficit.Journal of Neuroscience, 15(10), 6531e6541.

Benabid, A. L., Chabardes, S., Mitrofanis, J., & Pollak, P. (2009).

Deep brain stimulation of the subthalamic nucleus for the treatment of Parkinson's disease.Lancet Neurology, 8(1), 67e81.

Benis, D., David, O., Lachaux, J.-P. P., Seigneuret, E., Krack, P., Fraix, V., et al. (2014). Subthalamic nucleus activity dissociates proactive and reactive inhibition in patients with Parkinson's disease.NeuroImage, 91, 273e281.

Boucher, L., Palmeri, T. J., Logan, G. D., & Schall, J. D. (2007).

Inhibitory control in mind and brain: An interactive race model of countermanding saccades.Psychological Review, 114(2), 376e397.

Burbaud, P., Clair, A.-H., Langbour, N., Fernandez-Vidal, S., Goillandeau, M., Michelet, T., et al. (2013). Neuronal activity correlated with checking behavior in the subthalamic nucleus of patients with obsessive-compulsive disorder.Brain, 136(Pt 1), 304e317.

Chabardes, S., Polosan, M., Krack, P., Bastin, J., Krainik, A., David, O., et al. (2012). Deep brain stimulation for obsessive- compulsive disorder: Subthalamic nucleus target.World Neurosurgery, 80(3e4), 1e8.

Chikazoe, J., Jimura, K., Hirose, S., Yamashita, K.-i., Miyashita, Y.,

& Konishi, S. (2009). Preparation to inhibit a response complements response inhibition during performance of a stop-signal task.Journal of Neuroscience, 29(50), 15870e15877.

Crossman, A. R., Sambrook, M. A., & Jackson, A. (1984).

Experimental hemichorea/hemiballismus in the monkey:

Studies on the intracerebral site of action in a drug-induced dyskinesia.Brain, 107(2), 579e596.

Eagle, D. M., Baunez, C., Hutcheson, D. M., Lehmann, O., Shah, A. P., & Robbins, T. W. (2008). Stop-signal reaction-time task performance: Role of prefrontal cortex and subthalamic nucleus.Cerebral Cortex, 18(1), 178e188.

Endrass, T., Klawohn, J., Schuster, F., & Kathmann, N. (2008).

Overactive performance monitoring in obsessive-compulsive disorder: ERP evidence from correct and erroneous reactions.

Neuropsychologia, 46(7), 1877e1887.

Farooqui, A. A., Bhutani, N., Kulashekhar, S., Behari, M., Goel, V.,

& Murthy, A. (2011). Impaired conflict monitoring in Parkinson's disease patients during an oculomotor redirect task.Experimental Brain Research, 208(1), 1e10.

Frank, M. J. (2006). Hold your horses: A dynamic computational role for the subthalamic nucleus in decision making.Neural Networks, 19(8), 1120e1136.

Gauggel, S., Rieger, M., & Feghoff, T. (2004). Inhibition of ongoing responses in patients with Parkinson's disease.Journal of Neurology, Neurosurgery, and Psychiatry, 75(4), 539.

Genovese, C. R., Lazar, N. A., & Nichols, T. (2002). Thresholding of statistical maps in functional neuroimaging using the false discovery rate.NeuroImage, 15(4), 870e878.

Greenhouse, I., Gould, S., Houser, M., Hicks, G., Gross, J., &

Aron, A. R. (2011). Stimulation at dorsal and ventral electrode contacts targeted at the subthalamic nucleus has different effects on motor and emotion functions in Parkinson's disease.Neuropsychologia, 49(3), 528e534.

Greenhouse, I., Sias, A., Labruna, L., & Ivry, R. B. (2015).

Nonspecific inhibition of the motor system during response preparation.Journal of Neuroscience, 35(30), 10675e10684.

Hanes, D. P., Patterson, W. F., & Schall, J. D. (1998). Role of frontal eye fields in countermanding saccades: Visual, movement, and fixation activity.Journal of Neurophysiology, 79(2), 817e834.

Haynes, W. I. A., & Haber, S. N. (2013). The organization of prefrontal-subthalamic inputs in primates provides an anatomical substrate for both functional specificity and integration: Implications for basal ganglia models and deep brain stimulation.Journal of Neuroscience, 33(11), 4804e4814.

Hochberg, Y. (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing.Journal of the Royal Statistical Society, 57(1), 289e300.

Isoda, M., & Hikosaka, O. (2008). Role for subthalamic nucleus neurons in switching from automatic to controlled eye movement.Journal of Neuroscience, 28(28), 7209e7218.

Jha, A., Nachev, P., Barnes, G., Husain, M., Brown, P., & Litvak, V.

(2015). The frontal control of stopping.Cerebral Cortex, 25(11), 4392e4406.

Karachi, C., Yelnik, J., Tande, D., Tremblay, L., Hirsch, E. C., &

Franc¸ois, C. (2005). The pallidosubthalamic projection: An anatomical substrate for nonmotor functions of the subthalamic nucleus in primates.Movement Disorders, 20(2), 172e180.

Kuhn, A. A. (2004). Event-related beta desynchronization in human subthalamic nucleus correlates with motor performance.Brain, 127(4), 735e746.

Lambert, C., Zrinzo, L., Nagy, Z., Lutti, A. A., Hariz, M., Foltynie, T., et al. (2011). Confirmation of functional zones within the human subthalamic nucleus: Patterns of connectivity and sub-parcellation using diffusion weighted imaging.

NeuroImage, 60(1), 83e94.

Lidierth, M. (2009). sigTOOL: A MATLAB-based environment for sharing laboratory-developed software to analyze biological signals.Journal of Neuroscience Methods, 178(1), 188e196.

Lipszyc, J., & Schachar, R. (2010). Inhibitory control and psychopathology: A meta-analysis of studies using the stop signal task.Journal of the International Neuropsychological Society:

JINS, 16(06), 1064e1076.

Li, C. R., Yan, P., Sinha, R., & Lee, T. (2008). Subcortical processes of motor response inhibition during a stop signal task.

NeuroImage, 41(4), 1352e1363.

Logan, G. D., & Cowan, W. B. (1984). On the ability to inhibit thought and action: A theory of an act of control.Psychological Review, 91(3), 295e327.

van Maanen, L., Brown, S. D., Eichele, T., Wagenmakers, E.-J., Ho, T., Serences, J., et al. (2011). Neural correlates of trial-to- trial fluctuations in response caution.Journal of Neuroscience, 31(48), 17488e17495.

Mallet, N., Schmidt, R., Leventhal, D., Chen, F., Amer, N., Boraud, T., et al. (2016). Arkypallidal cells send a stop signal to striatum.Neuron, 89(2), 308e316.

Mallet, L., Schupbach, M., N'Diaye, K., Remy, P., Bardinet, E., Czernecki, V., et al. (2007). Stimulation of subterritories of the subthalamic nucleus reveals its role in the integration of the emotional and motor aspects of behavior.Proceedings of the National Academy of Sciences, 104(25), 10661e10666.

Mattis, S. (1976). Mental status examination for organic mental syndrome in the elderly patient.Geriatric Psychiatry, 11(77), e121.

Mink, J. W. (1996). The basal ganglia: Focused selection and inhibition of competing motor programs.Progress in Neurobiology, 50, 381e425.

Mirabella, G., Iaconelli, S., Romanelli, P., Modugno, N., Lena, F., Manfredi, M., et al. (2012). Deep brain stimulation of subthalamic nuclei affects arm response inhibition in Parkinson's patients.Cerebral Cortex, 22(5), 1124e1132.

Monakow, K. H., Akert, K., & Ku¨nzle, H. (1978). Projections of the precentral motor cortex and other cortical areas of the frontal lobe to the subthalamic nucleus in the monkey.Experimental Brain Research, 33(3e4), 395e403.

Nambu, A., Takada, M., Inase, M., & Tokuno, H. (1996). Dual somatotopical representations in the primate subthalamic nucleus: Evidence for ordered but reversed body-map transformations from the primary motor cortex and the supplementary motor area.Journal of Neuroscience, 16(8), 2671e2683.

Nambu, A., Tokuno, H., & Takada, M. (2002). Functional significance of the cortico-subthalamo-pallidal“hyperdirect”

pathway.Neuroscience Research, 43, 111e117.

Nishioka, H., Taguchi, T., Nanri, K., & Ikeda, Y. (2008). Transient hemiballism caused by a small lesion of the subthalamic nucleus.Journal of Clinical Neuroscience, 15, 1416e1418.

Obeso, I., Wilkinson, L., Casabona, E., Bringas, M. L.,Alvarez, M., Alvarez, L., et al. (2011). Deficits in inhibitory control and conflict resolution on cognitive and motor tasks in Parkinson's disease.Experimental Brain Research, 212(3), 371e384.

Obeso, I., Wilkinson, L., Casabona, E., Speekenbrink, M., Luisa Bringas, M., Alvarez, M., et al. (2014). The subthalamic nucleus and inhibitory control: Impact of subthalamotomy in Parkinson's disease.Brain, 137(5), 1470e1480.

Obeso, I., Wilkinson, L., & Jahanshahi, M. (2011). Levodopa medication does not influence motor inhibition or conflict resolution in a conditional stop-signal task in Parkinson's disease.Experimental Brain Research, 213(4), 435e445.

Obeso, I., Wilkinson, L., Rodrı´guez-Oroz, M.-C., Obeso, J. A., &

Jahanshahi, M. (2013). Bilateral stimulation of the subthalamic nucleus has differential effects on reactive and proactive inhibition and conflict-induced slowing in Parkinson's disease.

Experimental Brain Research, 226(3), 451e462.

Parent, A. (1990). Extrinsic connections of the basal ganglia.

Trends in Neurosciences, 13(7), 254e258.

Peron, J., Fru¨hholz, S., Verin, M., & Grandjean, D. (2013).

Subthalamic nucleus: A key structure for emotional component synchronization in humans.Neuroscience and Biobehavioral Reviews, 37(3), 358e373.

Piallat, B., Polosan, M., Fraix, V., Goetz, L., David, O., Fenoy, A., et al. (2011). Subthalamic neuronal firing in obsessive- compulsive disorder and Parkinson disease.Annals of Neurology, 69(5), 793e802.

Pillon, B., Dubois, B., Lhermitte, F., & Agid, Y. (1986). Heterogeneity of cognitive impairment in progressive supranuclear palsy, Parkinson's disease, and Alzheimer's disease.Neurology, 36(9), 1179e1185.

Quiroga, R. Q., Nadasdy, Z., & Ben-Shaul, Y. (2004). Unsupervised spike detection and sorting with wavelets and

superparamagnetic clustering.Neural Computation, 16(8), 1661e1687.

Rae, C. L., Hughes, L. E., Anderson, M. C., Rowe, X. B., & Rowe, J. B.

(2015). The prefrontal cortex achieves inhibitory control by facilitating subcortical motor pathway connectivity.Journal of Neuroscience, 35(2), 786e794.

Ratcliff, R., & Frank, M. J. (2012). Reinforcement-based decision making in corticostriatal circuits: Mutual constraints by

neurocomputational and diffusion models.Neural Computation, 24(5), 1186e1229.

Ray, N. J., Brittain, J.-S., Holland, P., Joundi, R. A., Stein, J. F., Aziz, T. Z., et al. (2012). The role of the subthalamic nucleus in response inhibition: Evidence from local field potential recordings in the human subthalamic nucleus.NeuroImage, 60(1), 271e278.

Ray, N. J., Jenkinson, N., Brittain, J., Holland, P., Joint, C., Nandi, D., et al. (2009). The role of the subthalamic nucleus in response inhibition: Evidence from deep brain stimulation for Parkinson's disease.Neuropsychologia, 47(13), 2828e2834.

Sarma, S. V., Cheng, M. L., Eden, U., Williams, Z., Brown, E. N., &

Eskandar, E. (2012). The effects of cues on neurons in the basal ganglia in Parkinson's disease.Frontiers in Integrative

Neuroscience, 6, 40. July.

Sauleau, P., Eusebio, A., Thevathasan, W., Yarrow, K., Pogosyan, A., Zrinzo, L., et al. (2009). Involvement of the subthalamic nucleus in engagement with behaviorally relevant stimuli.The European Journal of Neuroscience, 29(5), 931e942.

Schall, J. D., Stuphorn, V., & Brown, J. W. (2002). Monitoring and control of action by the frontal lobes.Neuron, 36(2), 309e322.

Schmidt, R., Leventhal, D. K., Mallet, N., Chen, F., & Berke, J. D.

(2013). Canceling actions involves a race between basal ganglia pathways.Nature Neuroscience, 16(8), 1118e1124.

Sudhyadhom, A., Bova, F. J., Foote, K. D., Rosado, C. A., Kirsch- Darrow, L., & Okun, M. S. (2007). Limbic, associative, and motor territories within the targets for deep brain stimulation:

Potential clinical implications.Current Neurology and Neuroscience Reports, 7(4), 278e289.

Swann, N., Poizner, H., Houser, M., Gould, S., Greenhouse, I., Cai, W., et al. (2011). Deep brain stimulation of the subthalamic nucleus alters the cortical profile of response inhibition in the beta frequency band: A scalp EEG study in Parkinson's disease.Journal of Neuroscience, 31(15), 5721e5729.

Swann, N., Tandon, N., Canolty, R., Ellmore, T. M., McEvoy, L. K., Dreyer, S., et al. (2009). Intracranial EEG reveals a time- and frequency-specific role for the right inferior frontal gyrus and primary motor cortex in stopping initiated responses.Journal of Neuroscience, 29(40), 12675e12685.

Tankus, A., Yeshurun, Y., & Fried, I. (2009). An automatic measure for classifying clusters of suspected spikes into single cells versus multiunits.Journal of Neural Engineering, 6(5), 56001.

Temel, Y., Blokland, A., Steinbusch, H. W. M., & Visser-

Vandewalle, V. (2005). The functional role of the subthalamic nucleus in cognitive and limbic circuits.Progress in

Neurobiology, 76(6), 393e413.

Thobois, S., Ardouin, C., Lhommee, E., Klinger, H., Lagrange, C., Xie, J., et al. (2010). Non-motor dopamine withdrawal syndrome after surgery for Parkinson's disease: Predictors and underlying mesolimbic denervation.Brain, 133(4), 1111e1127.

Ursu, S., Stenger, V. A., Shear, M. K., Jones, M. R., & Carter, C. S.

(2003). Overactive action monitoring in obsessive compulsive disorder: Evidence from functional magnetic resonance imaging.Gene Therapy, 9(4), 53e63.

Verbruggen, F., Chambers, C. D., & Logan, G. (2012). Fictitious inhibitory differences: How skewness and slowing distort the estimation of stopping latencies.Psychological Science, 24(3), 352e362.

Verbruggen, F., & Logan, G. D. (2008). Response inhibition in the stop-signal paradigm.Trends in Cognitive Sciences, 12(11), 418e424.

Welter, M.-L., Burbaud, P., Fernandez-Vidal, S., Bardinet, E., Coste, J., Piallat, B., et al. (2011). Basal ganglia dysfunction in OCD: Subthalamic neuronal activity correlates with symptoms severity and predicts high-frequency stimulation efficacy.Translational Psychiatry, 1, e5.

Wessel, J. R., & Aron, A. R. (2015). It's not too late: The onset of the frontocentral P3 indexes successful response inhibition in the stop-signal paradigm.Psychophysiology, 52(2014), 472e480.

Wessel, J. R., Conner, C. R., Aron, A. R., & Tandon, N. (2013).

Chronometric electrical stimulation of right inferior frontal cortex increases motor braking.Journal of Neuroscience, 33(50), 19611e19619.

Wiecki, T. V., & Frank, M. J. (2013). A computational model of inhibitory control in frontal cortex and basal ganglia.

Psychological Review, 120(2), 329e355.

van den Wildenberg, W. P. M., van Boxtel, G. J. M., van der Molen, M. W., Bosch, D. A., Speelman, J. D., & Brunia, C. H. M.

(2006). Stimulation of the subthalamic region facilitates the selection and inhibition of motor responses in Parkinson's disease.Journal of Cognitive Neuroscience, 18(4), 626e636.

Zaghloul, K. A., Weidemann, C. T., Lega, B. C., Jaggi, J. L., Baltuch, G. H., & Kahana, M. J. (2012). Neuronal activity in the human subthalamic nucleus encodes decision conflict during action selection.Journal of Neuroscience, 32(7), 2453e2460.