## Abstract

The detection and assessment of errors are a prerequisite to adapt behavior and improve future performance. Error monitoring is afforded by the interplay between cortical and subcortical neural systems. Ample evidence has pointed to a specific cortical error-related evoked potential, the error-related negativity (ERN), during the detection and evaluation of response errors. Recent models of reinforcement learning implicate the basal ganglia (BG) in early error detection following the learning of stimulus–response associations and in the modulation of the cortical ERN. To investigate the influence of the human BG motor output activity on the cortical ERN during response errors, we recorded local field potentials from the sensorimotor area of the internal globus pallidus and scalp electroencephalogram representing activity from the posterior medial frontal cortex in patients with idiopathic dystonia (hands not affected) during a flanker task. In error trials, a specific pallidal error-related potential arose 60 ms prior to the cortical ERN. The error-related changes in pallidal activity—characterized by theta oscillations—were predictive of the cortical error-related activity as assessed by Granger causality analysis. Our findings show an early modulation of error-related activity in the human pallidum, suggesting that pallidal output influences the cortex at an early stage of error detection.

## Introduction

Motor tasks, such as typing on a keyboard or playing piano, are prone to errors. Error detection and adaptive learning, which lead to the avoidance of future errors, are essential for successful motor performance. These error-monitoring mechanisms are implemented at different levels of the central nervous system, including cortical and subcortical areas. The focus of the present study is based on the investigation of cortico-basal ganglia (BG) interactions in processing response errors.

The error-related negativity (ERN) is a specific event-related cortical potential (ERP) peaking at 50–100 ms after erroneous outcomes that can be used to study the function of the error-monitoring system (Falkenstein et al. 1990; Gehring et al. 1993; Taylor et al. 2007; Ullsperger et al. 2007). An ERN is generally associated with error commission and interpreted as an index of (1) the mismatch between an expected and upcoming response (Gehring et al. 1993; Falkenstein et al. 1995) or (2) the degree of conflict between competing erroneous and correct responses (Botvinik et al. 1999; Coles et al. 2001). A multitude of neurophysiological, functional imaging, and computational modeling studies implicate the posterior medial frontal cortex (pMFC) and dorsal anterior cingulate cortex in the detection and evaluation of response errors and generation of the ERN/error-related activity (Ridderinkhof et al. 2004; Debener et al. 2005; Taylor et al. 2007; Ullsperger et al. 2007; Agam et al. 2011; Silvetti et al. 2011). However, more recent magnetoencephalography (MEG) and functional magnetic resonance imaging evidence from response-time conflict tasks point to a broader network of areas modulating the ERN. These areas include the supplementary motor area, insula, inferior and superior frontal gyrus, parietal lobe, precuneus, and thalamus (Kiehl et al. 2000; Hester et al. 2004, 2009; Holroyd et al. 2004; Doñamayor et al. 2011). The findings of abnormal ERN modulation and/or behavioral performance during error monitoring in patients with Parkinson's disease (Falkenstein et al. 2001; Stemmer et al. 2007; Willemssen et al. 2009), Huntington's disease (Beste et al. 2006, 2008), Tourette syndrome (Johannes et al. 2002), or focal lesions of the BG (Ullsperger and von Cramon 2006) and the thalamus (Seifert et al. 2011) all point to the BG as an important subcortical relay station in monitoring and adjusting motor performance.

These observations are in line with the notion of reinforcement learning (RL; Holroyd and Coles 2002; Cockburn and Frank 2011) where the mesencephalic dopaminergic system encodes reward prediction errors that teach the BG via temporal difference (TD) learning to link cues or actions to their estimated “future” outcome and reward (Sutton 1988; Schultz et al. 1997; Sutton and Barto 1998; Schultz 2002). As a consequence, BG may be able to predict upcoming nonrewarded actions, thereby influencing the ERN. To test this prediction, we investigated whether error-related activity can be directly recorded from the human pallidum and evaluated the timing and directionality of cortico-BG interactions during the processing of response errors arising from laterality errors in a reaction time task with automatic stimulus-response mapping.

BG activity may also influence the forward control of movements in order to maximize the ratio between cost and benefit based on the predicted costs (energy cost and error/accuracy cost) and rewards (Shadmehr and Krakauer 2008). Within this framework, so far formulated and tested in reaching movements (Mazzoni et al. 2007; Shadmehr and Krakauer 2008), prediction of costs and “implicit” rewards associated with upcoming actions by the BG would enable these nuclei to influence the cortical error-monitoring processes in advance of future outcomes. Thus, early motor BG activity predictive of upcoming errors would support that an implicit motivational circuit might operate within the motor frontal-BG-thalamocortical loop.

Animal data have shown a specific error-related single-cell neural activity in different nuclei of the BG such as the primate dorsal pallidum (GPi and external globus pallidus [GPe]) or rodent subthalamic nucleus (STN) during error commission, yet in tasks quite different than those probed in humans (Arkadir et al. 2004; Hong and Hikosaka 2008; Lardeux et al. 2009; Hong et al. 2011). It is noteworthy that the seminal works of Hikosaka (e.g. Hikosaka et al. 2008) emphasize that error- and reward-modulated activity can be recorded in the primate dorsal pallidum, part of the “sensorimotor” cortico-BG-thalamocortical loop.

Direct recordings of neuronal activity from the human BG are available in patients undergoing deep brain stimulation (DBS) for movement disorders. Up to now, invasive recordings exploring the generation of the ERN in humans have been limited to a single case in a patient with therapy resistant alcohol dependency and a single case in a patient with obsessive–compulsive disorder, where this ERP component was demonstrated in recordings from the nucleus accumbens (NAcc; Münte et al. 2008; Heinze et al. 2009). Interestingly, the error-related local field potential (LFP) activity in the NAcc precedes the cortical ERN by approximately 40 ms. The aim of the present study was to investigate the role of the human sensorimotor area of the globus pallidus internus (GPi)—the main BG motor output station—in error detection and modulation of the cortical ERN during the performance of a flanker task (Kopp et al. 1996). We selected this task to compare our results with those from previous studies using the flanker task, which suggested an influence of the BG on the cortical ERN (Ullsperger and von Cramon 2006; Münte et al. 2008; Seifert et al. 2011). A special emphasis was set on the assessment of the directionality in the dynamic interactions between the cortical and subcortical/pallidal activity during error-monitoring processes. To this end, we used the index of partial directed coherence (PDC), which was introduced for a frequency domain analysis of linear Granger causality to test for “direct” directed statistical temporal influence among recorded signals (Baccala and Sameshima 2001). Here, the concept of Granger causality (Granger 1969) refers to the determination of causal influences estimated as statistical temporal predictability among signals.

Direct recordings were performed in the GPi in 9 patients undergoing DBS for severe cervical or segmental dystonia. Importantly, dystonia did not involve the hands in any of the patients; thus all patients had clinically normal hand motor function.

## Materials and Methods

### Patients, Surgery, and Electrode Localization

Nine patients with idiopathic dystonia (49 [standard error of the mean, SEM 4] years old) participated in this study. All patients were right-handed. Six of the 9 patients suffered from cervical dystonia (CD) 1 patient had Meige's syndrome (oro-facial dystonia), and 2 suffered from segmental dystonia (SD), which did not affect the limbs. Clinical details are presented in Table 1. Patients gave informed consent prior to participation in the study, which was approved by the local Ethics Committee. DBS electrodes were targeted in all patients bilaterally in the posteroventral lateral “motor” portion of GPi. The macroelectrode used was model 3389 (patients #1–6 and 8: Berlin DBS Center) and model 3387 (Medtronic Neurological Division, MN, USA; patients #7, 9: Hanover DBS Center) with 4 platinum–iridium cylindrical surfaces (1.27 mm diameter and 1.5 mm length) and a contact-to-contact separation of 0.5 mm. Contact 0 was the most ventral and contact 3 was the most dorsal contact. Target coordinates were based on the direct visualization of the GPi in the individual stereotactic T2-weighted magnetic resonance imaging (MRI) or based on the intercommissural line as determined by stereotactic computed tomography (CT). The intended coordinates at the tip of electrode 0 were 17.4–22 mm lateral from the midline, 2–4 mm in front of the midcommissural point, and 2–4 mm below the midcommissural point as determined by MRI adjusted to the individual patient's anatomy. Correct placement of the DBS electrode was confirmed by intraoperative microelectrode recordings and location of typical side effects evoked by direct macrostimulation in all patients. Moreover, the localization of DBS electrodes was verified by postoperative MRI (cases #1–6, 8) or stereotactic CT (cases #7, 9) using automated normalization and contact localization in standard Montreal Neurological Institute (MNI) stereotactic space coordinates (accuracy of contact localization is 1–2 mm; see details in Schönecker et al. 2009) in all patients (Fig. 1). The evaluation of the clinical efficacy of chronic DBS on motor symptoms (31% mean improvement on Toronto Western Spasmodic Torticollis Rating Scale [TWSTRS, cases #1–4, 6, 9] or Burke–Fahn–Marsden Dystonia Rating Scale [BFMDRS, cases #5, 8] at least 3 months after DBS surgery) further supported correct placement of DBS electrodes within the motor area of GPi.

Table 1

Patients' demographics

Patient no. (sex) Diagnosis Age Onset age Preoperative medication Preoperative clinical scores (TWSTRS severity subscore) Postoperative clinical scores (TWSTRS severity subscore) Contact pairs selected for LFP analysis (PDC analysis) Stimulation settings
1 (M) Cervical dystonia 33 31 None 23 14 R: 01, 12, 23
L: 01, 12
(R: 12
L: 01)
R: 1−, 2−; 90 μs, 130 Hz, 2.3 V
L: 1−, 2−; 90 μs, 130 Hz, 1.5 V
2 (M) Cervical dystonia (Torticollis) 45 44 One year preop trihexyphenidyl 2 mg 24 21 R: 12
L: 12, 23
(R: 12
L: 12)
R: 0−; 90 μs,130 Hz, 2.3 V
L: 1−; 90 μs, 130 Hz, 2 V
3 (M) Cervical dystonia 44 42 None 14 12 R: 12, 23
L: 12, 23
(R: 23
L: 12)
R: 0−; 60 μs, 60 Hz, 3.8 V
L: 1−; 60 μs, 60 Hz, 5 V
4 (M) Cervical dystonia 33 19 None 17 14 R: 01, 12, 23
L: 12, 23
(R: 12
L: 23)
R: 1−; 60 μs, 160 Hz, 3.3 V
L: 2−; 60 μs, 160 Hz, 2.5 V
5 (M) Segmental dystonia, primarily cervical (ante-lateroflexion of the upper body) 57 12 Trimipramin 100 mg, Tetrazepam 50 mg n.a. (BFMDRS movement score) 14/120 R: 01, 23
L: 01, 23
(R: 01
L: 01)
R: 0−; 60 μs, 150 Hz, 3.5 V
L: 1−, 3−; 60 μs, 150 Hz, 4 V
6 (F) Cervical dystonia 51  23 R: 23
L: 01, 12, 23
(R: 23
L: 01)
R: 0−; 90 μs, 140 Hz, 3 V
L: 1−; 90 μs, 140 Hz, 3.5 V
7 (M) Blepharospasm-oromandibular dystonia 76 Amitryptiline 25 mg n.a. n.a. R: 01, 12

(R: 12
−)
n.a.
8 (F) Segmental dystonia (Torticollis with orolinguar dystonia) 48 15 Four months before: Tetrabenazin 25 mg, Pimozid 1 mg Biperiden (dosage N/At) 25 (BFMDRS movement score 12/120) 15 (BFMDRS movement score 5.5/120) R: 01, 12, 23
L: 01, 12, 23
(R: 01
L: 01)
R: 1−, 2−; 90 μs, 130 Hz, 3.7 V
L: 2−, 3−; 90 μs, 130 Hz, 3.7 V
9 (M) Cervical dystonia 58 37 Clonazepam
2 × 1 mg
27 20 R: 01, 12
L: 01
(R:12
L:01)
R: 1−, 2+; 180 μs, 130 Hz, 3.5 V
L: 1−, 2+; 180 μs, 130 Hz, 3.5 V
Patient no. (sex) Diagnosis Age Onset age Preoperative medication Preoperative clinical scores (TWSTRS severity subscore) Postoperative clinical scores (TWSTRS severity subscore) Contact pairs selected for LFP analysis (PDC analysis) Stimulation settings
1 (M) Cervical dystonia 33 31 None 23 14 R: 01, 12, 23
L: 01, 12
(R: 12
L: 01)
R: 1−, 2−; 90 μs, 130 Hz, 2.3 V
L: 1−, 2−; 90 μs, 130 Hz, 1.5 V
2 (M) Cervical dystonia (Torticollis) 45 44 One year preop trihexyphenidyl 2 mg 24 21 R: 12
L: 12, 23
(R: 12
L: 12)
R: 0−; 90 μs,130 Hz, 2.3 V
L: 1−; 90 μs, 130 Hz, 2 V
3 (M) Cervical dystonia 44 42 None 14 12 R: 12, 23
L: 12, 23
(R: 23
L: 12)
R: 0−; 60 μs, 60 Hz, 3.8 V
L: 1−; 60 μs, 60 Hz, 5 V
4 (M) Cervical dystonia 33 19 None 17 14 R: 01, 12, 23
L: 12, 23
(R: 12
L: 23)
R: 1−; 60 μs, 160 Hz, 3.3 V
L: 2−; 60 μs, 160 Hz, 2.5 V
5 (M) Segmental dystonia, primarily cervical (ante-lateroflexion of the upper body) 57 12 Trimipramin 100 mg, Tetrazepam 50 mg n.a. (BFMDRS movement score) 14/120 R: 01, 23
L: 01, 23
(R: 01
L: 01)
R: 0−; 60 μs, 150 Hz, 3.5 V
L: 1−, 3−; 60 μs, 150 Hz, 4 V
6 (F) Cervical dystonia 51  23 R: 23
L: 01, 12, 23
(R: 23
L: 01)
R: 0−; 90 μs, 140 Hz, 3 V
L: 1−; 90 μs, 140 Hz, 3.5 V
7 (M) Blepharospasm-oromandibular dystonia 76 Amitryptiline 25 mg n.a. n.a. R: 01, 12

(R: 12
−)
n.a.
8 (F) Segmental dystonia (Torticollis with orolinguar dystonia) 48 15 Four months before: Tetrabenazin 25 mg, Pimozid 1 mg Biperiden (dosage N/At) 25 (BFMDRS movement score 12/120) 15 (BFMDRS movement score 5.5/120) R: 01, 12, 23
L: 01, 12, 23
(R: 01
L: 01)
R: 1−, 2−; 90 μs, 130 Hz, 3.7 V
L: 2−, 3−; 90 μs, 130 Hz, 3.7 V
9 (M) Cervical dystonia 58 37 Clonazepam
2 × 1 mg
27 20 R: 01, 12
L: 01
(R:12
L:01)
R: 1−, 2+; 180 μs, 130 Hz, 3.5 V
L: 1−, 2+; 180 μs, 130 Hz, 3.5 V

Note: Individual patient's data including presurgical medication. Motor scores provided are Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS I severity scale, maximum 35). In the case of SD, additional Burke–Fahn–Marsden Dystonia Rating Scale (BFMDRS, movement scale; maximum 120) is provided. Postoperative scores were assessed at least 3 months after electrode implantation (8 months on average). Average improvement in TWSTRS scores of 31%.

Figure 1.

Normalized localization in the standard MNI stereotactic space coordinates of selected contact pairs (depicting center of contact pair). (A) Localizations of the center (+) of the bipolar pair of 31 contacts (7 Berlin patients, #1–6 and 8; for the Hannover patients, #7 and 9, stereotactic CT and no space coordinates were available) selected for LFP analysis are shown on corresponding horizontal slices (Z) of the MNI standard brain template (dimensions in mm: x: medio-lateral, y: antero-posterior, z: dorso-ventral direction). (B) Localizations depicted on corresponding coronal slices (Y). Put, putamen; GPe, globus pallidum ext.; GPi, globus pallidum int. Note that, in all selected contact pairs, at least one single contact was located within the GPi. Center locations at the GPi–GPe interface originate in contact pairs with one contact inside GPi and another outside of it. In relation to the center point of the anterior commissure, the horizontal level z = −3 mm is located approximately 1.3 mm more dorsal and the level z= −4.75 mm is located 0.45 mm more ventral. Mean nuclear boundaries of the globus pallidus are outlined (black) based on the Harvard-Oxford subcortical probabilistic structural atlas (N = 20 patients).

Figure 1.

Normalized localization in the standard MNI stereotactic space coordinates of selected contact pairs (depicting center of contact pair). (A) Localizations of the center (+) of the bipolar pair of 31 contacts (7 Berlin patients, #1–6 and 8; for the Hannover patients, #7 and 9, stereotactic CT and no space coordinates were available) selected for LFP analysis are shown on corresponding horizontal slices (Z) of the MNI standard brain template (dimensions in mm: x: medio-lateral, y: antero-posterior, z: dorso-ventral direction). (B) Localizations depicted on corresponding coronal slices (Y). Put, putamen; GPe, globus pallidum ext.; GPi, globus pallidum int. Note that, in all selected contact pairs, at least one single contact was located within the GPi. Center locations at the GPi–GPe interface originate in contact pairs with one contact inside GPi and another outside of it. In relation to the center point of the anterior commissure, the horizontal level z = −3 mm is located approximately 1.3 mm more dorsal and the level z= −4.75 mm is located 0.45 mm more ventral. Mean nuclear boundaries of the globus pallidus are outlined (black) based on the Harvard-Oxford subcortical probabilistic structural atlas (N = 20 patients).

We additionally recruited 10 healthy participants matched in age (52 [6] years) to the patient group in order to have reference values in terms of performance and ERN data with the parameter settings of the flanker task in the current study.

### Paradigm

The paradigm was a modification of the speeded flanker task (Kopp et al. 1996). Flanker stimuli consisted of vertical arrays of 4 horizontal arrowheads, whose onset was followed after 70 ms by the onset of a central arrowhead, the target stimulus (Fig. 2). The target arrows pointed either in the same direction as the flankers (congruent trials) or in the opposite direction (incongruent trials; 50% rate each). Flankers and targets vanished 20 ms after target onset. Patients were instructed to respond with maximal accuracy and speed to the target arrow by pressing down a button with the thumb of the corresponding hand (left/right). Importantly, patients were holding in each hand one device with a button, and therefore, the button press in the correct direction required only a small movement of the thumb. No instructions regarding the correction of erroneous responses were provided. Following their response, a fixation cross was presented at the center of the screen during an interstimuli interval randomized between 2500 and 3500 ms. Participants had to fixate the central cross to avoid eye movements.

Figure 2.

Time course of experimental paradigm. Flanker distractors presented around the fixation point were followed after 70 ms by a central target arrowhead. All stimuli vanished 20 ms after target onset. Adaptive behavioral feedback was given every 20 trials with regard to accuracy or speed. Finally, a variable fixation time between 2500 and 3500 ms preceded the next trial.

Figure 2.

Time course of experimental paradigm. Flanker distractors presented around the fixation point were followed after 70 ms by a central target arrowhead. All stimuli vanished 20 ms after target onset. Adaptive behavioral feedback was given every 20 trials with regard to accuracy or speed. Finally, a variable fixation time between 2500 and 3500 ms preceded the next trial.

Five blocks of 64 trials each were performed. Within each block, trials were selected at random and without replacement to yield 16 trials from each of 4 categories (the combination of left/right targets with congruent/incongruent flankers). Adaptive behavioral feedback was given every 20 trials with regard to accuracy or speed. The types of trials available for analysis were congruent correct, congruent erroneous, incongruent correct, and incongruent erroneous.

### Recording

Recordings were made 2–4 days postoperatively, while electrodes were externalized and before implantation of the pulse generator. LFP activity was recorded from GPi bipolarly from all 4 adjacent contact pairs along the macroelectrode in all patients. The 4 contacts on each macroelectrode in left and right GPi (contacts 0, 1, 2, and 3) corresponded to bipolar recordings from contact pairs 01, 12, and 23. Signals were bandpass ﬁltered between 0.5 and 250 Hz, ampliﬁed 50 000-fold using a Digitimer D360 (Digitimer Ltd., Welwyn Garden City, Hertfordshire, UK), and sampled at 1 kHz through a 1401 A-D converter (Cambridge Electronic Design, Cambridge, UK) onto a computer using the Spike 2 software (Cambridge Electronic Design). Simultaneously, continuous scalp electroencephalogram (EEG) was recorded from frontocentral, central and parietal electrodes at the midline (FCz, Cz and Pz, International 10-20 System) and referenced to linked earlobes. The selection of these EEG electrode positions was based on the involvement of the pMFC in the generation of the cortical ERN and the posterior so-called error positivity (Pe), which often follows the ERN 200–450 ms after incorrect responses (Pe; van Veen and Carter 2002; Ullsperger et al. 2007). In addition, transverse electrooculogram (EOG) was recorded to monitor for vertical and horizontal eye movements. EEG signals were also bandpass filtered at 0.5–250 Hz, amplified 25 000-fold, and sampled at 1 kHz.

### Analysis

#### Behavioral Data

For each subject, reaction times (RT) exceeding 2 times the standard deviation around their mean value for each response-type distribution (congruent and incongruent: Error and correct) were discarded from the analysis, leaving an average of 300 trials per subject. Data in Table 2 are provided as sample mean and SEM.

Table 2

Performance data in the 9 patients and 9 healthy participants are given as mean and SEM

Patient group

Control group

RT in ms Number trials Events percentage RT in ms Number trials Events percentage
Incongruent error 350 (20) 30 (20) 20 (10) 330 (30) 8 (3) 7 (4)
Incongruent correct 450 (30) 100 (20) 80 (10) 460 (20) 120 (2) 93 (3)
Congruent error 380 (60) 10 (6) 6 (4) 430 (20) 1.6 (0.7) 1.5 (0.7)
Congruent correct 390 (20) 140 (10) 94 (4) 390 (20) 127 (3) 98.5 (0.7)
Patient group

Control group

RT in ms Number trials Events percentage RT in ms Number trials Events percentage
Incongruent error 350 (20) 30 (20) 20 (10) 330 (30) 8 (3) 7 (4)
Incongruent correct 450 (30) 100 (20) 80 (10) 460 (20) 120 (2) 93 (3)
Congruent error 380 (60) 10 (6) 6 (4) 430 (20) 1.6 (0.7) 1.5 (0.7)
Congruent correct 390 (20) 140 (10) 94 (4) 390 (20) 127 (3) 98.5 (0.7)

#### Preprocessing of EEG and LFP Data

We used the EEGLAB Matlab Toolbox (Delorme and Makeig 2004) for visualization and ﬁltering purposes. Stimulus-locked and response-locked epochs were selected in a window spanning [−1000, 3000] ms around target onset and [−1000, 1000] ms around response onset, respectively. The notation [a, b] will be used to indicate a temporal window spanning from the first (a) to the second (b) time point with respect to either stimulus or response.

As assessed by visual inspection of the EEG, LFP, and EOG signals, data epochs containing artifacts due to eye movements, eye blinks, and/or muscular activity were removed from further analysis. The mean number (SEM) of remaining trials per subject averaged across EEG and LFP recordings for the different response types was: 140 ([10] congruent correct), 90 ([20] incongruent correct), 30 ([20] incongruent error), and 10 ([6] congruent error).

An important issue in the analysis of the electrical potentials from bipolar recordings is that the term “polarity” cannot be defined due to the measurement of the “differential” electrical potential from 2 local sources. Notwithstanding this issue, in analogy to the description of cortical ERP as the modulation of positive and negative amplitudes, we use the terms “negative” and “positive” polarity by convention to describe LFP–ERP waveforms. Importantly, however, this convention aims exclusively at a simpler description of the LFP–ERP data, and no true assignment of polarity is implied. Similarly, the polarity of the LFP–ERP modulations depends on the precise positioning of the bipolar contact pairs with respect to the generator of the activity. To normalize the polarity of the LFP–ERPs in all bipolar contacts in all patients, we proceeded as follows: (1) First, we averaged all bipolar “stimulus-locked” LFP–ERP waveforms across the left and right GPi and across patients for incongruent correct trials. Correct event types had the largest number of trials; therefore, we obtained a smooth stimulus-locked grand-average ERP waveform (Supplementary Fig. 1A) that (2) was used as a template against which we could compare a reference single-contact pair stimulus-locked LFP–ERP per GPi side (typically R01 and L01) in each patient. (3) Finally, we assessed whether there was a phase reversal between the stimulus-locked waveforms recorded from neighbor contact pairs in each GPi side (see e.g. in Supplementary Fig. 1B). To this end, we low-pass filtered at 20 Hz the waveform epochs (in analogy to ERP analysis, see next sections) and applied the Hilbert transform to obtain the phase values $$\varphi _{ik} (t,f)$$ for each bipolar recording i, at time point t, and trial number k. The index of pairwise phase coupling at time t between each pair of neighbor channels i and j can be assessed with the formula (e.g. Pikovsky et al. 2001): $$\bar R_{ij} = \left|{\displaystyle{1 \over n}\sum\nolimits_{k = 1}^n {\exp [i(\varphi_{\,jk} - \varphi_{ik} )]} } \right|$$ where n is the number of trials. According to that expression, which refers to the norm of the vectorial sum on the unit circle within the 1-dimensional complex space C1, the $$\bar R_{ij}$$ index is “strictly positive” and approaches 1 (0) for the strict (no) phase relationship between the considered electrode pair across the epochs. Here, however, we wanted to detect phase reversals, which would arise from phase differences within the range [π/2, 3π/2] and should lead to a maximum negative value $$\bar R_{ij} = - 1$$ at stable phase differences of π for each trial k (Supplementary Fig. 1C). Consequently, we assigned negative values to the vectorial sum for $$\bar R_{ij}$$ when the real part of the resultant vector $$\displaystyle{1 \over n}\sum\nolimits_{k = 1}^n {\exp [i(\varphi _{\,jk} - \varphi _{ik} )]}$$ was negative (negative cosine of the angle of resultant vector, angle within [π/2, 3π/2]). Finally, the index of pairwise phase coupling was averaged across time points within the interval [0, 600] ms poststimulus, reflecting the largest amplitude modulations (Supplementary Fig. 1A). The significance level for value $$c = \bar R_{ij}$$ was computed as $$e^{ - nc^2 }$$, according to the asymptotic formula given by Strutt (1905; see similar approach in Rodriguez-Oroz et al. 2011).

Following this procedure, 30% of the bipolar LFPs used for the analysis (R01 in patient #1, R12, L12, L23 in patient #2, R01, R23, L01, L23 in #5, and L01, L12 in #9; none of the contacts from these bipolar pairs was in the GPe) were flipped. Note that the stimulus-locked correct LFP epochs showed a prominent negative deflection in the grand-average waveform (e.g. see Fig. 3A), and this convention was kept throughout the paper. Importantly, the “response-locked” LFP–ERPs were not assessed during the flipping of the waveforms. Consequently, the polarity reversal procedure did not aim at maximizing response-related differences. Note that the polarity convention does not affect the indexes of spectral power or PDC (see next sections).

Figure 3.

Time course of the GPi activity during movement preparation. (A) Stimulus-locked ERP waveforms in the contralateral GPi relative to the button press. Waveforms to correct and erroneous responses are denoted by gray and black lines, respectively. (B) Same in the ipsilateral GPi. F and T denote onset of the flanker and target stimuli, respectively. Rcorr and Rerr denote the button press in correct and erroneous responses. Note that the time of response for correct and error trials differed by approximately 100 ms. Moreover, the SEM of the RT was 30 ms in correct trials and 20 ms in error trials.

Figure 3.

Time course of the GPi activity during movement preparation. (A) Stimulus-locked ERP waveforms in the contralateral GPi relative to the button press. Waveforms to correct and erroneous responses are denoted by gray and black lines, respectively. (B) Same in the ipsilateral GPi. F and T denote onset of the flanker and target stimuli, respectively. Rcorr and Rerr denote the button press in correct and erroneous responses. Note that the time of response for correct and error trials differed by approximately 100 ms. Moreover, the SEM of the RT was 30 ms in correct trials and 20 ms in error trials.

All bipolar contact pairs per electrode which had at least one contact within the GPi based on the postoperative electrode localization (see patients and surgery) were used for further LFP analysis (N = 36; 17 GPi, 9 patients). Left GPi in patient #7 was not included due to artifacts.

In this study, the terms “contralateral” and “ipsilateral” are used for GPi in relation to the overt button press (i.e. hand moved). For instance, during responses with the right hand—irrespective of whether the response is correct or erroneous—the contralateral GPi refers to the left GPi.

#### Event-Related Potentials

Stimulus- and response-locked epochs were averaged separately for erroneous and correct trials, and in congruent and incongruent trials separately, following previous subtraction of baseline activity (300–100 ms prior to the flanker onset). The analysis focused primarily on the incongruent epochs, because of an insufficient number of error trials in the congruent condition. Unless otherwise stated, these trials will be hereafter termed erroneous (E) and correct (C). Exclusively for the standard ERP analysis, data epochs were additionally low-pass filtered using a finite impulse response at a cutoff frequency of 20 Hz. GPi activity during movement onset was observed in the response-locked representation, whereas GPi activity during motor preparation was evident in the stimulus-locked representation.

#### Directionality Analysis Between GPi and pMFC Activity

We were particularly interested in analyzing the frequency-specific directionality of the interactions between pallidal and cortical activity during an early detection of errors in incongruent trials. To this end, the PDC was used (Baccala and Sameshima 2001), because it is a sensitive and reliable Granger causality-based method to assess direct statistical temporal influence between recorded signals in the frequency domain. Specifically, the main focus of this analysis was to investigate whether the “subcortical-to-cortical” direction of predictability was enhanced in incongruent erroneous trials when compared with correct trials. This outcome would imply that pallidal LFP activity influences cortical EEG activity more prominently in error trials. In addition, despite an expected pattern of bidirectional coupling between GPi and pMFC, we tested whether the cortical → GPi and GPi → cortical directions of influence were differentiated.

In the following, the procedure to compute the PDC in our study is outlined (for more details see Supplementary Material). The PDC is computed from the coefficients of the multivariate autoregressive (MVAR) modeling transformed to the frequency domain. The order of the MVAR was estimated with the Bayesian information criterion (Schwarz 1978). In each patient, the MVAR model was fitted on multiple trials to increase the reliability of the model parameters (Ding et al. 2000)—after appropriate normalization in order to achieve local stationarity in the single trials (Ding et al. 2000)—and solved in the 4- to 100-Hz frequency range for a set of 5 time series: FCz, Cz, Pz, one contact pair in the contralateral, and one in the ipsilateral GPi. In each patient, the single contact pair per GPi side was selected based on a correct localization and a prominent response-locked LFP–ERP modulation in error and correct trials (Table 2). The signals were downsampled to 250 Hz to improve frequency resolution, while not interfering with the time scale of interactions of interest (Schlögl and Supp 2006; Florin et al. 2010). At a sampling rate of 250 Hz, the time interval between consecutive points in the time series is 4 ms, which was acceptable since we did not expect an interaction between cortical and pallidal oscillatory activity at a shorter time scale due to the inherent delay in transmission times by the neural pathways (∼8 ms from M1 to GPi through the faster cortico-subthalamo-pallidal pathway in monkeys, see Nambu et al. 2000). The data for each channel consisted of the concatenation of all response-locked epochs corresponding to each response type (erroneous and correct) within [−300, 100] ms with respect to the button press. This time interval was selected based on the temporal modulation of early pallidal error-related potentials and cortical ERN (see Results). The PDC index was calculated for erroneous and correct trials separately from the fitted MVAR coefficients. MVAR and PDC computations were performed with the Biosig toolbox (http://biosig.sf.net/) and additional custom-made scripts for MATLAB®. Finally, for statistical analyses, a first single-subject assessment of the significant PDC values in each condition was performed. This analysis aimed at revealing the significant PDC values at the single-subject level, which were later used in the statistical group analysis. The statistical significance of single-subject PDC values was assessed by computing the 95% confidence level based on the estimated standard error of the Jackknife distribution (e.g. Efron 1992; Schlögl and Supp 2006). Next, at the group level, statistical differences in the PDC index across subjects were estimated by means of nonparametric permutation tests (see Statistics).

#### Time-Frequency Analysis of LFP Signals

We investigated changes in response-locked spectral power in the GPi focusing on the theta (4–8 Hz) and gamma (30–100 Hz) frequency ranges motivated by: (1) The general increase in theta band power observed over the pMFC during and following errors, a phenomenon which is at the basis of the generation of the cortical ERN (Luu et al. 2004; Trujillo and Allen 2007; Marco-Pallarés et al. 2008; Cavanagh et al. 2009). (2) Recent data pointing to a facilitation of movement gain via GPi gamma band oscillations (Brücke et al. 2012). To this end, we used the wavelet energy, which was computed as the average across epochs of the squared norm of the response-locked complex wavelet transform in [−1000, 1000] ms around the button press. After subtracting the prestimulus baseline level (from 300 to 100 ms prior to the flanker onset), we normalized the wavelet energy with the standard deviation of that prestimulus baseline period and expressed it as a percentage of power change (see details in Trujillo et al. 2005). The normalization procedure reduced the effects of intersubject and interelectrode variability (see Supplementary Material for more details).

The Morlet wavelet transform was computed as follows: Raw stimulus-locked LFP epochs were extracted in the interval [−1000, 3000] ms and then convolved with the complex Morlet wavelet to obtain time-frequency (TF) phases $${\varphi _{ik} (t,f)}$$, at an electrode i and epoch k, and amplitudes $$A_{ik} (t,f) = |{Wx_{ik} (t,f)} |$$ of the wavelet-transformed LFP signal x(t). The constant η characterizes the family of wavelet functions in use and deﬁnes the constant relation between the center frequency and the bandwidth $$\eta = f/\sigma _f$$. We selected a value η = 7, which provides a good compromise between high-frequency resolution at low frequencies and high temporal resolution at high frequencies (Mallat 1999). The frequency domain was sampled from 4 to 100 Hz with a 1-Hz interval between each frequency.

#### Statistics

The assessment of statistical differences between conditions was performed with the use of a nonparametric pairwise permutation test (Good 2005) across subjects, with the difference in sample means as test statistic. In this procedure, we first created the joint distribution of samples from both conditions (e.g. for N = 9 subjects per condition, the joint distribution has 2N = 18 samples) and then rearranged randomly the sample indexes prior to constructing replications of the 2 samples of size N with the first and second half of the rearranged joint distribution. We performed n = 5000 rearrangements, drawn at random from the complete permutation distribution (Monte Carlo Permutation Test; see Supplementary Material). The P-values were then obtained as the frequencies that the replications of the test statistic had a larger absolute value than the experimental difference.

We primarily investigated differences between erroneous and correct incongruent trials to evaluate the time course of pallidal activity during the processing and resolution of the conflict between coactivated responses. To this aim, 2-factorial tests with factors Laterality (2 levels: Ipsilateral GPi and contralateral GPi) and Response type (2 levels: Incongruent error [E] and incongruent correct [C]) were computed by means of 5000 synchronized rearrangements (Good 2005; Basso et al. 2007). Furthermore, statistical differences between ipsilateral and contralateral GPis were estimated by means of paired permutation tests.

In addition, despite the limitations of the small number of congruent erroneous trials, we analyzed differences between the ERPs in erroneous and correct congruent trials to detect modulations induced by errors when no explicit conflict is present. Moreover, to additionally disentangle the roles of conflict and correctness in the modulations of pallidal ERP activity, we performed a 2-factorial analysis with factors Conflict (congruent/incongruent trials) and Response type (error/correct) on the LFP–ERPs. This test was complemented by unifactorial tests performed via the permutation test separately on correct and erroneous trials by comparing congruent and incongruent stimuli conditions.

In the case of grand-average LFP–ERP analysis, the permutation test was evaluated across subjects, after averaging data from the left and right GPi. Similarly, LFP–ERP and TF analyses in contralateral or ipsilateral GPis were performed across subjects. Specifically, 9 patients were equivalent to 9 contralateral GPis after averaging epochs coming from both the left (when contralateral to the response: i.e. right-hand response) and right GPis (when contralateral: i.e. left-hand response). In patient #7 (Table 1), only epochs from the “right” contralateral GPi were used. Similarly, 9 patients were equivalent to 9 ipsilateral GPis. The statistical analyses of grand-average ERP waveforms focused on the temporal windows [−300, 400] ms for response-locked waveforms and [−200, 600] ms for stimulus-locked potentials. Statistical differences in response-locked spectral power were analyzed between −400 and 500 ms around the button press (broader time window due to the spread of wavelet functions) and in the 4- to 100-Hz frequency range. Within these time and frequency ranges of interest, statistical tests were assessed at each point. Differences were considered significant if P < 0.05. Multiple comparisons arising from TF analysis or temporal spread in ERP waveforms were corrected by controlling the false discovery rate (FDR) at level q = 0.05 by means of an adaptive 2-stage linear step-up procedure (Benjamini et al. 2006). The corrected threshold P-value obtained from this procedure, pth, was used to reject all null hypotheses fulfilling the condition: P-value < pth. Throughout the paper, pth is given when multiple comparisons are performed.

## Results

Behavioural data for patients and healthy age-matched controls are provided in Table 2 as mean and SEM. Patients made about 20% (10) errors in the incongruent condition, and 10% (6) errors in congruent trials. In accordance with the extent of response conflict, the RT of correct responses in incongruent trials was significantly longer than in congruent trials (P = 0.0130). In addition, incongruent erroneous responses were faster than correct responses (P = 0.0234). Patients made few corrections following incongruent errors (2 [3] trials) or congruent errors (1 [1] trials), except patient #7, who corrected 28 congruent errors and 29 incongruent errors (∼40%).

Reaction times and percentage of error corrections were similar between the patient group and healthy age-matched controls (no significant differences, see details in Supplementary Results), suggesting that the low error-correction rate in patients was not disease related. Note that the small percentage of corrections in the patient and control groups might be due to the speed–accuracy feedback being provided every 20 trials: No explicit trial-by-trial speed–accuracy pressure was exerted on the participants, which might have consequently led to the smaller coactivation of conflicting responses, such that only one response reached activation threshold.

Importantly, healthy participants did commit significantly fewer errors when compared with the patients (congruent errors: P = 0.0350, incongruent trials: P = 0.0164, permutation test).

Finally, there was a significant posterror slowing in incongruent trials, reflected in the larger RT in correct trials following errors when compared with the RT in correct trials following correct responses (posterror slowing = 20 ms; P = 0.02). This finding resembles the RT slowing subsequent to errors in our control group (30 ms, n.s. between-group differences) and in previous studies (Fiehler et al. 2005; Münte et al. 2008).

### Error-Related Cortical and Pallidal ERP Activity

For the sake of clarity, we report first the main findings from our analysis focusing on incongruent trials, termed erroneous (E) and correct (C) responses. At the end of this section, however, we outline the outcomes of the comparison between congruent and incongruent trials, which are expanded in Supplementary Material.

#### Stimulus-Locked Representation

During motor preparation following target onset, the stimulus-locked pallidal ERPs (Fig. 3) elicited an amplitude modulation of negative polarity (see polarity notation by convention in Materials and Methods) around 100–400 ms, which was observed in both GPi sides and in both response types. A 2-factorial analysis with factors Laterality and Response type demonstrated significant effects for Response type in [150, 190] ms, due to a larger amplitude increase in correct trials, and for Laterality in [350, 450] ms, due to more enhanced amplitude change in the ipsilateral GPi (P < pth = 0.01, control of FDR). The Response-type effect resembles the cortical N200, a negative stimulus-locked fronto-central ERP component with larger amplitude in incongruent trials leading to correct responses (Kopp et al. 1996). In addition, we found a significant interaction effect (P < pth = 0.01) in [150, 190] ms (larger correct-error difference in the ipsilateral GPi, post hoc analysis, P < pth = 0.008) and in the broad interval [400, 600] ms (larger correct-error difference in the ipsilateral GPi against larger error-correct difference in the contralateral GPi, post hoc analysis, P < pth = 0.004). These results related to the lateralization of the ERP to the ipsilateral GPi, and more prominently in correct trials, might imply “inhibitory mechanisms” of the corresponding response side (flanker direction is suppressed in correct trials, whereas target direction is suppressed in error trials).

#### Response-Locked Representation

We found similar peak-to-peak cortical ERN amplitude in patients and control subjects (Supplementary Results). In patients, the grand-average response-locked cortical ERP for erroneous trials revealed a negative amplitude deflection with a peak latency of 60 ms at FCz and Cz, corresponding to the ERN (representative electrode position Cz is shown in Fig. 4A), which was not present in correct trials. A significant negative difference between error and correct trials was obtained at FCz and Cz in [40, 80] ms (P < pth = 0.01, after control of FDR, Fig. 4A). This effect was followed by a significant positive difference at Cz and Pz within 150–185 ms (P < pth = 0.01, Fig. 4A), due to the positive deflection in error trials resembling the earlier component of Pe (see van Veen and Carter 2002 for earlier Pe at Cz).

Figure 4.

Time course of the ERPs to correct (C) and erroneous (E) responses in the incongruent condition. (A) Cortical response-locked ERPs at the electrode position Cz for E (black), C (magenta) trials revealed the significant components ERN and Pe (gray area: P < 0.05, controlled for FDR) (B) Response-locked GPi ERP waveforms averaged across 9 subjects (17 GPi) demonstrated a prominent significant difference between erroneous and correct trials prior to and around response onset (P < 0.05, controlled for FDR). (C) Response-locked GPi ERP waveforms separately for the ipsilateral (left panel) and contralateral (right panel) GPi. Note that the terms contralateral and ipsilateral are used for GPi in relation to the overt button press (i.e. hand moved). (D) To rule out the possibility that differences in GPi LFP responses were due to differences in reaction times between error and correct trials, correct trials were classified into 3 RT intervals with an equal number of epochs, thus leading to the mean RT values in each interval of 353 ms (short RT), 430 ms (medium), and 548 ms (long RT). No differences in the ERP profile of correct trials with respect to RT modulation were observed. (E) Directionality analysis in incongruent trials. Grand-average significant PDC estimates after subtraction of the single-subject confidence threshold (P < 0.05, assessed with the estimated standard error of the Jackknife distribution) are presented for the directions GPi to posterior median frontal cortex (pMFC: FCz, Cz; left panel), ipsilateral GPi to pMFC (middle panel), and contralateral GPi to pMFC (right panel). Significant group-level differences between erroneous and correct trials are denoted by the gray areas and were computed at alpha level 0.05 and controlled for FDR.

Figure 4.

Time course of the ERPs to correct (C) and erroneous (E) responses in the incongruent condition. (A) Cortical response-locked ERPs at the electrode position Cz for E (black), C (magenta) trials revealed the significant components ERN and Pe (gray area: P < 0.05, controlled for FDR) (B) Response-locked GPi ERP waveforms averaged across 9 subjects (17 GPi) demonstrated a prominent significant difference between erroneous and correct trials prior to and around response onset (P < 0.05, controlled for FDR). (C) Response-locked GPi ERP waveforms separately for the ipsilateral (left panel) and contralateral (right panel) GPi. Note that the terms contralateral and ipsilateral are used for GPi in relation to the overt button press (i.e. hand moved). (D) To rule out the possibility that differences in GPi LFP responses were due to differences in reaction times between error and correct trials, correct trials were classified into 3 RT intervals with an equal number of epochs, thus leading to the mean RT values in each interval of 353 ms (short RT), 430 ms (medium), and 548 ms (long RT). No differences in the ERP profile of correct trials with respect to RT modulation were observed. (E) Directionality analysis in incongruent trials. Grand-average significant PDC estimates after subtraction of the single-subject confidence threshold (P < 0.05, assessed with the estimated standard error of the Jackknife distribution) are presented for the directions GPi to posterior median frontal cortex (pMFC: FCz, Cz; left panel), ipsilateral GPi to pMFC (middle panel), and contralateral GPi to pMFC (right panel). Significant group-level differences between erroneous and correct trials are denoted by the gray areas and were computed at alpha level 0.05 and controlled for FDR.

Similarly, the grand-average response-locked pallidal ERP in erroneous trials showed an error-related negative amplitude deflection at response onset that was clearly distinct from the ERP in correct trials (Fig. 4B). Raw data and single-subject ERPs in one representative patient are presented in Supplementary Figure 3. It is noteworthy that there were no significant differences between the number of trials with the right- and left-hand movements for any event type (P = 0.65 for errors and P = 0.41 for correct trials).

Permutation tests revealed a significant difference between pallidal error and correct waveforms in [−20, 140] ms (P < pth = 0.015). This effect was due to prominent differences between potentials of different polarity and large amplitude emerging in erroneous and correct trials. These potentials peaked at response onset (0 ms) in erroneous trials and at 70 ms in correct trials, leading to peak latency at 70 ms in the difference waveform erroneous minus correct (Supplementary Fig. 2). Importantly, the peak latency (at 0 ms) of the pallidal ERP preceded the cortical “error” waveform by approximately 60 ms. An additional significant difference was obtained earlier in [−260, −150] ms (P < pth = 0.015) and was due to a negative amplitude increase in correct trials, not present in erroneous trials.

A 2-factorial analysis demonstrated the significant main effects for factors Laterality and Response type in the aforementioned time intervals (P < pth = 0.01; synchronized rearrangements). These main effects reflected: (1) A robust difference between erroneous and correct trials, and (2) a different ERP modulation between the ipsilateral and contralateral GPis. A significant interaction was obtained in [−30, 100] ms (P < pth = 0.01). Additional post hoc analyses revealed that the negative ERP component around the button press in erroneous trials was significantly larger in the ipsilateral GPi (erroneous: Contra vs. ipsi GPi, ERP in [−20, 140] ms; P < pth = 0.011; Fig. 4C), whereas the positive component in correct trials had a significantly larger amplitude in the contralateral GPi (correct: Contra vs. ipsi GPi, ERP in [−20, 140] ms; P < pth = 0.020). Note also that, in the stimulus-locked representation (Fig. 3B), a second negative deflection emerges in error trials around the time of the erroneous response at approximately 350 ms. Importantly, in the post hoc analysis, we found no significant differences around −200 ms between ipsilateral and contralateral GPis, neither in correct, nor in error, trials.

Thus, we obtained a specific pallidal ERP modulation around the onset of the erroneous response, peaking at button press, and which was more prominent in the GPi ipsilateral to the wrong overt response. The nature of the effect around −200 ms remains unclear, but see below (congruent trials).

To test the possibility that the differences between ERPs in both conditions arose from shorter RTs in erroneous trials, we classified correct trials in each patient into 3 RT intervals based on an equal number of trials (mean 30 [SEM 10] in each RT bin). The resulting bins had mean RT values at 353 ms (short RT), 430 ms (medium RT), and 548 ms (long RT), with the short RT bin being more closely matched to reaction times in erroneous trials (Table 2). As observed in Figure 4D, the ERP profile in correct trials was invariant with respect to RT modulation (Fig. 4D). An additional control test was performed aiming to disentangle the influence of proximal and distal errors on the early error-related LFP negative deflection. For that purpose, each single-subject distribution of differences between indices of consecutive error trials was binned into 2 samples (i.e. close indices: Proximal errors and distant indices: Distal errors) separated by the median value. If the error-related LFP deflection is affected by the degree of expectancy in error occurrence, the waveforms to both proximal and distal errors should differ. The results demonstrated that ERP-LFP waveforms in erroneous trials were identical in both proximal and distal errors (details in Supplementary Material and Supplementary Fig. 4). Thus, the frequency of error occurrence in the recent performance history did not modulate the early negative erroneous LFP deflection.

Finally, we present the results of the analysis of congruent trials, which can provide further understanding of the early modulations of pallidal ERP activity when there is no explicit conflict as a potential confounding factor. In addition, the RT in erroneous and correct congruent trials was similar (P = 0. 811). We found significant differences between erroneous and correct congruent trials in time windows similar to those reflecting the effects in incongruent trials: [−200, −170] ms and [−10, 60] ms (P < pth = 0.01, permutation test); the earlier effect was due to larger negative amplitude modulation in correct trials, whereas the later effect reflected a prominent negative deflection at button press exclusively in erroneous trials (Supplementary Fig. 5). There were no laterality effects in congruent correct trials (n.s. differences; this analysis was not available for errors due to the small number of trials). A 2-factorial statistical assessment with factors Conflict and Response type of the stimulus-locked ERPs revealed a significant interaction within [340, 420] and [470, 510] ms (P < pth = 0.012), related to the significant stimulus-locked ERP differences in these range between “correct” trials with and without conflict (within 380–510 ms, post hoc analysis, P < pth = 0.0078, permutation test, Supplementary Fig. 6A), not present in error trials. The analysis of response-locked ERPs demonstrated a main effect for Response type at [−300, −60] ms and [−20,140] ms (P < pth = 0.0234) and an interaction effect at [−10, 140] ms (P < pth = 0.0234). The significant interaction effect reflected different modulations around the button press between trials with or without conflict “exclusively” in correct responses (post hoc permutation test, P < pth = 0.01; Supplementary Fig. 6B, left), and not present in erroneous responses (n.s. differences). The response-locked differences between correct congruent and incongruent trials did not originate in the difference response latency of these types of responses (Supplementary Fig. 6B, right). In summary, whether trials had a explicit response conflict or not led to significantly different ERP patterns exclusively in correct trials (but not in error trials), at the time around button press. No difference was found in the premovement period. Erroneous responses lead to markedly similar response-locked ERP profiles independently of the degree of conflict (Supplementary Fig. 7, n.s. differences). These outcomes support that our main finding of error-related negative deflection peaking at the button press might be primarily reflecting an early error-detection signal as a result of monitoring the unfolding of response inhibition patterns, even in the absence of explicit conflict.

### Directionality Analysis Between Error-Related GPi and pMFC Activity

Single-subject directionality analysis between response-locked EEG and LFP signals in incongruent trials was performed by means of the PDC (Baccala and Sameshima 2001). Note that this index assesses in a frequency-specific manner how activity from one recorded source helps in predicting the temporal evolution of activity in a different source, which is interpreted as a temporal influence. The PDC analysis focused here on the range 4–100 Hz and in [−300, 100] ms, as a time window of interest spanning the modulations in the GPi ERPs by erroneous/correct responses and the emergence of the cortical ERN. The single-patient MVAR order parameter ranged from 2 to 7. PDC values in significant frequency bins were pooled together among contralateral and ipsilateral GPis for directions (1) GPi to pMFC (FCz and Cz) and (2) for the opposite direction. We observed a significant “bidirectional” statistical influence between GPi and pMFC in all patients at 4–100 Hz in both error trials (P < 0.05, based on the standard error of the Jackknife distribution) and correct trials (P < 0.05). Group analysis of PDC values demonstrated no significant differences between the 2 directions of influence in either erroneous or correct trials (permutation test, P > 0.05). However, when considering differences between erroneous and correct trials, rather than between the 2 directions of influence, a significant difference emerged for the PDC at 40–65 Hz, due to larger temporal predictability in error trials, but exclusively when considering the influence of GPi on the pMFC (P < pth = 0.02; Fig. 4E, left panel). For the reverse direction of influence, from pMFC to GPi, there were no significant differences between erroneous and correct trials (P > 0.05). Accordingly, for the direction GPi to pMFC, we investigated further the PDC index by means of a 2-factorial analysis with factors Laterality and Response type. This analysis pointed to a significant interaction in the 4- to 65-Hz band (P < pth = 0.0236; synchronized rearrangements; no main effects). The interaction was the result of larger and significant difference in the PDC index between error and correct trials when considering the ipsilateral GPi as a source of influence (P < pth = 0.02; 10–65 Hz; post hoc permutation test, Fig. 4E, middle panel; no significant difference for contralateral GPi).

### Modulation of Spectral Power in GPi

We investigated whether the index of spectral power in different frequency ranges dissociated between error-related and movement-related processes. A general pattern of enhanced theta-band and decreased beta-band spectral power prior to and following response onset was observed (Fig. 5AD). In addition, there was a prominent increase in gamma spectral power at button press, more salient in correct trials and in the contralateral GPi (Fig. 5C). The 2-factorial analysis confirmed a significant main effect for factors Laterality in [−50, 80] ms and 50–100 Hz (gamma band, P < pth = 10−5) and Response type within [−185, −120] ms at 24–30 Hz (beta range) and in [−250, −200] at 40–60 Hz (gamma range; P < pth = 10−5). The Laterality effect is consistent with the previous findings of larger gamma oscillations over the contralateral GPi (Brücke et al. 2008, 2012). Additionally, significant interaction effects P < pth = 0.0026) were found: 1) in the theta (4–7 Hz) band prior to the response from −300 to −100 ms, and at 70–90 Hz in the gamma band (2) from −270 to −250 ms, and (3) between −50 and 50 ms around the response onset. No interaction effects were found in the alpha or beta frequency ranges.

Figure 5.

Grand-average response-locked normalized spectral power within 4–100 Hz in the GPi during the incongruent condition. (A) Time-frequency representation (TFR) of the spectral power in error trials in the contralateral GPi. (B) Same in the ipsilateral GPi. TFR in correct trials in the contralateral (C) and ipsilateral GPis (D). TFR of the error-correct difference in oscillatory activity in the ipsilateral (E) and contralateral (F) GPis. Significant error-correct differences in panels E and F, provided by the post hoc permutation test, are denoted by the black contour (P < pth = 0.005, controlled for FDR). The index of normalized spectral power indicates difference values between wavelet energy and prestimulus baseline wavelet energy relative to the standard deviation of the baseline. Note that, exclusively in the ipsilateral GPi, the [−200, 0] ms pre-error theta-band spectral power was significantly higher than prior to correct responses.

Figure 5.

Grand-average response-locked normalized spectral power within 4–100 Hz in the GPi during the incongruent condition. (A) Time-frequency representation (TFR) of the spectral power in error trials in the contralateral GPi. (B) Same in the ipsilateral GPi. TFR in correct trials in the contralateral (C) and ipsilateral GPis (D). TFR of the error-correct difference in oscillatory activity in the ipsilateral (E) and contralateral (F) GPis. Significant error-correct differences in panels E and F, provided by the post hoc permutation test, are denoted by the black contour (P < pth = 0.005, controlled for FDR). The index of normalized spectral power indicates difference values between wavelet energy and prestimulus baseline wavelet energy relative to the standard deviation of the baseline. Note that, exclusively in the ipsilateral GPi, the [−200, 0] ms pre-error theta-band spectral power was significantly higher than prior to correct responses.

Post hoc unifactorial analyses were performed to test for differences between error and correct trials in the contralateral and ipsilateral GPis separately (Fig. 5E,F). Erroneous responses were associated with significantly larger theta oscillations in the ipsilateral GPi from −200 to 50 ms (P < pth = 0.005, Fig. 5F), an effect that resembles the cortical error-related theta increase typically observed in the postresponse interval (Luu et al. 2004; Trujillo and Allen 2007; Marco-Pallarés et al. 2008; Cavanagh et al. 2009). A similar significant enhancement of theta-band spectral power in erroneous trials was found at the cortical level (Supplementary Fig. 8).

Motor output at the button press led to larger gamma amplitude oscillations in correct trials than in error trials at [−20, 20] ms, exclusively in the contralateral GPi (P < pth = 0.005, Fig. 5E). Finally, additional significant differences were obtained bilaterally in the GPi in the beta and gamma bands within [−125, 150] and [−215, −290] ms, respectively (P < pth = 0.005, Fig. 5E,F), due to a more prominent decrease of activity prior to correct responses.

## Discussion

This is the first account in humans of prominent and specific modulation of pallidal LFP activity prior to erroneous responses in a flanker task, preceding the cortical ERN by 60 ms. This finding was complemented by directionality analysis between cortical and pallidal activity with PDC, showing pallidal influence on cortical activity preferentially in error trials. Moreover, early theta-band spectral power increased in the ipsilateral GPi with erroneous responses, in line with the larger negative GPi ERP component that emerged ipsilateral to overt errors. These data resemble the general increase in theta power over the pMFC during and following errors, a phenomenon that is at the basis of the generation of the cortical ERN (Luu et al. 2004; Trujillo and Allen 2007; Marco-Pallarés et al. 2008; Cavanagh et al. 2009). The implication of our findings is that the human GPi is involved in the initial stages of error detection and may influence the generation of the cortical ERN.

The oscillatory pallidal error-related modulations were distinct from activity patterns during (1) movement preparation and (2) movement execution. (1) The preparatory movement-related activity was reflected in bilateral decreased beta and gamma spectral power effects, which were more salient in correct trials. Previous findings in several BG structures have supported a general role of bilateral reduced beta-band spectral power in preparation and during execution of movements (Alegre et al. 2004; Kühn et al. 2004; Brücke et al. 2008). Our data point to a facilitation of movement preparation in correct trials indexed by the larger beta spectral power decrease. (2) Activity at movement onset was reflected in the increased gamma spectral power lateralized to the contralateral GPi, which was larger in correct trials. Lateralized movement-related gamma spectral power is an important feature of physiological BG function (Alegre et al. 2004; Kempf et al. 2007; Brücke et al. 2008; Liu et al. 2008) and cortical activity (Crone et al. 1998; Cheyne et al. 2008; Huo et al. 2010; Muthukumaraswamy 2010). More recently, pallidal gamma spectral power has been proposed to index BG influence on motor gain (Brücke et al. 2012). Accordingly, our findings of reduced gamma spectral power at overt erroneous responses imply that GPi activity might have facilitated a movement of reduced motor gain, in line with reduced force observed during overt erroneous responses in previous studies (Gehring et al. 1993).

Note that the specific error-related findings (ERP, PDC, and theta oscillations) indicate that activity in the ipsilateral GPi reflects to a larger extent the differences between error and correct trials. This is an interesting aspect, which is better understood with the findings of the stimulus-locked ERP analysis: The larger amplitude modulation at [150,190] ms in correct trials relative to error trials (significant main effect Response type) resembles the cortical N200, which is larger in incongruent correct trials and has been linked to the inhibition of competing responses (Kopp et al. 1996; Coles et al. 2001). This early putative inhibition-related effect approximately 200 ms after stimuli presentation might correspond to the significant differences observed within [−260, −150] ms in the response-locked representation. Moreover, also in the stimulus-locked representation, the interaction effect at [150,190] ms demonstrated larger correct-error modulation in the ipsilateral GPi, indicating that, at an early stage during motor preparation, there might be larger inhibition in the ipsilateral GPi in correct trials. In error trials, this early inhibitory process is weaker and is lateralized to the side ipsilateral to the “erroneous” response. These outcomes were paralleled by the later significant main effect of Laterality in the stimulus-locked ERP between 350 and 450 ms, reflecting also larger activation in the ipsilateral GPi. In summary, stimulus-locked pallidal ERP analysis emphasized the lateralization of the activity to the ipsilateral GPi.

To further understand the nature of the early pallidal ERP modulations, we performed analysis of congruent trials. The outcomes showed that, even in the absence of potential confounding factors, such as conflict and reaction time, there is a larger negative ERP modulation 200 prior to the button press in correct, but not in error, trials. Therefore, we hypothesize that the pallidal activity modulation emerging 200 ms prior to the button press might be related to an enhanced inhibitory level necessary for the correct response to be selected. This interpretation is, however, limited by the current experimental paradigm (see below), in which a potential overlap of several processes, such as general inhibition (“break”) of all responses, specific lateralized inhibition, response selection, and reward prediction, could modulate pallidal population activity.

Following on from this, interestingly, around the time of the erroneous response onset, a second negative deflection was observed in error trials, which was more enhanced in the ipsilateral GPi. This error-related signal was similar in amplitude and latency both in congruent and incongruent trials, which suggests that it was not affected by explicit conflict between competing responses or by the timing of the response. We speculate that, in the response-locked representation, the error-related modulations at button press might reflect the outcome of a BG mechanism monitoring the preceding ongoing pallidal activity influencing response selection, but which might be independent of explicit response conflict. This monitoring mechanism might trigger an error-detection process in association with the incorrectly suppressed response side. An early initiation of a corrective movement (Rodríguez-Fornells et al. 2002) is unlikely to account for the larger ipsilateral error-related modulations since participants were not encouraged to correct wrong responses and, therefore, did not correct in the vast majority of the trials.

Studies using directed transfer function, PDC, or partial coherence (PC) as measures of directionality between cortex and BG activity have primarily focused on Parkinson's disease patients, and the outcomes have revealed a variable pattern of directionality between cortex and STN depending on the frequency range (Williams et al. 2002; Fogelson et al. 2006; Lalo et al. 2008; Litvak et al. 2011). It is noteworthy that the patterns of bidirectional influences reported in these studies were not necessarily accompanied by changes in LFP, MEG, or EEG spectral power in the same frequency ranges. Consistent with this, here the PDC index between GPi and cortex was larger in error trials in the gamma frequency range (40–65 Hz), whereas the broad error-related changes in spectral power were localized to the theta frequency range. Moreover, since during dystonic movements the GPi drives muscle EMG activity in the theta frequency range (Sharott et al. 2008), the net PDC findings in the gamma band cannot be accounted for by the abnormally enhanced low frequency BG activity in dystonia.

What could the role of the human sensorimotor GPi be during this early stage of error monitoring?

Here, 2 frameworks can be useful to understand the outcomes of our study. The computational RL model of Holroyd and Coles (2002) proposes that error signals generated in the BG are modulated by action-related rewards encoded by the mesencephalic dopamine system prior to their projection to the anterior cingulate cortex (ACC) in the pMFC. An ACC acts as a “control filter” and chooses an appropriate motor response. In this view, the ERN reflects mismatch between the expected goal and predicted upcoming action (Gehring et al. 1993; Falkenstein et al. 1995). The RL model posits that, following learning of stimulus–response associations and their corresponding reinforcements in the striatum, reward prediction error signals propagate back from reward delivery to preceding actions and cues via TD signals (Sutton 1988; Sutton and Barto 1998; McClure et al. 2003). As a consequence, predictive signals linked in the BG to actions can be computed in advance of the future reward.

An expansion of the RL framework based on recent neurophysiological data from the primate sensorimotor GPi and lateral habernula (LHb) neurons has been put forward by Hikosaka and colleagues (Hong and Hikosaka 2008; Hikosaka et al. 2008; Hong et al. 2011). Here, it has been proposed that the sensorimotor GPi has 2 functionally distinct outputs, one involved in motor execution and the other involved in reward evaluation. Overall, these studies highlight the plausibility of recording error-related or negative reward prediction signals in the sensorimotor GPi, at an early stage of movement selection. Recently, Cockburn and Frank (2011) have linked the RL model by Holroyd and Coles (2002) and the conflict monitoring model in the ACC (Botvinick et al. 2001), in an explicit attempt to account for the feedback ERN (peaking 200 ms after negative feedback; Cohen and Ranganath 2007). In their model, learning of the “explicit” reward contingencies by the ACC eventually leads to an independence from the BG gating mechanisms.

This aspect leads us to the second framework to interpret our data: The contributions of the BG to the optimization of behavior and therefore to performance monitoring can be conceptualized within the optimal feedback control framework (Todorov and Jordan 2002). Within this computational framework, some studies have emphasized a possible role of the BG in optimal motor control by estimating the expected costs (energy and accuracy cost) of the motor commands and the expected implicit rewards of the predicted sensory states (Shadmehr and Krakauer 2008). Empirical data on the control of reaching movements have supported this view and have stressed that an implicit motivational circuit might operate within the motor frontal-BG-thalamocortical loop (Mazzoni et al. 2007; Niv et al. 2007). It remains yet unclear how cost functions can be implemented for motor tasks characterized by “discrete” movements such as thumb presses in the flanker task.

An additional important consideration is that an implicit motivational circuit within the BG motor loop might drive automatic or spontaneous behavior (Mazzoni et al. 2007; Niv et al. 2007). Here, we used a flanker task that is characterized by an automatic “overlearned” stimulus–response association, and performance was motivated by implicit (i.e. without awareness) reward. Therefore, predictive signals of an upcoming error are likely to be more relevant in our task than reward prediction error signals following outcomes. The former may contribute to the ERN modulation, whereas the later play a major role in learning stimulus–response associations (Pasupathy and Miller 2005) and modulating the cortical feedback ERN (Cohen and Ranganath 2007). Moreover, the frequency of error occurrence in the recent performance history did not modulate the early negative error-related LFP deflection (proximal = distal errors), which supports the interpretation that the reported pallidal ERPs in erroneous trials reflect automatic predictive processes in the BG and are not affected by the degree of expectancy of error occurrence.

It must be clarified, however, that the time course of the pallidal–cortical interactions during error-detection reported here is probably specific to response errors arising in the context of automatic stimulus-response mappings. This may account for the differences with the existing literature on error processing, particularly with studies focusing on online error correction during arm reaching or trajectory movements. In that context, some studies highlight the role of the posterior parietal cortex and cerebellum (Desmurgert and Grafton 2000; Desmurget et al. 2001), whereas others imply a contribution of the BG (Houk et al. 2007; Tunik et al. 2009) or contradict it (Desmurget et al. 2004). It might be that the BG contributions to these processes are intertwined with the modulation of force grip (Grafton and Tunik 2011). Together, the alternative hypotheses for the involvement of the different cortical and subcortical structures in error monitoring demand further elucidation in future research. In this line, a limitation of our study is that, during the flanker task, several processes might overlap in the recordings of the LFP activity in the GPi, therefore leading to potential confounding factors in the interpretation of the activity 200 prior to the button press. These processes might include: General inhibition of all responses, specific lateralized inhibition, response selection, and reward prediction. Future work should complement our novel findings by investigating the influence of pallidal activity on the cortical ERN with new paradigms designed to exclude the inhibition of competing responses as a potential confounding factor. Another possibility would be to use more experimental blocks in the flanker task to increase the number of congruent erroneous epochs and to focus the analysis exclusively on congruent trials.

Up to now, empirical evidence in support for the BG contribution to the human cortical ERN comes predominantly from studies in patients with BG dysfunction or lesions (Falkenstein et al. 2001; Johannes et al. 2002; Frank et al. 2004; Beste et al. 2006, 2008; Ullsperger and von Cramon 2006; Willemssen et al. 2009; Seifert et al. 2011). In particular, focal lesions in the GPe, GPi, ventral anterior, and ventral lateral anterior thalamic nuclei, the two latter being termination sites of pallidal efferents, lead to a reduction or even absence of the cortical ERN (Ullsperger and von Cramon 2006; Seifert et al. 2011) and corresponding behavioral consequences (Seifert et al. 2011). Invasive recordings from the NAcc during performance of a flanker task have revealed error-related modulation of LFP activity, which precedes the midline ERN by 40 ms (Münte et al. 2008; Heinze et al. 2009). Here, it was suggested that the NAcc integrates contextual information from motor, limbic, and cognitive areas (Mogenson et al. 1980) by weighting the error signals that are sent from different subcortical and cortical areas to guide motivated behavior (Münte et al. 2008). Alpha–gamma band cross-frequency coupling in the NAcc might be a neural mechanism facilitating the gating of information from the limbic system to the BG for reward-guided learning (Cohen et al. 2009a, 2009b). Following NAcc DBS, functioning of the performance monitoring system can be restored (Kuhn et al. 2011). Our GPi findings converged with the NAcc data on error monitoring (Münte et al. 2008; Heinze et al. 2009) in that the subcortical error-related modulation preceded the cortical ERN by 60 ms. Therefore, it seems that erroneous responses in the flanker task modulate LFP activity in subcortical structures earlier than pMFC activity. This is in parallel to the observation that thalamic LFP activity precedes cortical ERPs during target detection in an oddball paradigm (Klostermann et al. 2006).

It has to be considered that our findings were collected in patients, and motor performance might therefore be modulated by the underlying disease. However, we examined patients with cervical or segmental dystonia, who preserved normal hand motor function. The similar reaction times in patients and healthy participants suggested no major abnormality in motor output in our patients. That said, we need to consider that patients committed significantly more errors which, due to the implicit speed-accuracy trade off in the flanker task, reveals that, at a similar speed level, task performance was affected in the patient group at the level of “response selection”. Due to the reduced inhibition demonstrated at several levels of the central nervous system in patients with dystonia (e.g. Hallet 1998), it remains unclear whether the larger error rates in the patient group reflect: 1) A deficiency in the correct inhibition of conflicting responses, or (2) the less attentive postoperative state of the patients. However, the similar posterror slowing data, ERN peak-to-peak values, and number of corrected errors in both experimental groups imply that “error-monitoring” processes were engaged similarly in the task in the patient and control groups.

Animal data has not so far assessed the cortical–subcortical interactions during error monitoring. Evidence for a cortical ERN-like component can be found in macaques (Godlove et al. 2011), whereas specific error-related modulation of spiking activity in BG structures has been found in the GPe, STN, GPi, NAcc (Arkadir et al. 2004; Taha et al. 2007; Hong and Hikosaka 2008; Lardeux et al. 2009; Hong et al. 2011). A direct link to human BG data in relation to the ERN is yet unclear, as animal and human error-related data have typically been studied separately using different tasks. Of special relevance, here are the studies of Hikosaka and colleagues because of their emphasis on the modulation of activity in the primate dorsal pallidum—corresponding to the sensorimotor loop—by negative and positive reward predictions (Hikosaka et al. 2008). The current findings are likely to be associated with the posteroventral lateral GPi within the sensorimotor cortico-striatal-thalamocortical circuit, which was also the target area of pallidal DBS in our patients. Here, it has to be considered that only contact pairs that had at least one contact placed in the GPi according to postoperative imaging were selected for LFP analysis. Clinical efficacy of DBS further supports correct placement of the electrode in the sensorimotor GPi, although contact localization cannot be exclusively related to the sensorimotor part of the GPi. Parallel information processing has been postulated along the segregated (cognitive, limbic, and motor) cortical-BG-thalamocortical loops (Albin et al. 1989; DeLong 1990), which, however, converge to some extent, thus enabling the integration of information processing (Draganski et al. 2008; Turner and Desmurget 2010). Evidence for the parallel processing of error signals in the limbic and motor loops can be found in the LHb-projecting neurons located in the border of the GPi in monkeys (Hong and Hikosaka 2008). Note, however, that LFP recordings in the human pallidum do most likely not exclusively reflect activity paralleled by the border cells in the monkey.

Finally, the time course of the error-related pallidal and cortical activity modulations documented in the present study would suggest that pallidal high-frequency stimulation—due to interference with physiological activity within the cortico-striatal-thalamic loop—may alter the integrity of the error-monitoring system at the BG level leading to behavioral consequences. Specifically, we would predict that stimulation-induced disruption of the output activity from the BG interferes with the neural signature of early error detection observed in the sensorimotor GPi and its projection to the cortex, possibly leading to larger error rates and reduced ERN amplitudes, rather than changes in reaction time. Such investigations could complement preliminary results regarding the effect of focal lesions in the GPi on performance monitoring during a flanker task (1 patient, Ullsperger and von Cramon 2006) and should be tested in future studies.

In summary, our investigations into the role of the sensorimotor GPi in error processing revealed error-related pallidal activity that precedes the cortical ERN and drives cortical activity in the gamma frequency range, suggesting an early modulation of error-related activity in the human GPi. In addition, similarly to the studies emphasizing theta oscillations as the main contributors to spectral content of the cortical ERN, we found that pallidal theta oscillations characterize the early error-related oscillatory modulations. Finally, stimulus-locked ERP modulations during motor preparation highlighted the general role of the ipsilateral GPi in the initial inhibition of conflicting responses, which was weaker and lateralized to the wrong side in error trials. Consequently, we propose that the prominent influence of pallidal error-related activity on the cortical ERN results from monitoring the earlier pallidal inhibition processes necessary for adequate response selection. But future work should identify in a more mechanistic way the early processes that lead to the putative pallidal error-detection signal.

## Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/

## Funding

This work was supported by the German Research Association (DFG) through project HE 6103/1-1 to M.H.R. and project KFO 247 to A.A.K., A.K., J.H., and G.H.S.

## Notes

We thank Carsten Allefeld and Felix Blankenburg for their comments on the first version of this manuscript, Markus Ullsperger for advice with the flanker task in a previous study and suggestions, and Vadim Nikulin for fruitful discussions. Conflict of Interest: J.K.K. is a consultant to Medtronic.

## References

Agam
Y
Hämäläinen
MS
Lee
AK
Dyckman
KA
Friedman
JS
Isom
M
Makris
N
Manoach
DS
Multimodal neuroimaging dissociates hemodynamic and electrophysiological correlates of error processing
Proc Natl Acad Sci USA
,
2011
, vol.
108

42
(pg.
17556
-
17561
)
Albin
RL
Young
AB
Penney
JB
The functional anatomy of basal ganglia disorders
Trends Neurosci
,
1989
, vol.
12

10
(pg.
366
-
375
Review
Alegre
M
Gurtubay
IG
Labarga
A
Iriarte
J
Valencia
M
Artieda
J
Frontal and central oscillatory changes related to different aspects of the motor process: a study in go/no-go paradigms
Exp Brain Res
,
2004
, vol.
159
(pg.
14
-
22
)
Arkadir
D
Morris
G
Vaadia
E
Bergman
H
Independent coding of movement direction and reward prediction by single pallidal neurons
J Neurosci
,
2004
, vol.
24

45
(pg.
10047
-
10056
)
Baccalá
LA
Sameshima
K
Partial directed coherence: a new concept in neural structure determination
Biol Cybern
,
2001
, vol.
84

6
(pg.
463
-
474
)
Basso
D
Chiarandini
M
Salmaso
L
Synchronized permutation tests in replicated IxJ designs
J Stat Planning Inference
,
2007
, vol.
137

8
(pg.
2564
-
2578
)
Benjamini
Y
Krieger
AM
Yekuteli
D
Adaptive linear step-up procedures that control the false discovery rate
Biometrika
,
2006
, vol.
93

3
(pg.
491
-
507
)
Beste
C
Saft
C
Andrich
J
Gold
R
Falkenstein
M
Error processing in Huntington's disease
PLoS One.
,
2006
, vol.
1

1
pg.
e86

Beste
C
Saft
C
Konrad
C
Andrich
J
Habbel
A
Schepers
I
Jansen
A
Pfleiderer
B
Falkenstein
M
Levels of error processing in Huntington's disease: a combined study using event-related potentials and voxel-based morphometry
Hum Brain Mapp
,
2008
, vol.
29

2
(pg.
121
-
130
)
Botvinick
M
Nystrom
LE
Fissell
K
Carter
CS
Cohen
JD
Conflict monitoring versus selection-for-action in anterior cingulate cortex
Nature
,
1999
, vol.
402

6758
(pg.
179
-
181
)
Botvinick
MM
Braver
TS
Barch
DM
Carter
CS
Cohen
JD
Conflict monitoring and cognitive control
Psychol Rev
,
2001
, vol.
108
(pg.
624
-
652
)
Brücke
C
Huebl
J
Schönecker
T
Neumann
WJ
Yarrow
K
Kupsch
A
Blahak
C
Lütjens
G
Brown
P
Krauss
JK
, et al.  .
Scaling of movement is related to pallidal γ oscillations in patients with dystonia
J Neurosci
,
2012
, vol.
32

3
(pg.
1008
-
1019
)
Brücke
C
Kempf
F
Kupsch
A
Schneider
GH
Krauss
JK
Aziz
T
Yarrow
K
Pogosyan
A
Brown
P
Kühn
AA
Movement-related synchronization of gamma activity is lateralized in patients with dystonia
Eur J Neurosci
,
2008
, vol.
27

9
(pg.
2322
-
2329
)
Cavanagh
JF
Cohen
MX
Allen
JJ
Prelude to and resolution of an error: EEG phase synchrony reveals cognitive control dynamics during action monitoring
J Neurosci
,
2009
, vol.
29

1
(pg.
98
-
105
)
Cheyne
D
Bells
S
Ferrari
P
Gaetz
W
Bostan
AC
Self-paced movements induce high-frequency gamma oscillations in primary motor cortex
Neuroimage
,
2008
, vol.
42
(pg.
332
-
342
)
Cockburn
J
Frank
MJ
Mars
RB
Sallet
J
Rushworth
MFS
Yeung
N
Reinforcement learning, conflict monitoring, and cognitive control: an integrative model of cingulate-striatal interactions and the ERN
Neural basis of motivational and cognitive control
,
2011
, vol.
Vol. 9

Cambridge, MA: MIT Press
(pg.
311
-
331
)
Cohen
MX
Axmacher
N
Lenartz
D
Elger
CE
Sturm
V
Schlaepfer
TE
Good vibrations: cross-frequency coupling in the human nucleus accumbens during reward processing
2009a
, vol.
21

5
(pg.
875
-
889
)
Cohen
MX
Axmacher
N
Lenartz
D
Elger
CE
Sturm
V
Schlaepfer
TE
Nuclei accumbens phase synchrony predicts decision-making reversals following negative feedback
J Neurosci
,
2009b
, vol.
29

23
(pg.
7591
-
7598
)
Cohen
MX
Ranganath
C
Reinforcement learning signals predict future decisions
J Neurosci
,
2007
, vol.
27

2
(pg.
371
-
378
)
Coles
MG
Scheffers
MK
Holroyd
CB
Why is there an ERN/Ne on correct trials? Response representations, stimulus-related components, and the theory of error-processing
Biol Psychol
,
2001
, vol.
56

3
(pg.
173
-
189
)
Crone
NE
Miglioretti
DL
Gordon
B
Lesser
RP
Functional mapping of human sensorimotor cortex with electrocorticographic spectral analysis. II. Event-related synchronization in the gamma band
Brain
,
1998
, vol.
121
(pg.
2301
-
2315
)
Debener
S
Ullsperger
M
Siegel
M
Fiehler
K
von Cramon
DY
Engel
AK
Trial-by-trial coupling of concurrent electroencephalogram and functional magnetic resonance imaging identifies the dynamics of performance monitoring
J Neurosci
,
2005
, vol.
25

50
(pg.
11730
-
11737
)
DeLong
MR
Primate models of movement disorders of basal ganglia origin
Trends Neurosci
,
1990
, vol.
13

7
(pg.
281
-
285
Review
Delorme
A
Makeig
S
EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis
J Neurosci Methods.
,
2004
, vol.
134

1
(pg.
9
-
21
)
Desmurget
M
Gaveau
V
Vindras
P
Turner
RS
Broussolle
E
Thobois
S
On-line motor control in patients with Parkinson's disease
Brain
,
2004
, vol.
127
(pg.
1755
-
1773
)
Desmurget
M
Grafton
S
Forward modeling allows feedback control for fast reaching movements
Trends Cogn Sci
,
2000
, vol.
4

11
(pg.
423
-
431
)
Desmurget
M
Gréa
H
Grethe
JS
Prablanc
C
Alexander
GE
Grafton
ST
Functional anatomy of nonvisual feedback loops during reaching: a positron emission tomography study
J Neurosci
,
2001
, vol.
21

8
(pg.
2919
-
2928
)
Ding
M
Bressler
SL
Yang
W
Liang
H
Short-window spectral analysis of cortical event-related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variability assessment
Biol Cybern
,
2000
, vol.
83

1
(pg.
35
-
45
)
Doñamayor
N
Heilbronner
U
Münte
TF
Coupling electrophysiological and hemodynamic responses to errors
Hum Brain Mapp
,
2011
Draganski
B
Kherif
F
Klöppel
S
Cook
PA
Alexander
DC
Parker
GJ
Deichmann
R
Ashburner
J
Frackowiak
RS
Evidence for segregated and integrative connectivity patterns in the human basal ganglia
J Neurosci
,
2008
, vol.
28

28
(pg.
7143
-
7152
)
Efron
B
Jackknife-after-bootstrap standard errors and influence functions
J R Stat Soc B
,
1992
, vol.
54

1
(pg.
83
-
127
)
Falkenstein
M
Hielscher
H
Dziobek
I
Schwarzenau
P
Hoormann
J
Sunderman
B
Hohnsbein
J
Action monitoring, error detection, and the basal ganglia: an ERP study
Neuroreport
,
2001
, vol.
12

1
(pg.
157
-
161
)
Falkenstein
M
Hohnsbein
J
Hoormann
J
Blanke
L
Brunia
CHM
Gaillard
AWK
Kok
A
Effects of errors in choice reaction tasks on the ERP under focused and divided attention
Psychophysiological brain research
,
1990
Tilburg, The Netherlands: Tilburg University Press
(pg.
192
-
195
)
Falkenstein
M
Hohnsbein
J
Hoormann
J
Event-related potential correlates of errors in reaction tasks
1995
(pg.
287
-
296
In Karmos G, Molnar M, Csepe V, Czigler I, Desmedt JE, editors. Perspectives of event-related brain potentials research. Amsterdam: Elsevier
Fiehler
K
Ullsperger
M
Von Cramoncd
DY
Electrophysiological correlates of error correction
Psychophysiology
,
2005
, vol.
42
(pg.
72
-
82
)
Florin
E
Gross
J
Pfeifer
J
Fink
GR
Timmermann
L
The effect of filtering on Granger causality based multivariate causality measures
Neuroimage
,
2010
, vol.
50

2
(pg.
577
-
588
)
Fogelson
N
Williams
D
Tijssen
M
van Bruggen
G
Speelman
H
Brown
P
Different functional loops between cerebral cortex and the subthalamic area in Parkinson's disease
Cereb Cortex
,
2006
, vol.
16
(pg.
64
-
75
)
Frank
MJ
Seeberger
LC
O'reilly
RC
By carrot or by stick: cognitive reinforcement learning in parkinsonism
Science
,
2004
, vol.
306

5703
(pg.
1940
-
1943
)
Gehring
WJ
Gross
B
Coles
MGH
Meyer
DE
Donchin
E
A neural system for error detection and compensation
Psychol Sci
,
1993
, vol.
4
(pg.
385
-
390
)
Godlove
DC
Emeric
EE
Segovis
CM
Young
MS
Schall
JD
Woodman
GF
Event-related potentials elicited by errors during the stop-signal task. I. Macaque monkeys
J Neurosci
,
2011
, vol.
31

44
(pg.
15640
-
15649
)
Good
P
Permutation, parametric, and bootstrap tests of hypotheses
,
2005
2nd ed
Berlin
Springer
Grafton
ST
Tunik
E
Human basal ganglia and the dynamic control of force during on-line corrections
J Neurosci
,
2011
, vol.
31

5
(pg.
1600
-
1605
)
Granger
CWJ
s Investigating causal relations by econometric models and cross-spectral methods
Econometrica
,
1969
, vol.
37
(pg.
424
-
438
)
Hallet
M
The neurophysiology of dystonia
Arch Neurol
,
1998
, vol.
55
(pg.
601
-
603
)
Heinze
HJ
Heldmann
M
Voges
J
Hinrichs
H
Marco-Pallares
J
Hopf
JM
Müller
UJ
Galazky
I
Sturm
V
Bogerts
B
, et al.  .
Counteracting incentive sensitization in severe alcohol dependence using deep brain stimulation of the nucleus accumbens: clinical and basic science aspects
Front Hum Neurosci
,
2009
, vol.
3
pg.
22

Hester
R
Fassbender
C
Garavan
H
Individual differences in error processing: a review and reanalysis of three event-related fMRI studies using the GO/NOGO task
Cereb Cortex
,
2004
, vol.
14
(pg.
986
-
994
)
Hester
R
Madeley
J
Murphy
K
Mattingley
JB
Learning from errors: error-related neural activity predicts improvements in future inhibitory control performance
J Neurosci
,
2009
, vol.
29
(pg.
7158
-
7165
)
Hikosaka
O
Bromberg-Martin
E
Hong
S
Matsumoto
M
New insights on the subcortical representation of reward
Curr Opin Neurobiol
,
2008
, vol.
18

2
(pg.
203
-
208
)
Holroyd
CB
Coles
MG
The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity
Psychol Rev
,
2002
, vol.
109
(pg.
679
-
709
)
Holroyd
CB
Nieuwenhuis
S
Yeung
N
Nystrom
L
Mars
RB
Coles
MGH
Cohen
JD
Dorsal anterior cingulate cortex shows fMRI response to internal and external error signals
Nat Neurosci
,
2004
, vol.
7
(pg.
497
-
498
)
Hong
S
Hikosaka
O
The globus pallidus sends reward-related signals to the lateral habenula
Neuron
,
2008
, vol.
60
(pg.
720
-
729
)
Hong
S
Jhou
TC
Smith
M
Saleem
KS
Hikosaka
O
Negative reward signals from the lateral habenula to dopamine neurons are mediated by rostromedial tegmental nucleus in primates
J Neurosci
,
2011
, vol.
31

32
(pg.
11457
-
11471
)
Houk
JC
Bastianen
C
Fansler
D
Fishbach
A
Fraser
D
Reber
PJ
Roy
SA
Simo
LS
Action selection and reﬁnement in subcortical loops through basal ganglia and cerebellum
Philos Trans R Soc Lond B Biol Sci
,
2007
, vol.
362
(pg.
1573
-
1583
)
Huo
X
Xiang
J
Wang
Y
Kirtman
EG
Kotecha
R
Fujiwara
H
Hemasilpin
N
Rose
DF
Degrauw
T
Gamma oscillations in the primary motor cortex studied with MEG
Brain Dev
,
2010
, vol.
32

8
(pg.
619
-
624
)
Johannes
S
Wieringa
BM
Nager
W
Müller-Vahl
KR
Dengler
R
Münte
TF
Excessive action monitoring in Tourette syndrome
J Neurol
,
2002
, vol.
249
(pg.
961
-
966
)
Kempf
F
Brücke
C
Kühn
AA
Schneider
GH
Kupsch
A
Chen
CC
Androulidakis
AG
Wang
S
Vandenberghe
W
Nuttin
B
, et al.  .
Modulation by dopamine of human basal ganglia involvement in feedback control of movement
Curr Biol
,
2007
, vol.
17

15
(pg.
587
-
589
)
Kiehl
KA
Liddle
PF
Hopfinger
JB
Error processing and the rostral anterior cingulate: an event-related fMRI study
Psychophysiology
,
2000
, vol.
37
(pg.
216
-
223
)
Klostermann
F
Wahl
M
Marzinzik
F
Schneider
GH
Kupsch
A
Curio
G
Mental chronometry of target detection: human thalamus leads cortex
Brain
,
2006
, vol.
129

Pt 4
(pg.
923
-
931
)
Kopp
B
Rist
F
Mattler
U
N200 in the flanker task as a neurobehavioral tool for investigating executive control
Psychophysiology
,
1996
, vol.
33

3
(pg.
282
-
294
)
Kuhn
J
Gründler
TO
Bauer
R
Huff
W
Fischer
AG
Lenartz
D
Maarouf
M
Bührle
C
Klosterkötter
J
Ullsperger
M
, et al.  .
Successful deep brain stimulation of the nucleus accumbens in severe alcohol dependence is associated with changed performance monitoring
Addict Biol
,
2011
, vol.
16

4
(pg.
620
-
623
)
Lalo
E
Thobois
S
Sharott
A
Polo
G
Mertens
P
Pogosyan
A
Brown
P
Patterns of bidirectional communication between cortex and basal ganglia during movement in patients with Parkinson disease
J Neurosci
,
2008
, vol.
28
(pg.
3008
-
3016
)
Lardeux
S
Pernaud
R
Paleressompoulle
D
Baunez
C
Beyond the reward pathway: coding reward magnitude and error in the rat subthalamic nucleus
J Neurophysiol
,
2009
, vol.
102

4
(pg.
2526
-
2537
)
Litvak
V
Jha
A
Eusebio
A
Oostenveld
R
Foltynie
T
Limousin
P
Zrinzo
L
Hariz
MI
Friston
K
Brown
P
Resting oscillatory cortico-subthalamic connectivity in patients with Parkinson's disease. Brain
,
2011
, vol.
134

Pt 2
(pg.
359
-
374
)
Liu
X
Wang
S
Yianni
J
Nandi
D
Bain
PG
Gregory
R
Stein
JF
Aziz
TZ
The sensory and motor representation of synchronized oscillations in the globus pallidus in patients with primary dystonia
Brain
,
2008
, vol.
131
(pg.
1562
-
1573
)
Luu
P
Tucker
DM
Makeig
S
Frontal midline theta and the error-related negativity: neurophysiological mechanisms of action regulation
Clin Neurophysiol
,
2004
, vol.
115
(pg.
1821
-
1835
)
Mallat
S
A wavelet tour of signal processing
,
1999
San Diego
Academic Press
Marco-Pallarés
J
Camara
E
Münte
TF
Rodríguez-Fornells
A
Neural mechanisms underlying adaptive actions after slips
J Cogn Neurosci
,
2008
, vol.
20

9
(pg.
1595
-
1610
)
Mazzoni
P
Hristova
A
Krakauer
JW
Why don't we move faster? Parkinson's disease, movement vigor, and implicit motivation
J Neurosci
,
2007
, vol.
27

27
(pg.
7105
-
7116
)
McClure
SM
Daw
ND
Montague
PR
A computational substrate for incentive salience
Trends Neurosci
,
2003
, vol.
26

8
(pg.
423
-
428
)
Mogenson
GJ
Jones
DL
Yim
CY
From motivation to action: functional interface between the limbic system and the motor system
Prog Neurobiol
,
1980
, vol.
14
(pg.
69
-
97
)
Münte
TF
Heldmann
M
Hinrichs
H
Marco-Pallares
J
Krämer
UM
Sturm
V
Heinze
HJ
Nucleus accumbens is involved in human action monitoring: evidence from invasive electrophysiological recordings
Front Hum Neurosci
,
2008
, vol.
1
pg.
11

Muthukumaraswamy
SD
Functional properties of human primary motor cortex gamma oscillations
J Neurophysiol
,
2010
, vol.
104
(pg.
2873
-
2885
)
Nambu
A
Tokuno
H
Hamada
I
Kita
H
Imanishi
M
Akazawa
T
Ikeuchi
Y
Hasegawa
N
Excitatory cortical inputs to pallidal neurons via the subthalamic nucleus in the monkey
J Neurophysiol
,
2000
, vol.
84

1
(pg.
289
-
300
)
Niv
Y
Daw
ND
Joel
D
Dayan
P
Tonic dopamine: opportunity costs and the control of response vigor
Psychopharmacology (Berl)
,
2007
, vol.
191

3
(pg.
507
-
520
)
Pasupathy
A
Miller
EK
Different time courses of learning-related activity in the prefrontal cortex and striatum
Nature
,
2005
, vol.
433

7028
(pg.
873
-
876
)
Pikovsky
A
Rosenblum
M
Kurths
J
Synchronization–a universal concept in nonlinear sciences
,
2001
Cambridge
Cambridge University Press
Ridderinkhof
KR
Ullsperger
M
Crone
EA
Nieuwenhuis
S
The role of the medial frontal cortex in cognitive control
Science
,
2004
, vol.
306

5695
(pg.
443
-
447
Review
Rodríguez-Fornells
A
Kurzbuch
AR
Münte
TF
Time course of error detection and correction in humans: neurophysiological evidence
J Neurosci
,
2002
, vol.
22

22
(pg.
9990
-
9996
)
Rodriguez-Oroz
MC
López-Azcárate
J
Garcia-Garcia
D
Alegre
M
Toledo
J
Valencia
M
Guridi
J
Artieda
J
Obeso
JA
Involvement of the subthalamic nucleus in impulse control disorders associated with Parkinson's disease
Brain
,
2011
, vol.
134
(pg.
36
-
49
)
Schlögl
A
Supp
G
Analyzing event-related EEG data with multivariate autoregressive parameters
Prog Brain Res
,
2006
, vol.
159
(pg.
135
-
147
)
Schönecker
T
Kupsch
A
Kuhn
AA
Schneider
GH
Hoffmann
KT
Automated optimization of subcortical cerebral MR imaging-atlas co-registration for improved postoperative electrode localization in deep brain stimulation
Am J Neuroradiol
,
2009
, vol.
30
(pg.
1914
-
1921
)
Schultz
W
Getting formal with dopamine and reward
Neuron
,
2002
, vol.
36

2
(pg.
241
-
263
Review
Schultz
W
Dayan
P
Montague
PR
A neural substrate of prediction and reward
Science
,
1997
, vol.
275

5306
(pg.
1593
-
1599
)
Schwarz
G
Estimating the dimension of a model
Ann Stat
,
1978
, vol.
6

2
(pg.
461
-
464
)
Seifert
S
von Cramon
DY
Imperati
D
Tittgemeyer
M
Ullsperger
M
Thalamocingulate interactions in performance monitoring
J Neurosci
,
2011
, vol.
31

9
(pg.
3375
-
3383
)
Shadmehr
R
Krakauer
JW
A computational neuroanatomy for motor control
Exp Brain Res
,
2008
, vol.
185

3
(pg.
359
-
381
Review
Sharott
A
Grosse
P
Kühn
AA
Salih
F
Engel
AK
Kupsch
A
Schneider
GH
Krauss
JK
Brown
P
Is the synchronization between pallidal and muscle activity in primary dystonia due to peripheral afferance or a motor drive?
Brain
,
2008
, vol.
131

Pt 2
(pg.
473
-
484
)
Silvetti
M
Seurinck
R
Verguts
T
Value and prediction error in medial frontal cortex: integrating the single-unit and systems levels of analysis
Front Hum Neurosci
,
2011
, vol.
5
pg.
75

Stemmer
B
Segalowitz
SJ
Dywan
J
Panisset
M
Melmed
C
The error negativity in non-medicated and medicated patients with Parkinson's disease
Clin Neurophysiol
,
2007
, vol.
118
(pg.
1223
-
1229
)
Strutt
JW
The problem of the random walk
Nature
,
1905
, vol.
72
pg.
318

Sutton
RS
Learning to predict by the methods of temporal differences
Machine Learn
,
1988
, vol.
3
(pg.
9
-
44
)
Sutton
RS
Barto
AG
Reinforcement learning: an introduction
,
1998
Cambridge (MA)
MIT Press
Taha
SA
Nicola
SM
Fields
HL
Cue-evoked encoding of movement planning and execution in the rat nucleus accumbens
J Physiol
,
2007
, vol.
584

Pt 3
(pg.
801
-
818
)
Taylor
SF
Stern
ER
Gehring
WJ
Neural systems for error monitoring: recent findings and theoretical perspectives
Neuroscientist
,
2007
, vol.
13
(pg.
160
-
172
)
Todorov
E
Jordan
MI
Optimal feedback control as antheory of motor coordination
Nat Neurosci
,
2002
, vol.
5
(pg.
1226
-
1235
)
Trujillo
LT
Allen
JJ
Theta EEG dynamics of the error-related negativity
Clin Neurophysiol
,
2007
, vol.
118

3
(pg.
645
-
668
)
Trujillo
LT
Peterson
MA
Kaszniak
AW
Allen
JJ
EEG phase synchrony differences across visual perception conditions may depend on recording and analysis methods
Clin Neurophysiol
,
2005
, vol.
116

1
(pg.
172
-
189
)
Tunik
E
Houk
JC
Grafton
ST
Basal ganglia contribution to the initiation of corrective submovements
Neuroimage
,
2009
, vol.
47

4
(pg.
1757
-
1766
)
Turner
RS
Desmurget
M
Basal ganglia contributions to motor control: a vigorous tutor
Curr Opin Neurobiol
,
2010
, vol.
20
(pg.
704
-
716
)
Ullsperger
M
Nittono
H
von Cramon
DY
When goals are missed: dealing with self-generated and externally induced failure
Neuroimage
,
2007
, vol.
35

3
(pg.
1356
-
1364
)
Ullsperger
M
von Cramon
DY
The role of intact frontostriatal circuits in error processing
J Cogn Neurosci
,
2006
, vol.
18

4
(pg.
651
-
664
)
van Veen
V
Carter
CS
The anterior cingulate as a conflict monitor: fMRI and ERP studies
Physiol Behav
,
2002
, vol.
77

4–5
(pg.
477
-
482
)
Willemssen
R
Müller
T
Schwarz
M
Falkenstein
M
Beste
C
Response monitoring in de novo patients with Parkinson's disease
PLoS One
,
2009
, vol.
4

3
pg.
e4898

Williams
D
Tijssen
M
Van Bruggen
G
Bosch
A
Insola
A
Di Lazzaro
V
Mazzone
P
Oliviero
A
Quartarone
A
Speelman
H
, et al.  .
Dopamine-dependent changes in the functional connectivity between basal ganglia and cerebral cortex in humans
Brain
,
2002
, vol.
125

Pt 7
(pg.
1558
-
1569
)