Proactive inhibition is not modified by deep brain stimulation for Parkinson's disease: An electrical neuroimaging study

Abstract In predictable contexts, motor inhibitory control can be deployed before the actual need for response suppression. The brain functional underpinnings of proactive inhibition, and notably the role of basal ganglia, are not entirely identified. We investigated the effects of deep brain stimulation of the subthalamic nucleus or internal globus pallidus on proactive inhibition in patients with Parkinson's disease. They completed a cued go/no‐go proactive inhibition task ON and (unilateral) OFF stimulation while EEG was recorded. We found no behavioural effect of either subthalamic nucleus or internal globus pallidus deep brain stimulation on proactive inhibition, despite a general improvement of motor performance with subthalamic nucleus stimulation. In the non‐operated and subthalamic nucleus group, we identified periods of topographic EEG modulation by the level of proactive inhibition. In the subthalamic nucleus group, source estimation analysis suggested the initial involvement of bilateral frontal and occipital areas, followed by a right lateralized fronto‐basal network, and finally of right premotor and left parietal regions. Our results confirm the overall preservation of proactive inhibition capacities in both subthalamic nucleus and internal globus pallidus deep brain stimulation, and suggest a partly segregated network for proactive inhibition, with a preferential recruitment of the indirect pathway.

. The pre-SMA and IFG first react to stopping stimuli, followed by an early recruitment of STN through the 'hyperdirect' pathway (Aron, Robbins, & Poldrack, 2014).
While some authors advance that proactive inhibition depend merely on the modulation of the hyperdirect pathway (Stuphorn & Emeric, 2012), other accounts propose that the 'indirect pathway' has a specific role in this process. This latter view implies a more important role for the dorsolateral prefrontal cortex (DLPFC) and the caudate, external and internal globus pallidum in proactive response suppression (Jahanshahi, Obeso, Rothwell, & Obeso, 2015). Yet, the involvement of each these structure remains largely speculative. Deep brain stimulation of the STN improves motor symptoms of Parkinson's disease interfering with dysfunctional oscillations within basal ganglia circuitry, eventually favouring a prokinetic state (Chiken & Nambu, 2016). Despite the proposed centrality of STN in inhibitory control, the few studies having investigated the effects of STN-DBS on proactive inhibition reported no effect (Mancini et al., 2019;Mirabella et al., 2012), or even suggested a possible normalisation of proactive capabilities (Obeso, Wilkinson, Rodríguez-Oroz, Obeso, & Jahanshahi, 2013). Experimental design is probably key for capturing subtle modifications. For instance, Mirabella et al. (2013) could demonstrate that STN-DBS restores to normal the relationship between reaction time and movement time in the context of uncertainty, possibly modulating proactive inhibition mechanisms still active during movement execution. Likewise, comparisons of the effects of dopaminergic drugs and STN-DBS suggest that STN-DBS may specifically normalise the capacity of releasing proactive inhibition, contrary to dopaminergic drug treatment (Favre, Ballanger, Thobois, Broussolle, & Boulinguez, 2013).
The internal globus pallidus (GPi) is a more rarely chosen target for DBS in Parkinson's disease. Despite allowing for a lesser decrease of dopaminergic medication than STN-DBS, GPi-DBS has a comparable efficacy on motor symptoms, and in the reduction of fluctuations and dyskinesia (Williams, Foote, & Okun, 2014). As a common structure within multiple basal ganglia pathways (direct, indirect, hyperdirect), it is involved in inhibitory control, but has been less studied than STN. Current evidence suggest a positive effect of GPi-DBS in triggering the action, with no impairment in its suppression (Kohl et al., 2015), or even an improvement of the dysfunctional proactive inhibition after turning GPi stimulator ON (Pan et al., 2018).
In the present study, we aimed at resolving the inconsistencies of previous literature and examined thoroughly the involvement of STN and of the internal globus pallidus (GPi) in proactive inhibition. To this aim, we capitalised on Parkinson's disease patients previously operated for deep brain stimulation (DBS) of either of the two nuclei.
We developed a task requiring the deployment of varying levels of proactive inhibitory control, and studied the behavioural and electrophysiological effects of the stimulation of STN and GPi in Parkinson's Disease patients previously operated for DBS. Topographic and source estimation analyses of event-related potentials recorded between the presentation of a cue and the Go/NoGo probe were conducted to examine the functional effect of the stimulation.
Turning the device on and off, we contrasted the electrophysiological activity associated with the engagement of proactive inhibition in the two stimulation conditions. Given the above, we expected (a) that both STN-and GPi-DBS reduce an excessive proactive inhibition, related to PD symptoms, (b) that both STN-and GPi-DBS decrease the capacity of modulating proactive inhibition, interfering with the activity of the hyperdirect pathway and the indirect pathway, respectively; and (c) EEG evidence of the involvement of DLPFC, IFG, and pre-SMA during the task. two were suspected of drug-induced parkinsonism, one was not able to complete the task, and two had excessively noisy EEG signal resulting in too few ERP trials for averaging. As we focused on the ERPs time-locked to the cue and thus not contaminated by the activity related to the motor response, we did not exclude left-handed patients from the EEG analysis (two NO and two STN). In the STN group, one left-handed participant was tested as a right-handed. The second lefthanded participant was tested on the left side and thus, the stimulator was switched off on the opposite side as the other participants. As the analyses with and without this patient showed the same results, we kept this patient to maximise our statistical power. Previous studies have shown that reactive and proactive inhibition behavioural performance was not influenced by the hand employed to perform the task (Caprio, Modugno, Mancini, Olivola, & Mirabella, 2020;Mirabella, Fragola, Giannini, Modugno, & Lakens, 2017).
Demographic data are summarised in Table 1. Hoehn and Yahr staging was not assessed formally but we estimate it was for all patients in all groups between II and III. Group comparisons were performed using non-parametric Kruskal-Wallis One-Way ANOVAs.
We collected clinical and demographic information from the patients before the experiment and during the 30 min pause between sessions. Patients filled in questionnaires assessing mood (Beck Depression Inventory, BDI); and the presence and severity of impulse control disorders (Questionnaire for Impulsive-Compulsive Disorders in Parkinson's Disease, QUIP, (Weintraub et al., 2012). They underwent a Montreal Cognitive Assessment (MoCA) conducted by an experienced neurologist or neuropsychologist during the experimental session, if not available from medical records and not older than 3 months. All patients were examined by a trained neurologist (EAA, ID) for assessing the motor part of the Movement Disorders Society Unified Parkinson's Disease Rating Scale (0-132 points, MDS-UPDRS III, [Goetz et al., 2008]). All patients were tested under their usual antiparkinsonian medication.

| Stimuli and task
The cued Go/NoGo task involved a green circle as Go signal and a red circle as NoGo signal preceded by a cue informing on the probability of an upcoming NoGo signal (Figure 1). The cue was represented by a vertical bar, in three alternative configurations. If the bar was completely white, then there was a 0% chance that the following trial would be a NoGo. If the bar was only partially shaded, then there was a 25% chance that the following trial would be a NoGo. This probability reached 75% if the symbol was almost all shaded. The participants were not informed about the detailed probabilities, but only given general indications ('certainly go', 'maybe stop', 'likely stop').
Each trial started with a white fixation cross on light grey background (1500-1900 ms). During fixation, the participants were asked to keep their hand on a response box (E-Prime Chronos box), with the index finger ready to press the key. The cue was then presented for 1,000 ms. The delay between the cue and the probe was variable (600-1,100 ms) to reduce the predictability of the task. Participants were instructed to press the key as fast as possible as soon as they UPDRS-III score was calculated before each session in DBS patients.

| EEG recording and pre-processing
Electroencephalogram (EEG) was recorded at a sampling rate of 1,024 Hz over 64 channels following the extended 10-20 system, with a Biosemi ActiveTwo system (Biosemi, Amsterdam, Netherlands).
Offline pre-processing of the raw EEG signal was conducted using inhouse Matlab scripts and EEGlab (Delorme & Makeig, 2004), in order to obtain ERPs time-locked to the cue, over all electrodes.
Raw EEG data were first filtered using a 0.5-40 Hz band-pass, which removed DBS-related high frequency noise (Lio, Thobois, Ballanger, Lau, & Boulinguez, 2018;Sun et al., 2014). Removal of occasional, large amplitude non-brain noise such as eye and muscle artefacts was completed with the Artefact Subspace Reconstruction EEGlab plugin (ASR; Chang, Hsu, Pion-Tonachini, & Jung, 2018;Mullen et al., 2015). The EEG signal was then segmented into epochs from 100 ms pre-cue onset to 600 ms post-cue onset, with a baseline correction applied over the whole epoch window. Due to remaining eye blinks in the GPi group, individual data for this group went through an ICA procedure using the AMICA algorithm with default settings Palmer, Makeig, Kreutz-Delgado, & Rao, 2008). This step was not necessary in the NO and STN groups. Channels showing a bad signal were interpolated (mean: 2.3 channels) using multiquadric interpolation relying on radial basis functions (Buhmann & Jäger, 2019;Jäger, Klein, Buhmann, & Skrandies, 2016). Epochs with at least one time-frame ±80 μV were automatically rejected and the remaining epochs were averaged across trial for each cue separately.
Finally, ERPs were re-referenced to the common average reference.
The number of accepted epochs for each condition were controlled in order to ensure that any observed between-condition differences were not due to differences in signal-to-noise ratio (STN: 35.7 ± 0.5, p = .630; GPi: 33.5 ± 4.8, p = .613; NO: 35.7 ± 1.0, p = .523).

| Data analysis
Because of the heterogeneity of the groups (NO had no stimulation, GPi was limited in size), we analysed the three groups separately. For all statistical tests, our alpha threshold was set at 0.05, and effect sizes are reported for behavioural data.

| Behavioural analysis
Performance at the Go/NoGo task was assessed by extracting the response times (RT) to Go stimuli and the false alarm rate (FA) to NoGo stimuli, that is, the percentage of inaccurately responded NoGo trials. We removed trials with RTs ≤ 100 ms as they reflect implausible cognitive processing of the Go signal (Gabay & Behrmann, 2014).
We then identified univariate outliers using the median absolute devi- RT and FA were averaged for each Cue type and Stim session (only the dominant hand session for the NO group). Normality of the data was assessed using the Shapiro-Wilk test a criterion of skewness and kurtosis within a ± 2 range (Kim, 2013). Whenever Mauchly's test indicated sphericity violation, we reported corrected p-values using the Greenhouse-Geisser estimates.
RT in the STN and GPi groups was analysed using repeated mea-

| EEG analysis
All analyses were conducted using the Cartool software ( The randomisation additionally estimates a minimal duration threshold for contiguous significant effects to account for multiple tests and temporal autocorrelation (Koenig et al., 2011). Only periods of statistical significance longer than the duration threshold were interpreted.
For each significant period identified, we averaged the individual ERP signals across timeframes and computed brain source estimation.
For this, we applied a local autoregressive average (LAURA) distrib- In order to identify brain areas associated with the effects observed at the ERP level, we applied the same design as for ERP analyses over each node using STEN. We corrected for multiple testing and spatial autocorrelation by applying a spatial-extent threshold of at least 19 contiguous nodes with p < .05. This spatial criterion was calculated with the AlphaSim program (available from the Analysis of Functional NeuroImages website: http://afni.nimh.nih.gov). This program applies a cluster randomisation approach by computing 10,000 Monte Carlo permutations performed on our lead field matrix, assuming a spatial smoothing of 6 mm FWHM and a cluster connection radius of 8.5 mm. For clusters of at least 19 nodes, the output indicated a node-level false positive probability of p < .001 for a clusterlevel likelihood of p < .05.

| Behavioural results
The behavioural results are reported in Figure 2. Detailed behavioural results can be found in Table S1.

| Response time
All groups showed a main effect of Cue (Table 2), driven by a gradual increase in RT as the probability of having a NoGo signal increased ( Figure 2a). As the assumption of normality was not consistently met in the GPi group, we verified the result with a Friedman nonparametric repeated measures ANOVA on the Cue factor. It also reached significance, χ 2 (2) = 6.00, p = .05. Additionally, the STN group showed a main effect of Stimulation (

| NO group
At the scalp level, the one-way repeated measures ANOVA (Cue [CG; MS; LS]) on GMD revealed a period of topographic dissimilarities between 195 and 257 ms post-cue signal, and no periods of difference for GFP (Figure 3a,b). Source estimation localised this difference within the right temporal areas. In this area, current density was higher following MS cues compared to CG and LS cues ( Figure 3c).
However, the MS versus LS post-hoc test did not survive the Holm-Bonferroni correction (p = .048 uncorrected). Detailed source estimation results can be found in Table S3.

| STN group
At  Table S4.
Source estimation for the second period (383-431 ms) localised the difference centred around bilateral posterior areas, the thalamus, the left insula, and the right orbitofrontal gyrus and putamen (    (Table S4). For all clusters, activation was stronger in the ON versus OFF condition (Figure 5b).
Source estimation for the second period (453-508 ms) indicated stronger activation in the ON versus OFF condition within the bilateral anterior and posterior cingulate cortices, the left occipital cortex, and the right inferior parietal lobule (Table S4). Activation was stronger in the OFF versus ON condition within the left frontal and motor cortices, while the other areas showed stronger activity in the ON versus OFF condition (Figure 5b).

| GPi group
Analysis in the GPi group showed no periods of significant difference for neither GFP nor GMD ( Figure 6).

| DISCUSSION
We aimed at characterising the functional correlates of proactive inhibition. We examined the effects of the electrical stimulation of the STN and GPi subcortical structures on proactive inhibitory  Although our predictions were based on theoretical models and on (limited) previous experimental evidence, we could find no effect of DBS on proactive inhibition at the behavioural level. The absence of any effect could be explained by three main reasons: insufficient statistical power, experimental task design, and model inaccuracies.

| No evidence of DBS effects on proactive inhibition
Collecting sufficient data from clinical populations is always challenging, particularly so when the inclusion criteria are stringent and the suitable population is limited, as is the case in studies on DBS patients. It is therefore likely that increasing the number of patients could have unveiled more robust results, but the Bayesian analyses performed seem to confirm the validity of our results. Moreover, our cohort is similar to or larger than previous studies addressing this topic (Kohl et al., 2015;Mirabella et al., 2013;Obeso et al., 2013).
Our findings are in line with previous experimental work, which did not find significant effects either (Mancini et al., 2019;Mirabella et al., 2012). Proactive inhibition is a complex function, and our (simple) task probably does not capture all the involved components. For instance, the timing of proactive inhibition implementation and release appears to be particularly important. In Parkinson's disease (PD), plan updating of a prepared motor action occurs in part after movement initiation, perhaps due to deficits in proactive inhibition release (Leis et al., 2005). STN-DBS appears to restore this function to more physiological levels (Mirabella et al., 2013).
A recent paper could demonstrate that STN-DBS induces a longer stopping time of a continuous ongoing movement, a different approach than testing the suppression of planned, incumbent actions before their execution, as it is normally the case in SSRT or GNG tasks (Lofredi et al., 2020). In future experiments, it will be interesting to test whether this effect occurs through an impairment of proactive inhibition.
Although our task was not designed to capture the release of proactive inhibition during an ongoing movement (Lofredi et al., 2020), or after movement initiation (Leis et al., 2005;Mirabella et al., 2013), we are confident it recorded the most determinant aspects of the cognitive mechanisms underlying the preparation to movement inhibition.
The validity of our behavioural findings is strengthened by the results DBS is thought to remove the antikinetic effect of STN glutamatergic projections by inactivating neuronal somata as in a functional lesion (Chiken & Nambu, 2016). According to the model proposed by Frank (Frank, 2006;Wiecki & Frank, 2013), one of STN functions would be to send a general 'stop' signal to ongoing motor programs, in the presence of conflicting external stimuli. From a cognitive perspective, it would 'buy time' for other structures to elaborate appropriate responses to external stimuli ('hold your horses'). The sudden increase of STN antikinetic activity and the modulation of its threshold are thought to be key in reactive and proactive inhibition, respectively.
Areas of STN mediating inhibition are likely far from the electrode, and only slightly affected by the stimulation. This assumption is in line with the known organisation of STN, which is constituted by sub-areas with different functional specialisation (Accolla et al., 2016).
We speculate that the associative areas of STN are more involved in motor inhibition than pure motor areas, which are the target of DBS electrode implantation. This has been recently demonstrated in a task during intraoperative recordings (Mosher, Mamelak, Malekmoha mmadi, Pouratian, & Rutishauser, 2021). This is also compatible with the observation that DBS impairs motor inhibition only in high conflict situations (Georgiev, Dirnberger, Wilkinson, Limousin, & Jahanshahi, 2016) or under speed pressure (Pote et al., 2016), and that there is a role for the electrode localisation (more ventral within the STN; Hershey et al., 2010;Rodriguez-Oroz et al., 2011). A recently described direct connection between IFC and ventral STN confirms this interpretation (Chen et al., 2020). We did not reconstruct the electrode localisation due to a limited access to imaging for this project, but we assume that stimulating contacts were located in the latero-dorsal STN, given the preoperative planning and the good clinical outcome of the intervention (Horn, Neumann, Degen, Schneider, & Kühn, 2017).

| STN stimulation enhances attentional processes
We observed an improvement of RT under the STN-DBS condition, compared to the OFF state. However, this improvement occurred with a comparable magnitude in all cue conditions, and was thus not related to an effect of DBS on proactive inhibition mechanisms. We interpret this finding as a general motor improvement effect. Note that our task did not allow for differentiating reaction times from movement times.
At the EEG level, we found two epochs of effect of stimulation alone after cue presentation. The source localisation shows that this early effect is strongly left lateralised. This localization is compatible with the fact that in our protocol, we turned the stimulator ON and OFF only on one side (the left one for all except one left-handed patient).
The first effect occurred very early, at 100-120 ms after cue presentation a latency probably corresponding to modulation of primary sensory or attentional processes. We could speculate that STN DBS promotes an increased reactivity immediately after cue presentation (unrelated to the semantic content of the cue), and maybe that this translates into a faster motor response later on: source localisation was consistent with a brain executive attention network called 'cingulo-opercular network' in fMRI studies (Neta, Nelson, & Petersen, 2017), involving the anterior cingulate and anterior insula.

| Internal globus pallidus (GPi)
We In reactive inhibition studies (response to a Go or NoGo stimulus), the N2 and P3 components have been associated with conflict monitoring and decision making (Falkenstein, 2006;Hartmann, Sallard, & Spierer, 2016;Manuel, Grivel, Bernasconi, Murray, & Spierer, 2010;Verbruggen & Logan, 2015). Earlier latencies have been associated with improved neural efficiency after training (Benikos, Johnstone, & Roodenrys, 2013), and may reflect a better association between stimulus and response (De Pretto et al., 2017;Spierer, Chavan, & Manuel, 2013). Here, it is possible that the MS cue was considered by the participants as most uncertain regarding the upcoming Go or NoGo stimulus, increasing the time to decide whether to anticipate a go or an inhibition response. Indeed, even though the probability of having a NoGo stimulus after an MS cue was the same as the probability of having a Go stimulus after an LS cue, the participants might have considered the MS cue as an inbetween, uncertain, condition. This interpretation remains highly speculative, as unfortunately we did not ask them feedback questions on how they perceived the cues. In the STN group however, the parieto-occipital cluster that distinguishes MS from CG and LS has been involved in visual and attentional processes (Spay et al., 2018), and fits with this interpretation.

| Direct, indirect and hyperdirect pathway in proactive inhibition
It is worth underscoring that localised source estimation within deep brain structures must be interpreted with caution (Cohen, Cavanagh, & Slagter, 2011;Seeber et al., 2019). Even more so given the inhomogeneity of our population. We tentatively interpret our results within the framework of known cortico-subcortical circuitry.
At around 400 ms (380-430 ms) after cue presentation, the presence or absence of proactive inhibition per se could play a role in the observed recordings. At this moment, in the CG condition, and to a lesser extent in the LS condition the P3 component is almost at its peak, whereas it still lies at the transition between N2 and P3 following the MS cues (Figure 4b). This shift in latency is consistent with our interpretation of MS being more indecisive. The P3 component has been associate with the implementation of the inhibition command (De Pretto et al., 2017;Spierer et al., 2013). Here, it might reflect proactive inhibition mechanisms, such as best illustrated by the right superior (temporo-) occipital cluster. This cluster showed an increase in current density as the probability of a NoGo signal increases. Similarly, the posterior-thalamic cluster and the left insula showed stronger activation in the LS conditions. Within the right fronto-basal cluster, activation was stronger in the CG condition.
At around 510-550 ms, an additional significant peak was found.
Its topography was consistent with an increased activation of right pre-central gyrus in the LS condition, and of a left parietal region in the CG condition ( Figure 5a).
As stated above, subcortical source localization must be interpreted with caution and likewise, our conclusions about involved pathways or circuits definitely need further confirmation. We here propose that the peak at around 400 ms from cue presentation corresponds to the engagement of the indirect pathway in proactively adjusting inhibition reactivity, in preparation for the upcoming probe signal. The higher activation of the right putamen, possibly of the right caudate and of the thalamic regions in the CG vs MS condition ( Figure 5a, orange panel) could be interpreted as an early release of proactive inhibition, assuming striatum as a globally prokinetic structure within the direct pathway. Later on, at around 500 ms, the hyperdirect pathway mediates a further release of inhibition, by reducing the relative activity in the right precentral gyrus: the pattern of activation is consistent with a lower activity in conditions when the probability of a Go signal, thus of an upcoming movement, was higher.
The cluster located in the right temporo-occipital region (peak at around 400 ms) and in the left parietal regions ($500 ms) are harder to interpret, but could be related to anticipated movement planning (Manuel et al., 2010).
Previous studies have not identified specific brain areas responsible for governing proactive inhibition. Most of the evidence suggests that proactive inhibition acts through the modulation of the reactive inhibition network, pre-SMA and SMA being the most likely areas tuning its activation threshold (Stuphorn & Emeric, 2012). A recent fMRI study with dynamic causal modelling (DCM) analyses attempted at differentiating among reactive and proactive inhibition (Zhang & Iwaki, 2019). A network involving right DLPFC, left caudate, and right IFG was found to be more specifically related to proactive inhibition.
On the contrary, a network involving IFG -SMA -STN -M1 was a common pathway shared by both inhibition modalities. Although not conclusive, our data seem to confirm the central role of the indirect pathway in proactive inhibitory control, the direct and hyperdirect pathway possibly intervening at a later stage, releasing inhibition immediately before the go/no-go signal if chances of stopping are low.
Our task was designed to be easily understood and performed by patients, and to specifically unveil a prolongation of reaction times linked to the presence of proactive inhibition. It consistently succeeded in capturing this effect in all groups, even though intergroup differences in terms of duration of disease, dopaminergic medication and cognition wereas expectedpronounced in our cohort, and motivated us to analyse groups separately. Despite this premise, several limitations must be acknowledged. In our experimental design, intervention was limited to turning OFF and ON only one stimulation side, the left one in all cases but one. This might have limited our capacity to find relevant behavioural effects of stimulation. However, this choice was dictated by several considerations. We wanted to limit the discomfort caused by the off state, particularly in STN patients; this was also instrumental for keeping populations comparable, as GPi stimulation is often particularly effective on tremor and dyskinesias (Williams et al., 2014) but is less effective on bradykinesia and rigidity, and allows for a lesser reduction of dopaminergic medication (as confirmed in our cohort, see Table 1): a bilateral OFF is milder in GPi-DBS than in STN-DBS patients.
Another potential limitation is that our EEG analyses were timelocked to the cue presentation. Interesting information might be retrieved from analyses of the response-locked ERPs. However, the signal-to-noise ratio of the response-locked signal might be unbalanced given the short delay between the Go signal and the response, and the variable response times depending on the Cue.
The similar behavioural results associated with dissimilar ERP results might suggest unreliable EEG results. However, to our view, this phenomenon is better explained by different brain processes, either due to different neurophysiological mechanisms, or due to different strategies. Regarding the ERPs, the NO group, for which the duration of the disease is much shorter than the other groups, and which does not have DBS implanted, shows average ERPs (Figure 3) very similar to what might be observed in healthy participants (Angelini et al., 2016;De Pretto, Hartmann, Garcia-Burgos, Sallard, & Spierer, 2019). In the STN group, the ERP components are less pronounced (see Figure 4). However, the topographic maps indicate clear periods of stable brain states.
Results regarding the GPi population should be interpreted with particular caution. This group was difficult to recruit, given that this target is rarely chosen for implantation. Moreover, among the reasons to choose GPi, some cognitive and psychiatric considerations might have been taken into account at the time of surgery, which may influence both the intergroup and the GPi within-group subject variability ( Figure 6). Thus, the absence of a Cue effect in the GPi group may be related to a lack of power due to the low number of participants, as illustrated by the noisy average ERPs.
Finally, the small number of trials is probably the main limiting factor affecting the robustness of our results. A pilot phase with longer experiment duration was rapidly interrupted when it was clear that patients fatigue grossly impaired performance, particularly in the GPi group.For most of our patients, the task was performed with the dominant hand (the right one in all cases but one), while turning on and off the stimulator only on the contralateral side. This means that the right STN was for most of the subjects always stimulated.
The inhibition network is thought to be at least partially right lateralized (Aron & Poldrack, 2006;Lofredi et al., 2020). However, recent experimental work shows that the effects on inhibition are only observed when bilateral stimulation is active (Mancini et al., 2019).

| CONCLUSIONS
Our results suggest that proactive inhibitory control has dedicated brain networks, distinct from those governing reactive inhibition.
Together with previous evidence, our findings support the hypothesis that the indirect pathway is the main involved circuit, and that the hyperdirect pathway has likely a secondary role. Further studies are needed to confirm this interpretation, also considering that our electrical neuroimaging approach is relatively novel in this field and needs further confirmatory studies.
Despite being remarkably effective on the motor symptoms of Parkinson's disease, both STN and GPi stimulation do not appear to interfere with the mechanisms responsible for adapting the threshold of motor inhibition deployment. Our findings confirm the general safety of DBS, with few effects on executive functions despite the remarkable motor improvement. Future research work will be tasked to investigate if proven behavioural side effects of DBS -mostly caused by electrode misplacement (i.e., hypomania) are accompanied by impairment of proactive inhibitory control.

ACKNOWLEDGMENTS
We thank Dr Diego De Leon Rodriguez, and Dr Simone Canonica for their help in data collection. Dr MDP was supported by the Parkinson Switzerland Foundation. Additional financial support was provided by institutional funding from Fribourg University.

CONFLICT OF INTEREST
The authors declare that they have no competing interests.

ETHICS APPROVAL
The protocol was approved from the local ethics committee (Protocol PB_2016-01384).

PATIENT CONSENT
All patients signed an informed consent according to the Declaration of Helsinki.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to ethical restrictions.