Deep brain stimulation and recordings: Insights into the contributions of subthalamic nucleus in cognition

Recent progress in targeted interrogation of basal ganglia structures and networks with deep brain stimulation in humans has provided insights into the complex functions the subthalamic nucleus (STN). Beyond the traditional role of the STN in modulating motor function, recognition of its role in cognition was initially fueled by side effects seen with STN DBS and later revealed with behavioural and electrophysiological studies. Anatomical, clinical, and electrophysiological data converge on the view that the STN is a pivotal node linking cognitive and motor processes. The goal of this review is to synthesize the literature to date that used DBS to examine the contributions of the STN to motor and non-motor cognitive functions and control. Multiple modalities of research have provided us with an enhanced understanding of the STN and reveal that it is critically involved in motor and non-motor inhibition, decision-making, motivation and emotion. Understanding the role of the STN in cognition can enhance the therapeutic efficacy and selectivity not only for existing applications of DBS, but also in the development of therapeutic strategies to stimulate aberrant circuits to treat non-motor symptoms of Parkinson's disease and other disorders.


Introduction
With targeted interrogation using invasive and non-invasive brain stimulation and imaging techniques, researchers and clinicians have provided insights into the complex functions of the basal ganglia in humans. The therapeutic use of deep brain stimulation (DBS) to interact with pathological neural circuits has been one of the most important advances in neuroscience and has contributed greatly to our understanding of the basal ganglia ( Eisinger et al., 2019a ). DBS is a neurosurgical procedure that involves targeted implantation of electrodes in the brain and the delivery of electrical pulses from an implanted pulse generator for the treatment of neurological and psychiatric disorders . In this review, we will focus on the subthalamic nucleus (STN) and the novel insights provided by DBS into its role in cognition. While the STN is traditionally considered a critical node in the cortico-striatal-thalamo-cortical motor loops for modulation of motor functions ( Alexander and Crutcher, 1990 ), the cognitive and affective side effects seen with STN DBS (see Barboza e Barbosa and Fichman, 2019 ;Fasano et al., 2012 ;Kim et al., 2015 for reviews) have fueled the recognition of its role in cognition. This highlights the tight coupling between clinical and basic neuroscience, especially in the context of DBS, with the knowledge gained in one field informing and translating to the other. Abbreviations: Striatum (Str); subthalamic nucleus (STN); external segment of the globus pallidus (GPe); internal segment of the globus pallidus (GPi); substantia nigra pars reticulata (SNr). B) Anatomical organization of the dense and partially overlapping projections from various regions of the cortex to the STN. Colored meshes represent the projections from; ventromedial prefrontal cortex/orbitofrontal cortex (vmPFC/OFC), anterior cingulate cortex (ACC), dorsal prefrontal cortex (DPFC), supplementary motor areas (SMA), and motor cortex (M1). Coronal view of (i) anterior, (ii) central, and (iii) posterior STN. (iv) Axial, superior view of the same projections. Scale bar, 1 mm. Reused with permission from Haynes and Haber (2013) .
The classical models of the basal ganglia consist of parallel pathways that work in concert to modulate motor output through the thalamus, controlling the initiation and inhibition of movement ( Alexander and Crutcher, 1990 ). Thus, the STN has a traditional role in motor function by counteracting the direct pathway via activation of the hyperdirect and indirect pathways in order to modulate net motor cortical activity ( Fig. 1 A). While the STN is traditionally viewed as modulating motor output (via the motor loop) by virtue of its role in the above-mentioned circuits, there are also associative and limbic loops ( Alexander and Crutcher, 1990 ). These distinct parallel loops suggest that the direct, indirect and hyperdirect circuit functions proposed for movement are also relevant to non-motor control and support the role of the STN in cognitive processing.
The distinct parallel loops originate from functionally distinct regions of the cerebral cortex, pass through distinct regions of the STN, and modulate different areas of the thalamus, providing feedback to functionally distinct cortical regions ( Ma and Geyer, 2018 ). The human STN has been divided into three subterritories: a dorsolateral region that connects with sensorimotor pathways, a ventral region that connects with the associative pathways, and a medial region that connects with limbic pathways ( Parent and Hazrati, 1995 ). There is, however, a graded organization in STN afferents, both for the cortical ( Fig. 1 B; Haynes and Haber, 2013 ) and pallidal ( Karachi et al., 2005 ) inputs in a partially overlapping pattern that suggests both parallel processing and integration of information (see Simonyan, 2019 for review). Thus, the available evidence suggests that these regions and pathways are organized along a continuum rather than having distinct borders. With movement as the only means by which we as humans can interact with the world, the STN is anatomically connected and positioned perfectly to balance cognitive control from frontal cortical regions ( Aron et al., 2016 ;Kelley et al., 2018 ;Zavala et al., 2018 ) and action control from the basal ganglia ( Haber and Behrens, 2014 ).

Subthalamic nucleus deep brain stimulation
STN neuromodulation for the treatment of Parkinson's disease (PD) is now an established therapy, and over the past few decades has provided insight into the role of this once mysterious deep structure. While DBS is an invasive neurosurgical procedure, it is considered non-lesional. DBS uses an interventional approach by first inserting an electrode into a specific target, then determining the stimulation location (i.e. specific electrode contact[s]) and parameters that optimize benefit while minimizing adverse effects. Clinically, this results in substantial improvement in motor symptoms -primarily appendicular bradykinesia, tremor, rigidity, on-off motor fluctuations and dyskinesia ( Deuschl et al., 2006 ;Hamani et al., 2005 ). The effects of STN DBS on axial symptoms, balance, and gait remain controversial ( Ramirez-Zamora and Ostrem, 2018 ) and there is ongoing debate as to whether it results in cognitive decline (see Cernera et al., 2019 for review). The literature collectively indicates the importance of controlling for confounding factors when analyzing the cognitive effects of DBS, including but not limited to: follow-up times, surgical techniques, post-operative management, volume of tissue activated, electrode trajectory, and cognitive battery used (see Cernera et al., 2019 for review). Furthermore, the effects of STN DBS on cognitive functions may vary with different subgroups of PD patients (e.g., age, cognitive impairment, motor impairment, rate of progression, laterality) and should be considered when selecting suitable candidates for DBS ( Kehagia et al., 2010 ;Lawton et al., 2018 ). STN DBS has been used by neuroscientists to study the underlying physiology related to STN function since the implanted electrodes can record neural activity. Initially, these recordings were only possible for a limited time prior to internalization of the pulse generator, either during the surgery or for 1-5 days post insertion of DBS electrodes when the electrode leads were externalized. However, the recent development of implantable devices capable of bidirectional communications, such as the Activa PC + S and Percept PC devices manufactured by Medtronic ( Rouse et al., 2011 ), has expanded the window of recording from DBS electrodes. These long-term sensing capabilities not only provide insight into basic circuit functions but also enable scientists to determine how the aberrant neural signatures that correlated with disease symptoms (e.g., beta band (~20 Hz) power; Kühn et al., 2006 ) change over time with treatment and disease progression ( Little et al., 2016 ;Rosa et al., 2015 ).

Clinical effects of deep brain stimulation on cognition
This section will review data drawn from studies of PD, dystonia and obsessive compulsive disorder (OCD), though the majority is from the PD population for which the STN is a primary target for DBS treatment. The symptoms that improved, those that did not improve, and adverse effects as a result of DBS have informed us about the role of the STN. Given the broadly defined regions within the STN (i.e., sensorimotor, associated, and limbic), the location of the stimulation electrode and the volume of tissue activated resulted in different motor and non-motor effects ( Nguyen et al., 2019 ). The dorsolateral STN is generally considered the optimal target for STN DBS in PD ( Dembek et al., 2019 ;Herzog et al., 2004 ;Horn et al., 2017 ), resulting in improved motor symptoms. However, the non-motor effects of stimulating this target location have been inconsistent, specifically the side effects related to inhibitory and cognitive control ( Castrioto et al., 2015 ;Rossi et al., 2017 ;Fasano et al., 2012 ;Mirabella et al., 2012 ;Merola et al., 2014 ;Leimbach et al., 2020 ) The relationship between impulsivity and STN DBS is still under investigation, with conflicting results showing improvement, worsening, and no change ( Amami et al., 2015 ;Castrioto et al., 2015 ;Kim et al., 2013 ;Lhommée et al., 2012 ;Moum et al., 2012 ;Rossi et al., 2017 ). PD patients receiving high frequency continuous STN DBS have shown deficits in behavioral control related to impulsivity and lack of inhibition compared to the off-stimulation state ( Fasano et al., 2012 ;Jahanshahi et al., 2015 ;Paliwal et al., 2019 ;Zavala et al., 2015 ). However, several studies have shown the opposite effects, demonstrating improved stopping with bilateral STN DBS ( Mirabella et al., 2012 ;Swann et al., 2011 ;van den Wildenberg et al., 2006 ). This variability across studies may be attributed to the heterogeneity in PD patients, the role of dopaminergic medications or unintended, off-target stimulation ( Eisinger et al., 2019b ).
A review examining the effect of STN DBS on cognition in PD patients showed that most studies which evaluated global cognitive function found no significant change following STN DBS, while only three studies reported an overall decline ( Barboza e Barbosa and Fichman, 2019 ). When examining specific components of cognitive function including memory, executive function, perception and attention, language, the only components which consistently showed decline following STN DBS were verbal fluency, working memory, planning and cognitive flexibility ( Barboza e Barbosa and Fichman, 2019 ).
While there are decrements in verbal fluency after STN DBS, they appear to be short-term and recover over time to baseline levels ( Aono et al., 2014 ;Lefaucheur et al., 2012 ;Yamanaka et al., 2012 ). This is in line with a recent study which found that DBS surgery led to impairment in verbal fluency and that acute stimulation did not result in further verbal fluency changes ( Leimbach et al., 2020 ). However, this is complicated by reports of subtle effects of stimulation frequency and location of tissue activation (i.e., sub-region somatotopy) on verbal fluency ( Mikos et al., 2011 ;Wojtecki et al., 2006 ). Specifically, clinically used high frequency 130 Hz stimulation reduced verbal fluency compared to 10 Hz stimulation , and activation of the non-motor ventral subregion of the STN negatively impacted verbal fluency compared to activation of the central or dorsal motor subregion of the STN ( Mikos et al., 2011 ).
Compared to the GPi as the DBS target in PD, the STN is associated with a potentially higher or equal risk of cognitive declines ( Boel et al., 2016 ;Follett et al., 2010 ;Odekerken et al., 2015 ;Ramirez-Zamora and Ostrem, 2018 ;Weaver et al., 2012 ). Contradictory to this conclusion, a long term follow-up ( > 5 years) of PD patients undergoing STN DBS revealed no increased risk of cognitive decline beyond that expected by natural history, even in PD patients with mild cognitive impairment ( Merola et al., 2014 ). Furthermore, a study of a large cohort of medically refractory isolated dystonia patients treated with STN DBS reported no cognitive decline from baseline to 12 months post-surgery ( Ostrem et al., 2017( Ostrem et al., , 2011. Together, the heterogeneity across studies related to cognitive functions supports the idea that certain factors such as medications, disease pathology, genetic factors, co-morbidities, DBS location and setting may be important and that STN DBS alone is not strongly implicated in cognitive functions ( Barboza e Barbosa and Fichman, 2019 ). These findings also relate to the fact that the complex effects of DBS are currently not fully understood. DBS likely modulates both local and widespread dynamic network activities and plasticity which disrupts pathological brain circuits ( Hamani et al., 2017 ;Jakobs et al., 2019 ;Lozano et al., 2019 ;Udupa and Chen, 2015 ).

Action inhibition
There is growing evidence implicating the STN as a critical neural structure in situations which require the stopping or withholding of movements ( Zavala et al., 2015 ). Single unit recordings in humans have revealed increased activation of the STN during response inhibition . Local field potential (LFP) recordings from the STN during the performance of a stop-signal task has repeatedly shown that movement inhibition is associated with increased power of beta oscillations (10-30 Hz) ( Alegre et al., 2013 ;Benis et al., 2014 ;Kühn et al., 2004 ;Ray et al., 2009Ray et al., , 2012Wessel et al., 2016a ). Fig. 2 shows an example of the difference in STN LFP event-related spectral power of successful stop versus failed stop trials, revealing an increase in beta power prior to the estimate of the behavioral stop onset ( Wessel et al., 2016a ). The increased STN beta power associated with successful movement-inhibition appears to result in broad and global inhibition of both involved and uninvolved movement effectors ( Badry et al., 2009 ;Majid et al., 2012 ). For example, increased STN beta power associated with successful verbal-inhibition positively correlated with the degree of corticospinal output suppression to an uninvolved hand ( Wessel et al., 2016a ).
This increase in STN beta power has also been found in non-motor inhibition, for example, in response to cues indicating inhibition of future responses or when deciding whether or not to remember someone's name ( Oswal et al., 2012 ;Zavala et al., 2017 ). Specifically, LFP and spiking activities in the human STN during a selective memory encoding task revealed increased beta power when preventing the encoding of an item into working memory ( Zavala et al., 2017 ). Furthermore, the increase in STN beta power triggered by surprising auditory stimuli positively correlated with the disruption of working memory ( Wessel et al., 2016b ).
Interestingly, the timing of changes in beta activity during motor and non-motor inhibition appears to be similar ( Kühn et al., 2004 ;Ray et al., 2012 ;Zavala et al., 2017 ), suggesting the underlying mechanisms are similar and the STN plays a role of in global inhibition.
In addition to beta, theta oscillations (4-8 Hz) in the STN have also been linked to inhibition, specifically during tasks with a high degree of conflict ( Ghahremani et al., 2018 ;Wessel et al., 2019 ;Zavala et al., 2018Zavala et al., , 2013. In general, these conflict tasks require suppression of a prepotent response. Examples include the Eriksen flanker task ( Eriksen and Eriksen, 1974 ), Stroop task ( Stroop, 1935 ), Simon task ( Simon and Rudell, 1967 ), and Arrow task ( Zavala et al., 2018 ). LFP recordings from PD patients during the performance of a Eriksen flanker task showed that failure to ignore the incongruent flanking arrows, as indicated by prolonged reaction time, corresponded to impaired cue-locked theta phase alignment compared to congruent and successfully ignored incongruent trials ( Zavala et al, 2013 ). The cue-locked theta phase alignment was consistently followed by an increase in theta power, with slow incongruent trials having significantly elevated power levels compared to congruent and fast-incongruent trials. The authors suggest that the onset of theta synchronization subsequently sets the absolute amount of theta power required before a response can be triggered -i.e., response threshold ( Zavala et al., 2013 ). In other words, conflict delays theta synchronization resulting in a higher response threshold, hence a longer reaction time compared to non-conflict situations in which theta synchronization is facilitated resulting in a lower threshold and faster reaction time. The link between theta power and response threshold is further supported by the findings of increased theta power following high conflict trials and higher pretrial theta power corresponding to slower reaction times during conflict . This interpretation is consistent with current models of response selection under conflict, where the STN is afforded a critical role in adjusting the response threshold ( Cavanagh et al., 2011 ;Frank, 2006 ).
The role of the STN in resolving conflict is thought to be tightly coupled to the medial prefrontal cortex (mPFC) ( Cavanagh et al., 2011 ;Zavala et al., 2016Zavala et al., , 2014. Accordingly, mPFC theta frequency band (4-8 Hz) activity increases during conflict and correlates with conflictinduced fluctuations in reaction time with high conflict trials. STN DBS reversed this relationship resulting in impulsive, erroneous choices ( Cavanagh et al., 2011 ). Concurrent STN LFP and midline frontal EEG recordings in PD during dot motion and flanker conflict tasks revealed an increase in the theta-delta band (2-8 Hz) coherence between the two structures that was specific to high-conflict trials ( Zavala et al., , 2014. Of note, Granger causal analysis indicated that cortical oscillations over the midline frontal cortex drive those in STN during conflict ( Zavala et al., 2014 ). Together, these findings support the importance of mPFC-STN connectivity in the theta range during conflict and inhibitory processing, by which mPFC signals the STN to inhibit or delay response output by increasing the response threshold.
While these studies implicate the STN in inhibitory control and in conflict processing, they do not prove causation. To address this issue, Ghahremani and colleagues (2018) used a novel method of event-related DBS to deliver brief trains of STN-stimulation (130 Hz, 100 μs pulse width, ~80 ms duration) time locked to specific events during a Stroop task in PD patients, which created conflict between the color of the ink and the color that the word spells ( Stroop, 1935 ). Similar to previous studies, they found a significant increase in theta power prior to the response in conflict trials compared to non-conflict trials. Interestingly, high frequency stimulation at the time of elevated theta power resulted in increased errors during conflict trials ( Fig. 3 ). The disruptive effect of high frequency event-related DBS at the time of conflict related increase in theta power indicates that the low frequency oscillations in the STN for conflict related processing are functionally significant and are possibly related to setting the decision threshold.
While the evidence to date clearly implicates the STN as a critical structure in motor and non-motor inhibition, it is recognized as an important node in the fronto-subthalamic circuit which is proposed to sup-port inhibitory control (see Aron et al., 2016 for review). Given the location of the STN within the basal ganglia and its diffuse connections with cortical and subcortical brain areas, it is well suited to act as a gate and regulate motor and non-motor cognitive processes such as memory. While the exact time that STN mediates the stopping process is still unknown, there is evidence for both early and late STN-mediated inhibition in the stopping process ( Boucher et al., 2007 ;Schmidt and Berke, 2017 ;Wessel and Aron, 2014 ). Basal ganglia recordings from mice during a stop-signal task suggested that stopping involved pausethen-cancel subprocesses, with the STN mediating an early pause in behavior rather than fully canceling it ( Schmidt and Berke, 2017 ). Alternatively, modeling based on frontal eye field recordings from monkeys during a countermanding stop task suggested a late but potent interruption of behavior ( Boucher et al., 2007 ). A recent set of studies in healthy humans supports the latter, providing evidence of a temporal cascade of action-stopping processes beginning with activation of the right prefrontal cortex at approximately 120 ms after the stop-signal followed by activation of the STN and basal ganglia circuitry to inhibit motor output ( Jana et al., 2020 ). Techniques such as event-related DBS may be useful in elucidating the specific timing of STN activation in humans, as a recent optogenetic study in mice revealed that brief activation of the STN was sufficient to interrupt or pause self-initiated movement, while STN inhibition attenuated the ability to inhibit behavior ( Fife et al., 2017 ).

Decision making
A central part of behavioral control is decision making and optimizing it based on task goals. Given the role of the STN in inhibitory control, it is not surprising that it may also play a role in decision making. For a response to be selected, the alternatives must be inhibited. There are several computational models which attempted to account for the neurophysiology underlying decision making, most notably those based on drift diffusion models ( Gold and Shadlen, 2007 ;Ratcliff and McKoon, 2008 ). These models have two main components, evidence accumulation and decision threshold. As the evidence accumulates over time, it eventually reaches the decision threshold and the response is made. Two key computational models indicate that cortical areas accumulate evidence for response options ( Bogacz and Gurney, 2007 ;Frank, 2006 ). Based on the options, the STN sets the decision threshold, and the striatum drives the response option towards threshold based on the strength of the evidence. Accordingly, difficult decisions (e.g. highconflict) are slower than easy decisions due to the STN increasing the threshold, resulting in a delay in all responses but increases the likelihood that the correct response is selected ( Bogacz and Gurney, 2007 ;Frank, 2006 ).
There have been several STN DBS LFP studies that support these models, providing evidence of the functional role of STN and corticosubthalamic network mechanisms during decision making ( Hell et al., 2018 ;Herz et al., 2017 ). In a study by Herz and colleagues (2017) , PD patients performed a perceptual decision making task (dot motion task) in which the investigators manipulated the emphasis for speed ('Fast!') versus accuracy ('Accurate!') via pre-trial instructions and task difficulty with low (difficult) and high (easy) coherence dot cues ( Herz et al., 2017 ). It was found that prefrontal electrode (Fz) -STN low frequency activity (2-8 Hz) phase coupling predicted increased decision thresholds only after the subject was instructed to emphasize accuracy. These observations are in line with the idea that prefrontal cortex increases its influence over STN activity when caution is warranted ( Frank, 2006 ;, leading to increased decision thresholds in order to delay the response . Conversely, the motor cortex (C3/C4) -STN beta activity (13-30 Hz) phase coupling predicted decreased decision thresholds irrespective of instructions but was more strongly modulated during speed emphasis. These modulations in motor cortical-STN network add to the existing literature indicating that decision-related activity occurs in the STN as well as motor cortical areas ( Carland et al., 2019 ; Cisek and Kalaska, 2010 ).

Fig. 3.
A) The contrast between conflict and non-conflict conditions during the performance of a Stroop task shows an increase in low frequency oscillations (LFO: 2-8 Hz). Time 0 represents the response onset. B) Stimulation timing (Mean ± SD) in the pre-response period relative to significant LFO power changes. LFO power significantly increased in the conflict compared to the non-conflict trials in the timing interval that covers the Early, but not the Late stimulation. The vertical dotted lines represent the range of stimulation timing. The grey-shaded area represents the timing of the significant difference between conflict and non-conflict trials extracted from panel A-inset. Republished with permission from Ghahremani et al. (2018) .
The amplitude of STN beta oscillations (13-30 Hz) also changed according to instructions and task difficulty in a time dependent manner ( Herz et al., 2017 ). Early into a given task (150-400 ms after stimulus onset), speed versus accuracy instruction modulated beta power whereas later into the task (after > 500 ms), beta power was modulated according to task difficulty. In a follow-up study using the same task, Herz and colleagues (2018) probed these time specific changes in STN beta oscillations during decision making by delivering event-related DBS (130 Hz) based on beta power amplitude (also referred to as adaptive DBS). With drift diffusion modeling, they found that event-related DBS affected the patients' ability to adjust decision threshold according to task difficulty, but only when stimulation was applied 400-500 ms following the moving dots cue. This temporally confined window of stimulation reduced the slowing of response times during difficult trials without DBS, while stimulation outside this window had no behavioral effect. The temporal specificity of the disruptive effect of high frequency DBS on periods of elevated beta power provides causal evidence for the role of STN beta in modulating response latency as a function of task difficulty. The result supports the hypothesis that the event related DBS interfered with the ability of the STN to set the decision threshold to the required level according to task difficulty. These findings are in line with computational models ( Bogacz and Gurney, 2007 ;Frank, 2006 ) and provide evidence for a specific role of the STN in setting the decision threshold based of initial information gathered on task difficulty, while task instructions appear to alter the rate of evidence accumulation towards threshold.
Of note, the proposed role of the STN in adjusting decision thresholds may be specific to rapid decision-making when it is important to slow the process when conflicting information is present or high reward is at stake. A study demonstrated that STN DBS had no effect on the decision threshold during the performance of a relatively slow ex-panded judgement paradigm which had no conflict, time pressure, or reward ( Leimbach et al., 2018 ). This duality is further exemplified by the contradictory findings regarding the acute effects of STN stimulation on decision making, where some studies found deficit with DBS ( Cavanagh et al., 2011 ;Coulthard et al., 2012 ;Florin et al., 2013 ;Frank et al., 2007 ;Green et al., 2013 ;Oyama et al., 2011 ;Pote et al., 2016 ;Rogers et al., 2011 ) and others reported no change or benefit with DBS ( Boller et al., 2014 ;Brandt et al., 2015 ;Fumagalli et al., 2015 ;Torta et al., 2012 ;van Wouwe et al., 2011 ). Interestingly, a recent study found that STN beta oscillations are involved in a subjective decisionmaking task with a low attentional load and a low degree of conflict ( Al -Ozzi et al., 2020 ). The authors recorded intraoperative single unit and LFP activities in PD patients as they performed a two-choice decision task based on subjective preference of animal pictures. They found significant beta band (13-35 Hz) desynchronization in response to the favorite but not non-favorite pictures, followed by a short burst in beta activity prior to the decision.
Together, the available evidence supports the role of the STN in cognitive decision-making with theta and beta oscillations being critical to its function and likely driven by top-down cortical control based on evidence, precaution, or preference ( Frank, 2006 ;Herz et al., 2017 ;. The STN is well suited to this role given the organization of cortical projections to the STN with signals in multiple modalities allowing the integration of cognition, emotion, and sensorimotor inputs to guide decision-making ( Haynes and Haber, 2013 ).

Motivation and emotion
Beyond motor control and decision making, the STN is involved in motivational and emotional processing. Multiple studies have reported that STN stimulation can induce unusual emotions including euphoria, feelings of merriment, infectious laughter, and hilarity ( Coenen et al., 2009 ;Krack et al., 2001 ;Mallet et al., 2007 ;Mandat et al., 2006 ). Subsequent clinical studies have reported occasional motivational and emotional effects of STN DBS ( Fasano et al., 2012 ;Kim et al., 2015 ;Mallet et al., 2008 ). These observations may be mediated by connections from the STN to limbic structures, including the ventromedial prefrontal, orbitofrontal, anterior cingulate cortices, accumbens nucleus, ventral pallidum, and ventral tegmental area ( Alexander et al., 1986 ;Haynes and Haber, 2013 ;Lambert et al., 2012 ;Parent and Hazrati, 1995 ;Voon et al., 2017 ).
To investigate the role of the STN in emotional processing, several studies presented images from the International Affective Picture System (IAPS) in which two affective dimensions can be manipulated: valence (pleasant-unpleasant) and arousal (low-high) ( Lang et al., 2008 ). In PD patients on dopaminergic medications, the emotional LFP response in the STN at 1-2 s post stimulus showed significantly larger alpha power (8-12 Hz) desynchronization in response to pleasant and unpleasant stimuli compared to neutral stimuli matched for arousal ( Kühn et al., 2005 ). A follow-up study recorded STN LFP in PD patients (on dopaminergic medications) in response to pleasant or neutral visual stimuli ( Brücke et al., 2007 ). The results revealed significantly larger alpha power (7-13 Hz) desynchronization at 1-2 s post stimulus in response to the pleasant stimuli compared to neutral stimuli irrespective of arousal level. Furthermore, the degree of alpha power desynchronization highly correlated with individual valence ratings by patients, but not with arousal ratings ( Brücke et al., 2007 ). Importantly, while these studies focused on the effects of visual stimuli, modulation of STN LFP activities in response to emotional auditory stimuli (happy and angry voices) have also been seen in PD patients on dopaminergic medications ( Péron et al., 2017 ). These findings provide evidence for involvement of the STN in emotional processing related to the valence irrespective of sensory modality in a non-motor context.
To address the effects of motor demands and dopaminergic medications, a study extended the paradigm by Brücke et al. (2007) to include STN LFP recordings in PD patients in the on and off medication states during various motor response context (motor, non-motor, passive) ( Buot et al., 2013 ). Larger event-related potentials were observed with emotional stimuli compared to neutral stimuli, regardless of their relevance to motor response context. Dopaminergic medications only modulated event-related potential amplitude in response to pleasant stimuli, suggesting that the encoding of pleasant information in the STN may be dependent on the dopaminergic system ( Buot et al., 2013 ). The dependence on dopamine is further supported by a study that investigated STN LFP activity in PD and found a double dissociation of the alpha band response depending on dopamine state and stimulus valence in non-depressed patients ( Huebl et al., 2014 ). Specifically, alpha band event-related desynchronization for pleasant stimuli was greater in the on medication state compared to the off medication state, and this pattern was reversed with increased alpha event-related desynchronization for unpleasant stimuli in the off medication state.
The authors also examined the relationship between the amplitude of the event-related potentials and the anatomical location of the recording electrodes, and found the largest response in the ventral electrode contacts with a graded response along the dorsoventral axis ( Buot et al. 2013 ). This observation suggests that emotional encoding of stimuli is primarily located ventrally, which is in accordance with the existence of a limbic territory in the ventromedial part of the STN ( Parent and Hazradi, 1995 ). These findings are in line with highfrequency STN DBS studies targeting the ventral limbic portion which induces an increase in emotional experience ( Benedetti et al., 2004 ;Greenhouse et al., 2011 ;Polosan et al., 2019 ).
Although LFP recordings from DBS contacts have expanded our understanding of the role of the STN, they are limited by low spatial resolution compared to microelectrode recordings. A recent study that analyzed microelectrode recordings combined with image analysis from a total of 933 trajectories in 303 PD patients identified a subarea with high theta-alpha (7-10 Hz) oscillations in the ventromedial STN which overlapped with the limbic STN ( Rappel et al., 2020 ). However, intraoperative single unit recording from the STN in PD patients during affective image presentation revealed two spatially distinct populations of emotion neurons for valence and arousal ( Sieger et al., 2015 ). Interestingly, affective neurons with oscillations in the alpha band were not limited to the limbic region but were also found in the sensorimotor region, with valence-related neurons located more posteriorly and arousal-related neurons located more anteriorly. The identification of emotion-specific neurons in the traditionally motor region of the STN challenges previous findings and supports parallel and integrative processing within the STN ( Simonyan, 2019 ).
Together, the findings from LFP and microelectrode recordings provide evidence for involvement of the STN in dopamine dependant emotional processing related to valence irrespective of sensory modality and motor context. Consistent across studies is the late (1-2 s) STN alpha power desynchronization post stimulus which may reflect processing of the emotional stimuli, and is in line with modulations of cortical alpha power during passive viewing of emotional stimuli ( Popov et al., 2012 ). The central position of the STN in the cortical-basal ganglia circuit and multifunctional domains suggests that it may be an integrative node for emotional processing, possibly mediated by alpha synchronization and desynchronization ( Haynes and Haber, 2013 ;Nambu et al., 2002 ).
While the alpha frequency band appears to be important for emotional processing in the STN ( Brücke et al., 2007 ;Huebl et al., 2011 ;Kühn et al., 2005 ), motivational processing in the STN has been related to the theta band ( Pearson et al., 2017 ;Rosa et al., 2013 ;Zénon et al., 2016 ). STN LFP recordings from PD patients while they had to decide whether to engage in a task based on the level of effort required and the value of the reward promised in return revealed a robust low frequency response (1-10 Hz) to both reward and effort cues ( Zénon et al., 2016 ). Interestingly, the response was informative of the subjective value of the reward and level of effort rather than their actual quantities, such that they were predictive of the participant's decision ( Zénon et al., 2016 ). A similar increase in STN LFP low frequency power (2-12 Hz) was also observed in PD patients with and without pathological gambling during an economic decision-making task ( Rosa et al., 2013 ). In an attempt to dissociate the role of theta oscillations during economic decisions, microelectrode recordings in PD patients off dopaminergic medications were made during the performance of a self-paced risk-reward decision making task ( Pearson et al., 2017 ). The authors found that as opposed to an event-related increase in STN theta activity (4-8 Hz), there was a continuous increase over the course of the trial, favouring the interpretation that theta represents a motivation or urgency signal that tracks the decision process ( Pearson et al., 2017 ). While the evidence to date is limited, the findings suggest that the STN may encode motivation related processes required to make cost-benefit decisions. The STN is well positioned to play such a role, given that it receives direct input from cortical regions such as orbitofrontal, ventromedial prefrontal, dorsolateral prefrontal and anterior cingulate cortices thought to be central to reward based decision making, subjective valuation of rewards and effort processing ( Haynes and Haber, 2013 ;Padoa-Schioppa, 2011 ;Voon et al., 2017 ;Zénon et al., 2015 ).

Limitations of electrophysiological studies
It is important to consider the limitations when interpreting the findings across electrophysiological studies. The research presented was almost entirely conducted on PD patients who had undergone STN DBS surgery. Thus, results may not represent normal functions and neural activities even in studies performed while patients were on their regular medications in an attempt to approximate normal physiological state. Therefore, extrapolation of the results to the healthy population should be made judiciously. Furthermore, the electrophysiological signals collected from PD patients are typically limited to the target region for DBS which is primarily the dorsolateral STN. This constrains our under-standing of the STN, especially when studying the role of the STN in non-motor and cognitive functions. However, the knowledge gained to date from DBS as a probe and modulator of brain circuitry has led to investigations of the therapeutic potential of DBS in a broad range of disorders, including those affecting motor, limbic, memory and cognitive functions which are likely to further expand our understanding of the cognitive role of the STN .
It is also important to consider that LFP is a composite measure of brain activity that reflects multiple neural sources, with several confounding factors to be aware of when interpreting and analyzing the signal ( Herreras, 2016 ). For example, LFP are subject to volume conduction of electrical signals, such that they are a mixture of local potentials of interest and volume conducted potentials from distant sources ( Kajikawa and Schroeder, 2011 ). Mindful of the inherent constraints, studies using DBS and LFP recordings have expanded our understanding of the relation between electrophysiological patterns and the role of STN in cognition.

Conclusions and future directions
We outlined novel insights provided by DBS into the role of STN in cognition. Anatomical, clinical, and physiological data converge on the view that the STN is a pivotal node linking cognitive and motor processes. Anatomically, the sensorimotor, associative, and limbic subterritories of the STN may account for the heterogeneity in the clinical effects of DBS as well as the range of tasks that the STN has been found to be involved in physiologically. While the results aid in the development in a general understanding of STN function, there are still many unresolved questions regarding how information is processed within the STN, the precise timing of processing and how the STN interacts and communicates with other brain structures.
Answering these complex questions will no doubt require technological advancements in the field and continued collaborations between clinicians and basic neuroscientists. For example, the technology and the ability to deliver event or time locked stimulation (e.g., Ghahremani et al., 2018 ;Herz et al., 2018 ) is growing, which will enhance our ability to understand with greater temporal precision the role of the STN in cognitive processes. In addition, DBS technology capable of stimulation and recording with implanted pulse generators (e.g., Activa PC + S and Percept PC, Medtronic Inc.) will allow for continuous, long-term recording of STN LFP to expand beyond the operating room or a few days after insertion of DBS electrodes. This new technology will allow for large and varied recordings which will no doubt provide great insight into the role of the STN in a multitude of real-world cognitive tasks, shaping future therapeutics including the ongoing development of closed-loop and adaptive DBS.

Declaration of Competing Interest
The authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article.