Cognitive functions of the GPe

Abstract

The globus pallidus externus (GPe) is a central nucleus in the basal ganglia. Although traditionally defined as a component of the basal ganglia’s motor circuit, converging evidence from humans, rodents, and non-human primates has established the GPe as a key contributor to the cognitive control of behavior. In this role, the GPe dynamically regulates goal-directed behavior in response to changing environmental demands. In this review, we highlight recent studies demonstrating that the GPe encodes cue and action values, supports behavioral flexibility, and modulates attention, behavioral inhibition, and habit formation. The picture that emerges is of a nucleus that aligns behavior to internal goals by constraining, updating, and optimizing ongoing actions.

Expanding the role of the GPe to cognitive functions

The external globus pallidus (GPe) serves as a key nucleus in the basal ganglia network. Its tonically firing neurons provide continuous inhibition to all other basal ganglia nuclei as well as regions of the thalamus, cortex, and brainstem [1–4]. Since the establishment of the ‘rate model’, the GPe has been characterized as a relay nucleus in the indirect pathway whose activity primarily affects movement [5,6]. Support for the GPe’s role in motor function has come from studies of disease, where GPe neurons exhibit decreased activity in hypokinetic conditions such as Parkinson’s disease [2,7,8] and elevated activity in hyperkinetic conditions such as dystonia and Huntington’s disease [9–12].

As a result, the vast majority of studies on the GPe have focused on its motor functions and understanding its pathological state in disease. This is despite the fact that early studies perturbing GPe activity in vervet monkeys showed several non-motor effects, including limbic and attentional deficits [13]. Thanks to recent genetic and anatomical dissection of the GPe in mice and rats, which has revealed its cell-type heterogeneity and extensive projections within and outside the basal ganglia [3,14–19], there has been renewed interest in studying the GPe’s non-motor functions. These studies have uncovered a wide range of cognitive functions, repositioning the GPe as a central hub of the basal ganglia that plays a nuanced role in goal-directed behavioral regulation.

This review will examine evidence for GPe’s extensive repertoire of cognitive functions, drawing on studies in humans, rodents, and non-human primates, highlighting GPe’s role not merely as a relay but as a regulator of information through the basal ganglia. First, we will highlight GPe’s dual roles in response inhibition by describing how different GPe cell types contribute to reactive and proactive inhibition. We will then describe how value is represented in the GPe and how it is used to guide action selection. Next, we will discuss how GPe recruitment supports behavioral adaptation in response to changes in the animals’ environment. Following this, we will explore the ability of the GPe to regulate attentional processes, including arousal, to optimize behavior and the deployment of cognitive resources. Finally, we will discuss the contributions of the GPe to habit regulation and risky decision-making and how they relate to addiction.

GPe’s dual role in inhibitory control: reactive and proactive inhibition

Arkypallidal pathway and reactive inhibition of ongoing actions

The canonical model of basal ganglia function posits that decreases in GPe activity should suppress movement by disinhibiting basal ganglia output neurons [5,6]. Early optogenetic experiments in mice reinforced this model by demonstrating that activation of D2-expressing medium spiny neurons (D2-MSNs), whose sole target outside the striatum is the GPe, suppressed movement [20]. Since the vast majority of GPe prototypic neurons (defined as GPe neurons that project to the basal ganglia output nuclei) are strongly inhibited by D2-MSNs [21–24], reducing their activity was thought to suppress movement as well. However, when GPe neurons were directly inhibited using optogenetics, the result was a slight increase in movement rather than a decrease [21]. This result contrasts with the expected motor effects of disinhibiting basal ganglia output nuclei and may instead reflect the actions of a direct GPe projection to the mesencephalic locomotor region [25]. Further complicating the role of the GPe in motor suppression is the observation that GPe lesions in macaques have relatively modest effects on movement [26,27]. Overall, studies from mice and macaques have yielded mixed effects on movement following GPe manipulation, underscoring the cellular and functional complexity of the GPe [21,26–29].

Motor suppression has been induced at the level of the GPe in mice, but by stimulation of a non-canonical pathway that extends from arkypallidal GPe neurons back to the striatum [14] (Fig. 1). In rats and mice, arkypallidal GPe neurons make up ~15% of GPe neurons and are characterized by their expression of enkephalin and FoxP2 [3,8,14]. Their projections to the striatum inhibit both D1- and D2-MSNs and have been proposed to convey a general movement-cancellation signal [14,24,30,31]. Consistent with this framework, GPe arkypallidal neurons in rats are strongly active during reactive inhibition, the rapid cessation of ongoing actions in response to external stimuli [31,32]. Arkypallidal neurons are maximally activated during periods of coincident input from the striatum and subthalamic nucleus (STN), producing strong locomotor arrest in mice under this condition [24]. Interestingly, arkypallidal GPe neurons receive substantial striatal input from D1-MSNs [22,33,34], suggesting that GPe-mediated motor suppression might be more tightly controlled by the direct pathway (D1-MSNs) than the indirect pathway (D2-MSNs).

Figure 1: Response properties of GPe subpopulations during proactive and reactive stopping.

Schematic summarizing findings from studies in macaques [43] and rats [31,49] describing arkypallidal [31] and putative prototypic [43] responses during reactive and proactive stopping respectively and the value encoding properties of these GPe cell types [49]. (A): Circuit diagram of arkypallidal projections to the striatum, the primary pathway involved in reactive stopping. (B) (Top): Velocity trace during reactive stopping; movement ceases in response to a stop-cue. (Bottom): Schematized response of GPe arkypallidal neurons, which show transient modulation to the cue (value encoding) [49] and strong activation to the stop-cue contributing to reactive inhibition [31]. Orange represents arkypallidal neurons and the behavior of reactive inhibition. Grey lines represent changes in activity in response to cues with different values. (C): Circuit diagram of prototypic neuron projections to the SNr, the primary pathway involved in proactive inhibition. (D) (Top): Velocity trace during proactive inhibition; movement may slow at cue-onset in preparation for rapid cessation of movement at the stop-cue. (Bottom): Schematized absolute change in GPe prototypic neuron activity during proactive inhibition in response to a negative valence go cue [43] exhibits sustained value modulation [49] and the cessation of modulation after the stop-cue. (A-D): Blue represents prototypic neurons and the behavior of proactive inhibition. Grey lines represent changes in activity in response to cues with different values. Arky – arkypallidal; Proto – prototypic.

Proactive inhibition of future actions

Do the results described above mean that prototypic GPe neurons do not participate in movement inhibition? Not necessarily. Most GPe neurons exhibit some degree of movement-related activity, but their responses are heterogeneous, and individual neurons encode movement poorly [35–39]. Recordings of GPe neurons from mice, macaques, and vervet monkeys performing cued-evoked motor tasks indicate that movement-evoked responses in the GPe are highly sensitive to behavioral context and predictions of future reward [35,36,38,40]. Such context-dependent modulation suggests that the GPe may be more involved in aspects of motor planning rather than in pure motor execution, a notion supported by recent findings implicating the GPe in proactive inhibition.

Proactive inhibition is a cognitive-motor preparatory process in which anticipation of a stop cue slows the initiation of movements [32,41]. A behavioral indicator of proactive inhibition is slowed reaction times on trials when a stop signal is anticipated. In rats, a dynamical systems analysis of GPe activity during a stop-signal task found that proactive inhibition was broadly encoded across the GPe population [42]. The authors found that the GPe network occupied a state space farther from movement initiation when stop signals were anticipated than when they were not, facilitating proactive inhibition. A study in macaques found that putative prototypic activity of the GPe was reduced when animals rejected ‘bad’ objects in a sequential choice task but not during a simple fixation task [43]. Because rejection in the fixation task only required withholding reflexive saccades, this finding suggests that GPe prototypic neurons were involved in proactive rather than reactive inhibition (Fig. 1). Another study in macaques identified a cluster of neurons in the anterior striatum that were selectively activated during rejection in the sequential choice task [44], suggesting that striatal activity inhibits the GPe during proactive inhibition. These results are consistent with the view that the indirect pathway is recruited by higher-order cortical areas to facilitate top-down proactive control over behavior [32].

These findings challenge the traditional view that the GPe functions solely as a relay for motor-suppression signals in the indirect pathway. Instead, they reveal functional specialization within the GPe. Prototypic neurons, which directly influence substantia nigra pars reticulata (SNr) activity, also participate in goal-directed motor planning, as indicated by their preparatory activity during proactive inhibition [43]. In contrast, arkypallidal neurons, which primarily project to the striatum [14], appear to facilitate the termination of ongoing actions, as they are strongly engaged during reactive inhibition [31]. Together, this evidence underscores the diverse and active role of the GPe in behavioral control.

The GPe encodes value signals to guide action selection

Most GPe neurons exhibit some degree of movement-modulated activity, but the degree of heterogeneity in their responses suggests additional computations that extend beyond those of a simple motor structure. This view is supported by findings showing that GPe neurons in macaques and vervet monkeys encode a wide array of information, including contextual information, reward size, and salience [35,38,45]. Although it is tempting to ascribe these different functions to motor, limbic, and associative territories that are topographically organized throughout the basal ganglia [46–48], individual GPe neurons have been shown to encode multiple task-relevant variables, whose representations shift dynamically depending on which variable best supports task performance [35,45].

Recordings from the anterior and ventral regions of the GPe (considered associative and limbic territories, respectively [46–48]) revealed neurons modulated by both motor and reward variables, with no correlation between a neuron's location and its encoding of movement or reward [45]. Indeed, most GPe neurons, regardless of their location within the nucleus, showed cue-evoked responses before movement onset, highlighting the broad involvement of the GPe in motor planning rather than motor execution [35].

Furthermore, cue-evoked responses in the GPe of rats and macaques are not static but change dynamically over the course of an experiment, depending on reward probability and effort predictions, thus signaling the value of a given cue [35,49] (Fig. 1). In rats, both arkypallidal and prototypic neurons encode value, with the strongest and most sustained value encoding expressed in prototypic, parvalbumin+ GPe neurons (PV-GPe) [49]. PV-GPe neurons increased their firing rates in response to cues predicting a future reward, and cues with higher reward probabilities evoked a larger firing rate increase than cues associated with lower reward probabilities. Although arkypallidal GPe neurons also exhibited graded changes in firing rate for cues associated with different probabilities (albeit to a lesser degree than PV-GPe neurons), only PV-GPe neurons maintained elevated firing rates up to the point of reward delivery, suggesting these neurons encode both transient and sustained value signals [49] (Fig. 1).

In macaques, visually responsive GPe neurons in the caudal-ventral territory changed their responses to cues that stably predicted reward outcomes over several days, suggesting these GPe neurons encode stable value representations [50]. The effect was particularly pronounced for visual cues that indicated no reward would be delivered. After 4 days of training, GPe neurons exhibited prolonged inhibition of firing upon presentation of non-rewarded cues, thereby enabling monkeys to suppress saccades to these “valueless” objects [50]. Since an animal's environment contains many consistently ‘valueless’ objects but only a handful of potentially rewarding ones, the GPe likely helps efficiently direct gaze by reducing time spent on irrelevant objects [50].

A critical question that remains is how the GPe uses these different representations of value to support goal-directed actions, a process that remains incompletely understood. The GPe may dynamically adjust the values of action-outcome associations in response to reward history, driven by striatal MSNs, and use this information to guide action selection [51–56]. Indeed, D2-MSNs in rats were recently shown to encode the value of the suboptimal choice and to bias exploration toward it when the current action strategy was unreliable or when uncertainty was high [57]. Alternatively, as computational modeling suggests, a network of GPe neurons and reciprocal connections with the STN may support optimal action selection by computing a normalization signal that ensures only actions with the strongest evidence are selected [58]. In comparison, if the environment contains static values, their representations in the GPe may facilitate rapid rejection of actions toward consistently low-value options [50]. Taken together, these results highlight the pivotal role of the GPe in regulating action selection based on learned values in both dynamic and static environments (Fig. 2).

Figure 2: GPe’s position within value-encoding and attention circuits.

Schematic depicting GPe’s central position within value-encoding and attention circuits that help to optimize action-selection. (Blue): A fast attention circuit involving GPe-cortex and STN-GPe pathways. The GPe-cortex pathway is involved in information filtering, helping to efficiently direct attention and regulate working memory, particularly in response to changing environmental demands [see refs: 69,92-96]. The STN-GPe pathway regulates the automatic execution of actions during periods of heightened cognitive demand [see refs: 77,78,80,82]. (Yellow): A slow learning circuit that signals the value of objects acquired over time [see refs: 51-56]. GPe projections to dopamine neurons play a role in the development of addiction-related behaviors [59,60]. GPe arkypallidal projections back to the striatum regulate action-selection strategy [121]. STN = subthalamic nucleus, DA = midbrain dopamine neurons, SNr = substantia nigra pars reticulata, GPe = globus pallidus externus, RPE = reward prediction error.

GPe’s interaction with the dopamine system and GPe’s encoding of reward prediction error

Recent mouse anatomical work has demonstrated large GPe projections to midbrain dopamine regions, including the substantia nigra pars compacta (SNc) [15,17], the ventral tegmental area [59,60], and the retrorubral field [18] (Fig. 2). The majority of GPe projections to the SNc arise from Lim homeobox 6 (Lhx6)- expressing GPe neurons, a unique subclass of prototypic neuron [15,17]. This pathway may receive input from striatal striosomes, forming a non-canonical indirect pathway that regulates limbic function [61,62]. In mice, detailed mapping of striosome synapses within the GPe revealed that they preferentially target a central GPe subregion that directly targets SNc DA neurons [61]. Optogenetic activation of this non-canonical pathway elevated dopamine levels in the striatum, suggesting that the GPe can influence striatal dopamine tone. This pathway may help to offset the excessive activation of the canonical indirect pathway under conditions of low dopamine, such as in Parkinson’s disease [61]. The GPe also expresses topographically organized, functional D2-class receptors, allowing direct dopaminergic modulation of its activity [63,64]. These findings indicate a tighter functional coupling between the GPe and dopaminergic systems, positioning the GPe to fine-tune dopamine-dependent basal ganglia activity across motor and non-motor domains.

Notably, a subset of slow-pacemaking neurons in the rat GPe was found to encode reward prediction error [49]. Reward prediction error signals in dopamine neurons and their corresponding influence over goal-directed behavior have been well characterized [65,66], but their presence in a subset of GPe neurons was surprising and merits investigation into whether these signals are upstream or downstream of reward prediction error signals in the dopamine system. The identity of these slow-pacemaking reward prediction error neurons in the GPe remains to be clarified, but might include Npr3+ GPe neurons, which project to both motor and nonmotor cortical areas [19,29,49].

GPe activity promotes cognitive flexibility and attentional regulation

Cognitive flexibility

Cognitive flexibility is a resource-intensive process that enables rapid behavioral adaptation in response to environmental changes [67]. It is often studied with tasks that introduce reversals in action-outcome contingencies, requiring animals to update previously established associations [68]. Studies from both vervet monkeys [38,69] and mice [70] implicate GPe circuits in supporting cognitive flexibility following these reversals.

In mice, GPe projections to the parafascicular nucleus of the thalamus were selectively recruited on incorrect trials following contingency reversals [70]. Optogenetic activation of these projections impaired acquisition of the new contingency by preventing animals from maintaining the newly rewarded choices [70]. In vervet monkeys, recordings from putative prototypic GPe neurons found that their activity increased immediately following reversals in action–outcome contingencies and predicted subsequent choice switching [38,69]. Activity was higher on switch trials than stay trials and correlated with a greater likelihood of behavioral updating [38,69]. After learning the new contingency, GPe activity gradually declined [38,69], consistent with a role in cognitive flexibility, specifically when task rules change. However, unlike in the aforementioned study in mice [70], activation of GPe neurons using low-frequency deep brain stimulation, which has been shown to have an excitatory effect on neural activity [71,72], improved performance in vervet monkeys by facilitating switches following unrewarded outcomes (directed exploration) [69]. The improvement in directed exploration appeared to result from GPe modulation of the dorsolateral prefrontal cortex, as perturbing this signaling with phencyclidine administration reduced directed exploration [69]. Together, these results suggest that the differing effects of stimulation may reflect engagement of distinct GPe circuits with separable functions, both of which contribute to cognitive flexibility.

A significant question that remains regarding GPe’s role in cognitive flexibility is the upstream circuit driving its elevated activity on shift trials following contingency reversals. Surprisingly, the GPe does not appear to inherit this activity from the striatum, as striatal tonically activated neurons exhibit none of the switch-related activity nor any of the decline to repeatedly rewarded choices [38]. In fact, in rats, striatal activity appears to oppose the increase in GPe activity, as D2-MSN activity during directed exploration is elevated [54].

Another candidate is the STN, the primary source of excitation to the GPe [73–75], which is well-suited to counteract striatal inhibition during directed shifts in choice selection (Fig. 2). This is a reasonable role given the known involvement of the STN in conflict, impulsivity, inhibition of ongoing or automatic responses, and attention [76–81]. Computational modeling even suggests that STN-GPe recruitment may guide exploration following contingency reversals [82]. Based on these observations, the STN is a likely candidate for exciting the GPe during changes in action contingencies. Direct cortical inputs to the GPe are another possibility, though a Granger Causality analysis suggests that this is less likely, as the GPe was shown to exert stronger control over the dorsolateral prefrontal cortex than vice versa, and under conditions where the dorsolateral prefrontal cortex exerted greater control over the GPe, directed exploration was impaired [69].

GPe’s influence over attention

There is evidence that the GPe’s influence over cognitive flexibility is mediated by the regulation of top-down attentional circuits. Early theories of attention proposed that attention towards low-certainty options increases following unexpected outcomes [83,84] - events that commonly occur following reversals in action-outcome contingencies [85]. This suggests that GPe involvement in cognitive flexibility may be related to attention. Indeed, computational modeling of GPe activity in vervet monkeys following action-outcome reversals found strong correlations with factors related to attentional and working memory [69]. The finding that the GPe exerts influence over the DLPFC during directed exploration, a center of executive and attentional regulation [86–90], further implicates the GPe in the regulation of top-down attention [69] (Fig. 2).

An interaction between attention and the GPe has long been hypothesized. In early studies, bicuculline injections into the associative territory of the GPe of vervet monkeys reliably produced attentional deficits, with or without accompanying hyperactivity [13,46,48], providing causal evidence that the GPe modulates attentional processes. Consistent with this role, individuals with attention-deficit/hyperactivity disorder (ADHD) showed reduced connectivity between the GPe and prefrontal areas [91], suggesting that perturbations in GPe-frontal circuits impair attentional processes. Indeed, human fMRI work has proposed that the globus pallidus acts as a filter, limiting distracting information from reaching the prefrontal cortex and thereby supporting working memory capacity [92]. This aligns with working memory models in which the indirect pathway functions as an attentional gating mechanism that determines which information is admitted into working memory [93,94].

There has been comparatively less work in rodents on the role of the GPe in attentional processing. However, a study in mice found that the GPe mediates aspects of top-down attentional filtering driven by the prefrontal cortex [95]. Reductions in GPe activity improved auditory discrimination and performance during a no-go task [95]. In a separate study, pan-neuronal GPe inhibition was found to improve working memory [96], a process that also involves the prefrontal cortex. Together, these results highlight important contributions of the GPe to higher-order cortical processes that include attention and working memory. However, many questions remain about the cell types and pathways that mediate these effects, and about whether the GPe also helps direct attention in more naturalistic contexts.

Role of the GPe in arousal

With respect to attention, arousal modulates animals’ receptiveness [97,98] and selectivity [99,100] to environmental stimuli, and changes in arousal affect task performance and engagement [101,102]. In mice, the GPe’s role in arousal appears to be mediated by prototypic GPe neurons (PV+) located in the anterior portion of the GPe [103]. Specifically, chemogenetic inhibition of anterior GPe PV neurons or activation of D2-GPe axons reduced time spent in the wakeful state [103]. These results demonstrate that anterior GPe neurons can modulate an animal’s arousal state, providing a theoretical mechanism for the known modulation of GPe activity by task engagement [104]. The specificity of these functions to the anterior portions of the GPe and striatum is particularly intriguing, given that prefrontal cortical areas in both macaques [105] and mice [106,107] preferentially innervate the anterior striatum. Indeed, evidence indicates that connectivity between the GPe and prefrontal areas is weakened in people with arousal impairments associated with disorders of consciousness [108]. Further studies should determine whether the anterior GPe PV+ neurons that regulate wakefulness do so by regulating prefrontal cortical areas.

Extending findings from rodents, studies in vervet monkeys show a correlation between changes in arousal and GPe dynamics. Specifically, changes in arousal, which can be assessed by pupil size [109–112], coincided with the onset of pauses in GPe activity [113]. Moreover, GPe pause frequency is negatively correlated with motor activity [114], suggesting that the GPe may not be recruited during low-arousal states. The onset of GPe pauses also corresponds to increased exploratory behavior [113]. All told, there is a clear link between the GPe, arousal, and exploratory behavior, and it suggests that some of the GPe’s contribution to behavior may be mediated by its effects on arousal state.

GPe’s role in habitual and risky decision-making

Habitual action selection

Unlike cognitive flexibility, habitual behaviors aim to reduce reliance on resource-intensive cognitive processes, such as attention, by automating action selection [115–119]. Rather than being driven by action-outcome associations, habitual behaviors are outcome agnostic and rely on stimulus-response associations, rendering them resistant to extinction [120]. This resistance provides a behavioral assay for distinguishing goal-directed from habitual action selection. Using this framework, in mice, both GPe arkypallidal neurons [121] and GPe astrocytes [122], a non-neuronal population highly enriched in the GPe [123,124], have been implicated in regulating the transition between goal-directed and habitual behavior. Specifically, ablation of arkypallidal neurons promoted habitual nose-poking during extinction. This result appears to conflict with the finding that arkypallidal neurons are more active during the acquisition of habitual, rather than goal-directed, actions [121]. However, ablation of GPe arkypallidal neurons also resulted in increased cFOS expression in the dorsolateral striatum [121], an indicator of recent neural activity [125] in the territory of the striatum associated with habit formation and expression [119,126]. Conversely, no increase in cFos expression was observed in the dorsomedial striatum [121], which is involved in goal-directed behavior [119,127]. Since nose-poking during the habit formation phase is disadvantageous, the authors suggest that GPe arkypallidal neurons suppress habit formation by inhibiting the dorsolateral striatum—a hypothesis supported by the fact that activation of GPe arkypallidal neurons drove a global reduction in nose-poking [121]. Together with the evidence discussed earlier that GPe arkypallidal neurons inhibit striatal ‘go’ signals during reactive inhibition [31], these results support the idea that GPe arkypallidal neurons facilitate goal-directed cessation of actions.

In comparison, GPe astrocytes, which regulate GPe neuronal activity [128], show increased activation during goal-directed behavior and can promote the transition from habitual to goal-directed strategies [122]. These non-neuronal GPe cells also exhibit elevated activity in response to bottom-up attentionally salient stimuli, exposure to which can drive a shift from habitual to goal-directed action selection [122,129]. These results suggest that GPe astrocytes receive bottom-up attentional salience information and use it to influence GPe neuronal activity, thereby promoting goal-directed behavior. This further indicates that the relationship between the GPe and attention is significant in regulating goal-directed behavior. In addition, it suggests that the GPe’s connection to attention may not be limited to top-down attentional networks but also extends to bottom-up networks. Future studies should investigate how bottom-up salience influences specific GPe neuronal cell types, as this will indicate the downstream circuits that allow GPe to translate attentional salience into goal-directed action selection (see Outstanding Questions).

Outstanding Questions.

• What is the functional role of GPe’s projection to midbrain dopamine areas? How does GPe’s projection to the SNc influence dopamine release in the striatum, and how does this affect locomotion and reward-based learning? Does the GPe’s projection to the RRF act as a gateway for the GPe to be able to influence limbic-associated behaviors?

• Does the activity of GPe prototypical neurons reflect task difficulty? Does the GPe regulate cognitive resources based on task demand?

• How does salient or unexpected outcome information get conveyed to the GPe, and how does the GPe then modulate cortical attentional processes to support task performance?

• How does astrocyte regulation of GPe neuronal activity in response to attentionally-salient stimuli affect the receptiveness of GPe to excitatory inputs from the STN?

• Is the STN the source of excitation in the GPe during ‘shifts’ in action selection following unexpected outcomes? If so, does the STN input to the GPe provide cognitive control signals to the GPe that prevent the execution of automatic actions and allow for top-down attentional processes to be recruited to help optimize action selection?

• In rodents, are dynamic and static values encoded in the GPe based on topographic input from the striatum? Particularly, are dynamic values used in cognitive flexibility encoded and used by rostral portions of the GPe, and are stable representations used for more habitual behaviors encoded in the caudal GPe, as primate evidence has suggested?

• Does loss of proper GPe signaling play a role in conditions with significant attentional control impairments, such as OCD and ADHD?

Since maladaptive habits contribute to several neuropsychiatric conditions, including addiction and obsessive-compulsive disorder (OCD) [130,131], dysregulation of GPe activity may play a role in their pathophysiology. Consistent with this idea, drug-seeking behavior is often habitual and associated with elevated DLS activity [119,132]. Given that loss of GPe arkypallidal neurons leads to DLS hyperactivity, it is possible that impaired GPe regulation of the DLS biases behavior toward maladaptive, habit-based action selection, as seen in addiction (Fig. 2).

Addictive substances increase the excitability of midbrain dopamine neurons, leading to strong reinforcement of drug consumption and addiction [133–137]. Studies in mice have found that PV-GPe projections to ventral tegmental area neurons are causally involved in driving elevated dopamine neuron activity following cocaine exposure [59,60]. This effect is driven by PV-GPe hyperactivity, which disinhibits VTA dopamine neurons via the SNr [59,60]. PV-GPe hyperactivity results from a combination of reduced inhibition from D2-MSNs and decreased M-currents through KCNQ channels [60]. Normalization of PV-GPe activity reduced cocaine intake, suggesting a causal link between GPe activity and drug-related behavior [59,60]. These findings position the GPe as a key regulator of addiction-related processes and a potential therapeutic target, with arkypallidal neurons shaping habits and PV-GPe neurons regulating addiction circuitry (Fig. 2).

Risky decision making

Risky decision-making is another form of maladaptive action selection characterized by a predilection towards high-risk/high-reward actions. It is also a common side effect of dopamine replacement therapies administered to patients with Parkinson’s disease, such as the D2/D3 agonist pramipexole [138–140]. A recent study in mice found that pramipexole's affinity for D2/D3 receptors on D2-MSNs led to GPe hyperactivity and increased risk-taking [141]. Critically, when GPe’s hyperactivity was attenuated, the increase in risky decision making was ameliorated [141], causally implicating the GPe in this form of maladaptive decision making. Outside of Parkinson’s disease, risky decision-making is common in addiction, where drug-seeking behavior leads to an increased likelihood of making risky decisions [120,142,143], further implicating the GPe in addiction-related behaviors.

Concluding remarks and future directions

The canonical model of the GPe as a homogeneous relay of the indirect pathway involved in motor suppression needs to be refined, as recent findings across species have revealed that the GPe plays a much broader role in goal-directed behaviors. A common theme among these studies is that the GPe is selectively recruited during periods of increased attentional and cognitive demands. The picture that emerges is one in which the GPe functions to optimize behavior by regulating cognitive control over action-selection. Each of GPe’s heterogeneous cell populations appear to play a unique role in optimizing behaviors.

Canonical prototypic neurons appear to facilitate flexibility by influencing prefrontal top-down attention following unexpected events [69]. Astrocytes respond to salient bottom-up stimuli, facilitating the transition from habitual to goal-directed action selection [121]. Finally, arkypallidal neurons facilitate the cessation of ongoing actions [31,122] in response to signals from the environment. These populations work synergistically to allow for productive and efficient deployment of cognitive resources. Pathologies or addictive substances that can perturb this finely tuned system result in impulsivity [139–141,144] and an inability to effectively respond to changes in the environment [121,122]. Collectively, these findings depict the GPe as a nuanced and heterogeneous nucleus that plays a major role in regulating cognitive and attentional control over behavior.

Highlights.

• The GPe is a heterogeneous, tonically active hub with widespread projections within and beyond the basal ganglia.

• Distinct GPe cell types support complementary roles in behavioral control, including reactive and proactive inhibition.

• The GPe encodes value and integrates contextual information to guide goal-directed action selection.

• GPe circuits interact with dopaminergic, thalamic, and cortical systems to regulate cognitive flexibility, attention, and arousal.

• Dysregulation of GPe activity contributes to maladaptive behaviors, including impaired flexibility, habitual responding, and addiction.