The External Globus Pallidus: Progress and Perspectives

Abstract

The external globus pallidus (GPe) of the basal ganglia is in a unique and powerful position to influence processing of motor information by virtue of its widespread projections to all basal ganglia nuclei. Despite the clinical importance of the GPe in common motor disorders such as Parkinson’s disease, we have only limited information about its cellular composition and organizational principles. In this review, we describe recent advances in our understanding of the diversity in the molecular profile, anatomy, physiology, and corresponding behavior during movement of GPe neurons. Importantly, we attempt to build consensus and highlight commonalities of the cellular classification based on existing but contentious literature. Additionally, we provide an analysis of the literature concerning the intricate reciprocal loops formed between the GPe and major synaptic partners, including both the striatum and the subthalamic nucleus. In conclusion, the GPe has emerged as a crucial node in the basal ganglia macrocircuit. While subtleties in the cellular makeup and synaptic connection of the GPe create new challenges, modern research tools have shown promise in untangling such complexity and will provide better understanding of the roles of the GPe in encoding movements and their associated pathologies.

Introduction

The basal ganglia are an ensemble of subcortical nuclei that are critically involved in control of action (Albin et al., 1995; Graybiel, 2008; Redgrave et al., 2010; Turner & Desmurget, 2010; Costa, 2011; Gerfen & Surmeier, 2011; Kravitz et al., 2012). The external globus pallidus (GPe) of the basal ganglia is in a unique and powerful position to influence processing of motor information by virtue of its widespread projections to all basal ganglia nuclei (Difiglia et al., 1982; Beckstead, 1983; Walker et al., 1989; Kita, 1994; Kita & Kitai, 1994; Shammah-Lagnado et al., 1996; Nambu & Llinas, 1997; Bevan et al., 1998; Smith et al., 1998b; Kita et al., 1999; Sato et al., 2000; Kita & Kita, 2001; Kita, 2007). Consistent with this anatomy, phasic changes in the firing of GPe neurons are associated with both passive and active body movements (DeLong, 1971; Georgopoulos et al., 1983; Anderson & Horak, 1985; Mitchell et al., 1987; Filion et al., 1988; Nambu et al., 1990; Brotchie et al., 1991; Mink & Thach, 1991b; a; Jaeger et al., 1995; Mushiake & Strick, 1995; Turner & Anderson, 1997; Boraud et al., 2000; Arkadir et al., 2004; Adler et al., 2010; Schmidt et al., 2013; Dodson et al., 2015). Highly synchronous bursting in the GPe correlates with hypokinetic symptoms of Parkinson’s disease (PD) (Pan & Walters, 1988; Filion & Tremblay, 1991; Filion et al., 1991; Hutchison et al., 1994; Nini et al., 1995; Rothblat & Schneider, 1995; Hassani et al., 1996; Taha et al., 1996; Bergman et al., 1998; Boraud et al., 1998; Wichmann et al., 1999; El-Deredy et al., 2000; Magill et al., 2000; Magnin et al., 2000; Raz et al., 2000; Brown et al., 2001; Magill et al., 2001; Bar-Gad et al., 2003; Starr et al., 2005; Heimer et al., 2006; Wichmann & Soares, 2006; Kita, 2007; Tang et al., 2007; Zold et al., 2007a; Zold et al., 2007b; Mallet et al., 2008; Starr et al., 2008; Sani et al., 2009; Chan et al., 2011). Similarly, aberrant GPe neuron activity is also observed in Huntington’s disease (HD) and dystonia (Starr et al., 2005; Chiken et al., 2008; Starr et al., 2008; Baron et al., 2011; Nambu et al., 2011; Nishibayashi et al., 2011), arguing for the centrality of the GPe in motor function and dysfunction. Despite its critical role in regulating motor activity, the organization of GPe neurons within the basal ganglia circuitry remains poorly understood, preventing us from understanding how GPe activity is regulated in behavioral and disease contexts.

While the traditionally-held belief is that the GPe is a homogeneous population of neurons that act as a mere relay in the indirect pathway of the basal ganglia (Albin et al., 1989; Alexander & Crutcher, 1990; DeLong, 1990; Albin et al., 1995; Parent & Hazrati, 1995b; Joel & Weiner, 1997; Graybiel, 2000), recent studies are challenging this view. A number of important discoveries on neuron diversity were made in the past few years as a result of a resurgence of interest in the GPe. Furthermore, accumulating evidence suggests that striatal inputs to the GPe do not arise strictly from indirect pathway neurons. In this review, we aim to analyze the historical literature and provide a critical update on the recent progress regarding our understanding of the GPe. We will also discuss the fundamental biology of different GPe neuron classes, their synaptic partners, and their potential importance in motor function and disease etiology.

Heterogeneity of neurons in the GPe

Early evidence for distinct neuron types in the GPe

A large body of published work suggests the existence of multiple GPe neuron types. GPe neurons are diverse in their expression of molecular markers (Hontanilla et al., 1994; Kita, 1994; Bevan et al., 1998; Hoover & Marshall, 1999; Kita & Kita, 2001; Cooper & Stanford, 2002; Chan et al., 2004; Domaradzka-Pytel et al., 2007; Mallet et al., 2012; Mastro et al., 2014; Abdi et al., 2015; Hernandez et al., 2015), dendritic morphology (Fox et al., 1974; Danner & Pfister, 1981; Iwahori & Mizuno, 1981; Difiglia et al., 1982; Park et al., 1982; Francois et al., 1984; Yelnik et al., 1984; Millhouse, 1986; Kita & Kitai, 1994; Nambu & Llinas, 1997; Cooper & Stanford, 2000), axonal projections (Difiglia et al., 1982; Beckstead, 1983; Walker et al., 1989; Kita, 1994; Kita & Kitai, 1994; Parent & Hazrati, 1995b; Shammah-Lagnado et al., 1996; Bevan et al., 1997; Bevan et al., 1998; Smith et al., 1998a; Kita et al., 1999; Sato et al., 2000; Kita & Kita, 2001; Mallet et al., 2012), and electrophysiology (Nambu & Llinas, 1994; Kelland et al., 1995; Nini et al., 1995; Nambu & Llinas, 1997; Cooper & Stanford, 2000; Raz et al., 2000; Paz et al., 2005; Gunay et al., 2008; Mallet et al., 2008; Joshua et al., 2009; Bugaysen et al., 2010; Chan et al., 2011; Chuhma et al., 2011; Benhamou et al., 2012; Mallet et al., 2012; Schmidt et al., 2013; Jin et al., 2014).

Earlier studies were limited by non-categorical expression of a number of phenotypic markers. Analysis of GPe anatomical projections is complicated by the fact that single GPe neurons can project to multiple nuclei (Parent & Hazrati, 1995b; Bevan et al., 1998; Mallet et al., 2012). Both in vivo and ex vivo studies have demonstrated quantitative differences in the electrophysiological characteristics of GPe neurons. The first classification was provided by DeLong’s seminal in vivo recording study in behaving monkeys, in which GPe neurons were divided into “high frequency pausers” and “low frequency bursters” on the basis of their spontaneous firing patterns (DeLong, 1971). Several subsequent studies have provided evidence for the existence of subtypes of GPe neurons according to their electrophysiological properties in ex vivo slices (Kita & Kitai, 1991; Nambu & Llinas, 1994; Cooper & Stanford, 2000; Bugaysen et al., 2010; Chuhma et al., 2011). As the methodologies and measurements were unique to each experimental setup, the field still awaited reliable classification criteria for GPe neurons.

Recent advances delineate distinct neuron classes in the GPe

More recent multidisciplinary studies using adult rodents have provided compelling data that serve as the basis for newer classification schemes for GPe neurons. Except for a small population of cholinergic (ChAT⁺) neurons that make up ~5% of all GPe neurons (see below), GPe neurons are GABAergic and autonomously active (Kita & Kitai, 1991; Nambu & Llinas, 1994; 1997; Cooper & Stanford, 2000; Chan et al., 2004; Surmeier et al., 2005; Mercer et al., 2007; Deister et al., 2009; Bugaysen et al., 2010; Nobrega-Pereira et al., 2010; Chan et al., 2011; Miguelez et al., 2012; Mastro et al., 2014; Abdi et al., 2015; Hernandez et al., 2015). These GABAergic GPe neurons largely fall into one of two general categories. Neurons in the first category—‘prototypic’ GPe neurons—exhibit fast and regular firing rates in vivo (Abdi et al., 2015; Dodson et al., 2015), project strongly to the subthalamic nucleus (STN), and constitute ~70% of all GPe neurons (Mallet et al., 2012; Hernandez et al., 2015). Neurons expressing the calcium binding protein parvalbumin (PV) represent the majority of these prototypic neurons (Mallet et al., 2012), making up ~55% of all neurons in the GPe. Two recent studies show that these neurons express the transcription factor Nkx2.1 (Abdi et al., 2015; Dodson et al., 2015). Additionally, at least a subset of these PV⁺ neurons also express Lhx6 (Abdi et al., 2015; Hernandez et al., 2015) (but see below). The principal electrophysiological characteristics of PV⁺ GPe neurons were recently established in ex vivo slices and include robust and regular autonomous firing, narrower action potentials, and lower membrane resistance, as well as large persistent sodium current, HCN current, and Kv4 current (Mastro et al., 2014; Abdi et al., 2015; Hernandez et al., 2015). The Type I and Type A neurons described in early electrophysiological characterizations of GPe neurons (Nambu & Llinas, 1994; Cooper & Stanford, 2000) share these key characteristics with PV⁺ neurons and likely correspond to the same class (see Table 1).

Table 1.

Characteristics of GPe neurons

	General properties						Anatomical and circuit properties					Intrinsic properties
	Approximate historical neuron type	Chemical content	Abundance	Defining markers	Other markers	Origin	Cell body size	Primary projection target	Formation of local collaterals	Extrinsic synaptic inputs	Intrinsic synaptic inputs	Input resistance	Spike rate	Rate variability	Spike width	Ionic conductances
																NaP	HCN	Kv4	Delayed rectifier
PV+ (prototypic)	Type I, Type A	GABA	~55%	Pvalb	Nkx2-1, Er81, Shh	MGE	Large	STN	Yes	dStr	Yes	Medium	High	Low	Narrow	High	High	High	High
References	1, 2, 3, 4, 5, 6, 7	8	4, 5, 6, 7, 8	4, 5, 9	5, 8, 10	2, 8, 10	1, 6, 7	3, 4, 5, 9	1, 3, 11	1, 4, 7, 11, 12	11	4, 6, 7, 9	2, 4, 5, 9	4, 5, 9	4, 6, 9	4, 5	4, 6	4, 7	4
Npas1+- Foxp2+ (arkypallidal)	Type II, Type B	GABA	~25%	Foxp2, Penk, Ascl1, Meis2	Npas1, Pax6	LGE	Small	dStr	Yes	dStr	Likely	High	Low	High	Wide	Low	Low	Very low	Low
References	1, 2, 3, 4, 5, 6, 7	8	4, 5, 6, 7, 8	4, 5	4, 5, 8	2, 8	1, 6, 7	1, 3, 4, 5	1, 3, 11	1, 4, 7, 12	15	4, 6, 7	2, 4, 5	4, 5	4, 6	4	4, 6	4	4
ChAT+ (cholinergic)	Type III, Type C	Ach, GABA	~5%	Chat	Nkx2-1	MGE	Very large	Ctx	Yes	dStr, STN	n.d.	Low	Very low	High	Very wide	n.d.	Very low	High	n.d.
References	3, 4, 5, 6, 7	8, 13	4, 6, 7	4, 8, 13	5, 8	2, 8	6, 13	13	13	4, 13		4, 6, 7	4, 13	4	4, 7		6	6

Alternative names: Ascl1, Mash1; Chat, choline acetyltransferase; Ctx, cortex; dStr, dorsal striatum; Etv1, ER81; LGE, lateral ganglionic eminence; MGE, medial ganglionic eminence; NaP, persistent sodium current; n.d., not determined; Penk, PPE, preproenkephalin; Pvalb, PV, parvalbumin; STN, subthalamic nucleus.

¹Nambu, A. & Llinas, R. (1997) Morphology of globus pallidus neurons: its correlation with electrophysiology in guinea pig brain slices. J Comp Neurol, 377, 85–94.

²Dodson, P.D., Larvin, J.T., Duffell, J.M., Garas, F.N., Doig, N.M., Kessaris, N., Duguid, I.C., Bogacz, R., Butt, S.J. & Magill, P.J. (2015) Distinct Developmental Origins Manifest in the Specialized Encoding of Movement by Adult Neurons of the External Globus Pallidus. Neuron, 86, 501–513.

³Mallet, N., Micklem, B.R., Henny, P., Brown, M.T., Williams, C., Bolam, J.P., Nakamura, K.C. & Magill, P.J. (2012) Dichotomous organization of the external globus pallidus. Neuron, 74, 1075–1086.

⁴Hernandez, V.M., Hegeman, D.J., Cui, Q., Kelver, D.A., Fiske, M.P., Glajch, K.E., Pitt, J.E., Huang, T.Y., Justice, N.J. & Chan, C.S. (2015) Parvalbumin+ Neurons and Npas1+ Neurons Are Distinct Neuron Classes in the Mouse External Globus Pallidus. J Neurosci, 35, 11830–11847.

⁵Abdi, A., Mallet, N., Mohamed, F.Y., Sharott, A., Dodson, P.D., Nakamura, K.C., Suri, S., Avery, S.V., Larvin, J.T., Garas, F.N., Garas, S.N., Vinciati, F., Morin, S., Bezard, E., Baufreton, J. & Magill, P.J. (2015) Prototypic and arkypallidal neurons in the dopamine-intact external globus pallidus. J Neurosci, 35, 6667–6688.

⁶Cooper, A.J. & Stanford, I.M. (2000) Electrophysiological and morphological characteristics of three subtypes of rat globus pallidus neurone in vitro. J Physiol, 527 Pt 2, 291–304.

⁷Nambu, A. & Llinas, R. (1994) Electrophysiology of globus pallidus neurons in vitro. J Neurophysiol, 72, 1127–1139.

⁸Nobrega-Pereira, S., Gelman, D., Bartolini, G., Pla, R., Pierani, A. & Marin, O. (2010) Origin and molecular specification of globus pallidus neurons. J Neurosci, 30, 2824–2834.

⁹Mastro, K.J., Bouchard, R.S., Holt, H.A. & Gittis, A.H. (2014) Transgenic mouse lines subdivide external segment of the globus pallidus (GPe) neurons and reveal distinct GPe output pathways. J Neurosci, 34, 2087–2099.

¹⁰Flandin, P., Kimura, S. & Rubenstein, J.L. (2010) The progenitor zone of the ventral medial ganglionic eminence requires Nkx2-1 to generate most of the globus pallidus but few neocortical interneurons. J Neurosci, 30, 2812–2823.

¹¹Kita, H. (1994) Parvalbumin-immunopositive neurons in rat globus pallidus: a light and electron microscopic study. Brain Res, 657, 31–41.

¹²Chuhma, N., Tanaka, K.F., Hen, R. & Rayport, S. (2011) Functional connectome of the striatal medium spiny neuron. J Neurosci, 31, 1183–1192.

¹³Saunders, A., Oldenburg, I.A., Berezovskii, V.K., Johnson, C.A., Kingery, N.D., Elliott, H.L., Xie, T., Gerfen, C.R. & Sabatini, B.L. (2015) A direct GABAergic output from the basal ganglia to frontal cortex. Nature. 521, 85–89.

Neurons in the second category—‘arkypallidal’ GPe neurons—exhibit slower and more irregular firing rates in vivo (Abdi et al., 2015; Dodson et al., 2015), project heavily to the dorsal striatum (dStr), and constitute ~25% of all GPe neurons. These neurons are devoid of PV (Mallet et al., 2012; Hernandez et al., 2015); instead, they express the opioid precursor preproenkephalin (Mallet et al., 2012; Abdi et al., 2015; Dodson et al., 2015) and the transcription factor Foxp2. Most arkypallidal neurons also express the transcription factor Npas1. However, it is important to emphasize that not all Npas1⁺ neurons express Foxp2. Additionally, a very small fraction of Foxp2⁺ GPe neurons do not express Npas1 (Abdi et al., 2015; Dodson et al., 2015; Hernandez et al., 2015). By generating an Npas1-Cre-2A-tdTomato BAC transgenic mouse line and Npas1 antibodies de novo, Hernández and colleagues recently demonstrated that Npas1⁺ neurons are distinct from PV⁺ neurons (Hernandez et al., 2015)—a finding that is in agreement with previous studies (Flandin et al., 2010; Nobrega-Pereira et al., 2010). Electrophysiological characterization of Foxp2⁺ neurons and Npas1⁺ neurons ex vivo find that they have high input resistances and low and variable firing rates (Abdi et al., 2015; Hernandez et al., 2015). These neurons share basic features with the Type II and Type B neurons described in earlier studies and likely correspond to the same class (see Table 1) (Nambu & Llinas, 1994; Cooper & Stanford, 2000). Additionally, these neurons have smaller cell bodies (Kita, 1994; Nambu & Llinas, 1997; Waldvogel et al., 1999; Waldvogel et al., 2004; Kita, 2007). Importantly, in chronic 6-OHDA lesioned mice the autonomous pacemaking activity of Npas1⁺ neurons is decreased while that of PV⁺ neurons is unchanged, corroborating findings from earlier studies that showed disrupted pacemaking of a subset of unidentified GPe neurons (Chan et al., 2011; Miguelez et al., 2012).

The existence of cholinergic neurons in the GPe has been known for several decades. Rather than being part of the basal ganglia, ChAT⁺ GPe neurons have been considered displaced basal forebrain neurons because of their similar electrophysiological properties and their tendency to be located primarily at the medial and ventral borders of the GPe (Das & Kreutzberg, 1969; Mesulam et al., 1983; Rye et al., 1984; Ingham et al., 1985; Rodrigo et al., 1998; Unal et al., 2012; McKenna et al., 2013; Eid et al., 2014; Saunders et al., 2015). Although it can be argued that the small number of ChAT⁺ neurons in the GPe are merely a dorsal extension of the much larger group of basal forebrain cholinergic neurons, their inputs from the dStr and the STN (Hernandez et al., 2015; Saunders et al., 2015) suggest they are integrated with, and therefore a part of, the basal ganglia. The membrane properties of these ChAT⁺ GPe neurons have only been modestly investigated; they likely correspond to the Type III or Type C neurons on the basis of their relative scarcity, very low firing rates, longer-duration action potentials, and large somata (Bengtson & Osborne, 2000; McKenna et al., 2013; Abdi et al., 2015; Hernandez et al., 2015; Saunders et al., 2015).

Classifying GPe neurons: consensus and division

Although recent studies provide a strong foundation for the classification of GPe neurons, several questions remain. Neuronal birthplace was recently proposed to dictate some features of GPe neuron identity (Dodson et al., 2015). In brief, GPe neurons have distinct developmental origins, arising from the medial ganglionic eminence (MGE), lateral ganglionic eminence (LGE), or the preoptic area (PoA). While PV⁺ neurons arise from the MGE, Npas1⁺ neurons are derived from both the MGE and LGE (Flandin et al., 2010; Nobrega-Pereira et al., 2010; Abdi et al., 2015; Dodson et al., 2015). Though developmental origins likely influence transcriptional programs that control the specifications of neurons, data from Hernández et al. (2015) do not fully support this idea. Despite the overlap between the developmental origins of PV⁺ GPe neurons and Npas1⁺ GPe neurons, their electrophysiological properties and axonal projections are strikingly different. As previously shown in the hippocampus, neurons can converge on a single anatomical and physiological phenotype despite differences in origin (Chittajallu et al., 2013). It is likely that a combination of transcription factors, chromatin modifiers, and enhancers are critical for the establishment and maintenance of distinct neuronal phenotypes (Deneris & Hobert, 2014).

A subset of GPe neurons express the transcription factor Lhx6, which is a nominal marker for MGE-derived neurons, but the field has yet to come to an agreement on whether Lhx6⁺ neurons are likely to represent a functionally-unique neuron class within the GPe. While recent studies converge on the existence of a distinct class of Lhx6⁺ GPe neurons (see below), a substantial population of Lhx6⁺ GPe neurons express PV or Npas1. Accordingly, Lhx6⁺ GPe neurons display electrophysiological and anatomical properties that span the range between PV⁺ GPe neurons and Npas1⁺ GPe neurons (Mastro et al., 2014; Hernandez et al., 2015). In particular, these studies do not agree upon the extent of the overlap of Lhx6 with PV, covering the range from virtually no overlap to near-complete overlap (Mastro et al., 2014; Abdi et al., 2015; Dodson et al., 2015; Hernandez et al., 2015). Similarly, these studies describe PV⁺ neurons as constituting anywhere from 30% to 60% of all GPe neurons (Nobrega-Pereira et al., 2010; Mastro et al., 2014; Abdi et al., 2015; Hernandez et al., 2015). It is likely that the discrepancies arise from differences in the detection sensitivity of immunoreactions. In addition, the eGFP expression pattern in the Lhx6-eGFP mice is non-discrete and does not reliably label neurons that natively express the transcription factor Lhx6 (Mastro et al., 2014; Dodson et al., 2015). Differences in the PV-Cre driver lines employed have not been examined and may also contribute.

Figure 1 represents an attempt to bring the different schemes for classification of GPe neuron classes into congruence. Readers should keep in mind that these classifications are only approximations; subtle species differences between rats and mice may exist. However, it is difficult to separate species differences from methodological ones especially given that Magill and colleagues are currently the only group to have published data using rats. Figure 1 highlights the existence of at least four distinct classes of GPe neurons, including a PV^–-Npas1^–-Lhx6⁺ neuron class, which represents the distinct Lhx6⁺ GPe neuron class described in previous studies (Mastro et al., 2014; Dodson et al., 2015; Hernandez et al., 2015). While the precise properties of these neurons have yet to be determined, they may be PoA-derived. This notion is supported by a recent study showing that while MGE-derived neurons express Lhx6, the PoA is another potential source of Lhx6⁺ neurons (Kanatani et al., 2015). Similarly, as the classification of GPe neurons is based on recent data derived from rodent studies, how such a classification would apply to primates remains to be determined.

Figure 1. Diagrams summarizing the classification of GPe neurons.

(a) Six different GPe neuron classes are identified based on the expression of various molecular markers. For consistency, only gene names are used throughout (though for readability they are not italicized). Plus and minus signs denote positive and negative-expression, respectively. Alternative names are listed below to cross reference with the main text. Percentages given are only approximations because of the ranges in the results reported in the literature. The relative numbers of different classes of GPe neurons listed is a deduction based on comparative analysis across several experimental findings in Abdi et al. (2015), Dodson, et al. (2015), and Hernández et al. (2015). Though it is clear that a unique class of Lhx6⁺ neurons exist, there is evidence for the expression of Lhx6 in multiple classes of GPe neurons. The field has yet to come to a consensus on the abundance and origin of Lhx6⁺ GPe neurons. In contrast, it is now well-established that Npas1⁺ GPe neurons and Pvalb⁺ GPe neurons form two distinct classes that project heavily to the dorsal striatum and subthalamic nucleus, respectively. Their relationships with arkypallidal GPe neurons and prototypic GPe neurons are illustrated in (b). See Table 1 for further description on their electrophysiological characteristics. (b) Four unique GPe neuron classes are identified so far based on the molecular signature, electrophysiological characteristics, projection patterns, and developmental origins. An additional list of molecular markers are listed for arkypallidal neurons and prototypic neurons. Prototypic neurons are more broadly defined in the literature than in the figure, which highlights the most well-defined subclass.

Alternative names: Chat, ChAT, choline acetyltransferase; dStr, dorsal striatum; Etv1, ER81; GPe, external globus pallidus; LGE, lateral ganglionic eminence; MGE, medial ganglionic eminence; Penk, PPE, preproenkephalin; PoA, preoptic area; Pvalb, PV, parvalbumin; STN, subthalamic nucleus.

Principal GPe projections

STN-projecting GPe neurons

As the GPe-STN projection and its importance in both health and disease is well-established (Canteras et al., 1990; Parent & Hazrati, 1995b; Shink et al., 1996; Joel & Weiner, 1997; Smith et al., 1998a; Bolam et al., 2000; Bevan et al., 2002b; Francois et al., 2004; Nambu, 2004; Bevan et al., 2007; Wilson & Bevan, 2011), only critical updates on the topic are included in the following section. PV⁺ GPe neurons constitute the principal GPe projection to the STN (Kita, 1994; Hoover & Marshall, 1999; 2002; Mastro et al., 2014; Abdi et al., 2015; Hernandez et al., 2015), accounting for ~94% of total GPe-STN inputs (Abdi et al., 2015; Hernandez et al., 2015). Subtle differences exist between the projections from PV⁺ GPe neurons and those from Lhx6⁺ GPe neurons to the STN. While PV⁺ GPe neurons target primarily the motor area of the STN, Lhx6⁺ GPe neurons preferentially target the limbic and associative areas of the STN (Mastro et al., 2014). Npas1⁺ GPe neurons send only a small number of axons to the STN. The vast majority of Npas1⁺ axonal projections instead run dorsally to the STN, along the lenticular fascicle (Hernandez et al., 2015). The caudal projection pattern of Npas1⁺ axons has not been examined systematically. However, it should map onto a number of brain regions that were charted previously (Hattori et al., 1975; Bunney & Aghajanian, 1976; Kanazawa et al., 1976; Staines & Fibiger, 1984; Hazrati et al., 1990; Hazrati & Parent, 1991; Kincaid et al., 1991a; Shammah-Lagnado et al., 1996; Saunders et al., 2015). The GPe-STN projection is sparse and distributed; individual GPe neurons contact only 2% of STN neurons and neighboring STN neurons rarely receive input from the same GPe axon (Baufreton et al., 2009). In addition to the classic perisomatic baskets, GPe axons also terminate on the proximal and distal dendrites of STN neurons (Smith et al., 1990a).

The GPe-STN projection plays critical roles in regulating STN neuron activity via a number of mechanisms. In health, the GPe provides phasic inhibition that promotes decorrelated activity between the GPe and the STN (Atherton et al., 2013). Additionally, GPe input limits activation of STN neurons by cortical input through hyperpolarization and shunting inhibition (Chu et al., 2015). Dopamine, via presynaptic D2 receptors, inhibits GABA release at the GPe-STN synapse (Shen & Johnson, 2000; Baufreton & Bevan, 2008). Hypersynchronization of the GPe with the STN in PD is in part attributable to increased presynaptic release and postsynaptic strengthening of the GPe-STN input via an NMDA receptor-dependent mechanism (Fan et al., 2012; Chu et al., 2015). Strengthened GPe inputs then interact with intrinsic, active conductances on STN neurons to generate rhythmic bursting of STN neurons (Baufreton et al., 2005; Baufreton et al., 2009; Fan et al., 2012). A reverberating feedback loop formed between the GPe and STN was proposed to serve as an intrinsic oscillator that drives aberrant network activity throughout the basal ganglia (Bevan et al., 2002b) (see further discussion below).

dStr-projecting GPe neurons

A projection from the GPe to the dStr, the primary input center of the basal ganglia, was postulated over a century ago (Wilson, 1911; 1913), and its existence has since been confirmed in a variety of species. However, very little is known about the identity of the GPe neurons that provide this input or the postsynaptic neurons they target (Nauta, 1979; Staines et al., 1981; Beckstead, 1983; Jayaraman, 1983; Staines & Fibiger, 1984; Smith & Parent, 1986; Shu & Peterson, 1988; Walker et al., 1989; Kita & Kitai, 1991; Shinonaga et al., 1992; Rajakumar et al., 1994; Shammah-Lagnado et al., 1996; Spooren et al., 1996; Nambu & Llinas, 1997; Bevan et al., 1998; Kita et al., 1999; Sato et al., 2000; Kita & Kita, 2001; Mallet et al., 2012). A major impediment to our understanding of the pallidostriatal pathway arises from the cellular complexity in the dStr and the GPe, as each of these nuclei comprises several types of neurons (Kreitzer, 2009; Tepper et al., 2010; Gittis et al., 2014; Abdi et al., 2015). Therefore, it is evident that we need a systematic analysis to map the connectivity between specific pallidostriatal inputs and identified postsynaptic target neurons.

It was recently demonstrated that Npas1⁺ GPe neurons project heavily to the dStr but only sparingly to the STN (Hernandez et al., 2015). For this reason, we postulate that spiny projection neurons (SPNs), the principal neurons in the dStr (Kemp & Powell, 1971; DiFiglia et al., 1976; Somogyi & Smith, 1979; Dimova et al., 1980; Preston et al., 1980; Groves, 1983), receive input from Npas1⁺ GPe neurons. In support of this idea, ultrastructural data suggest that Npas1⁺-Foxp2⁺ (arkypallidal) axon terminals form synapses with spine-bearing dendrites in the dStr (Mallet et al., 2012). Although Lhx6⁺ GPe neurons also target the dStr, Npas1⁺-Foxp2⁺ neurons appear to do so to a much higher degree (Mastro et al., 2014; Hernandez et al., 2015); however, unlike Lhx6⁺ neurons, Npas1⁺-Foxp2⁺ neurons do not project to the STN (Abdi et al., 2015). This suggests Npas1⁺-Foxp2⁺ neurons and Npas1⁺-Lhx6⁺ neurons are distinct (see Figure 1). Considering that Lhx6 expression shows essentially no overlap with Foxp2 and that most Npas1⁺-Foxp2^– neurons are also Lhx6⁺ (Abdi et al., 2015; Hernandez et al., 2015), it is tempting to speculate that Npas1⁺-Foxp2⁺ neurons and Npas1⁺-Lhx6⁺ neurons are both dStr-projecting but preferentially target distinct subsets of striatal neurons—for example, SPNs and GABAergic interneurons, respectively. PV⁺ GPe neurons also provide a small number of projections to the dStr, where they appear to preferentially target interneurons (Bevan et al., 1998; Kita et al., 1999; Mastro et al., 2014).

We still lack a systematic and quantitative analysis of the targeting properties of the pallidostriatal inputs. However, it is possible to calculate the contact probability of cell-specific GPe axons with individual SPNs, as the number of neurons in both the dStr and the GPe has been previously determined (see Table 2). Similarly, the number of synaptic boutons formed by Npas1⁺ GPe neurons and PV⁺ GPe neurons in the dStr has also been estimated. Assuming 100% connectivity between GPe and dStr neurons, each SPN on average receives a small number of boutons from GPe neurons: roughly 40 from Npas1⁺ GPe neurons and less than ten from PV⁺ GPe neurons. In contrast, each SPN receives 2,500 symmetrical synapses (Wilson, 2013). From these estimates alone, pallidostriatal inputs arising from GPe neurons would appear unlikely to have an important impact on the output of SPNs. However, we do not yet know if pallidostriatal neurons make contact with striatal neurons in a target cell-specific manner, as is often observed between SPN classes (MacAskill et al., 2012; Wall et al., 2013; Deng et al., 2015; Guo et al., 2015b) (but see Kress et al., 2013), or if they exhibit strategic positioning on the dendrites of SPNs. Either of these features could substantially increase the efficiency of their influence on the network. Additionally, pre- and postsynaptic mechanisms may exist to provide anatomical and biophysical specialization at these connections. Finally, the temporal relationship between the activation of the pallidostriatal inputs and the excitatory inputs (e.g. from the cortex) will be a crucial factor in determining the impact of the postsynaptic effect.

Table 2.

Calculation of GPe-dStr connectivity

Cell count						Refs
1a	Total no. of dStr neurons		2,791,000			Oorschot (1996)
1b	Total no. of GPe neurons		45,960			Oorschot (1996)
1c	PV+ GPe neurons	(1b × 55%)	25,278			Abdi et al. (2015), Dodson et al. (2015), Hernández et al. (2015)
1d	Npas1+-Foxp2+ GPe neurons	(1b × 25%)	11,490			Abdi et al. (2015), Dodson et al. (2015), Hernández et al. (2015)
No. boutons produced by each GPe neuron			Approximation
2a	PV+ *		1,000			Bevan et al. (1998), Mallet et al. (2012)
2b	Npas1+-Foxp2+ **		10,000			Mallet et al. (2012)
Total no. of boutons formed by each GPe neuron class			Approximation
3a	PV+	(1c × 2a)	25,278,000			Bevan et al. (1998), Mallet et al. (2012)
3b	Npas1+-Foxp2+	(1d × 2b)	114,900,000			Mallet et al. (2012)
3c	Npas1+-Foxp2+/PV+	(3c/3a)	5
			Contact probability
Total no. of contacts per dStr neuron from GPe inputs			100%	10%	1%
4a	PV+	(3a/1a)/contact prob.	9	91	906
4b	Npas1+-Foxp2+	(3b/1a)/contact prob.	41	412	4,117
4c	Total no. symmetrical synapses per SPN		2,500			Wilson (2007)
4d	PV+ (% total)	(4a/4c)	0.36%	3.62%	36.23%
4e	Npas1+-Foxp2+ (% total)	(4b/4c)	1.65%	16.47%	164.67%
Total no. of contacts per SPN branchlet
5a	No. of primary dendrites per dSPN		8			Gertler et al. (2008)
5b	PV+ to dSPN	(4a/5a)	1	11	113
5c	Npas1+-Foxp2+ to dSPN	(4b/5a)	5	51	515
5d	No. of primary dendrites per iSPN		6			Gertler et al. (2008)
5e	PV+ to iSPN	(4a/5d)	2	15	151
5f	Npas1+-Foxp2+ to iSPN	(4b/5d)	7	69	686

^*The axonal arborization patterns in Bevan et al. (1998) are highly consistent with those of the more recently described TI neurons, which are primarily PV⁺ (Mallet et al., 2012). Minimum and maximum number of PV⁺ boutons described by the literature cited is 329 and 1,353, respectively.

^**Given that Foxp2 and preproenkephalin show completely overlapping expression in the GPe (Abdi et al., 2015; Dodson et al., 2015) and that essentially all Foxp2⁺ neurons are also Npas1⁺ (Abdi et al., 2015; Dodson et al., 2015; Hernández et al., 2015), it can be inferred that Npas1⁺-Foxp2⁺ neurons represent the arkypallidal neurons that exhibit extensive axonal arborizations in the dStr (Mallet et al., 2012). Minimum and maximum number of Npas1⁺-Foxp2⁺ boutons described by the literature cited is 9,085 and 13,789, respectively.

GPe-dStr inputs and sparse coding in the striatum

What is the biological significance of the pallidostriatal inputs? Could they provide subcellular compartment-specific inhibition? To date, there has been no consensus on the coding scheme employed by SPNs. ‘Sparse coding’ is one potential computational strategy whereby information is communicated by spatially- and temporally-distributed activity in a relatively small fraction of neurons. This form of neural coding is well-established in sensory systems and allows for efficient and flexible information processing (Vinje & Gallant, 2000; Hromadka et al., 2008; Isaacson, 2010; Wolfe et al., 2010). In line with this idea, the dStr shares several characteristic features with neural networks that support and use sparse coding (Olshausen & Field, 2004). In addition to the large spatial volume of the dStr (Rosen & Williams, 2001), SPNs exhibit burst activity in response to a limited range of stimuli (DeLong, 1973; Wilson & Groves, 1981; Kimura et al., 1990; Stern et al., 1997) as well as an absence of redundancy and a loose temporal correlation between responses of nearby SPNs (Jaeger et al., 1995; Ponzi & Wickens, 2010; Adler et al., 2012; Adler et al., 2013).

In addition, it has been previously demonstrated that a nonspecific inhibitory input can facilitate sparse coding by acting as a gain control (Laurent, 2002; Isaacson & Scanziani, 2011). By dampening excitatory responses across a broad area of the dStr, Npas1⁺ GPe neurons could potentially promote sparse coding in SPNs (Burrone & Murthy, 2003; Semyanov et al., 2004; Silver, 2010). Typically, SPNs rest close to the potassium reversal potential, spiking only when they are driven by glutamatergic input from the cortex (Parent & Hazrati, 1995a; Smith et al., 2004). Specifically, through activation of NMDA receptors, SPNs in the dStr display dendritic plateau potentials in distal dendritic compartments (Plotkin et al., 2011). It is thus intriguing to speculate that pallidostriatal input controls SPN output by limiting the summation of excitatory inputs, preventing the subsequent nonlinear-generation of dendritic plateaus; spiking would be limited to only those SPNs receiving robust or well-timed excitatory input. As the dStr is organized in a somatotopic fashion (Nambu, 2011), spatially-broad inhibition from the GPe could be used to suppress or reset somatotopically-complex motor sequences across the dStr (see below).

Synaptic and neuromodulatory control of the GPe

dStr forms the principal inhibitory input to the GPe

The dStr input to the GPe is topographically organized and highly convergent. In primates, the dStr-GPe projection displays a precise rostrocaudal, mediolateral, and dorsoventral topography. Furthermore, the injection of two different anterograde tracers into two small, adjacent areas of the striatum led to the formation of two clearly distinguishable sets of bands in the GPe (Hazrati & Parent, 1992; Parent & Hazrati, 1995a). In the rat basal ganglia there are roughly three million SPNs but only 46 thousand GPe neurons (Oorschot, 1996). Assuming all of the SPNs in the dStr are GPe-projecting, individual GPe neurons must on average receive input from at least 60 dStr SPNs. The high level of convergence in the dStr-GPe projection is supported by both anatomical and electrophysiological findings. Retrograde tracer injections into a small area of the primate GPe label neurons in spatially-broad areas of the putamen (Flaherty & Graybiel, 1993; 1994). Moreover, focal stimulation in the GPe induces antidromic activation of multiple striatal neurons over a wide area (Kimura et al., 1996). Anatomically, the dendritic arbor of GPe neurons is oriented perpendicularly to the incoming, radial striatal fibers, creating an ideal arrangement for intercepting axons from broad striatal regions (Chang et al., 1981; Percheron et al., 1984; Yelnik et al., 1984; Kawaguchi et al., 1990; Yelnik et al., 1997). Such an anatomical organization allows dStr axons to contact the dendrites of multiple GPe neurons. The sharing of dStr inputs by multiple GPe neurons provide an anatomical substrate for synchrony and pause-burst firing pattern in a large population of GPe neurons, as demonstrated by both experimental and computational studies (Terman et al., 2002; Elias et al., 2007; Zold et al., 2007a; Zold et al., 2007b; Kita & Kita, 2011b; a; Adler et al., 2012; Schwab et al., 2013; Wilson, 2013; Schechtman et al., 2015). As PD progresses, the activity of GPe neurons transitions from decorrelated, single-spike pacemaking to synchronous, rhythmic bursting (but see Mallet et al., 2008). This pathological network behavior is thought to be critical to the core motor symptoms of PD (Pan & Walters, 1988; Filion & Tremblay, 1991; Filion et al., 1991; Hutchison et al., 1994; Nini et al., 1995; Rothblat & Schneider, 1995; Hassani et al., 1996; Taha et al., 1996; Bergman et al., 1998; Boraud et al., 1998; Wichmann et al., 1999; El-Deredy et al., 2000; Magill et al., 2000; Magnin et al., 2000; Raz et al., 2000; Brown et al., 2001; Magill et al., 2001; Heimer et al., 2002; Bar-Gad et al., 2003; Starr et al., 2005; Heimer et al., 2006; Wichmann & Soares, 2006; Kita, 2007; Tang et al., 2007; Zold et al., 2007a; Zold et al., 2007b; Mallet et al., 2008; Starr et al., 2008; Cruz et al., 2009; Sani et al., 2009; Chan et al., 2011). Although active decorrelating processes have been proposed to prevent synchrony among neighboring GPe neurons in the healthy state (Nini et al., 1995; Bar-Gad et al., 2003; Chan et al., 2011), the exact mechanisms involved and why they collapse in the absence of dopamine remain to be explored.

The dStr inputs originating from SPNs account for 65–80% of the GABAergic synapses within the GPe (Smith et al., 1998a; Kita, 2007). Approximately two-thirds of these dStr-GPe inputs arise from the enkephalin and dopamine D2 receptor-expressing indirect-pathway SPNs (iSPNs), while the remaining one-third originate from collaterals of substance P and dopamine D1 receptor-expressing direct-pathway SPNs (dSPNs) that form en passant synapses (Feger & Crossman, 1984; Gerfen & Young, 1988; Kawaguchi et al., 1990; Parent et al., 1995; Wu et al., 2000; Levesque & Parent, 2005; Nadjar et al., 2006; Matamales et al., 2009; Fujiyama et al., 2011). Recent experimental data directly demonstrate the different roles played by dSPNs (movement facilitation) and iSPNs (movement suppression) in learned-behavior and motor dysfunction, in agreement with those proposed in the classic model (Kravitz et al., 2010; Cui et al., 2013; Freeze et al., 2013; Calabresi et al., 2014; Sippy et al., 2015). While it has yet to be demonstrated how striatal information is processed at the GPe level, Saunders and colleagues find that dSPN and iSPN inputs target both GABAergic and cholinergic GPe neurons (Saunders et al., 2015).

Both types of dStr inputs to the GPe are likely important players in PD, given the remarkable anatomical remodeling following the perturbation of dopaminergic signaling. Direct examination using electron microscopy has revealed pathological enlargement of iSPN terminals in the GPe following chronic dopamine depletion (Ingham et al., 1997). Similarly, alterations in the dSPN axonal arborization within the GPe are observed when dopamine signaling is disrupted (Cazorla et al., 2014; Cazorla et al., 2015). Consistent with this idea that dStr-GPe inputs are remodeled after dopamine depletion, compelling evidence from in vivo studies suggests that in PD, increased dStr-GPe input contributes to neuronal synchrony within the GPe. This subsequently leads to pathological network oscillations throughout the basal ganglia (Bevan et al., 2002b; Terman et al., 2002; Kita & Kita, 2011b). It is important to note that, even in the healthy state, coordinated dStr-GPe input exhibits a strong ability to reset—and therefore temporarily promote synchronization of—the pacemaking of GPe neurons in an HCN channel-dependent manner (Chan et al., 2004). While ex vivo studies so far do not support the theory of altered dStr-GPe transmission in a chronic 6-OHDA model of PD (Miguelez et al., 2012), they were confounded by co-activation of both dSPN inputs and iSPN inputs with conventional electrical stimulation. Additionally, the identities of the postsynaptic GPe neurons were undefined. A recent modeling study suggests dStr input onto PV⁺ (prototypic) GPe neurons is stronger than that onto Npas1⁺-Foxp2⁺ (arkypallidal) GPe neurons (Nevado-Holgado et al., 2014), highlighting that much investigation of dStr-GPe signaling still needs to be done, particularly regarding how dSPNs and iSPNs are connected with distinct GPe neuron classes.

Local collaterals are another major inhibitory input to GPe neurons

In addition to the dStr input, local collaterals are a second source of GABAergic input onto GPe neurons. Juxtacellular labeling and intracellular dye-loading of GPe neurons have revealed the presence of local axon collaterals with numerous varicosities, suggesting the presence of lateral GABAergic inhibition within the GPe (Millhouse, 1986; Okoyama et al., 1987; Kita, 1994; Kita & Kitai, 1994; Nambu & Llinas, 1997; Bevan et al., 1998; Sato et al., 2000; Sadek et al., 2007; Mallet et al., 2012). Most, if not all, GPe neurons exhibit local axon collaterals; it is estimated that a single local collateral axon gives rise to as many as 650 boutons within the GPe. However, this number varies with the identity and geographical location of the cell body (Park et al., 1982; Millhouse, 1986; Kita & Kitai, 1994; Nambu & Llinas, 1997; Bevan et al., 1998; Sato et al., 2000; Sadek et al., 2007; Mallet et al., 2012).

In the GPe, these local axon collaterals terminate on somata and proximal dendrites (Kita, 1994; Kita & Kitai, 1994; Nambu & Llinas, 1997; Bevan et al., 1998; Sato et al., 2000; Sadek et al., 2007; Mallet et al., 2012), positioning them to have a powerful influence on the firing of their postsynaptic targets. Early electrophysiological analysis describing the kinetics of putative intrapallidal inhibitory synaptic currents—faster than those arising from dStr inputs—is consistent with this perisomatic location (Sims et al., 2008; Gross et al., 2011). Although local collaterals are integral to GPe circuit dynamics and downstream network effects (Terman et al., 2002), they have not been studied in great detail due to the inherent difficulty in identifying and selectively activating individual classes of GPe neurons and their local collateral axons.

Functional connections between GPe neurons are demonstrated in a recent study with paired-recordings. Connections between GPe neurons are mediated by GABA_A receptors and strongly influence the firing rate of the postsynaptic GPe neuron, even at the level of unitary connections (Bugaysen et al., 2013). However, as connection probability is only ~1–2% (Sadek et al., 2007; Bugaysen et al., 2013), only a handful of recordings were obtained in this study (Bugaysen et al., 2013). In a chronic model of PD, Miguelez and colleagues discovered a strengthening of this intrapallidal connection (Miguelez et al., 2012). While these studies have provided important insights into the basic biology of intrapallidal signaling, the identities of both pre- and postsynaptic neurons were not determined (Sims et al., 2008; Gross et al., 2011). Recent advances in transgenic (Heintz, 2004; Madisen et al., 2010; Madisen et al., 2012; Gerfen et al., 2013; Hernandez et al., 2015; Madisen et al., 2015) and optogenetic approaches (Boyden, 2015; Deisseroth, 2015) will undoubtedly promote future discoveries concerning the intrapallidal signaling between specific GPe neuron classes. Though it remains to be tested empirically, Mallet and colleagues show the existence of various connection types between GPe neuron classes (Mallet et al., 2012). It is likely that the connectivity pattern varies in a cell-specific manner. This idea is supported by a recent computational analysis (Nevado-Holgado et al., 2014) that suggests the inputs from PV⁺ (prototypic) GPe neurons to Npas1⁺-Foxp2⁺ (arkypallidal) GPe neurons are relatively strong, whereas inputs from Npas1⁺-Foxp2⁺ neurons to PV⁺ neurons are weaker. This analysis also predicts that the connections between Npas1⁺-Foxp2⁺ neurons are modest and that the connections between PV⁺ neurons are negligible. As local collateral inhibition plays a pivotal role in governing network synchrony (Jefferys et al., 1996; Paz & Huguenard, 2015), it is tempting to speculate that intrapallidal connections between GPe neurons serve as a decorrelating mechanism in the healthy state. Inappropriate scaling of these connections, as occurs in the absence of dopamine, has been suggested to contribute to hypersynchrony in PD (Cruz et al., 2011). Lastly, gap junctions represent another means by which neurons can be electrically coupled and have been previously found on PV^– GPe neurons at the electron microscopy level (Kita, 1994). Accordingly, the molecular correlates of electrical synapses (connexins 26, 32, 36, and 43) are expressed in the GPe (Dermietzel et al., 1989; Vis et al., 1998; Condorelli et al., 2000; Rash et al., 2000; Schwab et al., 2014; Phookan et al., 2015). However, the existence of electrical coupling between GPe neurons awaits functional confirmation.

Postsynaptic GABA_A receptors

As previously discussed, the majority of synaptic inputs to the GPe are mediated by GABA_A receptors, which are ligand-gated Cl^– channels. Each receptor is a heteromeric structure composed of five out of at least 16 different subunits that are grouped into several classes. Eight subunit classes have been isolated to date (α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3). It is thought that most functional GABA_A receptors in vivo are formed by co-assembly of two α subunits, two β subunits, and an additional subunit from one of the remaining classes (Schofield, 1989; Mohler et al., 1995; Sieghart, 1995; McKernan & Whiting, 1996; Mohler et al., 1996; Barnard et al., 1998; Rudolph & Mohler, 2004; 2006; Olsen & Sieghart, 2008).

Although mRNAs for essentially all cloned GABA_A receptor-subunits are present in the GPe (Laurie et al., 1992; Pirker et al., 2000; Schwarzer et al., 2001), those subunits that are generally thought to be extrasynaptic and underlie “tonic” inhibition (e.g. α4–α6, β3 and δ subunits) (Danglot et al., 2003; Jacob et al., 2005) are present only at extremely low levels. This may indicate that transfer of information at the GABAergic synapses is accomplished by phasic, point-to-point signaling (Farrant & Nusser, 2005). Concordantly, the α1 subunit is expressed at very high levels in the GPe, richly investing the dStr-GPe synapses (Wisden et al., 1992; Hartig et al., 1995; Somogyi et al., 1996; Riedel et al., 1998; Waldvogel et al., 1998; Waldvogel et al., 1999; Pirker et al., 2000; Schwarzer et al., 2001; Sur et al., 2001; Waldvogel et al., 2004; Charara et al., 2005); zolpidem, an α1 subunit-selective imidazopyridine agonist, slows the decay kinetics of dStr-GPe postsynaptic inhibitory currents in the GPe without changing their frequency or amplitude. Zolpidem has a stronger impact on dStr-GPe inhibitory postsynaptic currents than those arising from local collaterals, suggesting synapse-specific enrichment of the α1 subunit (Chen et al., 2004b). Perhaps the most convincing evidence for an association between dStr-GPe synapses and parkinsonism is an observed downregulation of GPe α1 subunits in PD patients and animal models (Chadha et al., 2000a; Yu et al., 2001), as well as the therapeutic efficacy of zolpidem (Chen et al., 2008b; Huang et al., 2012).

At the same time, investigation of both the striatopallidal and local collateral inputs with TP003, an α3 subunit-selective agonist (Dias et al., 2005), suggests that the α3 subunit is uniquely present at local collateral synapses in the GPe and not at striatopallidal synapses (Gross et al., 2011). While these findings put the α3 subunit forward as a potential target for local collateral-specific therapy in motor disease, the data were collected using relatively young (postnatal 18–22 days) rats. Not only have α1 and α2 subunit expression levels in the GPe been shown to change considerably during the first month of development in rats (Fritschy et al., 1994), but in situ and immunohistochemical analyses have also confirmed that expression of the α3 subunit is present in the GPe in young rats but fades to low to undetectable levels once adulthood is reached (Laurie et al., 1992; Fritschy & Mohler, 1995). Interestingly, immunohistochemical evidence from adult human brains indicates that α3 subunit expression is present but restricted to PV⁺ neurons (Waldvogel et al., 1999). Given that GABA_A receptor pharmacology may allow specific therapeutic targeting of PV⁺ prototypic neurons via the α3 subunit, further investigation in a cell- and input-specific fashion is warranted.

GABA_B signaling in the GPe

GABA_B receptors are heteromeric G protein-coupled receptors composed of one GABA_BR1 subunit and one GABA_BR2 subunit (Kaupmann et al., 1998; White et al., 1998; Kuner et al., 1999; Ng et al., 1999). At the presynaptic sites, GABA_B receptors suppress release by inhibiting voltage-gated calcium channels (Dolphin & Scott, 1986) and directly impeding synaptic vesicle exocytosis (Blackmer et al., 2001; Yoon et al., 2007; Rost et al., 2011), while at the postsynaptic membrane, they constrain excitability by activating an inward-rectifying potassium conductance (Newberry & Nicoll, 1984b; a; Gahwiler & Brown, 1985; Luscher et al., 1997) in addition to inhibiting voltage-gated calcium channel activity and NMDA receptor calcium signaling (Mintz & Bean, 1993; Perez-Garci et al., 2006; Chalifoux & Carter, 2010; 2011; Lur & Higley, 2015).

Immunogold labeling and immunocytochemistry have shown that the GABA_BR1 subunit is present in monkeys at the presynaptic membrane of both symmetric and asymmetric synapses in the GPe (Charara et al., 2000; Charara et al., 2005). Using the same approach, it has also been demonstrated that GABA_B receptors are found at the postsynaptic membrane of both symmetric synapses and asymmetric synapses in the GPe, but that the majority are extrasynaptic (Chen et al., 2004a; Charara et al., 2005). Direct agonist of GABA_B receptors in the GPe in vivo produces ipsilateral turning behavior in rats (Chen et al., 2002; Ikeda et al., 2010). Activation of postsynaptic GABA_B receptors in the GPe neurons slows their pacemaking in ex vivo rodent brain slices (Chan et al., 2004; Kaneda & Kita, 2005). Additionally, in rat brain slices baclofen acts presynaptically in the GPe to reduce the release of glutamate (Chen et al., 2002; Kaneda & Kita, 2005; Jin et al., 2012) and GABA (Kaneda & Kita, 2005). It should be noted that electrophysiological assessment of GABA_B signaling on bona fide striatopallidal and subthalamic inputs to the GPe has yet to be documented; previous assessments have either measured unidentified miniature events (Chen et al., 2002; Kaneda & Kita, 2005; Jin et al., 2012) or used contamination-prone terminal field electrical stimulation to evoke IPSCs and EPSCs (Kaneda & Kita, 2005), leaving open the possibility that the measured IPSCs and EPSCs were of pallidal and thalamic origins, respectively (see further discussion below).

To date, concrete information concerning the expression of GABA_B subunits in GPe neuron subpopulations is lacking. Immunohistochemistry in human brains has indicated that the vast majority (98%) of PV⁺ GPe neurons express some combination of GABA_BR1 and GABA_BR2 while the same is true for two thirds of PV^– GPe neurons (Waldvogel et al., 2004), contrasting with results from Chen and colleagues, who reported postsynaptic inhibitory GABA_B currents in response to baclofen in only a minority of GPe neurons (Chen et al., 2002). In addition to detection sensitivity, this discrepancy between immunological labeling and electrophysiological function may be explained by receptor trafficking. The proportion of total GABA_BR1 located intracellularly as opposed to membrane-bound has been calculated to be 70% in rats (Chen et al., 2004a) and 80% in monkeys (Charara et al., 2005), perhaps limited by the availability of GABA_BR2, pairing with which is required for the localization of the GABA_B heteromeric receptor complex to the membrane surface (Couve et al., 2000).

Similarly, changes in trafficking may produce the sensitized response to GABA_B signaling that is seen in MPTP-treated monkeys (Galvan et al., 2011), as no increases in GABA_B receptor expression have been noted in human PD patients (de Groote et al., 1999) or MPTP-treated monkeys (Galvan et al., 2011). It is tempting to speculate that that the minority of GPe neurons that display a significant outward postsynaptic current in response to baclofen in the healthy state (Chen et al., 2002) corresponds to the Npas1⁺ GPe neuron population (Hernandez et al., 2015), with a sensitized GABA_B response contributing to the decrease in pacemaking of GPe neurons seen in 6-OHDA lesioned mice (Chan et al., 2011; Hernandez et al., 2015). Given that GABA_B receptor pharmacology could offer a method of therapeutic modulation of GPe excitability, the field will benefit from a thorough investigation of GABA_B receptor expression and signaling in identified GPe neurons in both healthy and PD model animals (Chen et al., 2002).

STN forms the principal excitatory input to the GPe

Both anatomical and physiological studies historically show that the principal glutamatergic input to the GPe is from the STN (Kita & Kitai, 1987; Smith et al., 1990b; Smith et al., 1998a). Recent estimations (Koshimizu et al., 2013), however, have found fewer STN-GPe synapses than expected, calling into question the prominence of the STN-GPe input (Wilson, 2013). The boutons of STN terminals form medium-sized, asymmetrical synapses on the largely aspiny dendrites of GPe neurons. Anatomical approaches have shown that both AMPA and NMDA receptors are present at these synapses (Bernard & Bolam, 1998). This observation is consistent with pharmacological studies in which local application of AMPA and NMDA receptor blockers reduce spontaneous activity of GPe neurons in awake monkeys (Kita et al., 2004). Furthermore, stimulation of the STN evokes fast excitatory postsynaptic potentials mediated by AMPA receptors and slower, strong excitatory postsynaptic potentials mediated by NMDA receptors in GPe neurons (Kita & Kitai, 1991). Computational modeling suggests a preferential connection from the STN to Npas1⁺-Foxp2⁺ (arkypallidal) GPe neurons (Nevado-Holgado et al., 2014). No experimental investigation yet delineates the difference in the connection strength and biophysical properties of STN inputs to distinct GPe neuron classes.

Of potentially major importance in understanding the influence of the STN on GPe neurons is the weak voltage-dependence of the NMDA component of the excitatory postsynaptic potentials induced by STN stimulation. Though NMDA receptor opening normally requires relatively depolarized membrane potentials to dislodge pore-blocking Mg²⁺ ions (Kutsuwada et al., 1992; Monyer et al., 1992; Ishii et al., 1993; Kuner & Schoepfer, 1996; Dingledine et al., 1999; Traynelis et al., 2010), this may not be a requirement in adult GPe neurons, as they express relatively high levels of the GluN2C and GluN2D subunits (Standaert et al., 1994; Wenzel et al., 1995; Wenzel et al., 1996; Kosinski et al., 1998), diminishing the efficacy of the Mg²⁺ block (Kuner & Schoepfer, 1996; Momiyama et al., 1996). NMDA channels containing GluN2C and GluN2D subunits also have a higher affinity for glutamate and slower deactivation kinetics than GluN2A- or GluN2B-containing NMDA receptors (Cull-Candy et al., 2001; Traynelis et al., 2010). Thus, prominent expression of GluN2C- and GluN2D-containing NMDA receptors could serve to enhance the impact of STN inputs on GPe neurons. We do not know if the expression of GluN2C and GluN2D at the STN inputs can be generalized across all GPe neurons.

The assertion that the STN-GPe synapse grows in functional significance in PD is consistent with two other lines of evidence. First, both AMPA receptors and NMDA receptors in the GPe are downregulated in animal models of PD, suggesting a compensatory response to increased glutamatergic input (Porter et al., 1994; Betarbet et al., 2000). Consistent with this, systemic administration of NMDA receptor antagonists is effective in ameliorating motor symptoms in animal models of PD (Starr et al., 1997; Kelsey et al., 2004). NMDA receptor antagonists also lessen parkinsonian tremor and levodopa-induced motor fluctuations (Butzer et al., 1975; Koller, 1986; Danysz & Parsons, 1998; Verhagen Metman et al., 1998; Chase et al., 2000; Marjama-Lyons & Koller, 2000). However, there have not been any detailed functional studies of the STN-GPe synapse in animal models of PD. In spite of their efficacy in alleviating motor symptoms, broad spectrum glutamate receptor antagonists are unlikely to be adopted therapeutically because of undesirable side-effects in other brain circuits. Recently, GluN2B-specific ligands have been developed, providing clinicians with an important tool for dissecting neural circuitry that has incidentally proven to be particularly important in treating pain (Chizh et al., 2001). The very restricted distribution of GluN2C- and GluN2D-containing subunits could make the side-effect profile of GluN2C- and GluN2D-selective antagonists and negative allosteric modulators very acceptable.

In addition to ionotropic receptors, metabotropic glutamate receptors (mGluRs) also exist in the GPe. On the basis of amino acid sequence homology, intracellular second messengers, and ligand selectivities, mGluRs are categorized into eight subtypes that are divided into: group I (mGluR1 and 5), group II (mGluR2 and 3), and group III (mGluR4, 6–8) (Nakanishi, 1994; Pin & Duvoisin, 1995; Conn & Pin, 1997).

In the GPe, group I mGluRs are abundantly expressed (Testa et al., 1994; Testa et al., 1998; Smith et al., 2000). Immunohistochemistry and electron microscopy studies show that group I mGluRs localize postsynaptically along dendritic processes (Testa et al., 1998; Hanson & Smith, 1999; Kaneda et al., 2005). The activation of group I mGluRs depolarize GPe neurons to increase their excitability (Stefani et al., 1998; Poisik et al., 2003; Kaneda et al., 2007). Group II mGluRs have a more modest expression (Ohishi et al., 1993; Poisik et al., 2005). Electron microscopy studies show group II mGluRs localize presynaptically on glutamatergic axon terminals (Poisik et al., 2005). However, as mGluR3 is robustly expressed in the striatum (Ohishi et al., 1993; Tanabe et al., 1993; Testa et al., 1994), it is also possible that they are targeted to the axon terminals in the GPe. Activation of group II mGluRs decreases neurotransmitter release from axon terminals (Poisik et al., 2005). Group III mGluRs are abundantly expressed in the GPe (Kinoshita et al., 1998; Bradley et al., 1999; Kosinski et al., 1999; Corti et al., 2002) and are primarily presynaptically localized (Kinoshita et al., 1998; Bradley et al., 1999; Corti et al., 2002; Bogenpohl et al., 2013). Functionally, group III mGluRs act as homo- and heteroreceptors by decreasing glutamate release from putative STN terminals and GABA release from inhibitory terminals (Marino et al., 2003; Matsui & Kita, 2003; Valenti et al., 2003; Gubellini et al., 2014). Much remains unknown about the purpose of mGluRs in the GPe. For example, the source of glutamate for these mGluRs has not been determined. As discussed below, the GPe receive a wide range of glutamatergic inputs in addition to the STN (see below). Coupled with the rich variety of mGluRs in the GPe, this suggests that these receptors may act as specific local regulators of network and neuronal activity. Astrocytes are abundant within the GPe and may play an important role in regulating the activation of mGluRs associated with neuronal elements within the GPe. This is likely in part through the detection of synaptic and ambient glutamate via the surface expression of mGluR3 and quite possibly mGluR5 on these cells (Testa et al., 1994; Sun et al., 2013; Panatier & Robitaille, 2015) (see further discussion below).

Intrapallidal and intracerebroventricular delivery of group III mGluR agonists have been explored as PD treatment with promising results (Valenti et al., 2003; MacInnes et al., 2004; Lopez et al., 2007; Agari et al., 2008). Given the therapeutic potential of mGluR pharmacology in PD, the field will benefit from addressing the current dearth of data on the origin of presynaptic terminals expressing mGluRs and the expression of mGluRs in specific GPe neuron subpopulations.

GPe-STN loop in health and disease

Topographically, the excitatory STN-GPe and inhibitory GPe-STN projections form a reciprocally-connected loop. Compelling evidence from experimental and modeling studies suggests that the GPe-STN loop supports oscillatory activity (Shink et al., 1996; Smith et al., 1998a; Plenz & Kital, 1999; Bevan et al., 2002b). Like GPe neurons, STN neurons are spontaneously active (Beurrier et al., 2001; Bevan et al., 2002a; Do & Bean, 2003), and the contribution of STN inputs to GPe firing has been examined in several preparations. In organotypic cultures, oscillatory activity in the GPe was abolished after cutting the input from the STN (Plenz & Kital, 1999). In awake rodents, silencing of the GPe after injection of the GABA_A receptor agonist muscimol in the STN was seen only in 6-OHDA lesioned animals; control animals instead had only a slight reduction in GPe neuron activity (Chan et al., 2011). In primates, while muscimol blockade of STN initially decreased and even silenced GPe neuron activity for five to ten minutes, the activity eventually settled into a pattern of high frequency active phases separated by pauses (Nambu et al., 2000; Kita et al., 2004), perhaps due to the effects of the inhibitory intranuclear collaterals that had been released from STN influence and regulation. Importantly, these studies highlight the role of excitatory STN input in regulating firing of GPe neurons.

In the normal state, activity in GPe neurons and STN neurons is uncorrelated and asynchronous, with complex spatiotemporal firing related to movement (Bevan et al., 2002b). In parkinsonian animals, however, activity in GPe and STN neurons becomes synchronous and correlated, with an increase in tremor-related (3–8 Hz) and beta-frequency (13–30 Hz) oscillations (Cruz et al., 2011; Tachibana et al., 2011). It is proposed that the Npas1⁺-Foxp2⁺ (arkypallidal) GPe neurons receive much stronger input from the STN than do the PV⁺ (prototypic) GPe neurons (Nevado-Holgado et al., 2014). When combined with the strengthening of intrapallidal connections between GPe neurons that is seen in ex vivo slices from PD model rats (Miguelez et al., 2012), this would likely contribute to the pathological synchrony of the GPe-STN loop that develops in PD. It is clear that the organization of the GPe-STN network is more complex than just the simple reverberating feedback loop that was proposed originally (Bevan et al., 2002b).

As GPe and STN projections also converge on neurons in the internal globus pallidus and substantia nigra pars reticulata (the basal ganglia output nuclei), pathological oscillatory activity in the GPe-STN loop could have a major influence on basal ganglia dysfunction in PD (Terman et al., 2002; Hashimoto et al., 2003). In fact, the GPe-STN loop has been implicated in the onset, progression, and maintenance of dysfunctional oscillatory activity in PD (Bergman et al., 1994; Nini et al., 1995). In support of this theory, high-frequency (130–180 Hz) electrical stimulation (HFS) of the STN improves motor symptoms and is the neurosurgical treatment of choice for mid- to late-stage PD (Starr et al., 1998; DeLong & Wichmann, 2001; Wichmann & Delong, 2006; Johnson et al., 2008; Bronstein et al., 2011; Wichmann & Delong, 2011; DeLong & Wichmann, 2012; DeLong & Wichmann, 2015). However, it is not clear that the therapeutic effects of HFS result from an increase in STN activity (McIntyre & Hahn, 2010), as lesioning of the STN alleviates motor symptoms in the MPTP primate model of PD (Bergman et al., 1990) while a similar effect is achieved by application of HFS directly to the GPe in both MPTP monkeys (Johnson et al., 2012; Vitek et al., 2012) and PD patients (Vitek et al., 2004). Furthermore, repetitive activation of STN neurons leads to a reduction in their excitability as a consequence of decreased voltage-gated sodium channel availability (Beurrier et al., 2001; Do & Bean, 2003). However, microdialysis measurements in the GPe have shown increased extracellular glutamate levels after STN-HFS, supporting the notion that HFS functions through activation of STN neurons instead (Windels et al., 2000). Consistent with this, STN-HFS induced c-fos expression in GPe neurons (Shehab et al., 2014). STN-HFS in awake monkeys also increased the average firing rate in the GPe (Hashimoto et al., 2003). Additionally, using an MPTP monkey model of PD, Bar-Gad and colleagues demonstrate that HFS of the STN induces phase-locking of GPe neuron to the stimuli (Bar-Gad et al., 2004).

While STN-HFS impacts the activity of the neurons in the STN, the subsequent effects on their synaptic partners may be a possible explanation for the results obtained in these studies, HFS in MPTP monkeys supports the notion that antidromic stimulation of GPe neurons can contribute to the relief of bradykinesia (Johnson et al., 2012). Similarly, recent evidence suggests that STN-HFS exerts a therapeutic effect by antidromically activating layer 5 neurons in the motor cortex (Gradinaru et al., 2009; Li et al., 2012). This idea is supported by the literature showing STN-HFS is effective in suppressing abnormal activity in the motor cortex of PD patients (Sabatini et al., 2000; Payoux et al., 2004; Haslinger et al., 2005) and transcranial stimulation of the motor cortex is also efficacious in ameliorating motor symptoms of PD (Cioni, 2007; De Rose et al., 2012; Broeder et al., 2015) and levodopa-induced dyskinesias (Ferrucci et al., 2015). In summary, it is very likely that STN-HFS alters the temporal structure and dynamics of a complex set of pathways within the entire cortico-basal ganglia-thalamocortical loop.

Multiplicity of excitatory inputs to the GPe

In addition to the STN, other sources of excitatory input to the GPe come from the thalamus, the cortex, and the pedunculopontine nucleus (PPN). Overall, very little is known about these projections, including whether or not they target distinct classes of GPe neurons.

Thalamic input to the GPe arises from the caudal intralaminar nuclei, consisting of centromedian (CM) and parafascicular (Pf) nuclei. Though these structures are anatomically less well-defined, the CM-Pf complex is conserved in rodents. As such, tracing experiments have shown that the CM-Pf projects topographically to the GPe in a manner that parallels the thalamo-dStr projections (Kincaid et al., 1991b; Sadikot et al., 1992; Deschenes et al., 1996; Smith et al., 2004; Yasukawa et al., 2004). These thalamic inputs to the GPe arise from collaterals that travel parallel to the GPe-dStr border, with some en passant boutons along the course, and terminate in dense aggregates of various sizes on the proximal dendrites of GPe neurons (Yasukawa et al., 2004). Electrophysiological data show that electrical stimulation of the thalamus evokes large excitatory postsynaptic potentials in some GPe neurons, suggesting that the thalamic input can have a powerful influence on a subset of GPe neurons (Yasukawa et al., 2004). It is tempting to speculate that at least a subset of Npas1⁺ neurons are the major recipient of thalamic (or cortical—see below) inputs; a recent modeling study further reinforces this notion (Nevado-Holgado et al., 2014). CM-Pf also receives inputs from PV⁺ GPe neurons (Shammah-Lagnado et al., 1996; Mastro et al., 2014). The importance of this complex feedback is not understood. Importantly, neuronal loss in the CM-Pf is observed in PD, and though it does not appear to correlate with the severity of motor symptoms, it may impact non-cognitive aspects of PD via altered GPe function (Brown et al., 2010; Kato et al., 2011; Bradfield et al., 2013; Smith et al., 2014).

While cortical input has traditionally been thought to reach the GPe through the cortico-dStr-GPe and the cortico-STN-GPe pathways (Ryan & Clark, 1991; Kita, 1992; Yoshida et al., 1993; Nambu et al., 2000; Nambu, 2004; Kita & Kita, 2011a), there is increasing evidence of a direct cortical input to the GPe. Imaging data from humans suggest a direct projection from motor, orbitofrontal, and dorsolateral prefrontal cortex to the GPe (Milardi et al., 2015). Tracer injections into the precentral medial and lateral cortices in rodents anterogradely label the ipsilateral but not contralateral GPe (Naito & Kita, 1994). In monkeys, cortical terminals labeled by VGluT1 target the dendritic spines and small dendrites throughout the GPe (Smith & Wichmann, 2015). At the same time, the cortex receives direct inputs from both GABAergic and cholinergic GPe neurons (Chen et al., 2015; Saunders et al., 2015); the neuronal identity of the former is at present unknown. As the cortex and its downstream synaptic influence undergo remodeling in models of PD (Kita & Kita, 2011a; Guo et al., 2015a), further investigation of the functional properties of the direct cortical projection to the GPe is warranted.

The PPN is reciprocally connected with the GPe, sending mixed glutamatergic and cholinergic projections (Saper & Loewy, 1982; Gonya-Magee & Anderson, 1983; Moriizumi & Hattori, 1992; Charara & Parent, 1994; Lavoie & Parent, 1994; Mena-Segovia et al., 2004; Dautan et al., 2014; Eid et al., 2014). Tracing experiments show that PPN input is sparse, but arborizes profusely in the ventral third of the GPe, with some poorly branched fibers found dorsally (Lavoie & Parent, 1994). These inputs synapse onto the soma and proximal dendrites of the GPe neurons and elicit action potentials after electrical stimulation of the PPN (Gonya-Magee & Anderson, 1983; Lavoie & Parent, 1994).

Dopaminergic and other neuromodulatory inputs to the GPe

Dopamine, its metabolites, and its associated metabolic enzymes are present in the GPe at relatively high levels (Carlsson, 1959; Bernheimer, 1964; Hornykiewicz, 1966; Hornykiewicz et al., 1968; Vogel et al., 1969; Broch & Marsden, 1972; Rosengren et al., 1985). Accordingly, dopaminergic fibers traverse the rodent, primate, and human GPe (Mettler, 1970; Fallon & Moore, 1978; Lindvall & Bjorklund, 1979; Arluison et al., 1984; Parent & Smith, 1987; Lavoie et al., 1989; Ciliax et al., 1995; Gaykema & Zaborszky, 1996; Rodrigo et al., 1998; Ciliax et al., 1999; Cossette et al., 1999; Gauthier et al., 1999; Hedreen, 1999; Jan et al., 2000; Kirik et al., 2000; Prensa et al., 2000; Prensa & Parent, 2001; Fuchs & Hauber, 2004; Eid & Parent, 2015). There is evidence that dopaminergic neurons in both the substantia nigra and the ventral tegmental area project to the GPe (Lindvall & Bjorklund, 1979; Smith et al., 1989; Kincaid et al., 1991b; Charara & Parent, 1994; Gauthier et al., 1999; Debeir et al., 2005); as diverse subtypes of dopaminergic neurons are intermingled in the midbrain (Poulin et al., 2014; Anderegg et al., 2015), having genetic access to distinct classes of midbrain dopamine neurons will help determine if the GPe-projecting dopaminergic neurons belong to a mixture of cell classes or a single cell class.

Ultrastructural analysis confirms the presence of direct dopaminergic synaptic contacts on pallidal neurons (Rodrigo et al., 1998; Smith & Kieval, 2000; Eid & Parent, 2015). Moderate levels of D1- and D2-class receptor binding are evident in rodent, primate, and human GPe (Martres et al., 1985; Boyson et al., 1986; Dawson et al., 1986; Dubois et al., 1986; Savasta et al., 1986; Charuchinda et al., 1987; Richfield et al., 1987; Beckstead, 1988; Beckstead et al., 1988; Dawson et al., 1988; Richfield et al., 1989; Mansour et al., 1990; Bouthenet et al., 1991; Fremeau et al., 1991; Joyce et al., 1991; Mansour et al., 1991; Mengod et al., 1991; Rao et al., 1991; Janowsky et al., 1992; Mansour et al., 1992; Mengod et al., 1992; Wamsley et al., 1992; Kessler et al., 1993; Levant et al., 1993; Herroelen et al., 1994; Murray et al., 1994; Hall et al., 1996; Carey et al., 1998; Gurevich & Joyce, 1999). In situ hybridization, tissue-level PCR, and immunohistochemical studies have argued that at least some of these receptors are postsynaptically expressed by cells residing within the GPe (Meador-Woodruff et al., 1989; Najlerahim et al., 1989; Mansour et al., 1990; Bouthenet et al., 1991; Joyce et al., 1991; Mansour et al., 1991; Weiner et al., 1991; Mansour et al., 1992; Fox et al., 1993; Larson & Ariano, 1995; Mrzljak et al., 1996; Ariano et al., 1997; Gurevich & Joyce, 1999; Marshall et al., 2001; Shin et al., 2003; Billings & Marshall, 2004; Hoover & Marshall, 2004). Nevertheless, an overwhelming majority of the dopamine D2 receptors are associated with dStr axons and their terminals (Hadipour-Niktarash et al., 2012). Local activation of dopamine receptors in the GPe produces stereotypy and increased locomotor activity, while local dopamine receptor blockade in the GPe produces profound akinesia and catalepsy (Costall et al., 1972a; b; Hauber & Lutz, 1999). Although the cellular effects of D1-class receptors on GPe neurons are poorly understood, the activation of D2-class receptors generally suppresses the responsiveness of GPe neurons to dStr GABAergic inputs via both pre- and postsynaptic mechanisms (Cooper & Stanford, 2001; Shin et al., 2003; Hernandez et al., 2006; Watanabe et al., 2009; Chuhma et al., 2011; Miguelez et al., 2012). Overall, the effects of dopaminergics on GPe neurons in vivo are more heterogeneous. In general, a decrease in the firing of GPe neurons is reported following systemic administration of haloperidol, a D2-class receptor antagonist. On the contrary, apomorphine, a non-selective dopamine receptor agonist, increases firing of GPe neurons (Bergstrom et al., 1982; Carlson et al., 1987; Walters et al., 1987; Napier et al., 1991; Kelland et al., 1995; Mamad et al., 2015). Finally, nigral dopaminergic neurons are subjected to feedback regulation by a heavy projection from the GPe itself (Celada et al., 1999; Paladini et al., 1999; Cobb & Abercrombie, 2003; Lee et al., 2004; Brazhnik et al., 2008; Watabe-Uchida et al., 2012). The circuit effects arising from this feedback regulation may be responsible for the variation in the responses of GPe neurons to dopaminergic agents.

Compelling evidence has suggested that disruption of dopamine signaling within the GPe is causally linked to motor symptoms. Severe loss (up to 90%) of dopamine (and its metabolites) as well as dopaminergic fibers within the GPe have been observed in human patients and animal models of PD (Jan et al., 2000; Kirik et al., 2000; Rajput et al., 2008). Accordingly, intrapallidal administration of dopamine is capable of restoring normal sensorimotor behavior and ameliorating the motor symptoms of dopamine-depleted animals (Galvan et al., 2001). As nigral dopamine neurons also release sonic hedgehog, brain-derived neurotrophic factor, and other trophic factors (Seroogy et al., 1994; Gonzalez-Reyes et al., 2012), it remains to be determined if toxin-animal models and human patients of PD suffer from additional complex alterations of neurochemistry following the loss of dopaminergic neurons. It will be important in the future to specifically manipulate different nodes along these signaling cascades to pinpoint where in a pathway and what molecules are crucial to proper cellular and circuit function.

In addition to dopaminergic inputs, the GPe receives abundant serotonergic innervation from the dorsal raphe (DeVito et al., 1980; Pasik et al., 1984; Vertes, 1991; Charara & Parent, 1994; Di Matteo et al., 2008; Bang et al., 2012; Eid et al., 2013; Ogawa et al., 2014; Pollak Dorocic et al., 2014) and expresses a plethora of serotoninergic receptors (Appel et al., 1990; Hoyer et al., 1990; Sijbesma et al., 1990; Sijbesma et al., 1991; Waeber et al., 1994; Waeber & Moskowitz, 1995; Wright et al., 1995; Compan et al., 1996; Vilaro et al., 1996; Bonaventure et al., 1998; Castro et al., 1998; Morales et al., 1998; Sari et al., 1999; Bonaventure et al., 2000; Clemett et al., 2000; Riad et al., 2000; Neumaier et al., 2001; Li et al., 2004; Martin-Cora & Pazos, 2004; Sari, 2004; Mostany et al., 2005). Accordingly, stimulation of the dorsal raphe nucleus evokes an increase of serotonin levels in the GPe (McQuade & Sharp, 1997). Serotoninergic receptor activation controls GPe neuron activity via complex mechanisms (Chadha et al., 2000b; Kita et al., 2007; Chen et al., 2008a; Hashimoto & Kita, 2008; Zhang et al., 2010; Miguelez et al., 2014). Hyperinnervation by serotonergic axons and increased serotonergic receptor responses occur following dopamine depletion (Di Matteo et al., 2008; Zhang et al., 2010) (but see Zeng et al., 2010). However, serotoninergic neurons eventually degenerate in advanced stages of PD (Halliday et al., 1990; Jellinger, 1990; Chinaglia et al., 1993; Kish, 2003). Finally, additional neuromodulatory inputs such as those from cholinergic neurons also target the GPe while there is no evidence for the existence of a noradrenergic innervation (Rodrigo et al., 1998).

Astrocytic regulation of the GPe

Glia are the most abundant cell type in the GPe. Astrocytes alone are estimated to outnumber neurons by an order of magnitude (Lange et al., 1976; Karlsen & Pakkenberg, 2011; Salvesen et al., 2015). This abundance implies that astrocytes play a crucial role in regulating GPe function. Astrocytes are important integral elements of neural circuits, where they integrate local and long-range modulatory signals through the expression of a myriad of surface receptors and transporters (Theodosis et al., 2008; Perea et al., 2009; Nedergaard & Verkhratsky, 2012; Araque et al., 2014). It has been shown that GPe astrocytes express glutamate (Glt1 and Glast) and GABA (GAT1 and GAT3) transporters (Furuta et al., 1997; Galvan et al., 2010; Jin et al., 2011; Scimemi, 2014). Astrocytes in turn regulate the spatiotemporal dynamics of activation, deactivation, and desensitization of neuronal receptors. In addition, astrocytes have the potential to release neuroactive substrates onto neurons (Halassa et al., 2007; Perea et al., 2009). While we have begun to understand the role of astrocytes in a few selective brain areas (Halassa et al., 2007; Araque et al., 2014), the biological importance and disease relevance of astrocytes in the basal ganglia is largely unexplored (Maragakis & Rothstein, 2006; Sofroniew & Vinters, 2010; Villalba & Smith, 2011; Chan & Surmeier, 2014; Tong et al., 2014; Martin et al., 2015). In light of this, it will be important to study how astrocytic regulation of synaptic inputs and chemical homeostasis in the GPe is altered in neurological diseases, such as PD.

Behavioral and clinical relevance of the GPe

The role of the GPe in movement

Results from lesion and pharmacological activation studies provide conflicting evidence as to the role of the GPe in movement. Unilateral ibotenic acid or kainic acid lesion of the GPe leads to ipsilateral turning behavior (Ossowska et al., 1983; Konitsiotis et al., 1998), bilateral quinolinic acid lesion of the GPe leads to a decrease in spontaneous movement (Hauber et al., 1998), and selective activation of the GPe by GABA_A antagonist microinjection increases spontaneous movement (Grabli et al., 2004) and dyskinesias (Crossman et al., 1984; Matsumura et al., 1995) in monkeys; all of these phenotypes are predicted by the classic basal ganglia model. Conversely, ibotenic acid GPe lesions in monkeys produce no motor deficits (Soares et al., 2004), and increased spontaneous movement has also been observed in rats with bilateral GPe lesion (Norton, 1976; Joel et al., 1998), findings that are inconsistent with the classic model. Furthermore, rats with bilateral GPe lesions display increased inaccuracy in a reaching task but not gross movement, suggesting a specific deficit in performance of organized limb movements (Schneider & Olazabal, 1984). There are several possibilities for the discrepancies between findings from different research groups, such as differences in the extent of lesion or the spread of drugs.

Studies of movement-related electrophysiological activity in the GPe have provided additional insight into its role in movement. Like neurons of other basal ganglia nuclei, GPe neuron firing patterns are related to movement amplitude, velocity, and direction (Georgopoulos et al., 1983; Mitchell et al., 1987; Turner & Anderson, 1997; Gage et al., 2010). This activity is context-dependent and can be modulated by the presence of external cues that signal whether a particular movement should be performed (Turner & Anderson, 1997; 2005; Gage et al., 2010). This cue-related activity in the GPe is consistent with the involvement of the basal ganglia in the integrative processing of movement with sensory information.

It is proposed that the basal ganglia control sequences of behavior via “chunking” (Graybiel, 1998; Smith & Graybiel, 2014). In other words, the basal ganglia piece together related individual movements into single movement sequences to accomplish complex movement patterns (Levesque et al., 2007; Tremblay et al., 2009; Tremblay et al., 2010). Previous studies have found that GPe neurons can have distinct patterns of firing activity within different temporal phases of a movement or movement sequence, preferentially firing before, after, or throughout a movement (Anderson & Horak, 1985; Shi et al., 2004; Turner & Anderson, 2005; Jin et al., 2014); these movement-related firing behaviors have been found to correlate with the molecular identity of the GPe neurons (Dodson et al., 2015) and are thought to collectively represent a chunked motor sequence (Jin et al., 2014).

The integrative role of the GPe—a circuit perspective

The GPe may play a role in reactive action cancellation, or suppression of planned actions (Schmidt et al., 2013; Gittis et al., 2014; Leventhal et al., 2014). Action cancellation is often studied in animals performing a stop-signal task, in which a “Go” cue is given to cue a rapid, specific movement, and a “Stop” cue is given just after the “Go” cue in a subset of trials to indicate the subject should cancel the movement (Schmidt et al., 2013; Leventhal et al., 2014). Studies suggest that the behavioral response to these signals—whether subjects are able to successfully stop after hearing the “Stop” cue—involves a competition between distinct basal ganglia circuits (Schmidt et al., 2013). Specifically, dSPNs in the dStr mediate relatively slow “Go” signals, whereas STN neurons, likely by receiving direct input from the cortex, mediate faster “Stop” signals. However, as “Stop” signals from the STN lead to only transient inhibition of substantia nigra pars reticulata neurons, it is thought that an additional pathway must mediate inhibition of “Go” signals for full behavioral stopping to occur (Schmidt et al., 2013). A recent study suggests that the projection from the GPe to the dStr could provide this inhibition of “Go” signals (Mallet et al., 2016). However, this raises the question of whether the GPe projection to the dStr is also involved in the facilitation of action by appropriately inhibiting “No-go” signals immediately prior to movement initiation. Conversely, it is possible that excessive inhibition of the dStr by the GPe could underlie the hypokinetic symptoms (i.e., bradykinesia and akinesia) that are characteristic of PD.

The GPe is clinically important

As we have only begun to reliably identify GPe neurons, very little is known about how different classes of GPe neurons are involved in the symptomatology of movement disorders. An overview of the relevant literature has been provided throughout this review. The following section highlights a number of movement disorders in which the GPe is critically involved.

Dopamine loss leads to altered physiology in the GPe in PD. The altered firing behavior of GPe neurons is one of the most striking and consistent electrophysiological signatures of PD. Compelling evidence suggests that disruption of dopamine signaling within the GPe is causally linked to hypokinetic symptoms of PD. The loss of dopamine shifts the firing pattern of GPe neurons from decorrelated spiking to synchronized, oscillatory bursts (Pan & Walters, 1988; Filion & Tremblay, 1991; Filion et al., 1991; Hutchison et al., 1994; Nini et al., 1995; Rothblat & Schneider, 1995; Hassani et al., 1996; Taha et al., 1996; Bergman et al., 1998; Boraud et al., 1998; Wichmann et al., 1999; El-Deredy et al., 2000; Magill et al., 2000; Magnin et al., 2000; Raz et al., 2000; Brown et al., 2001; Magill et al., 2001; Bar-Gad et al., 2003; Starr et al., 2005; Heimer et al., 2006; Wichmann & Soares, 2006; Kita, 2007; Tang et al., 2007; Zold et al., 2007a; Zold et al., 2007b; Mallet et al., 2008; Starr et al., 2008; Sani et al., 2009; Chan et al., 2011). Accordingly, administration of dopamine to the GPe is capable of reversing abnormal synchrony among GPe neurons (Heimer et al., 2002) and restoring normal sensorimotor behavior and ameliorating the motor symptoms of dopamine-depleted animals (Galvan et al., 2001). In addition, high frequency stimulation of the GPe leads to symptomatic improvement in parkinsonian monkeys (Johnson et al., 2012; Vitek et al., 2012) and PD patients (Vitek et al., 2004).

In addition to PD, HD is another major basal ganglia disorder. It is an autosomal dominant neurodegenerative disease that leads to progressive impairments in motor function, cognition, and behavior. A characteristic feature of HD is the presence of chorea (Zuccato et al., 2010; Plotkin & Surmeier, 2015). There is no apparent cell death in the GPe in HD until late in the disease progression (Reiner et al., 2011; Waldvogel et al., 2015). Instead, loss of GABAergic SPNs results in a dramatic upregulation of GABA_A receptor-subunits in the GPe (Faull et al., 1993; Waldvogel & Faull, 2015). Currently, there is very little information available concerning the firing behavior of GPe neurons in HD. Consistent with its presumptive hyperactivity (Temel et al., 2006; Starr et al., 2008), GPe lesioning or electrical stimulation leads to symptomatic relief in both animal models and human patients (Ligot et al., 2011; Beste et al., 2015; Nagel et al., 2015). Further investigation of the changes that occur in the GPe in HD will be facilitated by the array of genetic mouse models that have been developed (Plotkin & Surmeier, 2015).

Finally, though its etiology has yet to be discovered, there is compelling evidence from both human patients and animal models of dystonia that aberrant activity is present in the GPe (Starr et al., 2005; Chiken et al., 2008; Baron et al., 2011; Nambu et al., 2011; Nishibayashi et al., 2011). Dystonia is characterized by involuntary repetitive twisting and sustained muscle contractions that result in abnormal movements and postures (Breakefield et al., 2008; Schwarz & Bressman, 2009; Tanabe et al., 2009; Albanese et al., 2013; Jinnah & Factor, 2015). Recent evidence further suggests that altered cortico-dStr-GPe signaling may underlie the altered firing of GPe neurons in dystonia (Nishibayashi et al., 2011).

Concluding remarks

In summary, despite the clinical importance of the GPe, we have only limited information about its cellular composition and organizational principles. This undermines our understanding of the GPe in motor function and dysfunction.

This article reviews the literature on neuron diversity in the GPe. The discovery of novel cellular markers revealed that the heterogeneity in GPe neurons’ anatomical and electrophysiological properties observed in early studies could be correlated to their molecular signatures. We have only begun to understand the cellular makeup of the rodent GPe. An important next step is to refine the classification schemes that have been developed and to identify the specific inputs and outputs of distinct GPe neuron classes. We can begin to accomplish this goal using currently available genetic tools, including the Npas1 mouse line developed by our group. The identification of novel cell-specific markers will undoubtedly continue to shape future research. Neuronal diversity is an emerging theme in the basal ganglia (Kreitzer, 2009; Tepper et al., 2010; Antal et al., 2014; Barrot, 2014; Poulin et al., 2014; Anderegg et al., 2015; Xiao et al., 2015), and it will be important to fully understand how distinct classes of GPe neurons are integrated into large-scale computations.

In conclusion, this review provides an overview of the complex reciprocal loops formed between GPe neurons and their synaptic partners in addition to neuronal diversity in the GPe. As a whole, the literature now argues that the GPe should no longer be considered a simple relay in which information flows unidirectionally from the dStr to GPe.

Abbreviations

ChAT

choline acetyltransferase, cholinergic

CM

centromedian nucleus of thalamus

dSPNs

direct-pathway spiny projection neurons

dStr

dorsal striatum

GABA

γ-aminobutyric acid

GPe

external globus pallidus

HD

Huntington’s disease

HFS

high-frequency stimulation

iSPNs

indirect-pathway spiny projection neurons

LGE

lateral ganglionic eminence

MGE

medial ganglionic eminence

mGluRs

metabotropic glutamate receptors

PD

Parkinson’s disease

Pf

parafascicular nucleus of thalamus

PoA

preoptic area

PPN

pedunculopontine nucleus

PV

parvalbumin

SPNs

spiny projection neurons

STN

subthalamic nucleus