Updating the striatal-pallidal wiring diagram

Abstract

The striatal and pallidal complexes are basal ganglia structures that orchestrate learning and execution of flexible behavior. Models of how the basal ganglia subserve these functions have evolved considerably, and the advent of optogenetic and molecular tools has shed light on the heterogeneity of sub-circuits within these pathways. However, a synthesis of how molecularly diverse neurons integrate into existing models of basal ganglia function is lacking. Here, we provide an overview of the neurochemical and molecular diversity of striatal and pallidal neurons and synthesize recent circuit connectivity studies in rodents that takes this diversity into account. We also highlight anatomical organizational principles that distinguish the dorsal and ventral basal ganglia pathways in rodents. Future work integrating the molecular and anatomical properties of striatal and pallidal subpopulations may resolve controversies regarding basal ganglia network function.

Introduction

The basal ganglia are a collection of subcortical nuclei, whose coordinated activity supports learning and flexible selection of behavior. The striatum is the primary input nucleus of the basal ganglia, integrating excitatory input from cortex, thalamus, hippocampus and amygdala along topographically-organized information streams. The striatum is comprised predominantly of GABAergic spiny projection neurons (SPNs) stratified into equal numbers of “direct pathway” SPNs (dSPNs) and “indirect pathway” SPNs (iSPNs) based on their projection targets. According to classical basal ganglia models, dSPN firing promotes motor actions while iSPN activity inhibits motor actions. Specifically, dSPN activity inhibits the GABAergic entopeduncular nucleus (EPN; homologue of the primate internal segment of the globus pallidus (GPi) and substantia nigra pars reticulata (SNr), thereby relieving tonic inhibition from GPi/SNr onto motor thalamic nuclei and brainstem structures, permitting motor actions. Conversely, iSPN activity inhibits the external segment of the globus pallidus (GPe), which relieves GPe-mediated inhibition of the glutamatergic subthalamic nucleus (STN,(1–3); in turn, STN neurons excite the EPN and SNr, thus opposing the direct pathway (Fig. 1). While this model has long been recognized to be oversimplified, it has been enormously influential, having informed computational models of action selection and learning and inspired neuromodulation therapies for basal ganglia disorders (4–6).

Figure 1. Basal ganglia pathways.

(a) Classic basal ganglia pathway model; direct and indirect striatal output pathways are segregated and innervate midbrain and pallidal structures, respectively. (b) Excitatory inputs to the striatal complex are highlighted. Cortico-striatal inputs arise from MO, SS and OFC and are topographically organized. Ventral striatal inputs arise from PrL, IL, VH, BLA, and PVT. (c) Striatal output pathways are highlighted. iSPNs from the dorsal and ventral striatum preferentially target the pallidal complex, dSPNs target the midbrain with bridging collaterals innervating the GP and STN in the dorsal stream and VP and LH in the ventral stream. (d) Pallidal output pathways are highlighted. Arkypallidal neurons in the GP and VP target the DSt and NAc, respectively, with a subpopulation traversing striatum to innervate motor cortex. In the GP, discrete populations of prototypical GABAergic neurons innervate the PFT and midbrain. Within the VP, glutamatergic and GABAergic prototypical neurons innervate the LHb, RMTg and VTA. Abbreviations: DSt: Dorsal striatum, NAc: nucleus accumbens/ventral striatum, MO: motor cortex, SS: somatosensory cortex, OFC: orbitofrontal cortex, IL: infralimbic and PrL: prelimbic cortex, VH: ventral hippocampus, BLA: basolateral amygdala, LH: lateral hypothalamus, STN: subthalamic nucleus, PF: parafascicular thalamus, LHb: lateral habenula, RMTg: rostromedial tegmental nucleus, VTA: ventral tegmental area, SNr: substantia nigra, MDT = midline thalamus.

Genetic tools that enable monitoring and manipulation of the basal ganglia with cell-type specific resolution have substantially advanced our understanding of the molecular and functional diversity within basal ganglia circuits. Here, we propose an updated ‘wiring diagram’ of the striatal and pallidal complexes, synthesized from contemporary studies, with a focus on rodents. This updated wiring diagram will increase our understanding of how these nuclei work in concert to orchestrate reward-guided decision making and motor action selection in health and disease. The advent of circuit-specific neuromodulation and cell-type-specific pharmacology may enable us to leverage this molecular heterogeneity into new therapies for movement, mood and substance use disorders, which are all characterized by altered function of basal ganglia circuits.

Striatal composition and microcircuitry

The striatum consists of GABAergic SPNs and interneurons. In both the dorsal striatum (DSt) and nucleus accumbens (NAc), SPNs comprise >95% of the neuron population, and segregate into equal proportions of D1-SPNs and D2-SPNs, according to their expression of either the dopamine D1- or D2- receptor (7). Classically, D1-SPNs are synonymous with dSPNs, and D2-SPNs with iSPNs. The remaining cell types are interneurons, which have been functionally classified as fast-spiking interneurons, low-threshold spiking interneurons, and tonically active interneurons (Fig. 2a). While the molecular identity of striatal interneurons is complex (8), single-cell sequencing has confirmed the existence of discrete classes of parvalbumin (PV)-expressing fast-spiking interneurons and cholinergic interneurons, the latter presumably overlapping with the tonically active interneuron population (9,10).

Figure 2. Neural circuit diagram of the striatum.

(a) Proportion of neurochemically-defined neuronal subpopulations in the striatum, with >95% corresponding to D1- (blue) and D2-SPNs (pink), and the remainder constituting functionally-distinct interneurons [cholinergic tonically active interneurons (ChAT; orange), parvalbumin-expressing fast-spiking interneurons (PV; gray), low-threshold spiking interneurons (LTS; yellow)]. (b) Neural circuit ‘wiring diagram’ of the striatum. Organizational principles that apply to the dorsal and ventral striatum are shown. D1- and D2-SPNs receive equivalent excitatory inputs from cortex, hippocampus, and thalamus. The strength of these contacts is functionally stronger onto interneurons than SPNs. PV interneurons dynamically regulate SPN firing by inhibiting PV-negative interneurons, distal PV interneurons, and CINs. Pathway trajectories are not anatomically accurate, pathway weights represent reported synaptic strength or density of monosynaptic projections. Abbreviations: D1 = D1-SPNs (blue), D2 = D2-SPNs (pink), CIN = cholinergic interneuron (gray), IN = non-cholinergic interneuron (orange), PV = PV-positive neuron (orange), VH = ventral hippocampus, Blue = glutamate, Pink = GABA, Black = Dopamine.

Striatal SPNs integrate excitatory input from multiple sources.

SPNs, also referred to as ‘medium sized spiny neurons’ due to their extensive spiny dendritic arbors, are the principal cell type of the striatum. These dendritic arbors provide a substrate for the synaptic integration of thousands of long-range excitatory inputs from upstream neurons (11). In vivo, the membrane potential of SPNs fluctuates between hyperpolarized ‘down-states’ (~-70–80 mV) and relatively depolarized (~45–60 mV) ‘up-states’ (12,13). Convergent excitatory drive onto the spiny dendrites of SPNs is required to generate depolarizing plateau potentials that drive the SPN cell body into an ‘up-state’ from which it can fire action potentials (14,15). Within individual SPNs, rapid desensitization of AMPA receptors prevents increased calcium influx and subsequent changes in membrane potential by repeated activity at a single excitatory synapse (16,17). The influence of asynchronous excitatory transmission in distributed synapses on SPN membrane potential is further reduced by the compartmentalization of calcium, which is imposed by the geometry of dendritic spines. Rather, synchronous excitatory transmission onto distal dendrites at multiple spatially-clustered synapses is required for local membrane depolarization (17,18). Even relatively small increases in membrane voltage are sufficient to locally relieve the magnesium block from NMDA receptors and activate T- and R-type voltage-dependent calcium channels. The resulting calcium increase drives prolonged depolarized up-states in the cell body (18). Despite the hyperpolarized resting membrane potential of SPNs and tens of thousands of dendritic spines on each neuron, only 10–15 simultaneously activated spines are required to initiate upstate transitions (14,15). Together, this indicates that coincident activity in few upstream excitatory neurons, potentially from discrete brain regions, are sufficient to drive SPNs into the upstate.

The striatum integrates inputs from diverse brain regions, including cortical, thalamic and limbic structures (Fig 2b). Each of these excitatory inputs exhibits gradients of connectivity along the dorsal–ventral and medial–lateral axes of the striatum. Specifically, the frontal associative cortex, primary and secondary motor cortices preferentially project to DSt, whereas prelimbic, infralimbic, amygdala and hippocampal excitatory inputs project to the NAc (19,20). Convergent thalamic inputs follow an anterior–posterior gradient throughout the striatum, with parafascicular (PF) and neighboring midline thalamus projecting to the DSt, and paraventricular thalamus (PVT) innervating the NAc (20).

Unique information is thought to be conveyed by excitatory inputs originating from these distinct upstream regions. For example, somatosensory and motor cortices convey sensory-motor contextual information (21,22) while infralimibic and prelimbic inputs have been suggested to encode representations of action–outcome associations (23–25). With respect to thalamic inputs, projections arising from the PF complex are critical for sensory-driven motor responses that are modulated by expectation (26,27), while PVT inputs relay proprioceptive and interoceptive information (28,29). Hippocampal (30–32) inputs communicate environmental and sensory context information, while amygdala (33–35) inputs carry information related to stimulus valence in decision-making tasks. Thus, although the striatum lacks a laminar organization, functional sensorimotor, associative, and limbic striatal domains (corresponding loosely to dorsolateral, dorsomedial and ventral striatum) are instantiated by the topographically-organized connection patterns of excitatory inputs (20).

Beyond this macroscale topographical organization, viral genetic approaches have revealed the microcircuitry of these anatomical arrangements. Reconstruction of D1- and D2- SPNs with labeling of cortical and thalamic inputs reveals that both populations receive equivalent input from cortex and thalamus impinging on their distal dendrites. Rabies tracing confirms that most inputs to DSt SPNs arise from cortex and thalamus, preferentially PF and midline thalamus, with somatosensory cortex and amygdala preferentially synapsing on D1-SPNs, and motor cortex on D2 SPNs (36,37). A similar pattern emerges in the ventral striatum: here, cortex, thalamus, amygdala and ventral hippocampus have proportionally equivalent inputs to D1-SPNs and D2-SPNs (30,36,38,39), although subcortical inputs onto D1-SPNs have considerably higher functional strength (40,41). This difference in excitatory drive arises from the anatomical arrangement of ventral hippocampal inputs, which contact proximal spines on D1-SPNs and distal dendrites of D2-SPNs (42,43). In addition to their unique proximal location, ventral hippocampus to D1-SPN synapses express NMDA receptors that are relatively insensitive to magnesium blockade at hyperpolarized membrane potentials (30). These two features allow ventral hippocampal inputs to shift D1-SPNs into a depolarized up-state. In contrast, PFC and amygdala inputs are subject to greater gating by subthreshold depolarization, requiring multiple convergent inputs to drive SPN up-state transitions and spiking (12). Together, these anatomical and synaptic properties imbue excitatory inputs to SPNs with the ability to potently regulate SPN membrane potential. Consequently, SPN spiking represents unique combinations of sensory, affective and cognitive information encoded in excitatory SPN-projecting neurons distributed across multiple upstream structures (Fig 2b).

Local and long-range inhibition sculpt striatal output.

Although parvalbumin-positive interneurons (PVINs) represent only ~1% of striatal interneurons and are distributed sparsely throughout the striatum, they potently modulate SPN output. Since PVINs lack dendritic compartmentalization conferred by spines, burst firing can be triggered in response to weak or asynchronous excitatory inputs (44–46). The axonal arbors of PVINs are complex yet compact, with each fast-spiking interneuron making synaptic contacts onto hundreds of SPN somas within a spatially discrete cluster (100–250 μm;(47,48). However, the modulation of striatal output by PVINs is non-linear, as both decreasing and increasing PVIN activity is associated with increased SPN firing at the population level (49,50). This is because PVINs both inhibit proximal SPNs and disinhibit SPN activity through multiple di-synaptic circuits. First, PVINs impinge on GABAergic PV-negative (i.e. neuropeptide Y-expressing) interneurons, which in turn inhibit SPNs (49). Second, PVINs themselves are interconnected by electrical synapses (51). In effect, the activity of individual PVINs can not only suppress firing and regulate plasticity within local SPN ensembles (50), but can also disinhibit distal SPN ensembles by inhibiting distal PVINs (52,53). Finally, PVINs provide strong feed-forward inhibition onto tonically active cholinergic interneurons (CINs,(50,54,55). This CIN inhibition and reduced cholinergic tone regulates SPN excitability through muscarinic receptors on SPNs (56), and promotes local dopamine release through the nicotinic-receptor mediated depolarization of dopaminergic axon terminals (57,58). Thus, PVIN activation not only potently regulates the activity of SPNs, but also governs neuromodulator-dependent plasticity and excitability within local striatal territories.

Striatal PVINs receive excitatory input from the same cortical, thalamic and hippocampal inputs as their surrounding SPNs, however, the strength of synaptic inputs onto PVINs is relatively stronger (Fig 2b)(59,60). Therefore, under conditions of strong afferent excitatory drive to striatum, PVIN activity may further promote the excitability of local spatial clusters of SPNs through disinhibition and altering local neuromodulator tone. In parallel, activated PVINs robustly inhibit distal PVINs, which indirectly suppresses activity in associated spatial clusters of distal SPNs. As a result of this circuit arrangement, activity in individual PVINs is de-correlated (52,61), supporting the role of PVINs in shaping activity and activity-dependent plasticity of local SPN ensembles, rather than synchronizing activity across the striatum (48,50). Consistent with their role in shaping striatal plasticity, PVIN activity is critical for the initial learning of action-selection tasks and sequence learning, but is dispensable for suppression or execution of those actions once associations are acquired (50).

In addition to these intra-striatal sources of inhibition, long-range GABAergic inputs from the pallidum and midbrain also modulate striatal activity in both the dorsal and ventral basal ganglia. While the GPe and VP are classically considered output targets of the DSt and NAc, respectively, recent work has coined the term ‘arkypallidal’ to refer to a subpopulation of pallidal neurons that cast large nets of influence throughout the striatum, forming mono-synaptic contacts with both populations of SPNs and interneurons (62–64). The ventral tegmental area (VTA) and substantia nigra in the midbrain are also important exogenous sources of GABAergic input to the striatum (65–68). Unlike arkypallidal neurons, VTA_GABA neurons nearly exclusively target CIN cell bodies in the NAc, with much sparser innervation of accumbal SPNs (66,68). VTA_GABA neurons are involved in signaling salience in the striatum, by pausing CIN firing to enhance stimulus–outcome association learning. An analogous pathway exists in the DSt, as stimulation of nigrostriatal neurons also induces a pause in CIN firing. However, co-release of GABA and dopamine and subsequent D2R-mediated inhibition of CINs has been implicated in this mechanism in the DSt, rather than purely GABAergic inhibition (65).

Inhibitory input from arkypallidal neurons to striatum occurs under conditions where prepotent or competing actions must be suppressed (63,64). By contrast, the conditions under which inhibitory midbrain GABAergic projections to the striatum are active during behavior are not well understood. Studies have been complicated by the immense diversity of striatum-projecting midbrain neurons, with certain populations capable of co-releasing multiple neurotransmitters (65,69,70), which are independently regulated even within a single pre-synaptic neuron (71). Future work will be needed to understand the microcircuitry of long-range inhibitory inputs, and the conditions under which specific neurotransmitters are released into the striatum. However, long-range inhibitory inputs clearly modulate striatal SPN output and are thus poised to play critical roles in associative learning and motor functions subserved by the striatum.

Competition between D1- and D2-SPNs orchestrates action selection.

Classic basal ganglia models proposed that dSPNs promote while iSPNs inhibit movements (3,4). These models were bolstered by early optogenetic studies, in which bulk activation of D1- and D2- pathways respectively promoted and inhibited locomotion (72) and approach behavior (73). Facilitation of movement by D1-SPNs was proposed to occur through disinhibition of brainstem, cortical and thalamic neurons, while inhibition of movement via D2-SPNs was proposed to occur through inhibition of the same output neurons through an intermediate relay of the GPe (2,3,72,73). These opposing effects of D1- and D2-SPNs on behavior were formalized as the ‘opponent parallel pathway’ hypothesis (3,4). In vivo calcium imaging has challenged this model by showing that D1- and D2-SPN populations are co-active during movement and action selection (74,75). This co-activation could be interpreted the following way: D1-SPNs are tuned to specific actions while D2-SPNs inhibit a variety of competing actions to allow execution of specific behaviors, in line with the ‘focused selection and inhibition of competing motor programs hypothesis’ (76). However, within a given spatially-defined striatal region, D1-SPN and D2-SPN activity are tightly correlated, and activity in both populations is sufficiently specific to allow decoding of behavior (77–79). This argues against the idea that D1-SPNs are tuned to specific motor programs and D2-SPNs provide blanket inhibition of competing actions. A 'competitive model' of striatal action selection has been proposed to reconcile these observations (80).

The 'competitive model' of striatal action selection bridges anatomical principles of striatal organization; it suggests that spatially proximal D1-SPNs and D2-SPNs are tuned to the same combination of excitatory inputs, and that it is the balance of activity between the two within an ensemble that determines whether a given action will be selected in the sensory–affective context encoded by these excitatory inputs (80). In addition to sharing common excitatory inputs, anatomically proximal SPNs exhibit predominantly directional lateral inhibition from D2-SPNs onto neighboring D1-SPNs (Fig 3a) (81). Critically, due to increased voltage-gated sodium channel conductance, D2-SPNs exhibit a lower spike threshold and greater depolarization in response to current injections relative to D1-SPNs (82,83). Consequently, under conditions of asynchronous excitatory drive, D2-SPNs are more readily excited, and shunt subthreshold depolarizations arising from excitatory drive onto D1-SPNs through lateral inhibition (Fig 3a). By contrast, synchronous or convergent excitatory input can overcome relatively weak lateral inhibition and shift D1-SPNs into an up-state from which they can fire. Consistent with this interpretation, the balance between D1-SPN and D2-SPN activation determines whether ongoing actions are aborted or executed, with higher D2-SPN activation associated with action termination or behavioral arrest (84,85). This 'competitive model' of striatal action selection also explains why D1-SPNs and D2-SPNs are co-activated in specific sensory-motor contexts (74,75,77,84,85), and why bulk stimulation of D1-SPNs promotes behavioral selection and repetition (73) whereas D2-SPN stimulation promotes behavioral switching or inhibition (73,86).

Figure 3. Dopamine asymmetrically modulates D1- and D2-SPNs.

(a) Basal conditions of asynchronous excitatory drive and low dopamine tone. The balance of D2-SPN activity over D1-SPN activity is favored, due to intrinsic differences in excitability. (b) Dopamine increases the probability of D1-SPN spiking through inhibition of voltage- and Ca²⁺-dependent K⁺ conductance. (c) Dopamine inhibits synaptic output of D2-SPNs through D2Rs, reducing lateral inhibition onto D1-SPNs. (d) Dopamine signaling through D1Rs enhances PKA-dependent insertion of AMPA receptors, strengthening excitatory drive onto D1-SPNs while simultaneously reducing PKA-dependent signaling in D2-SPNs. Abbreviations: PKA – protein kinase A, D1R – dopamine D1 receptor, D2R – dopamine D2 receptor. AMPA - α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid.

Dopamine bidirectionally modulates D1-SPNs and D2-SPNs.

This 'competitive model' is also consistent with the established role of DA in invigorating action selection by modulating the balance of activity between D1- and D2-SPNs (7,87). Because D1 receptors couple to excitatory G_αs/olf proteins and D2s couple to inhibitory G_αi proteins, dopamine has been proposed to bidirectionally modulate the excitability of these two populations (88). However, the effect of dopamine on SPN excitability has been controversial, as signaling through D1R has been reported to both potentiate (89,90) and inhibit (91) SPN Kir2 channels, which is the primary determinant of SPN membrane voltage in the hyperpolarized resting state (92). Recent work sought to resolve this controversy by performing perforated patch-clamp recording of D1-SPNs to prevent dialysis of second-messenger signaling molecules while optogenetically emulating physiologically-relevant DA release (93). Under these conditions, phasic DA only negligibly altered resting membrane voltage, arguing against D1-mediated modulation of Kir2 channels. Instead, phasic DA robustly increased firing frequency and decreased latency to fire in response to current injections, suggesting facilitated upstate transitions (Fig 3b). This effect was dependent on D1-mediated negative modulation of outward potassium currents mediated by Kv1.2 channels (94,95) (Fig 3b). By facilitating D1-SPN firing frequency, dopamine would facilitate spike-timing-dependent plasticity of excitatory input through back-propagation of action potentials and relief of magnesium block from NMDA receptors in D1-SPNs (96). Over slower timescales, D1R signaling through canonical PKA-dependent cascades can potentiate excitatory transmission onto D1-SPNs by promoting the insertion and stabilization of AMPARs into the post-synaptic membrane (Fig 3b) (97,98).

There is only weak evidence for DA-mediated regulation of D2-SPN excitability through inhibitory D2Rs. However, under conditions of excitatory drive that would sustain SPN firing and induce spike-timing-dependent potentiation, signaling through D2Rs modulates intracellular calcium through L-type calcium channels and promotes synaptic depression to oppose this excitatory influence (98,99). In parallel, D2R signaling suppresses synaptic output of D2-SPNs; genetic ablation of D2Rs selectively from D2-SPNs increases their transmitter release onto neighboring D1-SPNs and pallidal structures (100,101), while supraphysiological dopaminergic stimulation attenuates their synaptic output to these same areas (101,102). This suggests that dopamine signaling through D2Rs reduces both the excitatory drive onto D2-SPNs and their synaptic output (Fig 3c).

Taken together, phasic dopamine can alter excitatory drive onto an SPN ensemble to bias activation of D1-SPNs over D2-SPNs. This occurs through increased D1-SPN excitability (Fig 3b) and decreased lateral inhibition of D2-SPNs onto D1-SPNs (Fig 3c), ultimately favoring activation of D1-SPNs to D2-SPNs, and to execution of actions gated by that SPN ensemble. This coincident SPN activity and phasic DA release potentiates the strength of excitatory synapses onto D1-SPNs while simultaneously de-potentiating the same temporally active inputs onto D2-SPNs (Fig 3d). This bi-directional plasticity is predicted to increase the probability of executing actions gated by that D1-SPN ensemble as subsequent activation of excitatory inputs would more readily activate D1-SPNs.

Summary.

The striatum is the principal input nucleus of the basal ganglia. Due to the morphology and biophysical properties of SPNs, their firing represents a synthesis of topographically-organized excitatory inputs to striatum. SPN firing is sculpted by local and long-range inhibition, as well as neuromodulatory tone. Ultimately, the balance between dSPNs and iSPN firing is thought to determine whether a given action will be executed.

Striatal output pathways

While the opposing effects of dSPN and iSPN firing on motor activity may arise partly from lateral inhibition at the level of the striatum, the prevailing view is that dSPNs promote motor action via direct inhibition of EPN and SNr, which in turn relieves inhibition in motor thalamus and cortex to facilitate actions. In parallel, iSPNs converge on GABAergic GPe neurons, inhibiting EPN and SNr, resulting in inhibition of downstream cortical and thalamic nuclei. However, the advent of transgenic mice has revealed unappreciated complexity in this anatomical arrangement, including the presence of bridging collaterals from dSPNs into pallidal structures and the lack of ‘direct’ and ‘indirect’ pathway segregation in the ventral striatal output pathways.

D1-SPNs modulate GPe activity through bridging collaterals.

Optogenetic tools have functionally confirmed canonical ‘direct’ and ‘indirect’ output pathways of the dorsal striatum in mice. Selective stimulation of striatal D1-SPNs inhibits SNr firing, whereas D2-SPN stimulation decreases firing of GPe neurons with monosynaptic latency (72,103,104). Likewise, both D1-SPN stimulation and D2-SPN inhibition increase spontaneous motor activity and promote action initiation in reward-based tasks (72,73,105). This arrangement offers a parsimonious explanation for the ‘go’ and ‘no-go’ functions of direct and indirect striatal output pathways (Fig 1). However, this segregation of D1- and D2- ‘direct’ and ‘indirect’ pathways is an oversimplification, as the existence of 'bridging collaterals' of D1-SPN axons in the GPe was already suggested by early tracing studies (11,106). It has since been confirmed that these bridging collaterals make functional synapses in GPe (Fig 1c), and that activation of D1-SPNs inhibits GPe firing in vivo, albeit to a lesser extent than D2-SPN-activation (103). D2-SPN terminals inhibit the majority of GPe neurons, whereas D1-SPNs activation appears to selectively impact subpopulations of GPe neurons. Specifically, the GPe has been divided into molecularly-defined populations, and GPe neurons expressing the transcription factors Lhx6+ and NPas1+ are subject to robust monosynaptic inhibition from D1-SPNs activation, whereas neighboring PV+ neurons are not (107,108).

Bridging collaterals from D1-SPNs to GPe are highly plastic even into adulthood. Their tight regulation by striatal D2R-mediated dopamine signaling is evident as enhanced D2R signaling is associated with increased density of bridging collaterals from D1-SPNs into the GPe (103). In parallel, enhanced D2R-signaling attenuates GABA release from D2-SPNs, reducing the functional impact of these D2-SPNs on GPe activity (100,101). It has thus been proposed that D1-SPN collaterals reinforce the actions of D2-SPNs at the level of the GPe. This is supported by the near abolition of the pro-locomotor response — which is normally elicited by D1-SPN activation — after genetic upregulation of the density of bridging collaterals. While this suggests an important allostatic mechanism for maintaining balance of direct and indirect pathway output to GPe, no studies to date have monitored activity of D1-SPN collaterals to GPe to establish under which conditions they are activated, nor have studies been able to selectively manipulate activity of these bridging collaterals without altering striatal excitability (103). Such investigations, perhaps facilitated by terminal-specific optogenetic tools, are required to establish causality between activity of bridging collaterals and behaviors subserved by the basal ganglia.

Ventral striatal output pathways do not follow direct vs. indirect pathway organization.

In the classic direct and indirect pathway model, accumbal D1-SPNs innervate GABAergic midbrain neurons in the VTA, and D2-SPNs innervate the VP (Fig 1a). This is indeed the case, as the majority of accumbal SPNs terminating within the midbrain are D1-SPNs, whereas the majority of accumbal D2-SPNs terminate at the level of VP (109,110). Consistent with the inhibitory influence of D2-SPNs on the VP, stimulation of these D2-SPNs increases firing of VP neurons, despite substantial heterogeneity of individual neural responses (111,112). However, D1-SPNs make up a substantial fraction of the accumbal-pallidal pathway, and over 50% of VP neurons are mono-synaptically innervated by accumbal D1-SPNs (102,113). Both bridging D1-SPN collaterals and the existence of a D1-SPN population projecting selectively to VP have been proposed to account for this large influence of D1-SPNs on the VP (Fig 1c). Adding to the controversy, dual retrograde tracing studies have argued for both distinct populations of D1-SPNs that innervate the VP or VTA on one hand (41), and nearly complete overlap of VP- and VTA-projecting D1-SPNs on the other (114,115). These discrepancies may be due to both the low efficiency of retrograde tracer uptake leading to underestimation of co-labelled neurons and the topographic organization of accumbal output pathways within the VP and VTA (116), which would also lead to underestimation of co-projecting SPNs when non-overlapping terminal fields are seeded. Single-cell reconstruction support the hypothesis of bridging collaterals, and indicate that the majority of midbrain-projecting SPNs form bridging collaterals in the VP and lateral hypothalamus (115). However, the low number of reconstructed neurons make it impossible to exclude the possibility of a unique D1-SPN population projecting selectively to VP.

There has been no definitive demonstration of which VP cell types receive innervation from D1- and D2-SPNs, leaving an additional open question regarding the organization of accumbal output pathways. The VP contains both ‘relay’ and ‘output’ neurons; ‘relay’ neurons project within the basal ganglia (to STN and VTA), while ‘output’ neurons send projections to the mediodorsal thalamus, cortex or lateral habenula. An early prediction proposed that ‘relay’ neurons are preferentially innervated by D2-SPNs, and ‘output’ neurons by D1-SPNs. While this arrangement would preserve the anatomical pathway organization of ‘indirect’ D2-SPNs and ‘direct’ D1-SPNs, this is not the case: both midbrain-projecting ‘relay’ neurons and thalamic-projecting ‘output’ neurons are innervated by D1- and D2-SPNs, with individual neurons likely receiving input from both pathways (102,113). Consistent with the relatively higher degree of convergence of these two pathways in the ventral basal ganglia relative to the dorsal, results of bulk optogenetic manipulation do not necessarily support a Go/No-Go model of D1- and D2-SPN function in the ventral striatum (105,117).

Summary.

In both the dorsal and ventral basal ganglia, there is substantial convergence in the striatal–pallidal pathway. The high proportion of D1- and D2-SPN connectivity onto post-synaptic pallidal neurons suggests individual pallidal neurons receive input from both populations. Contemporary approaches have corroborated the classic model whereby D1-SPNs and D2-SPNs exert their opposing influence on behavior through inhibition of distinct downstream nuclei. However, models must integrate D1-SPN collaterals as an additional layer to be reconciled to understand how SPN output shapes downstream basal ganglia network activity.

Pallidal composition and output pathways

The pallidal complex comprises the GPe, GPi/EPN and VP. The classical box-and-arrow model posits that the pallidum is a relay nucleus of the basal ganglia, receiving input from the striatum and sending inhibitory projections to basal ganglia output nuclei: the midbrain, thalamus and STN (Fig 1a). However, the GPe exhibits a dichotomous anatomical organization, with prototypical neurons that follow canonical output pathways and arkypallidal neurons that project back to the striatum (Fig 1d) (62,118). Beyond the anatomical segregation, ‘arkypallidal’ and ‘prototypical’ neurons have distinct developmental origins. Arkypallidal neurons derive from the lateral ganglionic eminence and express the transcription factor forkhead box protein P2 (FoxP2), whereas prototypical neurons originate in the medial ganglionic eminence (MGE) or dorsal preoptic area (POA) (119). Homeobox protein Nkx-2.1 (NKx2–1) is restricted to subregions of the subpallium and is necessary for expression of genetic programs of MGE-derived lineages (120), with negligible overlap between FoxP2 and NKx2–1 (121,122). Adding to this complexity, there are at least four distinct molecularly-defined populations of GP neurons sending ‘back-projections’ to the forebrain. Moreover, there is considerable heterogeneity within the prototypical output pathways, which has only recently been incorporated into models of basal ganglia function.

Molecular logic defines prototypical GPe neurons.

Prototypical GPe neurons align most closely with the role of the GPe proposed in canonical models of the basal ganglia, innervating the thalamic motor nuclei, STN and SNr. The heterogeneity of these output neurons has relatively recently been appreciated, using molecular markers to classify distinct populations of prototypical neurons. Within the transcription factors controlling the fate of MGE-derived neurons, lim homeobox protein 6 (Lhx6) is a downstream effector of NKx2–1, and is upstream of the transcription factor Sox6 (123). As predicted based on these complementary roles in transcriptional programs, there is considerable overlap between NKx2–1, Lhx6 and Sox6 expression in the GPe (121). Characterization of GPe neurons by the expression of each of the markers Sox6, Lhx6 and neuronal pas domain 1 (NPas1) also report convergence of electrophysiological properties, with these populations exhibiting low-frequency, irregular firing (124). As the GPe neurons defined by NKx2–1, Lhx6 or Sox6 expression exhibit a high, but not complete, degree of overlap (122,124), Lhx6 has been adopted as a molecular identifier of ‘low-firing’ prototypical GPe neurons (125,126), which constitutes ~50% of the GPe (122,124), Fig 4).

Figure 4. Neural circuit diagram of the GPe.

(a) Proportion of neurochemically-defined neuronal subpopulations in the GPe. (b) Neural circuit ‘wiring diagram’ of the GPe. Pathway trajectories are not anatomically accurate, pathway weights represent reported synaptic strength or density of monosynaptic projections. PV neurons impinge on Lhx6+ cells; however, whether this input arises from PF- or basal ganglia-projecting PV+ neurons is unknown. Key neurochemically defined populations of the entopeduncular nucleus (EPN) receive input from striatum and innervate downstream basal ganglia nuclei. Abbreviations: D1 = D1-SPNs (blue), D2 = D2-SPNs (pink), Arky = arkypallidal neurons (yellow), CIN = cholinergic interneuron (gray), IN = non-cholinergic interneuron (orange in GPE; green/orange in EPN), PV = parvalbumin-positive neuron (orange), Lhx6 = Lhx6-positive neuron (black), MC = motor cortex, PF = parafascicular nucleus of the thalamus, STN = subthalamic nucleus, LHb = lateral habenula, Thal = Thalamus.

Early molecular classifications of the GPe defined PV+ neurons as a subset of prototypical neurons distinct from Lhx6+ neurons (122,124,125). Like NKx2–1-derived neurons, PV+ cells do not significantly overlap with FoxP2+ cells, suggesting that PV is a valid marker of prototypical GPe neurons (121,127). However, a caveat with the Lhx6 vs PV distinction of GP outputs is the lack of molecular segregation between these populations. Lhx6 is required for the maturation of MGE-derived PV+ neurons in cortex (128), and the level of PV expression can be modulated by developmental stage, neural activity, and by transcription factor binding (129,130). Perhaps unsurprisingly, estimates of overlap between Lhx6 and PV vary widely between laboratories and preparations, with the estimate of overlap as high as 13% of all GP PV+ cells expressing Lhx6, and 21% of Lhx6+ neurons expressing PV (Fig 4a).

Despite this caveat, the dissociation of prototypical GP neurons into ‘PV+’ and ‘Lhx6+’ has enabled important insight into the molecular and anatomical organization of GP output circuits (Fig 4b). PV is a calcium buffering protein, and its expression confers the ability to sustain high firing rates. A direct comparison of firing properties of Lhx6+ and PV+ GP neurons confirms that PV+ cells have significantly higher firing rates and narrower waveforms, supporting the functional segregation of these populations (124,125). However, within this PV+ subclass, there are multiple populations which innervate discrete basal ganglia areas, and electrophysiological recordings determined that this canonical ‘fast-spiking’ PV+ profile is restricted to SNr-projecting PV+ neurons (131). By contrast, PV+ neurons projecting to thalamus exhibit slower firing rates and wider action potentials, comparable to Lhx6+ neurons (131). The factors dissociating intrinsic properties in discrete populations of these PV+ neurons are not yet understood.

At the population level, Lhx6+ and PV+ prototypical GPe populations exhibit distinct circuit topology (Fig 4b). Both Lhx6+ and PV+ neurons receive excitatory drive from the STN and inhibitory drive from D2-SPNs, whereas activation of striatal D1-SPNs preferentially suppresses firing of Lhx6+ neurons, with minimal effects on PV+ GPe neurons. Again, at the population level, both Lhx6+ and PV+ neurons innervate potentially non-overlapping territories of downstream basal ganglia and thalamic structures (125). There are two primary differences in the projection patterns of these subclasses. First, while both subclasses display some arborizations in the striatum, axonal arborizations of Lhx6+ neurons are more extensive than PV+ terminals. Second, the PF is innervated exclusively by PV+ neurons (125). Despite these otherwise overlapping output targets, activation of the two populations has opposing effects on locomotor behavior, as Lhx6+ neuron activation suppresses and PV+ neuron activation drives locomotion (126). Given the more extensive ramifications of Lhx6+ neurons in the striatum, the anti-kinetic effects of Lhx6+ neuron activation could be partially due to inhibition of striatal neurons. Alternatively, Lhx6+ neurons may inhibit prototypical PV neurons locally in the GP, opposing the pro-kinetic effects of PV+ neuron activation (107,126). Finally, these two prototypical neurons may target distinct post-synaptic cell types in their target regions, which could also provide a substrate for their opposing locomotor effects. Future work will be required to dissociate these possibilities.

Subsets of prototypical VP neurons mirror projection neurons of EPN.

Dissection of projection populations of the VP is more nascent than of the GPe. At the molecular level, VP PV+ neurons analogously release GABA within the VTA, and receive monosynaptic inputs from the NAc, midbrain and STN (132), akin to their prototypical counterparts in the GPe. It is not established whether low-firing Lhx6+ cells constitute a subclass of prototypical VP neurons and to what extent these cells may overlap with PV+ VP neurons. At the anatomical level, unlike the GPe, prototypical VP populations project both internally within the basal ganglia to the ventromedial STN and GABAergic neurons of the VTA ('relay' neurons), and outside the basal ganglia to lateral habenula (LHb) and mediodorsal thalamus ('outpuť neurons). Neurochemically, all GPe prototypical neurons are GABAergic (Fig 4a), while prototypical VP neurons are both glutamatergic (VP_Glu; ~18%) and GABAergic (VP_GABA; ~49%), with minimal overlap between these populations (0.9% - 3%, Fig6;(133,134).

Figure 6. Neural circuit diagram of the VP.

(a) Proportion of neurochemically-defined neuronal subpopulations in the VP. (b) Neural circuit ‘wiring diagram’ of the ventral pallidum. Pathway trajectories are not anatomically accurate, pathway weights represent reported synaptic strength or density of monosynaptic projections. While Lhx6 is expressed in mature VP neurons, how the Lhx6 marker overlaps with other populations of VP neurons is not known. Abbreviations: D1 = D1-SPNs (blue), D2 = D2-SPNs (pink), Arky = arkypallidal neurons (yellow), CIN = cholinergic interneuron (gray), IN = non-cholinergic interneuron (orange), PV = PV-positive neuron (green and/or orange), Lhx6 = Lhx6-positive neuron (black), PFC = medial wall frontal cortex, PVT = paraventricular nucleus of the thalamus, LHb = lateral habenula, STN = subthalamic nucleus, VTA GABA = GABAergic neurons of the ventral tegmental area, VTA DA = dopaminergic neurons of the ventral tegmental area.

This neurochemical identity carries important functional implications, as VP_Glu and VP_GABA neurons synapse in overlapping territories in the LHb and midbrain, oppositely influencing activity in these structures (133,134). Specifically, stimulation of VP_Glu neurons drives activity in LHb, rostromedial tegmental nucleus (RMTg) and VTA_GABA neurons, and inhibits dopaminergic VTA (VTA_DA) neurons through polysynaptic inhibition (134). By contrast, stimulation of prototypical VP_GABA neurons inhibits LHb and VTA_GABA neurons, and disinhibits VTA_DA cell firing through the same polysynaptic network (133). Consequently, VP_Glu and VP_GABA projection neurons exert opposing roles on appetitive behavior. VP_Glu neurons are activated in response to reward omission and punishment and are inhibited by reward and reward-predictive cues (135). Activating VP_Glu is acutely aversive, and their activity is necessary to update reward seeking behavior in response to conflict-induced negative outcomes, which depends on LHb and RMTg activity (134,135). Conversely, stimulation of VP_GABA supports behavioral reinforcement, and VP_GABA activity is necessary for the expression of reward-seeking (133,136,137). While endogenous activity of VP_GABA neurons is more heterogeneous than that of VP_Glu neurons, a large proportion (~82%) exhibit activity patterns which are diametrically opposed to that of VP_Glu neurons, as they are inhibited by punishment and activated by reward and reward-predictive cues (135).

These opposing effects of VP_Glu and VP_GABA neurons on behavior have been ascribed to opposing influences on downstream structures, however their antagonistic effects are also likely amplified by local connectivity within the VP. Specifically, VP_GABA neurons synapse onto VP_Glu neurons, and would attenuate the suppression of appetitive behavior imposed by VP_Glu neurons. Likewise, VP_Glu neurons innervate and excite neighboring VP_Glu neurons (133), synchronizing and amplifying downstream actions. It is unclear how the opposing patterns of endogenous activity in prototypical VP_Glu and VP_GABA neurons arises, as both populations receive input from largely overlapping upstream structures, including the NAc, medial wall cortex, basolateral amygdala, ventromedial STN and VTA (133,134,138). However, VP_Glu neurons do receive proportionately more input from accumbal D1-SPNs than VP_GABA neurons (138). Thus, under conditions where D1-SPNs are active, VP_Glu neurons would be inhibited relative to neighboring VP_GABA neurons, providing one mechanism through which accumbal D1-SPN activity disinhibits appetitive behavior at the level of the VP.

As mentioned above, parallel GABAergic and glutamatergic pallidal output neurons that characterize the VP are not observed in the GPe; rather they are reminiscent of output pathways of the GPi/EPN. Within the EPN, three primary populations of output neurons have been described from unbiased clustering of single cell transcriptional profiles: two separate PV+ populations that release either glutamate or GABA, and a third population co-releasing GABA and glutamate that also express the peptide somatostatin (139). In both the EPN and VP, the population of neurons projecting to LHb co-releases GABA and glutamate, while an exclusively GABAergic population preferentially innervates the midbrain SNr and VTA_GABA neurons, respectively (132,139). As in the VP, glutamatergic EPN cells are nearly exclusively innervated by D1-SPNs, and not D2-SPNs (139). Thus, some unique populations of prototypical VP neurons may be considered analogous to GABAergic, glutamatergic or co-releasing output neurons of the EPN. Molecular profiling of the VP and careful anatomical validation will be required to resolve this question.

Arkypallidal neurons inhibit the striatum.

Arkypallidal neurons in both the GPe and VP send dense, filigree-like processes through large territories of striatum and their activity peaks during conditions requiring rapid suppression of action selection (63,64). While arkypallidal neurons comprise only ~18% of pallidal neurons, their ability to inhibit striatum is robust, with individual neurons forming over 10,000 striatal synaptic contacts over terminal fields in excess of 1 mm in rats (62). As a consequence, >80% of SPNs and >25% of striatal interneurons receive direct synaptic input from arkypallidal cells (64,140). Arkypallidal neurons in the GP co-release GABA and enkephalin (62,63,141), although this co-release has not been confirmed in the VP. This confers arkypallidal neurons with the ability to inhibit striatal neurons through fast ionotropic transmission (GABA_A), slower GPCR transmission (GABA_B), or mu- and delta-opioid receptors (142). The conditions under which arkypallidal neurons release these distinct neurotransmitters is not known, although burst firing is typically required for enkephalin release (143). Indeed, while arkypallidal neurons exhibit slow firing relative to other pallidal populations, they are capable of firing up to ~75 Hz in the VP (64) and over 100 Hz in the dorsal pallidus (127), which is predicted to be sufficient for peptide release.

Arkypallidal neurons are predominantly innervated by D1-SPNs (108), only sparsely innervated by D2-SPNs, and are under potent tonic inhibition from prototypical pallidal neurons (Fig 5,6) (104). Thus, arkypallidal activity is also regulated indirectly by D2-SPNs and glutamatergic neurons in the STN through these prototypical intermediaries (63,104). Integrating this into a functional model (Fig 4), when D1-SPNs are inhibited relative to D2-SPNs, the subsequent disinhibition of arkypallidal neurons would promote inhibition of large territories of striatal SPNs (Fig 5). This arkypallidal-mediated striatal inhibition may therefore amplify the anti-kinetic effects of D2-SPN activity by suppressing inappropriate or competing actions at the level of striatal ensembles. Alternatively, arkypallidal-mediated inhibition may preferentially suppress firing in weakly-excited SPNs while still allowing activity in subsets of strongly-activated SPNs, serving as a low-pass filter on SPN firing. Future studies will be needed to resolve these possibilities. Relatively little is known about the molecular expression profile of arkypallidal neurons in the VP. Viral tracing has established that enkephalinergic VP neurons are almost exclusively innervated by D1-SPNs from the NAc (114,138). If it is confirmed that enkephalin is a selective arkypallidal marker (as in the DSt), this would support a similar anatomical arrangement of preferential D1-SPN projections to arkypallidal neurons in the ventral and dorsal basal ganglia.

Figure 5. Predicted effects of SPN balance on pallidal output.

Innervation of D1-SPNs has been proposed to be relatively stronger onto Arky and glutamatergic Proto pallidal neurons. Arky neurons are under tonic inhibition from GABAergic prototypical pallidal neurons (left). Increased D1 output would favor increased output of inhibitory proto pallidal neurons to midbrain (middle). High D2 SPN output would favor output of glutamatergic prototypical pallidal neurons and amplify D1 SPN inhibition through arkypallidal activation (right). Abbreviations: D1 = D1-SPNs (blue), D2 = D2-SPNs (pink), Arky = arkypallidal neurons (yellow), Proto = Prototypical pallidal neuron (green, orange).

To date, no monosynaptic excitatory inputs to arkypallidal neurons in the GPe or VP have been conclusively identified, although input from thalamic or cortical nuclei have been proposed to account for the rapid onset of arkypallidal activity in response to task-relevant cues (63). Characterization of cell-type-specific inputs will determine whether arkypallidal firing is inherited from specific upstream structures or arise within the arkypallidal population through disinhibition, neuromodulation or integration of multiple excitatory inputs.

Non-arkypallidal neurons project rostrally to forebrain.

In addition to the arkypallidal neurons, there are at least three molecularly-distinct GPe populations (Fig 6a) that project rostrally to the forebrain rather than following prototypical output pathways. These populations can be defined based on their anatomical projections, and combinatorial expression of transcription factors. Original reconstructions of individual pallidal neurons identified a minority of bipolar GPe neurons that follow ‘prototypical’ output pathways to EPN and STN, but also innervate the striatum. The striatal axons of these ‘prototypical’ neurons exhibited fewer and less complex striatal arborizations, clearly distinguishing them from arkypallidal morphology. While classified as ‘prototypical’ based on firing pattern, the expression of commonly used ‘prototypical’ cellular markers (i.e. PV and LHx6) is unclear. It was subsequently proposed that multiple classes of prototypical GPe neurons exhibit bipolar or pseudo-bipolar morphology to innervate both striatum and downstream basal ganglia structures (144). To date, functional studies have not dissociated these striatal-ramifying ‘prototypical’ neurons from ‘arkypallidal’ neurons in either the dorsal or ventral basal ganglia, although their existence must be considered when interpreting studies employing retrograde labeling to define arkypallidal neurons.

In addition to the aforementioned pallidal populations, approximately 5% of GPe and VP neurons are cholinergic (ChAT+) (Fig 6a), with GPe_ChAT+ projecting to motor and orbitofrontal cortex, and VP_ChAT+ to medial frontal cortex and basolateral amygdala (145). These ChAT+ neurons are spontaneously active and arborize throughout cortical layers 1–6. Activation of cholinergic projections decorrelates neural activity in sensory and associative cortices to enhance attention to and discrimination of behaviorally-relevant stimuli (146). Although this has been ascribed to acetylcholine release, these ChAT+ neurons also co-release GABA from both overlapping and acetylcholine discrete release sites (147). To date, the significance of GABA co-released from ChAT+ neurons has not been elucidated. Classically, ChAT+ pallidal cells have been considered extensions of the basal forebrain cholinergic system rather than part of the basal ganglia per se. However, this distinction has been questioned, given that activity of cholinergic GPe neurons is under the control of striatal SPNs and STN neurons, which sets them apart from classical nucleus basalis cholinergic neurons (145,147).

In parallel, a separate population of non-cholinergic, GABAergic GPe neurons sends monosynaptic inhibitory projections to interneurons in layers 2/3, 5 and 6 along with principal projection neurons in layer 5 and 2/3 of frontal cortex (147). This pattern is distinct from the projection targets of pallidal ChAT+ neurons. However, like GPe_ChAT+, this non-cholinergic population receives excitatory drive from the STN and direct inhibitory input from both D1-SPN and D2-SPNs. To date, no specific motor or cognitive behaviors have been attributed to this non-cholinergic, cortical-projecting population within the GPe or VP. However, their activation entrains cortical activity, and they likely modulate attention and cortical plasticity necessary to support learning (148). This population has been distinguished from prototypical GPe neurons by their co-expression of Lhx6, NPas1 and Nkx2.1, along with an absence of PV (121,122,127). Leveraging this knowledge for intersectional genetic strategies would elucidate the role of each of these subpopulations on behavior and establish whether analogous molecular logic defines cortical-projecting neurons in the VP.

Summary.

The pallidum can be divided into ‘prototypical’ and ‘arkypallidal’ neurons, which exert distinct effects on basal ganglia network activity by virtue of their unique projection patterns and released neurotransmitters. Only prototypical neurons receive excitatory drive from cortex and subthalamic nucleus (104); no inputs to arkypallidal neurons outside the striatal-pallidal complex have been identified. However, arkypallidal neurons are under inhibitory control of both D1-SPNs and prototypical pallidal neurons. This anatomical arrangement positions these two populations to coordinate activation throughout the entire basal ganglia network (Fig 6b).

Outlook

The classic box-and-arrows diagram of the basal ganglia suggests a degree of homogeneity of striatal and pallidal projection neurons. While enormously influential, this model has always been recognized as an oversimplification, and the advent of modern genetic tools has allowed for dissection of molecular, neurochemical and anatomical heterogeneity of these nuclei. The molecular expression profile of a neuron determines its development, migration, recruitment into neural circuits, its intrinsic electrophysiological properties, as well as how it is affected by neuromodulators and synaptic input. Therefore, understanding this resolution is necessary for informing models of circuit function within the striatal-pallidal complex and the encompassing basal ganglia network.

Many functional principles discussed were established using optogenetic-assisted circuit mapping using patch-clamp electrophysiology. It is important to be cautious when inferring circuit properties from slice preparations, in which network connectivity has been disrupted and intrinsic membrane properties are sensitive to recording conditions. Similarly, optogenetic activation of pre-synaptic neurons while recording the post-synaptic responses can confirm the presence of synapses between these structures. However, caution is needed when predicting how these pre-synaptic neurons modulate downstream activity based on the existence of monosynaptic connections. For example, activation of excitatory synaptic input can suppress firing of post-synaptic neurons if local poly-synaptic inhibitory circuits are engaged by this excitatory drive, as is the case in both the striatum and VP (134,149). Therefore, several principles of basal ganglia circuitry and plasticity remain to be thoroughly corroborated in vivo.

While we focused this review on circuit function in rodents in the absence of disease states, contemporary work has begun to identify how synaptic plasticity and functional changes in these molecularly defined striatal-pallidal subcircuits contribute to altered behavior in models of movement, mood and substance use disorders. This is important as both the pallidum and ventral striatum are emerging as promising targets for neuromodulation therapies for these disorders (5,150). However, optimizing and refining the specificity of neuromodulation therapies is precluded, in part, by a lack of understanding of the specific adaptations within these striatal-pallidal circuits that are causally related to aberrant behavioral symptoms of these disorders. Accurate and comprehensive models of basal ganglia information flow, along with an understanding of circuit adaptations arising in disease states, is a necessary step in conceptualizing novel neuromodulation therapies for these disorders.