Description

Functional anatomy

The basal ganglia form a tightly connected and relatively phylogenetically conserved brain subsystem consisting of deep gray matter structures that form recurrent loops with the cortex and thalamus. These nuclei include the striatum (caudate and putamen forming the dorsal striatum, nucleus accumbens, and olfactory tubercle forming the ventral striatum), globus pallidus (external and internal segments), subthalamic nucleus, and substantia nigra (pars compacta and pars reticulata) (121). In “standard” models, the striatum is the sole input nucleus, though newer models also incorporate the hyperdirect cortical-subthalamic nucleus pathway (120).

The globus pallidus externa (GPe) is essentially isolated from extra-basal ganglia structures, and the output nuclei comprise the globus pallidus interna (GPi, entopeduncular nucleus in rodents), ventral pallidum, and the substantia nigra pars reticularis (SNr). The substantia nigra pars compacta (SNc) modulates cortical-basal ganglia-thalamic circuits with dopaminergic innervation of the striatum principally but also other basal ganglia nuclei, frontal cortex, and thalamus.

Striatum. The striatum is the largest subcortical brain structure in the mammalian brain. It contains both projection (inhibitory GABAergic medium spiny neurons, 90%) and interneurons (cholinergic, 10%). Due to their volume, most information processing is accomplished by the medium spiny neurons (MSNs). They receive most of the glutamatergic excitatory input from the whole neocortical mantle and related structures, such as the hippocampal formation and the amygdala as well as thalamic intralaminar (CM/Pf) and basal ganglia-recipient relay nuclei (107). Some GABAergic striatal inputs come from the GPe (94). The striatum sends massive GABAergic efferents to GPe and the basal ganglia output nuclei.

The organization of corticostriate projections is complex. Lesion studies and functional imaging demonstrate the topographic organization of corticostriatal projections, suggesting that different striatal regions are functionally specialized depending on their cortical inputs (59). Tract tracing studies also indicate that cortical regions project to the striatum in overlapping parasagittal zones of considerable rostrocaudal extent, suggesting some interdigitation of corticostriatal projections (60). Other work suggests some convergence of corticostriatal projections from cortical regions that are functionally linked by corticocortical connections (60). A study identified six parallel subnetworks making up the cortico-basal ganglia-thalamic system (49). The thalamus acts as a “relay domain,” sending outputs back to the originating corticostriatal neurons of each subnetwork and creating a closed loop. Diffusion tensor imaging demonstrates separate but overlapping corticostriatal connections in sensorimotor (posterior putamen), associative (anterior putamen and caudate), and limbic (ventral striatum) loops (150). Resting-state fMRI functional connectivity studies also correlate with the “parallel loop model” (68). In Parkinson disease, resting-state fMRI studies suggest remapping of these networks possibly secondary to dopamine depletion, with decreased coupling in the cortico-striatal sensorimotor network and between the striatum and brainstem. There may also be increased coupling in the associative network, which could be compensatory (150). Whole-brain white matter structural connectivity studies utilizing diffusion MRI in early-stage and medication-naïve Parkinson disease patients have found increased structural connectivity between Parkinson disease-relevant brain regions, including the striatum, caudate, putamen, and supplementary motor area, and suggest reduced functional network flexibility (114).

Resting-state functional connectivity in early Parkinson disease also shows that greater motor deficits correlate with weaker coupling between the anterior putamen and midbrain, whereas cognitive decline in memory and visuospatial domains correlates with stronger coupling between the dorsal caudate and the rostral anterior cingulate cortex (97).

The striatum is composed of GABAergic medium spiny neurons and several interneuron types in 2 interdigitated compartments known as “patches” (or striosomes) and “matrix.” The striosomal compartment incorporates connections mainly from limbic areas, whereas the matrix is related to sensorimotor and associative regions (30). They represent 90% to 95% of neurons depending on the species, with lower percentages in primates (157; 146). Medium spiny neurons are characterized by the presence of multiple dendritic spines (small mushroom-shaped protuberances) that are the site of most synapses.

They fire sparsely, requiring coordinated excitatory synaptic input to initiate spiking (85). Medium spiny neurons are divided into two groups of roughly equal proportions based on projection patterns (52). The monosynaptic connection from the striatum to GPi/SNr is termed the “direct” pathway, whereas “indirect” pathway medium spiny neurons (iMSNs) synapse in the GPe. Direct pathway medium spiny neurons (dMSNs) also send collaterals to the GPe. dMSNs express D1 dopamine receptors with relatively low affinity for dopamine, whereas iMSNs express higher affinity D2 dopamine receptors (52). Thus, phasic increases or decreases in striatal dopamine concentration are likely to exhibit distinct effects on dMSNs or iMSNs.

The interneurons are characterized by smooth dendrites and can be categorized into groups based on neurochemical profiling . Cholinergic interneurons are the most abundant group of interneurons (87). There are at least three subtypes of striatal GABAergic interneurons. Parvalbumin-positive fast-spiking interneurons selectively and powerfully inhibit medium spiny neurons (157), though their precise role in striatal microcircuitry remains to be determined (55). The physiology of other GABAergic interneurons (nitric-oxide synthase and calretinin-positive) is less well-studied (157).

Cholinergic interneurons probably correspond to tonically active neurons recorded in vivo; these tonically active neurons fire spontaneously at low frequencies (lower than 10 Hz) (144). Striatal cholinergic signaling may be primarily via “volume neurotransmission” because there are relatively few typical cholinergic synapses within the striatum (129). Striatal cholinergic neurons directly influence GABAergic interneuron and medium spiny neuron activity and interact with dopamine to regulate synaptic plasticity (52). Their axons arborize densely and diffusely in the striatal matrix, with sparse crossings into the patch compartment (146). Tonically active neuron firing is correlated in awake, behaving animals and may transmit coordinated signals to large striatal regions. An example is the “TAN pause,” a transient, coordinated decrease in tonically active neuron firing in response to salient behavioral events dependent on afferents from CM/Pf (101). Of relevance to Parkinson disease, the tonically active neuron pause is lost after striatal dopamine depletion (144). Cholinergic interneuron inhibition in the dorsal striatum normalizes the burst firing initiated in the SNr by dopamine depletion (similar to the effects of levodopa or subthalamic nucleus deep brain stimulation) (145). Therefore, reduction of striatal cholinergic tone by muscarinic acetylcholine receptor blockade may be expanded as a therapy for Parkinson disease (176).

The nigrostriatal projection features many terminals diverging from a relatively small number of neurons; each dopaminergic nigral neuron innervates many striatal neurons. The relative lack of specificity of nigrostriatal projections suggests that this pathway performs a diffuse “biasing” function for striatal neurons, though nigral dopamine neuron activity may not be as uniform as previously assumed (100). Dopaminergic terminals tend to synapse on the necks of medium spiny neuron spines, with cortico- and thalamo-striatal terminals on spine heads. These “triadic” synapse complexes allow tight regulation of the contributions of specific cortical and thalamic afferents to medium spiny neuron firing.

Dopamine modulates striatal output on multiple timescales via pre- and postsynaptic mechanisms. Presynaptic D2 receptors on multiple cell types regulate dopamine, acetylcholine, glutamate, and GABA release (52). Postsynaptic D1 receptor stimulation acutely enhances medium spiny neuron excitability, whereas D2 receptor activation has opposite effects (52). The differential expression of D1 and D2 receptors by medium spiny neurons (D1-dMSN; D2-iMSN) results in contrasting effects of acute changes in dopamine levels on the direct and indirect basal ganglia pathways (see below). Striatal interneurons also express multiple dopamine receptor subtypes with complex responses to dopaminergic stimulation. Inhibition of striatal cholinergic interneurons in parkinsonian mice has been shown to regulate the excitability of striatal D1 and D2 medium spiny neurons, increase the functional weight of the direct striatonigral pathway in cortical information processing, and normalize pathological basal ganglia output bursting activity to improve motor control. These effects are not seen in nonlesioned mice, demonstrating the importance of dopamine tone on the regulation of cholinergic interneuron activity in motor control (102). In addition to direct effects on striatal neuronal activity, dopamine influences synaptic plasticity at cortico- and thalamostriatal synapses of dMSNs and iMSNs in distinct ways (52). Furthermore, chronic loss of dopaminergic terminals changes striatal microanatomy. Dendritic spines and glutamatergic synapses onto iMSNs are greatly diminished (34), though dMSN excitability is diminished and iMSN excitability is enhanced to cortical stimulation (93). This may be partially explained by the strengthening of remaining corticostriatal synapses (160). Fast-spiking interneurons, which under normal circumstances have a slight preferential innervation of dMSNs compared to iMSNs, become much more strongly connected to iMSNs after dopamine loss (54).

Subthalamic nucleus (STN). The subthalamic nucleus is located ventral to the zona incerta and rostral to the substantia nigra. It has become increasingly relevant as a target for deep brain stimulation in Parkinson disease. Glutamatergic input to subthalamic nucleus comes from frontal regions, including primary, presupplementary, and supplementary motor cortex (120). Additional excitatory input comes from CM/Pf (140), which may give the subthalamic nucleus access to ascending “salience” signals from the brainstem (113). GABAergic input is primarily from the GPe (14), and sparse dopaminergic afferents come from SNc (137).

The subthalamic nucleus projects to GPe, forming a GPe-STN loop that is strengthened after dopamine depletion (44). The subthalamic nucleus also projects to the basal ganglia output nuclei, where it synapses throughout the dendritic trees of GPi and SNr neurons. STN activity has been shown to correlate with specific basal ganglia activity patterns (172). A disynaptic projection to the cerebellar cortex via the brainstem has been identified (18). Subthalamic nucleus projections diverge to innervate large regions of other basal ganglia nuclei, suggesting that it cannot encode detailed information in population activity but may broadcast information-poor timing signals. Principal subthalamic nucleus neurons are spontaneously active, with tonic in vivo firing rates in the 10 to 30 Hz range in nonhuman primates (12). They express T-type (Cav 3) calcium channels that render them prone to “rebound” bursting, and the likelihood of bursting may be modulated by subthalamic dopamine (137).

Globus pallidus, pars externa (Gpe). GPe receives GABAergic afferents from striatal iMSNs and glutamatergic input from the subthalamic nucleus and CM/Pf (140) and sends feedback GABAergic projections to the striatum and STN (66). The GPe also receives striatal projections with significant expression of adenosine type 2A receptors (87). Distinct populations of GPe neurons project to striatum and downstream nuclei (subthalamic nucleus, GPi, and SNr) (94), where they typically form large perisomatic synapses. This anatomy suggests a central role for the GPe in coordinating basal ganglia-wide activity.

In vivo recordings describe two populations of GPe neurons; one group that consistently fires at high rates with occasional pauses and another group that tends to fire sparsely with intermittent bursts (10). GPe neurons can also be classified based on their synchronization with local field potentials (LFPs) (94), but it is not known whether these two classification schemes are related to each other. In dopamine-depleted rats, the LFP-based classification identifies populations that project uniquely to striatum or subthalamic nucleus/GPi/SNr. Interpretation of rates and patterns of neuronal firing, however, is complicated in light of theoretical models that bring into question our understanding of their meaning (23), including in the context of the rate model (Montgomery 2011).

Dopaminergic projections from SNc have pre- and post-synaptic effects in GPe, though the dopaminergic innervation is much weaker than in the striatum (137). How dopamine influences pallidal activity in vivo is not well understood, though local lesions of GPe dopamine terminals recapitulate some features of parkinsonism (01). As in the striatum and subthalamic nucleus, chronic dopamine loss changes GPe anatomy and physiology. The intrinsic pacemaking ability of pallidal neurons is lost (25), intrapallidal collateral effects are strengthened (112), and subthalamopallidal connections are strengthened (44). Further complicating understanding, including in the context of the rate model, deep brain stimulation of multiple basal ganglia structures improves parkinsonian symptoms (Montgomery 2011).

The output nuclei (GPi and SNr). The GPi and SNr share structural properties with similar afferent and efferent systems. The basal ganglia output nuclei receive afferents primarily from dMSNs, GPe, and subthalamic nucleus, though afferents from CM/Pf also exist (140). Striatal afferents tend to synapse on the distal portions of dendrites, external pallidal afferents on proximal portions of dendrites, and subthalamic afferents throughout the dendritic tree (175). There is dopaminergic input to both nuclei, with a unique dendritic release mechanism in SNr (137). GPi projects primarily to relay nuclei in the thalamus, as well as CM/Pf and the pedunculopontine nucleus (02). The SNr has similar projections but also sends efferents to the superior colliculus and SNc (175). SNr neurons are consequently associated with orienting movements and saccades and regulating nigrostriatal dopaminergic neurons. These processes occur through connectivity with the superior colliculus. A detailed study showed that the SNr contains neuronal populations that target specific brainstem effectors beyond the superior colliculus. The closed-loop networks of the basal ganglia described above are linked with projections to brainstem premotor areas through relays with the thalamus, pedunculpontine formation, and reticular formation. The SNr also has an extensive range of targets in the spinal cord, regions for orofacial and vocal movements. Further, the SNr also shows electrophysiological specificity with neurons targeting different anatomical regions that have unique electrophysiological signatures (106).

Basal ganglia output neurons exhibit spontaneous activity with high firing rates during quiet wakefulness but without the pauses that characterize most GPe neurons (12). These neurons can burst at very high rates, which is enhanced after dopamine depletion (12; 147). Enhanced bursting in the dopamine-depleted state may depend on both altered afferent activity and changes intrinsic to the output nuclei (83).

Substantia nigra, pars compacta. The principal dopaminergic innervation of basal ganglia circuits comes from SNc (118). As described above, the SNc gives rise to a massive nigrostriatal projection with sparser projections to other basal ganglia nuclei, the thalamus, and an important projection to the frontal cortex. Over 70% of SNc input is GABAergic, including afferents from the rostromedial tegmental nucleus (RMTg), striatal patches, and SNr (50; 72). SNc receives glutamatergic afferents from diverse subcortical structures, including the subthalamic nucleus and pedunculopontine nucleus, as well as cholinergic inputs from pedunculopontine nucleus (118). Local somatodendritic dopamine release from SNc neurons influences both SNc and SNr (118). SNc neurons exhibit intrinsic pacemaker activity, firing below 10 Hz in vivo to maintain tonic dopamine levels in the striatum (and other targets). Phasic changes in nigrostriatal signaling are tightly regulated by upstream structures (eg, lateral habenula via RMTg) and are believed to be critical for striatum-based implicit learning processes (72; 08).

Primate tract tracing studies indicate that striatal and SNc limbic, associative, and motor subregions form primarily reciprocal connections (61). More recent imaging studies in humans also suggest a tripartite pattern of connectivity with medial SNc, lateral SNc, and ventral SN related to limbic, motor, and cognitive functional organization, respectively (174). There is, however, also an “ascending spiral” relationship in which SNc neurons project to striatal subregions adjacent to the source of their striatal afferents. This pattern is repeated in ways that, theoretically, allow limbic subregions of the SNc to influence motor and associative regions of the striatum.

Thalamic interactions with the basal ganglia. Basal ganglia afferents form large GABAergic perisomatic synapses in thalamic relay nuclei (15), which are relatively unique in that “driver” inputs to other thalamic nuclei are glutamatergic (138). These relay nuclei form recurrent loops with cortex believed to generate oscillatory brain rhythms (78). Cortex and thalamic relays send collaterals to the reticular nucleus of the thalamus, which in turn project back to thalamic relays.

Two major thalamocortical projection patterns are recognized. “Matrix” neurons tend to project diffusely to cortical layer I, whereas “core” neurons have more focused projections to layers III to IV. Basal ganglia-receiving areas consist primarily of matrix-like cells, whereas cerebellar recipient regions consist of core-like cells (27). Thalamocortical neurons express T-type (Cav 3) calcium channels that allow transitions from tonic to burst-firing modes in response to hyperpolarization (78). Modeling studies suggest that tonic firing allows the transmission of afferent signals to the cortex, whereas bursting modes may inhibit cortico-cortical communication (139). Therefore, the basal ganglia may modulate information flow in other circuits rather than transmitting information from the basal ganglia to the cortex (139).

The parafascicular nucleus receives cortical pathways and projects to the basal ganglia. The parafascicular nucleus can regulate movement initiation through projections to the subthalamic nucleus and is an important aspect in orienting movement. Stimulation of this pathway significantly restored movement in akinetic mice (6-OHDA model). This pathway could explain the therapeutic effect of deep-brain stimulation (166). CM/Pf nuclei receive dense afferents from the brainstem (eg, pedunculopontine nucleus and superior colliculus) and basal ganglia output nuclei (140), as well as cortical afferents. The pedunculopontine nucleus input is part of a major pedunculopontine nucleus-thalamic cholinergic projection. These midline intralaminar thalamic nuclei project primarily to the striatum, with synapses on medium spiny neurons, GABAergic interneurons, and cholinergic interneurons. There are weaker projections to the cortex, subthalamic nucleus, GPi, and SNr. CM/Pf degenerates in Parkinson disease and the MPTP model of parkinsonism (161), potentially contributing to the clinical and physiological features of Parkinson disease.

Pedunculopontine nucleus. Located at the junction of the midbrain and pons, the pedunculopontine nucleus is a phylogenetically ancient structure playing a crucial role in gait and postural control. More recently, it has also been implicated in the modulation of behavioral flexibility (109). The cuneate nucleus and pedunculopontine nucleus form the mesencephalic locomotor region (119), a functionally-electrophysiologically defined part of the mesopontine tegmentum crucial for normal gait and posture. Neurons of the pedunculopontine nucleus help regulate anticipatory postural adjustments preceding initiation of a step, as well as the step itself. Diffusion tensor imaging of Parkinson disease patients with freezing of gait shows reduced connectivity of the pedunculopontine nucleus with the cerebellum, thalamus, and regions of the frontal cortex, all of which are part of the locomotion pathway (47). Deep brain stimulation of this region of the pedunculopontine nucleus has been performed in parkinsonian patients for freezing of gait and to improve axial symptoms, with varying results (158). The pedunculopontine nucleus has connections with several components of the basal ganglia, thalamus, cerebellum, and caudal brainstem structures that regulate spinal central pattern generators. Pedunculopontine nucleus neurons are heterogeneous, mainly cholinergic, GABAergic, and glutamatergic neurotransmitters (165). Cholinergic neurons have historically defined the borders of the pedunculopontine nucleus. The axons of these cholinergic neurons are highly collateralized with connections in the midbrain, forebrain, and lower brainstem (109). The cholinergic neurons are mainly located in the subnucleus compactus of the pedunculopontine nucleus and are severely depleted in both Parkinson disease and progressive supranuclear palsy (65). The GABAergic neurons of the pedunculopontine nucleus are less extensive than the cholinergic system and are more densely populated in the rostral pedunculopontine nucleus, extending into the substantia nigra pars compacta and rostromedial tegmental nucleus (110). The innervation of the SNc is likely heterogeneous, with connections to both dopaminergic and nondopaminergic neurons (33).

Cortex. A comprehensive review of cortical anatomy and physiology is beyond the scope of this article. First, corticostriatal projections originate in all areas of the cortex (99), but corticosubthalamic projections originate in frontal areas (120). Second, the densest projection to the primary motor cortex originates from the GPi, where outputs are likely somatotopically organized (18). Third, cortical efferents can be broadly classified as intratelencephalic and pyramidal tract-like neurons (134).

Intratelencephalic neurons form slow-conducting corticocortical and corticostriatal pathways that project bilaterally (eg, from primary motor cortex) or ipsilaterally (eg, from primary somatosensory cortex) and receive minimal direct input from the basal ganglia-recipient relay thalamus (126). Pyramidal tract neurons send at most weak collaterals to the ipsilateral striatum, receive strong inputs from the basal ganglia-recipient relay thalamus, and may provide corticosubthalamic innervation (99). Thus, the “closed loop” architecture of cortico-basal ganglia-thalamic circuits may not be entirely closed (152).

Clinical applications

The rate model and electrophysiological changes in animal models of Parkinson disease

Modern understanding of basal ganglia anatomy and physiology permitted the development of a model of basal ganglia pathophysiology accounting for many clinical phenomena. Neurotoxins that selectively target monoaminergic neurons, primarily 6-OHDA and MPTP, have allowed detailed study of the physiologic consequences of striatal dopamine loss (32). These include changes in neuronal firing rates, burst activity, oscillations, and synchrony (117).

Rate changes. According to the prevailing “rate” model of basal ganglia physiology (121), striatal dopamine increases dMSN firing, decreases iMSN firing, and, ultimately, suppresses basal ganglia output to release thalamocortical circuits from tonic inhibition.

In parkinsonism, dMSN activity may decrease, and iMSN activity may increase, resulting in “excessive” basal ganglia inhibition of thalamocortical activity. A key feature of the rate model is the modulation of subthalamic nucleus neurons, which act as governors of neuronal activity in the basal ganglia output nuclei. In support of this model, lesions (11) or transient inactivation (154) of the “overactive” subthalamic nucleus in MPTP-treated monkeys improves parkinsonism. More persuasively, selective activation of dMSNs or iMSNs as well as bilateral activation of striatonigral neurons using optogenetic techniques dramatically alters locomotor activity in a manner consistent with the rate model (84). The standard rate model also explains the chorea-athetosis-ballism spectrum of involuntary movements and the fact that such involuntary movements occur with both striatal and subthalamic nucleus pathologies. In Huntington disease, there is disproportionate degeneration of iMSNs compared to dMSNs early in the disease course. The downstream consequence is decreased subthalamic nucleus activity and diminished basal ganglia output, disinhibiting thalamocortical activity. Destruction of the subthalamic nucleus produces similar effects, accounting for phenomenological similarity of chorea-athetosis-ballism in Huntington disease and acute subthalamic nucleus lesions.

Despite some predictive value, however, the rate model is not consistent with all observations. For example, lesioning or stimulation of the GPi produces therapeutic benefit in both hyperkinetic and hypokinetic disorders. Lesions in the thalamus and GPe should result in hypokinesis per this model; however, this has not been observed in animal studies (147). Studies have also demonstrated coactivation of both direct and indirect pathways (31; 173). These observations may support a more nuanced version of the rate model in which movement is driven by an action selection model.

Chronic dopamine depletion generally leads to changes in the firing rates of basal ganglia neurons in awake, behaving animals that match rate model predictions. In MPTP-treated monkeys, firing rates increase in subthalamic nucleus and GPi/SNr, but decrease in GPe, thalamus, and motor cortex (12; 147; 126). The cortical firing rate decrease may be specific to pyramidal tract neurons (126). In anesthetized rats, dMSNs and iMSNs decrease and increase their firing rates, respectively, after 6-OHDA lesions (93; 82). Furthermore, firing rate changes after chronic levodopa (124), dopamine agonist (17), or combined (67) treatment in MPTP-treated monkeys are consistent with the standard model. Other studies, however, are inconsistent with the rate model. Some show no or inconsistent changes in firing rates after MPTP despite clinical parkinsonism (133; 128; 89; 154). Also, subthalamic nucleus deep brain stimulation at therapeutic frequencies increases GPi firing rates in parkinsonian monkeys despite clinical improvement (62). Thus, the firing rates of basal ganglia output neurons cannot be the sole determinant of motor behavior.

Burst firing. Bursts are brief episodes of high-frequency firing against slower background activity. For many neurons, bursting is a distinct firing mode supported by specific ion channels, not simply a transient firing rate increase. Identifying firing mode transitions requires intracellular recordings, which are technically difficult in behaving animals (92). The cellular mechanisms underlying extracellularly recorded “bursts” in Parkinson disease models cannot, therefore, be established definitively and likely differ across regions.

Burst firing in GPe, subthalamic nucleus, and GPi. Several studies in the primate MPTP and rodent 6-OHDA models describe enhanced bursting in GPe, subthalamic nucleus, and GPi/SNr (12; 147; 170; 89; 96; 154). Bursting is reduced by intravenous levodopa or chronic dopamine agonist/levodopa combination therapy (67; 154), suggesting a relationship between burst-firing and clinical parkinsonism. In humans treated with apomorphine, however, subthalamic nucleus and GPi bursting increases compared to the “off” state, sometimes associated with dyskinesias (91). Therapeutic, but not subtherapeutic, GPe and subthalamic nucleus deep brain stimulation reduces GPi bursting (62; 162). Conversely, low-frequency subthalamic nucleus deep brain stimulation, which may exacerbate parkinsonism, increases GPi bursting (42). In humans, effective GPi deep brain stimulation reduced GPi burst-firing (28), though conflicting results were obtained in MPTP-treated monkeys (104). Overall, it seems that burst-firing along the indirect pathway increases after dopamine depletion and decreases with effective therapy. Emerging theories of the meaning of neuronal firing rate further complicate this understanding (23).

Thalamic bursts. There are limited data comparing thalamic activity between dopamine-depleted and intact subjects. In one study comparing thalamic activity before and after MPTP, a nonsignificant increase was found in the proportion of burst-firing neurons in the basal ganglia-recipient thalamus. Bursting was significantly increased, however, in the cerebellar-recipient thalamus (128). Others have found that clinically effective subthalamic nucleus or GPe deep brain stimulation is associated with decreased basal ganglia-recipient thalamic bursting compared to the “off” state (42; 171; 162). Results in the cerebellar recipient thalamus are less consistent (42; 171). Human studies indicate that thalamic bursting is common in Parkinson disease (115), but the studies lack healthy controls. Because these are all extracellular recordings, it is uncertain whether the bursts reflect true low-threshold spike bursts mediated by T-type Ca2+ channels or brief periods of rapid tonic firing (92). Excessive thalamic relay burst-firing is probably characteristic of Parkinson disease, though the evidence is less robust than changes in bursting behavior in the indirect pathway.

Cortical bursts. Dopamine depletion increases burst-firing in primary motor cortex as well as the percentage of time spent in bursts. As with firing rates, bursting increases among pyramidal tract neurons, but not intratelencephalic neurons (126). Because thalamic basal ganglia-recipient relay nuclei project preferentially to pyramidal tract neurons, these data suggest that cortical bursting is driven by the thalamus. On the other hand, the long duration of cortical bursts (greater than 1 second) is quite different from the brief bursts observed in the basal ganglia and thalamus.

Neuronal oscillations. “Oscillatory activity” is a nonspecific term indicating that some measure of neuronal function repeats itself periodically. This may be reflected in single-unit spiking or local field potentials.

Local field potentials are generally easier to measure than single-unit activity, are believed to represent the summation of local transmembrane currents and synaptic activity, and may reflect afferent activity.

In vivo recordings in nonhuman primates demonstrated exaggerated single-unit oscillations at tremor frequencies and harmonics, though the relationship between neuronal and mechanical oscillations is complex. Bimodal distributions (approximately 5 and 10 Hz peaks) of single unit oscillation frequencies are found in the GPi and subthalamic nucleus of tremulous MPTP-treated monkeys (12; 133; 135). Tonically active neuron firing also becomes oscillatory after MPTP lesions (132). Interestingly, higher frequency oscillations at 10 to 15 Hz may correlate best with the typical 5 Hz rest tremor of Parkinson disease (135; 154). Inactivation of the subthalamic nucleus in tremulous monkeys eliminates tremor and 8 to 20 Hz, but not 4 to 8 Hz, pallidal oscillations (169).

Neuronal oscillations are correlated also with bradykinesia and rigidity (70). MPTP-treated monkeys lacking the typical rest tremor of Parkinson disease exhibit low-frequency oscillations in single unit activity similar to those with tremor (108; 67). Chronic levodopa, combined levodopa/dopamine agonist treatment, and therapeutic deep brain stimulation increase spontaneous movement and suppress these low-frequency oscillations (108; 67; 171; 53). Higher frequency “beta” (approximately 15 to 30 Hz) oscillations are correlated with rigidity and bradykinesia in humans and rodent models of Parkinson disease (86; 76; 45). Single unit and local field potential beta oscillations were recorded from the subthalamic nucleus and GPi of humans with Parkinson disease, and beta power reductions were correlated with clinical improvement (22; 168; 86). Enhanced beta frequency oscillations are observed in the cortex, striatum, globus pallidus, subthalamic nucleus, and SNr of dopamine-depleted rodents (95; 20). However, acute dopamine depletion using reserpine in rodent models showed enhanced beta oscillations without involvement of the primary motor cortex (07). The enhanced beta oscillations are suggested to reflect a basal ganglia network-wide abnormality that causes some features of parkinsonism (90). Beta oscillations are directly associated with acute and chronic regulations of dopaminergic tone. However, power and coherence between single units and LFPs did not show a similar relationship. This suggests that beta frequencies specifically are an accurate marker of changes in dopaminergic tone (74). One study compared the proportion of beta oscillatory neurons in the subthalamic nucleus, GPi, and ventrolateral thalamus of patients with Parkinson disease. A significantly higher proportion of beta oscillatory neuronal activity was noted in both subthalamic nucleus and GPi, suggesting abnormal synchronization of these neurons in the pathologic state (43). Subthalamic nucleus LFPs have been used as a physiological marker to distinguish between tremor-dominant and postural-instability-gait-difficulty subtypes of Parkinson disease (156). It has been noted that enhanced neuronal oscillations emerge after motor deficits in the dopamine-deficient state (89; 96; 36), arguing against a causal role for “pathologic” oscillations in the genesis of bradykinesia and rigidity. In humans, abnormal oscillatory activity in basal ganglia thalamocortical loops has been studied using depth recordings during deep brain stimulation, and advances toward closed-loop devices allow for local field potential recordings directly through implanted leads (131). More recent recordings incorporating signals from the surface of the cortex using electrocorticography have provided improved capacity to better understand the phase-amplitude coupling of cortical-subcortical structures and abnormal synchrony of activity in Parkinson disease (130).

Levodopa-induced dyskinesias are also associated with characteristic neuronal oscillations. The subthalamic nucleus and GPi/SNr of dyskinetic humans and rats show elevated local field potential power at low frequencies (5 to 10 Hz) contralateral to the dyskinetic hemibody (48; 04; 167). As rats became dyskinetic, a striatal shift from beta to high gamma (approximately 70 to 100 Hz) oscillations was observed (63). A similar beta-to-high gamma shift is observed in the subthalamic nucleus and GPi of levodopa-treated humans, though no mention was made of whether the patients experienced dyskinesias (22). High gamma oscillations are also observed in nondyskinetic healthy rats treated with apomorphine or amphetamine (13).

Synchronization. “Synchrony” describes 2 or more events that occur nearly simultaneously and may refer to single unit spikes, local field potential (LFP) phase across recording sites, or consistent relationships between single unit spikes and LFP phase. The LFP itself represents coordinated transmembrane potential fluctuations and is, thus, a measure of local synchrony. The Parkinson disease literature often refers to “burst synchrony” or “oscillatory synchrony” as if these concepts are inseparable. However, not all bursts occur in an oscillatory pattern, tonically (non-burst) firing neurons may fire synchronously, and not all synchronized activity is periodic (26). For example, GPe neurons exhibit more nonoscillatory synchrony than GPi neurons after MPTP (67).

Under normal physiologic conditions in awake animals, basal ganglia and motor cortical neurons rarely fire synchronously (06). After dopamine depletion, however, single unit oscillations in the GPi, GPe, and subthalamic nucleus become highly synchronized with each other (108; 67; 95; 96; 135; 43). Tonically active neuron and GPi oscillations are also synchronized in dopamine-depleted primates (132). Deep brain stimulation and chronic combined levodopa/dopamine agonist treatment tend to reduce interneuronal synchrony (108; 67). Despite the strong evidence for a role of single-unit synchrony in the pathophysiology of Parkinson disease, however, at least one study found that synchrony develops after motor deficits (89). On the other hand, in the basal ganglia-recipient thalamus, synchrony increased before clinical parkinsonism became evident (128).

Related to synchrony is the loss of specificity of basal ganglia neurons for passive and active movement (21). The proportion of neurons with firing rate changes during movement is larger in MPTP-treated monkeys compared to controls (88), and an increased proportion of basal ganglia and thalamic neurons encode for movement about multiple joints (16; 128). Collectively, these data suggest that information transfer into and out of the basal ganglia may be compromised in the dopamine-depleted basal ganglia.

LFP oscillations throughout cortico-basal ganglia-thalamic circuits also synchronize after dopamine depletion (40). The exaggerated beta oscillations of the dopamine-deficient state are highly coherent across the cortex and basal ganglia (96; 20). Single-unit activity is also strongly entrained in “pathologic” LFPs (57; 168; 96; 20).

Relating physiology to clinical symptoms

Anatomical changes have already been shown to have utility in tracking Parkinson disease progression and specific Parkinson disease subtypes. Patients with freezing of gait-predominant Parkinson disease showed greater cortical and subcortical atrophy compared with non-freezing of gait Parkinson disease whereas non-freezing of patients with gait Parkinson disease had greater basal ganglia and sensorimotor atrophy (141). How physiological changes in cortico-basal ganglia-thalamic circuits described above translate into the motor features of movement disorders remains a central question. Much of this literature is dominated by discussions of Parkinson disease pathophysiology due to the greater availability of deep brain recordings in Parkinson disease patients. We discuss the relationship of the available data to the pathophysiology of Parkinson disease and key areas for future investigation.

Dysfunction of basal ganglia connectivity in prodromal and early Parkinson disease. Resting-state fMRI shows aberrant connectivity within the basal ganglia network (bilateral caudate, putamen, and globus pallidus) and frontal lobes in patients with early Parkinson disease (153). The dysfunction in connectivity has similarities with that seen in patients with REM behavioral disorder (RBD), a prodromal symptom of Parkinson disease that tends to occur 10 to 15 years prior to the onset of motor symptoms (136). Reduced dopaminergic transmission in REM behavioral disorder compared to more significant dopaminergic deficiency in early Parkinson disease may account for differences in connectivity between the 2 groups (136). Patients with prodromal Parkinson disease in the form of REM behavior disorder or hyposmia have also been shown to have reduced striato-thalamo-pallidal intra- and interhemispheric connectivity without alteration between other cortical and subcortical structures at this stage of the disease (35). Diffusion MRI studies in the early stage have also suggested decreased flexibility between functional networks implicated in Parkinson disease (114).

The relationship of dopamine depletion to Parkinson disease. The 6-OHDA and MPTP models are quite good at modeling the physiologic and behavioral effects of dopamine loss. Nonetheless, there are caveats. First, the toxins act quickly, creating rapid dopamine loss that is clearly different from the slow progression of Parkinson disease. Second, MPTP and 6-OHDA selectively kill monoaminergic neurons but do not recapitulate the progressive degeneration of all neuronal types (eg, cholinergic) affected by Parkinson disease (19). Third, they do not recapitulate the currently presumed causative pathologic lesion thought to underlie the development of Parkinson disease, that being abnormal synuclein present diffusely throughout the nervous system (37). Lastly, Parkinson disease involves pathophysiology of multiple neurotransmitter systems beyond dopamine (79).

The origins of “pathologic” physiology. Because of the looped nature of cortical-basal ganglia-thalamic circuits, it is difficult to distinguish cause from effect among the many changes observed in the dopamine-depleted state. Starting with the striatum, it is possible to suggest a sequence of events that could lead to electrophysiological and clinical parkinsonism.

Consistent with the “rate” model, iMSN firing rates in anesthetized rats increase after dopamine depletion (93). Because striatopallidal synapses exhibit short-term facilitation, rapid iMSN firing may drive pause-and-burst activity in GPe (80). This is an attractive hypothesis, as it explains how selectively driving iMSNs can cause akinesia and excessive basal ganglia bursting (84). In essence, information may be translated from a rate code in the striatum to a pattern code in GPe. Microanatomic changes in the striatum and GPe also likely contribute to the excessive synchrony and loss of somatotopy observed in the parkinsonian state (54; 112).

The GPe-STN network has been proposed to generate pathologic rhythms (14).

Bursting GPe neurons may hyperpolarize subthalamic nucleus neurons, which fire rebound bursts that stimulate GPe neurons in recurrent cycles. The dopamine-denervated subthalamic nucleus may also be intrinsically more “bursty” (14). In support of the importance of a GPe-STN oscillator in the generation of pathologic rhythms, local pharmacologic manipulations and lesions of the GPe and subthalamic nucleus decrease oscillations and bursting at the basal ganglia output (169; 154). Modeling studies indicate that the GPe-STN oscillator may be capable of generating beta rhythms (71), though it may not be able to do so in isolation (148). Aberrant cortical activity could also drive pathologic subthalamic nucleus activity via the hyperdirect pathway, as suggested by reduced oscillatory activity after local blockade of glutamate receptors in the subthalamic nucleus (154). Recordings from patients with Parkinson disease provide more information on this network-high frequency (21 to 30 Hz) coherence between the supplementary motor area and subthalamic nucleus correlated with the hyperdirect pathway. Cortico-subthalamic nucleus coherence and cortico-subthalamic nucleus tract density correlated spectrally to upper beta frequencies, implying a strong functional relationship within the hyperdirect pathway from the cortex to subthalamic nucleus. Importantly, this pathway did not relay via the striatum. Deep-brain stimulation may be effective, at least in part, by suppressing this hyperdirect pathway (123). The GPe-STN circuit is well-positioned to transmit pathological patterns to the basal ganglia output nuclei.

The rate model predicts that excessive basal ganglia output should suppress thalamic firing, which does not seem to be the case (128). Instead, pallidal output may influence the fine timing of thalamic spikes driven by cortical inputs (58) and gate cortico-cortical communication by regulating transitions between burst (“closed” gate) and tonic (“open” gate) firing modes (139). Whether excessive/bursty GPi activity forces bursting in thalamocortical neurons in unknown.

Given the role of thalamocortical circuits in generating oscillatory synchrony in sensory pathways (78), it is likely that these circuits play a role in the generation of pathologic rhythms in Parkinson disease. This may occur at a network level in the reciprocal loops between the cortex and thalamus or within local cortical circuits (24).

Thalamocortical bursts evoke long-lasting, spatially distributed activity (09), suggesting that synchronized thalamocortical bursts could drive an intracortical beta resonance.

Plastic changes in cortical-basal ganglia-thalamic circuits clearly play a role in the circuit-level physiology of Parkinson disease (34; 54; 160; 08; 44; 112). Beta oscillations and synchrony emerge with chronic, but not acute, decreases in dopamine signaling (89; 96; 36). Further, motor deficits develop progressively after dopamine loss/receptor blockade (08), presumably due to altered regulation of cortico- and thalamo-striatal synaptic plasticity and the microanatomic changes described above.

Research has focused on basal ganglia-thalamocortical loops, but the basal ganglia also influence motor output via direct connections to brainstem nuclei, including the pedunculopontine nucleus and superior colliculus (02; 175). The pedunculopontine nucleus gives the basal ganglia access to important brainstem motor centers and creates yet another cortical-basal ganglia-thalamic loop. Deep brain stimulation of the pedunculopontine nucleus can increase thalamic perfusion, even when performed unilaterally (151). The pedunculopontine nucleus has been difficult to study in animal models, however, because it comprises multiple cell types and lacks clear borders (02).

The relationship between “pathologic” physiology and clinical parkinsonism. Determining how network-level changes relate to clinical parkinsonism remains an elusive goal. No doubt, many physiologic changes have direct causal links with distinct motor abnormalities, but it is equally likely that some do not.

Parkinsonian tremor is an excellent illustration of this problem (69). Parkinson disease patients with tremor demonstrate increased functional connectivity between the GPi, putamen, and cerebello-thalamic circuit, indicative of increased oscillatory activity in these networks (70). Deep brain stimulation of the cerebello-thalamo-cortical network reduces tremor in Parkinson disease (29). The cerebellar recipient thalamus is a likely tremor generator,but is not directly connected to the basal ganglia.

It is not obvious how striatal dopamine loss is translated into tremor or why DRT, anticholinergics, and STN/GPi deep brain stimulation effectively suppress tremors. Indeed, striatal dopamine deficiency does not correlate with the presence or severity of tremor (142). Subthalamo-cerebellar connections link the basal ganglia and cerebellum but do not readily explain the efficacy of GPi deep brain stimulation. Alternatively, the basal ganglia-recipient thalamus may influence the cerebellar recipient thalamus through recurrent thalamocortical connections (69).

Beta oscillations are consistently correlated with bradykinesia and rigidity (but not tremor). Transient cortical and basal ganglia beta oscillations also occur in healthy subjects and appear to represent a subset of oscillatory basal ganglia states characterized by transitions between combinations of theta (approximately 8 Hz), beta, low gamma (approximately 50 Hz), and high gamma (approximately 80 Hz) oscillations (13; 159). These beta “bursts” are associated with resistance to altering current motor plans (76; 90).

Local field potentials are generally easier to measure than single-unit activity and may reflect afferent activity. Pathologically persistent beta activity may, therefore, prevent initiation of new actions (13; 105; 90). From a mechanistic standpoint, behavioral inertia could be related to excessive entrainment of single neurons throughout the motor system to stable beta frequency states (96; 13; 20; 90). In patients with Parkinson disease, exaggerated coupling between the phase of beta oscillations in the basal ganglia and the amplitude of broadband activity in the primary motor cortex results in excessive beta phase locking of cortical motor neurons, manifesting clinically as bradykinesia and rigidity (38). This hypothesis implies that beta oscillations cause bradykinesia and rigidity by enhancing neuronal synchrony, which reduces the information that can be transmitted by neuronal ensembles (05). In Parkinson disease, this may be reflected in imprecise movements or rigidity as small groups of muscles cannot be activated independently. Deep brain stimulation of the subthalamic nucleus in parkinsonian patients reduces phase-amplitude interactions with a subsequent reduction in bradykinesia and rigidity (39). A similar effect has been demonstrated with pallidal deep brain stimulation as well (164).

The physiology of other movement disorders has also been studied, though not in as much detail. For example, dystonia shares many physiologic features with Parkinson disease. In recordings from human subjects, both disorders exhibit enhanced neuronal oscillations, synchrony, and burst-firing in the subthalamic nucleus (143) and GPi (149; 155; 21). Thus, none of these changes in and of themselves can drive parkinsonism. In dystonia, however, firing rates are lower (149; 155) as are peak oscillation frequencies (149; 167). After several studies demonstrating clinically significant improvement, deep brain stimulation of the GPi for chronic, medically intractable dystonia was approved by the FDA in 2003 (41). The presumed mechanism is “normalization” of neuronal patterns, which prevents pathologic bursting and oscillatory activity within the network. This results in improved sensorimotor information processing and reduction of disease symptoms (73). Reduction of cortical synchronization by subthalamic deep brain stimulation has been seen in studies of isolated dystonia and Parkinson disease as well (130).

Levodopa-induced dyskinesias share electrophysiological features with dystonia. In both cases, low frequency (approximately 4 to 10 Hz) LFP and single-unit oscillations are prominent at the basal ganglia output, and firing rates are relatively low (124; 48; 04; 67; 03). Burst-firing may also be increased during dyskinesias, though results are inconsistent (17; 91). It has also been shown that levodopa-induced dyskinesia murine models (6-OHDA) showed that the dyskinetic state is largely mediated through the direct basal ganglia pathway and via suppression of the indirect pathway (163). Teasing out the physiology of dyskinesias is difficult, however, as only a handful of studies have compared the “on” and “on with dyskinesias” states (04; 67).

A final major problem for understanding basal ganglia physiology and the pathophysiology of movement disorders is linking what we know of basal ganglia physiology, the nature of movement disorders, and normal functions of the basal ganglia. Considerable animal experimental data and human clinical studies indicate that a major function of the basal ganglia is the acquisition of habitual action patterns via reward-based learning mechanisms mediated at least in part by striatal dopamine. Phasic dopamine signaling may act as a “reward prediction error” signal, indicating the difference between expected (predicted) and actual behavioral outcomes (56). Some interesting experimental work suggests that bradykinesia following striatal dopamine loss is the consequence of the misperception of the value of motor acts (103; 173).

The constellation of motor abnormalities and neuropsychiatric dysfunction seen in Tourette syndrome highlights the interaction of the basal ganglia and associative loops. Motor and vocal tics may manifest due to deficient GABAergic synaptic inhibition in corticobasal ganglia connections involving the sensorimotor, premotor, possibly supplementary motor, and striatopallidothalamic networks (51). Common comorbid neuropsychiatric diseases such as attention deficit disorder or obsessive-compulsive disorder may be related to deficient GABAergic synaptic inhibition in the cortex and striatum. The influence of emotion on tics is apparent and highlights the role of the limbic system and amygdala on the corticobasal ganglia network (122). Deep brain stimulation of various targets in the motor and limbic pathways (most commonly the centromedian thalamic nucleus, GPi, and nucleus accumbens) for Tourette syndrome is increasingly successful and may regulate abnormal neuronal activity in these networks (75; 77; 98). The promising results of deep brain stimulation targeting the bilateral STN or nucleus accumbens for obsessive-compulsive disorder demonstrate the overlap between motor and limbic pathways in the corticostriatothalamocortical network (64; 81).

Concluding remarks

Driven by the advent of deep-brain stimulation, functional and structural neuroimaging, and the availability of animal models of dopamine deficits, new data have been collected regarding the physiology of the basal ganglia. Although these findings have spawned numerous theories regarding the pathophysiology of Parkinson disease and other movement disorders, the only consensus is that the “standard” model is incomplete. Now that the clinical and electrophysiological phenomenology of Parkinson disease are well-described, the next steps will be to test hypotheses linking neuropathology with physiological changes and physiological changes with clinical symptoms and therapeutic response. This will require a range of models, taking advantage of cell-type specific manipulations available in rodents (eg, optogenetics) and the close homology between humans and non-human primates. Many important questions remain whose answers may not only directly impact patient care but also elucidate the normal functions of the basal ganglia. It remains to be seen which lines of inquiry will provide the critical insights that allow the rational manipulation of neural circuits to achieve the ultimate goal of transforming “pathological” neural activity into patterns compatible with normal motor, cognitive, and emotional function.

Table 2. Selected Areas for Future Research