Basal Ganglia

Abstract

The basal ganglia are a group of interconnected subcortical nuclei at the base of the forebrain that sit at the heart of the loops the cortex runs to select and shape action. Their central input structure, the striatum, receives projections from nearly the whole cortex and, through two opposing pathways whose balance is set by dopamine, adjusts the tonic inhibition that the output nuclei impose on the thalamus and brainstem. This arrangement lets the basal ganglia solve a selection problem: releasing one motor or cognitive program while suppressing its competitors. The same circuitry supports reinforcement learning through the dopamine reward-prediction-error signal, the gradual formation of habits, and aspects of cognition and motivation. This article surveys their anatomy, the direct and indirect pathways, action selection, reward learning, and habit formation, with three interactive demonstrations.

Keywords: basal ganglia, striatum, direct and indirect pathways, dopamine, action selection

The basal ganglia are a set of deep grey-matter nuclei, shared by all vertebrates, that lie beneath the cerebral cortex and surround the thalamus. Once regarded as a purely motor system because their damage produces the tremor of Parkinson's disease and the involuntary movements of Huntington's, they are now understood as a general-purpose selection device embedded in parallel loops that link the cortex, basal ganglia, and thalamus and return to the cortex (Alexander et al., 1986). Each loop takes a broad swath of cortical input, funnels it through the striatum, and returns a focused signal that gates the activity of its cortical target. The organizing question the circuit answers is not how to generate a movement but which of many competing programs to allow at any moment (Redgrave et al., 1999), and the same logic extends from limb movements to eye movements, thoughts, and choices.

What the Basal Ganglia Are

The basal ganglia are a collection of nuclei buried in the forebrain and upper midbrain. The largest is the striatum, comprising the caudate nucleus and putamen, which is the principal input structure and receives glutamatergic projections from almost the entire cerebral cortex. The striatum projects to the output nuclei, the internal segment of the globus pallidus and the substantia nigra pars reticulata, which in turn send inhibitory projections to the thalamus and brainstem. Between input and output lie the external segment of the globus pallidus and the subthalamic nucleus, and modulating the whole system is the dopaminergic substantia nigra pars compacta (DeLong, 1990).

A defining feature of the output nuclei is that they fire tonically at high rates, so at rest they impose continuous inhibition on their thalamic and brainstem targets. Movement is not produced by activating the basal ganglia but by transiently lifting this inhibition from the desired target, a release that the striatum orchestrates. This is why the system is best described as gating rather than driving: it holds everything in check by default and selectively removes the brake (Mink, 1996). The nuclei are organized not as a single circuit but as several parallel loops, each keeping to its own territory as it passes through the basal ganglia, so that motor, oculomotor, associative, and limbic streams remain segregated from cortex to thalamus and back (Alexander et al., 1986).

Figure 1

The Direct and Indirect Pathways Through the Basal Ganglia

Types of Basal Ganglia

In the Medical Subject Headings (MeSH) classification the basal ganglia are filed under the Cerebrum, the division of the forebrain to which they belong, and the descriptor has four narrower structures, listed below. These are anatomical components of one interconnected system rather than independent organs, and they are partly a matter of indexing convention: MeSH is a vocabulary for cataloguing the biomedical literature, so the grouping reflects how work is filed rather than a single theoretical claim about which subdivisions matter. The list also mixes structures that are functionally central to the motor and reward circuits described here, such as the corpus striatum, with structures conventionally grouped with the basal ganglia on anatomical grounds, such as the amygdala, which is functionally part of the limbic system.

The Direct and Indirect Pathways

The striatum influences the output nuclei through two routes with opposite effects. In the direct pathway, striatal neurons project straight to the output nuclei and inhibit them; because the output nuclei themselves inhibit the thalamus, activating the direct pathway disinhibits the thalamus and promotes movement. In the indirect pathway, striatal neurons project to the external globus pallidus, which inhibits the subthalamic nucleus, which in turn excites the output nuclei; activating this route therefore deepens the thalamic brake and suppresses movement (Albin et al., 1989). The two pathways express different dopamine receptors: the direct pathway carries excitatory D1 receptors and the indirect pathway inhibitory D2 receptors, so a rise in dopamine simultaneously strengthens the movement-promoting route and weakens the movement-suppressing one.

This scheme explains the two great families of basal-ganglia disorder as opposite imbalances. In Parkinson's disease the loss of dopamine tips the balance toward the indirect pathway, over-inhibiting the thalamus and producing poverty of movement; in Huntington's disease the early loss of indirect-pathway striatal neurons releases movement into the involuntary writhing of chorea (DeLong, 1990). The causal role of the two pathways was confirmed directly when optogenetic activation of the direct pathway in mice initiated movement and relieved parkinsonian slowness, while activation of the indirect pathway induced freezing (Kravitz et al., 2010). The demonstration below lets the dopamine level be varied and shows how the balance of the two pathways sets the net gate on the thalamus.

Action Selection

Why route action through a structure that inhibits by default and must be selectively released? The influential answer is that the basal ganglia solve a selection problem. At any moment many potential actions compete for the same motor apparatus, and an animal that tried to execute several at once would be paralyzed by conflict. A system that tonically inhibits every option and lifts the inhibition from just one provides a clean arbitration: the winner is expressed and the losers are held back (Redgrave et al., 1999). On this view the basal ganglia are a general-purpose selector that evolved to resolve competition between whatever the vertebrate brain needed to commit to.

The circuitry implements this with a center-surround arrangement. The focused direct-pathway projection releases the selected program while the diffuse indirect pathway, acting through the broad projections of the subthalamic nucleus, raises inhibition on the competing programs, sharpening the contrast between the chosen action and its rivals (Mink, 1996). This scales the model naturally to decision-making: because parallel loops carry oculomotor, associative, and limbic streams alongside the motor one, the same selection operation applies to choosing where to look, which rule to apply, and which goal to pursue. The demonstration below models several competing action channels and shows how the basal ganglia select one while suppressing the surround.

Reward Learning and Dopamine

The basal ganglia do not select at random; they learn which actions are worth selecting, and dopamine is the teaching signal. The landmark finding is that midbrain dopamine neurons do not simply report reward but encode a reward-prediction error: they fire above baseline when an outcome is better than expected, stay at baseline when it is exactly as predicted, and pause below baseline when it is worse (Schultz et al., 1997). This is precisely the error term that reinforcement-learning algorithms use to update predictions, and it arrives at the striatum where dopamine gates the plasticity of corticostriatal synapses, strengthening the connections that led to better-than-expected outcomes. The link to operant conditioning is direct: the phasic dopamine signal is a neural implementation of the law of effect, reinforcing the actions that pay off.

Modern circuit work has refined this picture. Striatal circuits do not encode a single value but support a family of computations for reward-based learning and choice, with the direct and indirect pathways contributing distinct roles in learning from positive and negative outcomes (Cox & Witten, 2019). Because the reward-prediction-error account also grounds the role of the basal ganglia in motivation, it connects the structure's motor and cognitive functions: the vigor with which an action is performed, as well as whether it is selected at all, is shaped by expected reward (Dudman & Krakauer, 2016). The demonstration below models the prediction-error signal and shows how a predicted value converges on the received reward across trials.

Habits and Procedural Learning

As a behavior is repeated and reliably rewarded, control over it changes hands within the basal ganglia. Early in learning an action is goal-directed, sensitive to the current value of its outcome, and depends on the associative loop through the caudate. With extended practice the same action becomes a habit, elicited by the situation and relatively insensitive to whether the outcome is still wanted, and control shifts to the sensorimotor loop through the putamen (Graybiel, 2008). This transfer is the neural basis of procedural memory: the gradual, incremental acquisition of skills and habits that proceeds without awareness and is preserved even when the conscious memory system is damaged.

The dissociation is clear in human patients. Amnesic patients with medial temporal damage but intact basal ganglia can acquire a probabilistic habit-learning task at a normal rate while having no conscious memory of the training, whereas patients with basal-ganglia disorders such as Parkinson's disease show the reverse pattern, revealing a neostriatal habit-learning system distinct from the declarative memory of the medial temporal lobe (Knowlton et al., 1996). Habits, rituals, and the automatization of learned sequences all draw on this striatal machinery, which packages a well-practiced routine into a unit that can be launched as a whole (Graybiel, 2008).

Cognitive Implications

Because the basal ganglia operate through parallel loops that reach far beyond the motor cortex, the selection operation they perform on movements applies equally to cognition. The associative loop through the dorsolateral prefrontal cortex lets the same gating logic select among rules, plans, and goals, linking the basal ganglia to executive function and the updating of working memory: the striatal gate determines when new information is admitted into the prefrontal store and when the current contents are protected (Alexander et al., 1986).

A second implication follows from the reward-prediction-error signal. Because dopamine trains corticostriatal synapses, the basal ganglia are central to how organisms learn the value of options and bias choice toward the rewarding ones, making the system a substrate for value-based decision-making as well as motor control (Cox & Witten, 2019). A third is timing and initiation: the basal ganglia contribute to deciding not only which action to take but whether and when to start it, so their dysfunction can present as difficulty initiating self-paced action rather than as weakness (Klaus et al., 2019).

Worked Example

Consider the reward-prediction-error signal the third demonstration models. Let an action have a learned predicted value V, and let the reward actually received be r. The dopamine prediction error is their difference, δ = r − V, and the prediction is updated toward the outcome by a fraction α, the learning rate: V ← V + α · δ. Take a novel action whose value starts at V = 0, a reward r = 1, and a learning rate α = 0.5.

On the first trial the outcome is a complete surprise: δ = 1 − 0 = 1, a large positive prediction error, so dopamine neurons fire a burst and the value is updated to V = 0 + 0.5 × 1 = 0.5. On the second trial the same reward is now half expected: δ = 1 − 0.5 = 0.5, a smaller burst, and V becomes 0.5 + 0.5 × 0.5 = 0.75. On the third trial δ = 1 − 0.75 = 0.25 and V rises to 0.875, and the errors continue to halve, so the predicted value converges geometrically on the true reward while the dopamine response shrinks toward baseline. Once V has reached 1 the reward is fully predicted and evokes no dopamine response at all; if the reward is then unexpectedly omitted, r = 0 and δ = 0 − 1 = −1, a negative error expressed as a dopamine pause below baseline. The arithmetic makes concrete why dopamine tracks surprise rather than reward: the signal is large when a lesson remains to be learned and vanishes once the outcome is anticipated, which is exactly the property a teaching signal for reinforcement learning must have.

Discussion

The basal ganglia illustrate how a single circuit motif can serve an extraordinary range of functions. The motif is gating by disinhibition: tonically active output nuclei clamp their targets, and behavior is released when the striatum, through the balance of its direct and indirect pathways, selectively lifts the clamp (Mink, 1996). Casting this operation as the solution to a selection problem explains why the same anatomy handles movements, eye movements, and cognitive acts alike, since all require committing to one option among competitors (Redgrave et al., 1999). Layered on top of selection is learning: the dopamine reward-prediction-error signal tells the striatum which selections were worth making and tunes the corticostriatal synapses accordingly (Schultz et al., 1997), and with practice the balance of control shifts from deliberate, goal-directed loops to automatic, habitual ones (Graybiel, 2008; Knowlton et al., 1996). The clinical picture confirms the framework from the opposite direction: because the two pathways push in opposite directions, too little dopamine impoverishes movement and too little indirect-pathway restraint releases it, so the disorders of the basal ganglia are the signatures of a selection-and-gating system thrown out of balance (Albin et al., 1989; DeLong, 1990).

Current Directions

Contemporary work has moved beyond the classic direct/indirect scheme toward a finer account of what the basal ganglia compute. One influential reframing holds that the basal ganglia control not merely the selection of an action but its vigor: the speed and force with which a chosen movement is executed, tied to the reward expected for it, which reinterprets parkinsonian slowness as a disorder of movement invigoration rather than of selection alone (Dudman & Krakauer, 2016). A parallel line of circuit-level research, using optogenetic and imaging tools that can read out and manipulate the two pathways separately, has shown that the direct and indirect pathways are co-active around movement rather than simply opposed, and that striatal circuits carry a richer set of signals for reward learning and choice than a single value term (Cox & Witten, 2019). A third strand asks how the basal ganglia govern the initiation of self-paced action, dissecting the striatal dynamics that determine what to do, whether to do it, and when to begin, and casting the circuit as a controller of the timing of spontaneous behavior (Klaus et al., 2019). Together these advances retain the selection-and-reinforcement core while recasting the basal ganglia as a controller of the vigor, value, and timing of behavior, not just its choice.

Common Misconceptions

The basal ganglia generate movement.

They gate rather than generate: their output nuclei tonically inhibit the thalamus and brainstem, and behavior is released by selectively lifting that inhibition, not by driving the muscles directly (Mink, 1996).

The basal ganglia are purely a motor system.

Parallel loops carry associative and limbic streams as well as motor ones, so the same selection operation extends to reward learning, habit formation, decision-making, and executive control (Alexander et al., 1986; Cox & Witten, 2019).

Dopamine is the brain's pleasure signal.

Phasic dopamine encodes a reward-prediction error, the difference between the reward received and the reward expected, not pleasure as such; it is large for surprising outcomes and silent for fully predicted ones (Schultz et al., 1997).

Glossary

Action selection.

The problem of choosing one behavior from many competing options, proposed as the general function the basal ganglia evolved to solve.

Caudate nucleus.

One of the two components of the striatum, especially associated with the associative loop that supports goal-directed action and cognition.

Cortico-basal-ganglia-thalamic loop.

The circuit in which a region of cortex projects to the striatum and receives a processed signal back through the output nuclei and thalamus; several such loops run in parallel and stay segregated.

Direct pathway.

The striatal route that inhibits the output nuclei, disinhibiting the thalamus and promoting the selected action; carries excitatory D1 dopamine receptors.

Disinhibition.

The release of a target from tonic inhibition; the basal ganglia gate behavior by disinhibiting the thalamus rather than by exciting it.

Globus pallidus.

A pallidal nucleus with an internal segment (a major output of the basal ganglia) and an external segment (a relay of the indirect pathway).

Goal-directed action.

Behavior controlled by the current value of its expected outcome, depending on the associative loop; contrasted with a habit, into which it can convert with practice.

Habit.

A well-learned behavior triggered by the situation and relatively insensitive to the current value of its outcome, supported by the sensorimotor loop through the putamen.

Indirect pathway.

The striatal route through the external pallidum and subthalamic nucleus that excites the output nuclei, deepening the thalamic brake and suppressing movement; carries inhibitory D2 receptors.

Putamen.

One of the two components of the striatum, especially associated with the sensorimotor loop that supports habitual and automatic movement.

Reward-prediction error.

The difference between the reward received and the reward expected, encoded by phasic dopamine firing and used to update the value of actions.

Striatum.

The principal input structure of the basal ganglia, comprising the caudate nucleus and putamen, receiving projections from nearly the entire cortex.

Substantia nigra.

A midbrain structure whose pars compacta supplies dopamine to the striatum and whose pars reticulata is an output nucleus of the basal ganglia.

Subthalamic nucleus.

A small nucleus in the indirect pathway that excites the output nuclei through broad projections; a principal target of deep brain stimulation for Parkinson's disease.

Tonic inhibition.

The continuous high-rate firing of the output nuclei that keeps their thalamic and brainstem targets suppressed by default until the striatum lifts the brake.

Key Researchers

Rui M. Costa. Director of the Zuckerman Mind Brain Behavior Institute at Columbia University; his work dissects the corticostriatal circuits for action initiation, skill learning, and self-paced movement. ORCID - Google Scholar

Mahlon R. DeLong. Professor of Neurology at Emory University; he developed the rate model of basal-ganglia disorders and identified the subthalamic nucleus as a target for deep brain stimulation in Parkinson's disease. Wikipedia

Ann M. Graybiel. Institute Professor at the Massachusetts Institute of Technology and investigator at the McGovern Institute; her work revealed the striosome organization of the striatum and the neural basis of habits. ORCID - Wikipedia

Okihide Hikosaka. Senior investigator at the National Eye Institute; his work established the basal-ganglia control of eye movements and its role in reward-based decision-making. ORCID

C. David Marsden (1938–1998). Neurologist at the Institute of Neurology, University College London; his clinical and physiological work characterized the basal ganglia's role in the automatic execution of learned motor plans. Wikipedia

Wolfram Schultz. Professor of Neuroscience at the University of Cambridge; his recordings of midbrain dopamine neurons established the reward-prediction-error signal that underlies reinforcement learning. ORCID - Wikipedia

Frequently Asked Questions

What are the basal ganglia?
The basal ganglia are a group of interconnected subcortical nuclei at the base of the forebrain, arranged in parallel loops with the cortex and thalamus, that select and gate motor and cognitive programs rather than generating movement directly (Alexander et al., 1986).

What do the basal ganglia do?
They solve a selection problem: their output nuclei tonically inhibit the thalamus, and the striatum releases a chosen action while suppressing its competitors, a logic that extends from movement to eye movements, decisions, and habits (Redgrave et al., 1999).

What are the direct and indirect pathways?
The direct pathway inhibits the output nuclei and promotes movement, while the indirect pathway excites them and suppresses movement; dopamine excites the direct and inhibits the indirect pathway, setting the balance between the two (Albin et al., 1989).

How do the basal ganglia relate to dopamine?
Midbrain dopamine neurons encode a reward-prediction error and act on the striatum to strengthen the synapses of actions that turned out better than expected, making dopamine the teaching signal for reinforcement learning (Schultz et al., 1997).

What happens to the basal ganglia in Parkinson's disease?
The loss of dopamine tips the balance toward the indirect pathway, over-inhibiting the thalamus and producing the slowness and poverty of movement characteristic of the disease (DeLong, 1990).

Are the basal ganglia only involved in movement?
No; parallel associative and limbic loops let the same selection operation apply to reward learning, habit formation, decision-making, and the updating of working memory (Cox & Witten, 2019).

How do the basal ganglia support habits?
As a rewarded behavior is repeated, control shifts from the goal-directed associative loop to the sensorimotor loop, producing a habit that is triggered by the situation and relatively insensitive to the current value of its outcome (Graybiel, 2008).

Do the basal ganglia have a role in memory?
Yes; they house a neostriatal habit-learning system that acquires skills and habits incrementally and independently of the declarative memory of the medial temporal lobe, as shown by the double dissociation between amnesic and Parkinson's patients (Knowlton et al., 1996).

References