Cerebellum

Abstract

The cerebellum is a densely folded structure at the base of the brain that contains more neurons than the rest of the brain combined, yet occupies only about a tenth of its volume. Long regarded as a purely motor organ that smooths and coordinates movement, it is now understood to contribute to timing, sensorimotor prediction, and a range of cognitive and affective functions. Its cortex is built from a single, endlessly repeated microcircuit, which led theorists to propose that it performs one canonical computation wherever it is applied. This article surveys its anatomy, its uniform circuit and the classic learning theories built on it, its role in motor adaptation and the timing of conditioned responses, its cognitive and affective contributions, and the modern circuit work extending it into reward and social behavior, with three interactive demonstrations.

Keywords: cerebellum, motor learning, internal models, Purkinje cell, cerebellar cognitive affective syndrome

The cerebellum (Latin for 'little brain') is a hindbrain structure lying behind the brainstem in the posterior cranial fossa, connected to the rest of the brain by three pairs of peduncles. Though it accounts for roughly 10% of brain volume, it holds on the order of 69 billion neurons, about 80% of the brain's total, the overwhelming majority of them tiny granule cells (Herculano-Houzel, 2009). Its surface is folded into fine parallel ridges called folia, packing an enormous cortical sheet into a compact volume. For more than a century the cerebellum was studied as the organ of motor coordination, because its damage produces ataxia, a breakdown of smooth, accurately timed movement. That view has broadened: the same circuitry participates in attention, language, working memory, and emotion, and its lesions can produce a recognizable cognitive and affective syndrome (Schmahmann & Sherman, 1998; Buckner, 2013).

What the Cerebellum Is

The cerebellum sits dorsal to the pons and medulla, forming the roof of the fourth ventricle. It is divided into two hemispheres joined by the midline vermis, and its cortex is a three-layered sheet whose cellular architecture is strikingly regular across its entire extent. Two afferent systems carry information in: mossy fibres, arising from the spinal cord, brainstem, and pontine nuclei that relay cerebral cortex, and climbing fibres, arising exclusively from the inferior olive (Marr, 1969). Mossy fibres synapse onto the vast population of granule cells, whose axons ascend and bifurcate into parallel fibres running through the dendritic trees of the Purkinje cells. Each Purkinje cell, the sole output of the cerebellar cortex, is contacted by on the order of 100,000 parallel fibres but by only a single climbing fibre, an anatomical asymmetry that is central to every theory of cerebellar learning. Purkinje cells inhibit the deep cerebellar nuclei, which provide the cortex's output to the rest of the brain (Herculano-Houzel, 2009).

Figure 1

The Cerebellar Cortical Microcircuit

Types of Cerebellum

In the Medical Subject Headings (MeSH) classification the cerebellum is filed under the Metencephalon, the upper hindbrain division it shares with the pons, and it is subdivided into three direct anatomical subtypes. These subtypes are structural partitions rather than functional modules: an ordinary cerebellar computation runs through the cortex and out through the nuclei together, so the divisions below describe where tissue sits, not separable jobs. MeSH is an indexing vocabulary used to organize the biomedical literature, so the list reflects how work on the cerebellum is catalogued rather than a theoretical claim about its only meaningful divisions.

The Uniform Microcircuit and Its Theories

The cerebellar cortex's regularity invited a computational reading. In an influential theory, Marr proposed that the immense granule-cell layer performs a pattern-separation function: by recoding each mossy-fibre input onto a much larger, sparsely active population of parallel fibres, it renders overlapping inputs more discriminable, so that a Purkinje cell can learn to respond selectively to many distinct contexts (Marr, 1969). Albus formalized a closely related model and made the critical prediction about the direction of learning: the climbing fibre acts as a teacher whose activity produces long-term depression of the parallel-fibre synapses that were recently active, weakening the connections that led to an error (Albus, 1971). Ito's experimental work later identified long-term depression at the parallel-fibre–Purkinje synapse as a physiological substrate for exactly this kind of climbing-fibre-instructed plasticity, giving the Marr–Albus theory a cellular mechanism (Ito, 2008).

The expansion recoding at the heart of Marr's proposal is easy to see. A small set of active mossy fibres is mapped onto a far larger granule population in which each cell fires only when several of its inputs coincide, so the active fraction shrinks and patterns that shared inputs become more separable. The demonstration below lets the input drive be varied and shows the resulting sparse granule code.

Motor Learning and Timing

The Marr–Albus–Ito framework predicts that the cerebellum should support supervised learning, in which a teaching signal drives error-correcting change. The delay eyeblink conditioning preparation provides the cleanest test. When a neutral conditioned stimulus, such as a tone, is repeatedly followed by an air puff to the eye, an animal learns to blink in anticipation, and crucially the learned blink is timed to peak just as the puff arrives. Lesion, recording, and stimulation studies established that this association and its precise timing depend on the cerebellum and its associated brainstem circuitry, with the conditioned response abolished by damage to the relevant cerebellar region (McCormick & Thompson, 1984). The conditioned eyeblink became a model system precisely because its neural pathway could be traced from the conditioned and unconditioned stimuli through to the motor output.

Adaptive timing is a general property. The learned response rescales when the interval between stimuli changes, a signature the cerebellar cortex is required to produce. The demonstration below models the timed conditioned eyelid closure and how its peak tracks the moment the unconditioned stimulus is expected.

Internal Models and Prediction

A unifying account of what the cerebellum computes is that it builds internal models — neural representations that predict the sensory consequences of actions and supply corrections faster than sensory feedback allows (Ito, 2008). On this view the climbing-fibre error signal trains a forward model whose predictions cancel the reafferent sensory noise of self-generated movement and anticipate its outcomes, which is why cerebellar damage degrades the smoothness and accuracy of movement rather than the ability to move at all. The same predictive machinery, applied to representations held in the cerebral cortex rather than to limb dynamics, is proposed to underlie the cerebellum's cognitive contributions (Sokolov et al., 2017).

Sensorimotor adaptation makes the internal-model idea concrete and measurable. When vision of a reaching hand is rotated, the first movements miss in the direction of the perturbation; over trials the error shrinks roughly exponentially as the cerebellar forward model recalibrates the mapping from motor command to expected sensory result, driven by the sensory-prediction error on each trial (Sokolov et al., 2017). The demonstration below models this trial-by-trial adaptation.

Cognitive Implications

Because the cerebellar microcircuit is anatomically uniform, its computation should not care whether the information routed through it concerns a limb or a thought — a logic that reframes the cerebellum as a contributor to cognition wherever cerebral association areas project into it. Anatomical tracing in primates showed that the cerebellum is reciprocally connected not only with motor cortex but with prefrontal area 46, a region central to executive function, establishing a substrate for nonmotor influence (Strick et al., 2009). Clinically, lesions restricted to the cerebellum can produce the cerebellar cognitive affective syndrome: impairments of executive function, spatial cognition, and language together with flattening or dysregulation of affect, described from a series of patients whose motor signs did not account for their deficits (Schmahmann & Sherman, 1998).

Neuroimaging mapped where these functions sit. A meta-analysis of functional imaging found a reliable division of labour: the anterior lobe is engaged by sensorimotor tasks, while posterior-lobe regions are recruited by language, working memory, and other cognitive demands, mirroring the motor-to-association gradient of the cerebral cortex it connects with (Stoodley & Schmahmann, 2009). Resting-state connectivity then showed that most of the human cerebellum maps onto cerebral association networks rather than motor ones (Buckner, 2013), and dense individual mapping with a multi-domain task battery resolved distinct functional boundaries for movement, cognition, and social processing within a single cerebellum (King et al., 2019). A principal-gradient analysis further placed cerebellar cognitive representation along a continuum from motor to abstract, transmodal processing (Guell et al., 2018).

Worked Example

Consider how reaching error falls during adaptation to a visuomotor rotation, the process the third demonstration models. Let the error on trial n be e(n) = r(1 − α)ⁿ, where r is the imposed rotation and α is the fraction of the remaining error the cerebellum cancels each trial. This is the discrete exponential decay expected when a forward model corrects a fixed proportion of its sensory-prediction error per movement. Take a rotation of r = 30° and a learning rate of α = 0.16, so each trial retains 84% of the previous error.

Trial 0: error = 30.0°. Trial 1: 30 × 0.84 = 25.2°. Trial 2: 21.2°. Trial 3: 17.8°. Trial 5: 12.5°. Trial 10: 5.2°. The error is halved about every four trials, and solving r(0.84)ⁿ < 2° gives n = ⌈ln(2/30) / ln(0.84)⌉ = 16 trials to fall below two degrees. The curve is steep at first, when the prediction error is large, and flattens as the model converges — the same negatively accelerated shape seen in the eyeblink acquisition of the previous section, because both are error-correcting learning driven by a teaching signal. If the rotation is then removed, the recalibrated model produces an aftereffect error in the opposite direction that washes out by the same rule, a hallmark of genuine cerebellar adaptation rather than a strategic correction (Sokolov et al., 2017).

Discussion

The cerebellum illustrates a principle that runs against the intuition that different mental functions need different machinery: one anatomically uniform circuit, applied to whatever information is routed into it, can serve motor coordination, associative timing, and cognition alike. Its lesions do not abolish movement or thought wholesale; they degrade the precision, timing, and calibration of both, producing ataxia in the motor domain and the dysmetria-like disorganization of the cerebellar cognitive affective syndrome in the mental one (Schmahmann & Sherman, 1998). This computational uniformity is what makes the internal-model account attractive: a single learning rule, taught by climbing-fibre error and expressed as synaptic depression, can be read as calibrating a forward model whether its target is a limb trajectory or a chain of reasoning (Ito, 2008). The clinical reach is broad, because cerebellar dysfunction is implicated in ataxias, in dyslexia, and in the motor and social features of several neurodevelopmental conditions, where the same predictive machinery that calibrates ordinary behavior is disturbed.

Current Directions

Recent work has pushed the cerebellum well beyond its motor and cognitive roles into the domain of reward and motivation. Population imaging of granule cells during a conditioning task found that they encode not only sensory and motor variables but the expectation of reward, and even its unexpected omission, showing that the input layer carries signals long assumed to belong to the forebrain (Wagner et al., 2017). Circuit tracing then demonstrated a direct projection from the deep cerebellar nuclei to the ventral tegmental area that is necessary for social preference, placing the cerebellum inside the brain's reward and social circuitry (Carta et al., 2019). In parallel, high-resolution functional mapping continues to refine the cerebellum's internal geography: individual-subject task batteries resolve separate territories for movement, language, and social cognition (King et al., 2019), and gradient analyses describe a smooth progression from motor to transmodal representation across the cerebellar cortex (Guell et al., 2018). Together these findings recast the cerebellum as a general-purpose predictive engine embedded in circuits for emotion and reward, not merely a coordinator of movement.

Common Misconceptions

The cerebellum is only for movement.

Most of the human cerebellum connects to cerebral association networks rather than motor ones, and its lesions can impair executive function, language, and affect with little motor deficit (Buckner, 2013; Schmahmann & Sherman, 1998). The motor label persists because ataxia is the most visible cerebellar sign.

Its small size means few neurons.

The cerebellum occupies about a tenth of brain volume but contains roughly 80% of all its neurons, on the order of 69 billion, most of them granule cells (Herculano-Houzel, 2009).

The cerebellum learns by strengthening the synapses that worked.

The classic cerebellar learning rule is the opposite: climbing-fibre error signals produce long-term depression of the recently active parallel-fibre synapses, weakening the connections that led to an error (Albus, 1971; Ito, 2008).

Glossary

Ataxia.

The loss of smooth, accurately timed and coordinated movement that follows cerebellar damage.

Cerebellar cognitive affective syndrome.

A pattern of executive, spatial, linguistic, and affective impairment produced by cerebellar lesions in the absence of adequate motor explanation.

Climbing fibre.

An afferent from the inferior olive that contacts a single Purkinje cell and carries the error or teaching signal for cerebellar learning.

Deep cerebellar nuclei.

The grey-matter nuclei that receive inhibition from Purkinje cells and form the output of the cerebellum.

Forward model.

An internal representation that predicts the sensory consequences of a motor command, allowing correction faster than sensory feedback permits.

Granule cell.

The smallest and most numerous cerebellar neuron, whose parallel-fibre axon carries recoded mossy-fibre input across the Purkinje dendrites.

Inferior olive.

A brainstem nucleus that is the sole source of climbing fibres, supplying the cerebellum with its error or teaching signal.

Internal model.

A neural system that emulates the input–output behavior of the body or the world to predict outcomes and guide action.

Long-term depression.

A lasting weakening of a synapse; at the parallel-fibre–Purkinje synapse it is the proposed substrate of cerebellar motor learning.

Mossy fibre.

A cerebellar afferent from the spinal cord, brainstem, and pontine nuclei that synapses onto granule cells.

Parallel fibre.

The bifurcated axon of a granule cell, running through and synapsing on the dendritic trees of many Purkinje cells.

Purkinje cell.

The large output neuron of the cerebellar cortex, integrating parallel-fibre and climbing-fibre input and inhibiting the deep nuclei.

Sensorimotor adaptation.

The trial-by-trial recalibration of movement to a perturbation, such as a visuomotor rotation, driven by sensory-prediction error.

Vermis.

The midline region joining the two cerebellar hemispheres, associated with axial motor control and, in its posterior part, affect.

Key Researchers

James S. Albus

(1935–2011). Engineer at the National Institute of Standards and Technology; he formalized a theory of cerebellar function in which climbing-fibre error signals adjust parallel-fibre synapses, a foundation of cerebellar learning theory. Wikipedia - Wikidata

Masao Ito

(1928–2018). Neuroscientist at RIKEN and the University of Tokyo; he discovered long-term depression at the parallel-fibre–Purkinje synapse and developed the internal-model account of cerebellar function. Wikipedia - Wikidata

Richard B. Ivry

Professor of Psychology and Neuroscience at the University of California, Berkeley; he studies the cerebellum's role in timing, prediction, and sensorimotor adaptation across motor and cognitive domains. Faculty Page - Wikipedia

David Marr

(1945–1980). Neuroscientist and computational theorist at the Massachusetts Institute of Technology; his theory of cerebellar cortex proposed granule-cell pattern separation and climbing-fibre-taught learning. Wikipedia

Jeremy D. Schmahmann

Professor of Neurology at Harvard Medical School and director of the Ataxia Center at Massachusetts General Hospital; he described the cerebellar cognitive affective syndrome and mapped the cerebellum's cognitive topography. Faculty Page - ORCID

Catherine J. Stoodley

Professor of Neuroscience at American University; her neuroimaging meta-analyses established the functional topography dividing motor from cognitive regions of the human cerebellum. Faculty Page

Peter L. Strick

Distinguished Professor of Neurobiology and director of the Brain Institute at the University of Pittsburgh; his tracing studies revealed reciprocal cerebellar connections with prefrontal cortex, a substrate for nonmotor function. Faculty Page - Wikipedia

Richard F. Thompson

(1930–2014). Neuroscientist at the University of Southern California; he identified the cerebellar circuitry essential for the classically conditioned eyeblink response. Wikipedia - Google Scholar

Frequently Asked Questions

What is the cerebellum?

The cerebellum is a densely folded hindbrain structure that coordinates movement and contributes to timing, prediction, and cognition; it holds most of the brain's neurons in a small fraction of its volume (Herculano-Houzel, 2009).

What does the cerebellum do besides movement?

It contributes to executive function, language, working memory, and emotion, and its damage can produce the cerebellar cognitive affective syndrome (Schmahmann & Sherman, 1998; Buckner, 2013).

How does the cerebellum learn?

By adjusting parallel-fibre synapses onto Purkinje cells under the instruction of climbing-fibre error signals, expressed physiologically as long-term depression (Albus, 1971; Ito, 2008).

Why does the cerebellum have so many neurons?

Its granule cells, the most numerous neuron type in the brain, form an enormous expanded layer that recodes inputs for pattern separation, as Marr's theory proposed (Herculano-Houzel, 2009; Marr, 1969).

What is an internal model?

It is a neural system that predicts the sensory consequences of actions; the cerebellum is thought to build such forward models to calibrate movement and, by extension, thought (Ito, 2008; Sokolov et al., 2017).

What happens when the cerebellum is damaged?

Movement becomes uncoordinated and poorly timed (ataxia), and cognitive and affective functions can be disrupted even when motor signs are mild (Schmahmann & Sherman, 1998).

How is the cerebellum studied in learning experiments?

The classically conditioned eyeblink is a standard model, because its precisely timed response and its neural pathway through the cerebellum can be traced in detail (McCormick & Thompson, 1984).

Does the cerebellum have a role in reward?

Yes; its granule cells encode reward expectation, and a direct projection to the ventral tegmental area contributes to social behavior, placing it within reward circuitry (Wagner et al., 2017; Carta et al., 2019).

References