The Brainstem and Cerebellum

Two structures, two jobs

The brainstem and the cerebellum sit together at the base and back of the brain, but they do very different work. The brainstem is a compact stalk, only a few centimetres long, through which every message travelling between the brain and the body must pass. Packed into it are the controls for the functions we never think about but cannot live without: the drive to breathe, the beat of the heart, the tone of the blood vessels. The cerebellum, by contrast, is dedicated to movement. It does not usually decide to act, but it makes action work, smoothing, timing, and correcting movement so that it is graceful rather than jerky.

What the two share is that they operate largely outside awareness. We do not consciously command our breathing rate, and we do not consciously calculate the muscle adjustments that keep us upright as we walk. Both structures handle these matters automatically, freeing the higher brain to attend to thought, perception, and choice.

Their position in the brain reflects their ancient origins. Both lie below and behind the cerebral hemispheres, closer to the spinal cord than to the reasoning cortex, and both are conserved across a wide range of animals. A fish, a bird, and a person all possess a recognisable brainstem and cerebellum, because the tasks these structures perform, staying alive and moving well, are as old as animal life itself. The great expansion of the human brain happened above them, in the cortex, but it was built on this stable, ancient foundation.

The brainstem: the brain-to-body pathway

The brainstem is the oldest part of the brain in evolutionary terms, and the most fundamental. It forms the continuous link between the spinal cord below and the rest of the brain above, and every ascending sensory pathway and every descending motor pathway runs through it. If the brainstem were a bottleneck in a road network, it would be the single bridge that all traffic must cross.

Beyond serving as a thoroughfare, the brainstem is the source of most of the cranial nerves, the twelve pairs of nerves that supply the head and neck and carry vision-related, hearing, facial, and other signals. Ten of the twelve cranial nerves arise from the brainstem, which is why brainstem injury can disturb eye movements, facial sensation, swallowing, and speech. It also houses centres that generate essential reflexes, including swallowing, coughing, gagging, and vomiting.

A further point is worth stressing. Because so many different functions are packed into such a small volume, the brainstem has no room to spare. In the cortex, a small area of damage may spare much of a function, since neighbouring tissue can sometimes take over. In the brainstem there is no such redundancy: a lesion of a few millimetres can knock out breathing, or eye movement, or consciousness, depending on precisely where it falls. This density is what makes the brainstem at once so efficient and so vulnerable.

The three parts of the brainstem

The brainstem is divided, from top to bottom, into three regions. Each has its own emphasis, though they work as a continuous whole.

The medulla oblongata deserves special note. Small as it is, it contains the vital centres that set the rhythm of breathing and regulate the heart and blood vessels. This is why the medulla is often called the most life-critical few centimetres of the entire body: damage here can stop breathing or the heartbeat outright.

Crossed findings: the localising rule

The last section made a claim and left it hanging: the brainstem is so densely packed that a lesion of a few millimetres produces a specific, predictable syndrome, and a clinician can name the level of the damage from the signs at the bedside. How? The answer is a single rule, and it is the most useful fact in the whole of brainstem anatomy.

Two kinds of pathway run through the brainstem, and they cross the midline at different places.

Put the two together. A lesion in the brainstem catches a cranial nerve nucleus or its exiting fibres, which report the same side, and it catches the long tract, which has not yet crossed and reports the other side. The patient therefore presents with cranial nerve signs on one side and limb signs on the other. This pattern is called crossed findings, and it is close to pathognomonic: no lesion of the cerebral hemispheres can produce it, because in the hemispheres everything is contralateral together. Crossed findings mean the brainstem, and they mean it with near-certainty.

This is what the density of the brainstem buys. In the cortex, the same volume of damage tells you little, because the tissue is redundant and the maps are broad. In the brainstem every millimetre carries different traffic, and the very feature that makes injury there so dangerous, the absence of any spare room, is the feature that makes the injury so precisely readable.

Arousal and consciousness

Running through the core of the brainstem is a loose, net-like web of neurons called the reticular formation. Rather than forming a tidy nucleus, it is a diffuse network that reaches upward to the thalamus and cortex and downward to the spinal cord. One of its most important jobs is to control arousal: the general level of wakefulness and alertness of the whole brain.

It is usually said that the reticular formation "keeps the cortex awake", which leaves out the interesting part: the route. The reticular formation does not project to the cortex directly in any great volume. Its principal ascending path, the ascending reticular activating system, runs from the brainstem up into the intralaminar nuclei of the thalamus, which are the nuclei scattered within the thalamus's internal medullary lamina, and those nuclei then project diffusely and non-specifically across the whole cortical surface. Brainstem, then intralaminar thalamus, then cortex. Arousal reaches the cortex through the same structure that its sensory content does, but by a different set of nuclei, and it arrives as a broadcast rather than as a message.

That route is why the thalamus appears in every account of consciousness, and it is why bilateral damage to the medial and intralaminar thalamus can abolish consciousness while leaving the cortex structurally intact. A second, parallel set of ascending pathways runs from small brainstem nuclei that release noradrenaline, serotonin, and acetylcholine directly onto the cortex, changing its state rather than its content; see neuromodulation.

Because of these routes, injury to the reticular formation or the upper brainstem can cause a loss of consciousness or coma even when the cortex itself is untouched. Consciousness depends not only on the thinking cortex but on the brainstem switch that keeps it turned on, and on the thalamic wiring through which that switch is thrown.

The cerebellum: coordination and balance

The cerebellum sits behind the brainstem, beneath the back of the cerebral hemispheres, and its name, Latin for little brain, suits it well: it looks like a smaller version of the whole brain, with its own finely folded surface. Its surface is even more densely pleated than the cortex above, packing an enormous sheet of tissue into a compact space.

The cerebellum's central role is the coordination of movement. It does not usually initiate actions; instead it takes the movement commands issued by the cortex, compares them against streams of information about the body's actual position and motion from the muscles, joints, and balance organs of the inner ear, and issues moment-to-moment corrections. The result is movement that is smooth, accurately timed, and well aimed. It also maintains balance and posture, keeping us upright and steady without conscious effort.

The signs of cerebellar damage make its role vivid. A person with cerebellar injury is not paralysed but becomes clumsy and unsteady: movements overshoot or fall short, gait becomes wide and staggering, and fine actions grow tremulous and imprecise. This pattern, known as ataxia, shows that the cerebellum is the brain's instrument for calibration and fine control rather than for raw strength.

1. Receiving the plan

   The cerebellum receives a copy of the movement commands sent out by the motor cortex, telling it what the body is about to attempt.

2. Comparing with reality

   At the same time it receives sensory feedback about the body's actual position and motion from the muscles, joints, and the balance organs of the inner ear.

3. Correcting the movement

   It computes the difference between intended and actual movement and sends corrections that keep the action smooth, timed, and on target.

4. Learning for next time

   Over repeated attempts it adjusts its own output, so that skills become more accurate and automatic, the basis of motor learning.

That four-step summary is the standard account, and it is not wrong, but it hides two enormous questions. What does "compares intended with actual" mean, mechanically? And how does the fourth step, learning, actually happen? Both have answers, and both answers are anatomy.

The circuit, and the learning rule inside it

The cerebellar cortex is famous for being the same everywhere. Slice it anywhere on its surface and you find the identical arrangement of the same handful of cell types, repeated with a regularity found nowhere else in the brain. That uniformity is the reason the cerebellum was the first large brain structure whose wiring diagram was worked out in full, and it is the reason its circuit can be described in a few paragraphs. Five elements carry the whole argument.

1. Mossy fibres bring the context

   The great mass of input to the cerebellum arrives on mossy fibres, carrying everything the cerebellum needs to know about the current situation: where the limbs are, what the muscles are doing, what the balance organs report, and, by way of the pons, what the cerebral cortex intends. Mossy fibres do not describe an error. They describe a state.

2. Granule cells recode it

   Mossy fibres synapse onto granule cells, the tiny, astronomically numerous cells that account for most of the neurons in the brain. Each granule cell samples only a handful of mossy fibres, so the population as a whole represents an enormous number of distinct combinations of the incoming context. This is the recoding step, and it is what the sheer number of granule cells is for.

3. Parallel fibres broadcast it

   Each granule cell sends its axon up toward the surface, where it splits in two like a T and runs for millimetres in both directions. These are the parallel fibres, and they run in strict alignment, like the wires of a loom. A single parallel fibre passes through and contacts hundreds of the cells in the next layer.

4. Purkinje cells read it

   The Purkinje cell is the output of the cerebellar cortex, and its dendritic tree is a magnificent flat fan oriented precisely at right angles to the parallel fibres, so that the fibres pass through it like wires through a comb. A single Purkinje cell may receive on the order of a hundred thousand parallel-fibre synapses. It is the most heavily innervated neuron known. Purkinje cells are inhibitory, and their axons carry the cerebellum's verdict to the deep cerebellar nuclei, which are the structure's actual output to the rest of the brain.

5. One climbing fibre, from the inferior olive

   Crossing all of that at right angles comes an entirely separate input. Each Purkinje cell receives exactly one climbing fibre, an axon from a nucleus in the medulla called the inferior olive, which wraps itself around the Purkinje cell's dendrites like ivy round a tree. It fires rarely, roughly once a second, but when it fires the Purkinje cell cannot ignore it: a single climbing-fibre input produces an overwhelming, unmistakable burst of activity. One hundred thousand whispers, and one shout.

Now the payload, and it is the reason all of this was worth setting out. The climbing fibre is not just loud. It changes the Purkinje cell's other synapses. When the climbing fibre fires, it depresses precisely those parallel-fibre synapses that were active in the moments just before. Those synapses become weaker. The ones that were quiet are left alone.

Consider what that rule does if the climbing fibre carries an error signal, which is what the inferior olive appears to supply: a report that the movement just made did not come out as it should have. The parallel fibres active in the moments before the error were the ones representing the context in which the faulty command was issued. The rule therefore weakens exactly the input combinations that led to the error, and leaves untouched every combination that did not. Repeat it a few hundred times and the Purkinje cell has learned to stop responding to the contexts that produce mistakes.

And it explains the page's own claim about uniformity. The circuit is identical everywhere because the operation is identical everywhere: take context, take an error, adjust. What differs from one patch of cerebellum to the next is only what the mossy fibres are carrying in and where the deep nuclei send the answer out. A patch wired to motor cortex learns to refine a reach. A patch wired to prefrontal cortex receives a different context and a different error, and refines whatever that cortex is doing. The tiling is the point.

The forward model, and why you cannot tickle yourself

The other loose end is the phrase "compares intended movement with actual movement", which is repeated in every account of the cerebellum, this page's earlier sections included, and which on inspection does not quite make sense. An intention is not the same kind of thing as a sensory signal. You cannot compare a command to leave the hand at a certain place with a stream of nerve impulses from the skin and the joints; they are not in the same units. Something has to translate.

That something is a forward model, and naming it is what turns the phrase into a mechanism.

1. Take a copy of the command

   When the motor cortex issues a command, a copy of it, an efference copy, is sent to the cerebellum by way of the pons. The cerebellum now knows what the body has been told to do.

2. Predict the consequences

   The cerebellum uses that copy to predict what the senses ought to report if the command works: this is what your fingertips will feel, this is where your arm will be, this is when. The prediction is in sensory units, not in motor ones, and it is available before the real sensation arrives, because prediction is faster than the body.

3. Compare prediction against reality

   The actual sensory feedback then arrives. The cerebellum compares it not with the intention but with the prediction. The difference between them is the error, and it is a well-defined quantity because both terms are now the same kind of thing.

4. Send the error where it can be used

   That error is what the inferior olive reports up the climbing fibres, and the learning rule described above then does its work. Prediction, comparison, error, adjustment: the loop closes, and next time the prediction is better.

This is also why the cerebellum can correct a movement faster than sensory feedback would permit. Waiting for the limb to report where it actually is costs tens of milliseconds, which is far too slow for a fast reach. A system that can predict where the limb will be does not have to wait, and the clumsiness of cerebellar ataxia, the overshooting, the tremor that grows as the hand nears its target, is what movement looks like when the prediction is gone and the brain is reduced to correcting on real feedback alone.

A little brain that holds most of the neurons

One of the most striking facts about the cerebellum is its cell count. Although it accounts for only about a tenth of the brain's total mass, it contains the large majority of the brain's neurons. Careful modern counts put the figure at roughly 69 billion of the brain's estimated 86 billion neurons in the cerebellum alone, most of them tiny, densely packed granule cells arranged in a regular, repeating circuit.

This abundance of cells is not wasteful, and it is not, as it is sometimes made to sound, simply a matter of needing lots of neurons to control lots of muscles. The circuit described above tells you exactly what the granule cells are for. Each one samples a few mossy fibres, so the granule cell population encodes the combinations of the incoming context, and the number of distinguishable combinations rises steeply with the number of cells. A learning rule that adjusts synapses on the basis of context can only be as discriminating as the code that represents that context. The cerebellum has tens of billions of granule cells because it needs an enormous number of separately adjustable input patterns, and every one of those patterns must run out along a parallel fibre and be offered to the Purkinje cells for weighting.

The cell count, in other words, is the capacity of the learning machine. That is why it is so large, and why the number is a feature of the design rather than an anomaly to be marvelled at.

Beyond movement: an emerging role in cognition

For most of its history the cerebellum was regarded as a purely motor structure. That view is changing. Anatomical and imaging studies show that the cerebellum is connected not only to the motor systems but also to regions of the cortex involved in language, attention, and emotion, and that different parts of the cerebellum map onto these different functions.

The current understanding is that the cerebellum applies the same kind of fine-tuning to thought that it applies to movement, smoothing and coordinating mental operations much as it smooths physical ones. People with certain cerebellar injuries can show subtle difficulties with language, planning, and emotional regulation, a pattern that has drawn growing attention. The cerebellum, once dismissed as merely a movement machine, is increasingly seen as a general-purpose coordinator that contributes to cognition as well.

This broader view is no longer hand-waving, because the circuit gives it an argument. The cerebellar cortex repeats the same wiring across its whole surface, and that wiring implements one operation: build a prediction from a copy of the command, compare it with what arrives, take the error on the climbing fibre, and reweight the parallel-fibre synapses that produced it. Nothing in that operation is specifically motor. It requires only that something upstream can issue a command, that something downstream can report an outcome, and that the difference between them can be computed. A patch of cerebellum wired to the motor cortex will therefore learn to predict and refine a reach. A patch wired to the prefrontal cortex receives a different context on its mossy fibres and a different error on its climbing fibres, and will learn to predict and refine whatever that cortex is doing.

The difference between a motor cerebellum and a cognitive cerebellum, on this account, is not a difference in the computation at all. It is a difference in the address book: which part of the cortex a given patch is wired to, and what that cortex counts as an error. The anatomy supports it, since the cerebellum's loops with prefrontal and parietal cortex are as real as its loops with motor cortex, and the clinical picture supports it, since cerebellar lesions can produce difficulties with planning, language, and emotional regulation that look, in their character, oddly like ataxia moved indoors: overshooting, poor timing, a failure of smoothness.

What remains genuinely contested is how far the analogy runs. It is not settled what the climbing fibre's "error" corresponds to in a non-motor task, nor whether the inferior olive can supply such a signal for thought at all, nor whether the cognitive symptoms of cerebellar damage reflect a lost computation or merely the loss of a well-timed contribution to a cortical process that does the real work. The purely motor picture is dead. What replaces it is not yet agreed.

Related topics