What is the cellular organisation of the cerebellum?

How might this organisation relate to function?

The cerebellum shows a very ordered, repeating, cellular structure. It is sometimes
called the “neuronal machine”. This has led many to speculate on how its function is
related to this: with mathematicians and computer scientists often joining
neuroscientists in proposing models for how the structure might work. The cerebellum
is responsible for a higher level of motor control than motor cortex, since lesions here
lead to a more general disorder of function, rather than a paralysis or weakness of
movement of the type that might be seen with a motor cortex lesion. Since birds and
reptiles lack motor cortex but still have cerebella, the cerebellum is probably
evolutionarily an older structure.

The cerebellum seems to have arisen as an outgrowth of the brainstem as an adjunct

to the vestibular system. It is divided into three parts by Fulton and Dow (in
monkeys): archicerebellum, paleocerebellum and neocerebellum, also known as the
vestibulocerebellum, spinocerebellum and cerebrocerebellum respectively. The first
set of names gives some idea of the evolutionary order of the regions, while the
second gives more idea of their function. The division is both on the basis of anatomy,
and on the varying types of disorder which occur when parts of the cerebellum are
lesioned. More recently, however, it has increasingly been realised that the division is
not absolute, and there is considerable overlap both in terms of connections and
functions. In fact, there also exists another idea entirely for the organisation of the
cerebellum, stemming from the observation that there is a somatotopic map in the
anterior lobe of the cerebellum, with trunk in the midline, limbs lateral, tail anterior
and head posterior1. It can be argued from this2 that each region of the cerebellum
looks after all aspects of cerebellar function, but then how are we to explain the
varying motor effects of lesions in different areas?

The deep nuclei of the cerebellum, that is to say the fastigius, globose and emboliform
(collectively known as the interpositus), and dentate, generate all the output of the
cerebellum. Each contains a separate somatotopic representation of the body, with the
head, tail, trunk and extremities represented in the caudal, rostral, lateral and medial
regions of the nucleus, respectively3. They project to and control the movement of the
ipsilateral side of the body, and between them output to almost every motor centre of
the CNS, by glutaminergic excitatory synapses. The cells of these deep neurons can
seem to fire along with certain behaviour. For example, the fastigial nucleus neurons
fire during walking, consistent with the theory that the fastigius is concerned with
stance and gait.4 The neurons of the interpositus fire when the holding position of a
limb is disturbed, and change their activity in response to sensory feedback. These
observations support the idea that the neurons of these nuclei are concerned with limb
position, and in particular its retention, by modulating the activity of both agonist and
antagonist muscles. Finally, excitation in the neurons of the dentate nucleus precedes
the onset of movement, which has given rise to the idea of their involvement in
planning these movements.
1
R. S. Snider and Edgar Adrian observed this in the 1940’s.
2
At least by Zigmond and Bloom, albeit fleetingly.
3
Asanuma, Thach and Jones
4
Andersson and Armotrong 1987
The archicerebellum, consisting of the flocculonodular lobe, has extensive
connections with the vestibular system and the reticular formation: it is concerned
with the maintenance of balance and the control of eye movements, and its effects are
mediated by the descending vestibulospinal and reticulospinal projections. It is the
oldest area. Damage to it through lesions or disease impairs ability to control eye
movements, and impairs patients’ sense of balance. But when they lie down, they
have no difficulty in moving arms or limbs. It is this type of evidence that leads one to
reject the idea of somatotopic function of the cerebellum, and instead embrace the
Fulton and Dow model.

The paleocerebellum consists of the vermis and paravermis. It influences muscle tone
and posture. Ventral and dorsal spinocerebellar fibres serve as afferents, carrying
information from muscle, joint and cutaneous receptors, particularly from the legs.
There are analogous pathways conveying information from the upper limbs. There is
also afferent information from the visual system, via the superior colliculus. It is
organised into somatotopic maps, as first described by Adrian and Snider. But more
modern techniques have shown that the somatotopy is in fact segmented: that is to say
that afferent input from a particular area will go to several little patches of granule
cells: “fractured somatotopy”. From here, the output varies according to the
anatomical part of the spinocerebellum. The vermis neurons output to the fastigial
nucleus, and the paravermal areas output via the interposed nuclei. This output is to
the red nucleus of the midbrain, affecting the rubrospinal tract, as well as to primary
motor cortex affecting the corticospinal tracts. Its actions are thus focused on limb
muscles and axial musculature, regulating such functions as balance and posture
control during voluntary tasks, as well as the voluntary tasks themselves. Lesions of
this part of the cerebellum result in a number of characteristic symptoms. There are
pendular reflexes, whereby the limb oscillates about its position after a reflex: this is
due to problems in co-ordinating action of agonist and antagonist muscles. There is a
terminal tremor (after attaining the desired position), for the same reason, as well as a
failure to anticipate the need for altered activity of the muscles as the desired position
is approached.

The neocerebellum is the newest part of the cerebellum, and is concerned with
muscular co-ordination, including the trajectory, speed and force of movements,
collectively grouped under the umbrella term of “planning”. In keeping with so many
of these high functions, the lateral cerebellar hemispheres that make up the
neocerebellum are greatly expanded in humans. The afferent fibres come from the
pons: the ponto-cerebellar fibres, but the information originally comes from the
cortex. Output is via the dentate nucleus to either the ventral lateral nucleus of the
thalamus or the red nucleus of the midbrain. These red nucleus neurons project to the
inferior olive, whence fibres project back to the cerebellum via the climbing fibres.
These neurons also receive input from the lateral premotor area, and there is the idea
that this loop could be involved in the mental rehearsal of movements, and with motor
learning. Lesions of this area disrupt motor planning, and prolong reaction time.

Since the three areas of the cerebellum have different functions, but the cellular
machinery appears the same, the idea has arisen that the cerebellum is performing the
same general function on all three pathways. The cerebellum influences the motor
systems by evaluating disparities between intentions and action, and by adjusting the
operation of motor centres in the cortex and brain stem while a movement is in
progress, as well as during repetitions of the same movement. This was suggested by
Lisberger. That this feedback information is very important is borne out by the fact
that 40 times more neurons enter the cerebellum than leave it. At least some areas of it
seem to contribute to memory, since the synaptic connections can be modified. The
exact details of this are much contested, but before this is visited, the detailed cellular
structure of the cerebellum will first be examined.

The cortex cerebellum has three cell layers, namely the molecular cell layer, Purkinje
cell layer and granule layer (dominated by the granular cell). The Purkinje cells are
the dominant cell type. They are lined up in orderly rows with their cell bodies in the
Purkinje cell layer. Each has a huge dendritic tree, typically containing the area of two
front doors5, at 90° to the surface of the cortex. Normally, the dendritic tree is in the
AP axis. These cells are the sole output of the cerebellar cortex, forming inhibitory
GABA-ergic synapses with the four main nuclei of the cerebellar cortex. These are
the fastigial, globose, emboliform and dentate nucleus, and through these pass all
output of the cerebellum.

The cerebellum receive two forms of afferent input, the mossy fibres and the climbing
fibres. Both are excitatory, but the two terminate in histologically different ways on
different areas of the cerebellar cortex, and evoke different firing patterns. Each
Purkinje cells receives a unique single afferent climbing fibre6. Each climbing fibre
will synapse with between one and ten Purkinje fibres. These come from the inferior
olive, and form strongly excitatory synaptic contacts. The inferior olive, in turn,
receives afferent input largely from the cerebral cortex, but also from the spinal cord
and special senses (sight and vestibular senses particularly). The fibres wrap around
the Purkinje neuron, as presumably a “climbing” fibre might. As well as exciting the
Purkinje fibres, they also excite the deep nuclei to which these fibres project.
Experiments show that an action potential in a climbing fibre never fails to excite the
associated Purkinje cell. That said, given that the Purkinje cells are firing almost
continuously, while the climbing fibres seem only to fire about once a second on
average. The Purkinje cell has voltage gated calcium channels in its dendrites, so this
infrequent firing of the climbing fibre can have quite a far reaching impact on the
Purkinje cell, triggering a whole series of action potentials followed by a period of
quiescence: a “complex spike”. These connections are thought to have some role in
laying down motor memory. The terminals of the climbing fibres in the cerebellar
cortex are arranged topographically, so that clusters of olivary neurons are connected
to parasagittal strips of Purkinje cells.

The other type of afferent received by the cerebellum is from the mossy fibres, which
arise from the nuclei of the spinal cord and brain stem. Information comes from all the
other areas (not the inferior olive) that input to the cerebellum, among them the
vestibular and spinal afferents, indirectly through the precerebellar nuclei of the
reticular formation, and from the cortex via the pons. The fibres enter via the lower
cortical levels and synapse in glomeruli with a number of dendrites from granule cells

5
Carpenter’s analogy. I could not find the actual numbers.
6
Although the term unique is used to imply a surjection (many to one correspondence) between
climbing afferents and Purkinje cells, the term could equally be used to describe this arrangement in
general. Where else at such high levels is there such a one to one correspondence? Certainly not in the
cerebral cortex, where there tends to be integration of information.
in the cortex. The granule cells send axons, the parallel fibres, to the surface. The
axons can run up to one third of the cerebellar surface, along the way contacting many
Purkinje cells. Each Purkinje cell will receive input from parallel fibres of up to one
million granule cells, indeed this is the use of the enormous dendritic surface area, and
the parallel fibres are controlled in turn by the input of many mossy fibres. Excitation
of a Purkinje cell by a parallel cell fibre causes a brief excitatory postsynaptic
potential, the “simple spike”, and a single action potential at best, although the
Purkinje cell will in general rely on summing many inputs temporally and spatially
before firing. As might be expected from such a broad input, the receptive fields of
the Purkinje cells are very large indeed, sometimes extending over a whole limb, and
can be multimodal. Presumably this complexity of arrangement has some implications
for the success or otherwise of the attempts to model the cerebellum!

The difference in firing patterns between the climbing and mossy fibres has been
touched upon. The mossy fibres demonstrate spontaneous activity, triggering a
stream of simple spikes (hundreds per second) in the Purkinje cells. Somatosensory,
vestibular or other sensory stimuli change the frequency of these simple spikes, as do
voluntary eye or limb movements. The frequency of complex spikes induced by the
climbing fibres, on the other hand, is only ever between 1 and 3 per second. It if
difficult to imagine what such infrequent events might signal. Rodolfo Llinas
suggested that the spatial pattern of activity of several climbing fibres, all firing at the
same time, might have coding properties. Neurons in the inferior olive are often
connected to each other by dendrodendritic electrical synapses, and thus could fire in
synchrony in this manner. These signals might give some feedback about the timing
of peripheral events, that is to say the success of movements, and seem to occur when
an animal encounters novel conditions or tries new movements7.

This then is the basic model of cellular organisation in the cortex. The mossy cells can
be seen as giving general information to the Purkinje cells, and this is being balanced
by the specific information, particularly concerning errors, being signalled by the
climbing fibres. The quite beautiful basic structure, though, is complicated by the
presence of interneurons in the cerebellum. These include the stellate cells, the basket
cells, excited by parallel fibres and, and Golgi cells, also excited by parallel fibres, but
inhibiting the granule cells instead.

Both the stellate and basket cells are excited by parallel fibres. The short axons of the
stellate cells contact and inhibit the nearby dendrites of the Purkinje cells, while the
axons of the basket cells inhibit a parasagittal row of Purkinje cells (in the same axis
as the parallel fibres). Thus these two cell types facilitate lateral inhibition, or the
centre surround inhibition seen in the visual system. They are found in the molecular
layer. The Golgi cells, found in the granular layer, offer some form of negative
feedback, shortening the duration of bursts in the parallel fibre.

There are several theories8, of varying usefulness, on how the cellular structure of the
cerebellum might be related to the general function described above. They differ
largely because no-one has a definite idea of what the function of the cerebellum
actually is. In the tonic reinforcer model, the cerebellum is seen as exerting a tonic
excitatory (reinforcing) influence over “motor pattern generators”, MPG’s, (areas of
7
Gilbert and Thach 1977, Simpson and Alley 1974.
8
Zigmond and Bloom catalogue these.
cells that stimulate reflex, stereotyped movement. These are combined for more
complex movement patterns) in the vestibular and reticular nuclei, and the cerebral
cortex. This theory stemmed from the observation that one of the major symptoms of
cerebellar ablation in experimental animals is a generalized atonia. This theory was
originally championed by Luciani. But it seems that the story is more complicated
than this, since the output of the cerebellum changes depending on posture, and it also
affects the α nerve efferents, in addition to the γ efferents one would expect for
modulation of stimuli.

The next idea was the timer model: where the cerebellum is seen as a cerebral
eggtimer. Braitenberg et al. suggested that each line of Purkinje cells functions to
slow down the signal: describing it as a “tapped delay line”. People with lateral
cerebellar damage are impaired in their ability to perceive differences in intervals
between tone pairs of the order of half a second9, which certainly implies some timer
function. The idea was that the cerebellum watched over the motor pattern generator
movements initiated by the motor cortex, and turned them off at the right time. But
unfortunately there is no evidence of any regular clock-like discharge from the
neurons of the cerebellum. Instead, as already mentioned their discharge is graded
depending on posture and limb activity, amongst other things. At one time it was
thought that the inferior olive and the cerebellum between them had this clock
function: application of the drug harmaline causes a 10Hz tremor in experimental
animals that is abolished by lesioning the inferior olive. But the normal discharge of
the neurons of the inferior olive is not merely non-clocklike: it is in fact random!

In the command-feedback comparator model, the cerebellum is seen as refining motor

movements “on the job”, as it were. The neocerebellum receives commands from the
association cortex and feedback from the motor cortex, and it projects back to the
motor cortex. It helps to initiate movements, which is consistent with the observation
that the dentate neurons fire up to 100ms before movement occurs. The
paleocerebellum receives command from the motor cortex and feedback from the
spinal cord, and projects to the red nucleus, and thus modifies the ongoing
movements. This is how computer feedback systems, such as for the guidance of
some missiles,10 work. Here, however, feedback occurs very fast, as it travels through
wires. Neurons are not wires, and it is difficult to reconcile the slowness of
information transfer with this theory.

The combine-coordinator model is the most recent attempt, and suggests that the role
of the cerebellum is the co-ordination of movements. Indeed, another of the specific
symptoms of cerebellar damage is asynergia, where the individual reflexes still
function, but movement has the appearance of being disjointed. Patients with
cerebellar damage have been known to report that they feel as if they have to stop and
think about each movement. The parallel fibres give feedback and allow the
combination of the activities of MPG’s to give co-ordinated movement. The cellular
arrangement of the cerebellum gives further support for this idea. The rostrocaudal
axis of the body is mapped onto the sagittal axis of the cerebellum. So the myotomes
are probably in the coronal dimension of the cerebellum, and thus parallel to the long
axons of the parallel fibres. These can thus influence the output of all the muscles in a
9
Zigmond and Bloom, after the experiments of Ivry et al.
10
Perhaps heat seeking missiles, where, unlike ballistic missiles, the trajectory of the rocket is
continuously modified depending on the source of heat, are a good analogy?
myotome, co-ordinating and synergising their actions, via the Purkinje cells and the
deep nuclei. Parallel fibres that connect the deep nuclei could provide further motor
co-ordination.

Marr and Albus suggested independently in the early 1970’s that the cerebellar cortex
might be involved in motor learning (through trial and error), and this is another
important feature of this model. They proposed that the climbing fibre input to
Purkinje cells modifies their response to the parallel fibre input: specifically that firing
of the climbing fibres within 100ms of the parallel fibre causes a long term depression
in the parallel fibre. That is to say, during a movement the climbing fibres provide
some sort of error signal, signalling the difference between the intended pattern of
movement (as signalled by motor cortical areas) and the actual results. They then
depress the effect of the parallel fibres that are concurrently active, to provide an
immediate feedback. In the sense of longer term learning, the parallel fibres
“responsible” for erroneous movement are suppressed.

Initial experiments to show the role of the cerebellum in motor learning were
performed by Masao Ito et al on the vestibular ocular reflex, the reflex which allows
the eyes to be trained on an object as the head is moved. The subject (both humans
and experimental animals) wore prism glasses that reversed right and left fields. The
vestibule ocular reflex became progressively reduced, and after a few days, reversed.
Lesions to the archicerebellum (in the experimental animals rather than the humans)
abolished this re-organisation of the reflex, and in humans cerebellar damage is
associated with impaired motor learning11.

The cerebellum is thus seen as governing unconsciousness movement. An example

might be learning to play a piano piece. Some go so quickly that it would be
impossible to play them merely by reading the music, and thinking about which finger
to use where12! Initial attempts to play the piece are slow and stuttering. The only way
to learn the piece is systematically going through the piece, hands separately,
rehearsing the sequence of movements again and again13. At some point the
movements become subconscious: they move from being controlled by the motor
cortex to being controlled by the cerebellum. This is in general how most pianists will
memorise music. There is sometimes the (undesirable) sensation of playing quite
robotically: of being almost unaware of having been playing the piece. But one way to
“freeze” during a very fast passage is to start to “think” about the movement of one’s
fingers. The memorisation of a piano piece, in addition to showing this role of the
cerebellum, also highlights the complexity of the stereotyped actions that can be
learnt. The cellular organisation of the cerebellum does seem well equipped to
provide this sort of learning function, although our state of knowledge at the moment
seems limited to the observation that the sort of complexity observed in the
cerebellum seems suitable for the sort of complexity of the motor movements that
might be learnt, as well as the changes in synaptic strength that are required for motor
learning.

11
There is the urban myth of the person born with severe cerebellar damage who has grown up to be a
ballet dancer/ international footballer/ concert pianist etc., but it is probably a myth and no more.
12
Coming back to the speed of firing of the neurons again.
13
Imagining oneself playing the piece is also useful, and again links in to the motor learning function
of the neocerebellum.
Finally, it is worth remembering that although all discussion of the cerebellum’s
function has so far been centred on its motor role, it has more recently been realised
that it has a general role in thought, behaviour and emotion, since it is connected with
the associative and paralimbic cortices. There is a cerebellar cognitive affective
syndrome in humans with diseases confined to the cerebellum, which manifests with
impairments of working memory, visual spatial reasoning, language disturbances and
impairments of executive processing14. It is not clear whether these effects are
mediated by the same properties so useful for the control of motor function, or
whether they are due to something else completely.

The cerebellum currently straddles both the study of motor functions and behavioural
neuroscience. It has a simple yet quite beautiful structure, which presumably helps it
to perform its functions. What that function is has yet to be clearly identified, but it
helps to co-ordinate body parts, eye movements and assists in motor learning. Its
relation to cognitive function represents a particularly exciting area for future
research.

14
Schmahmann: Cerebellum-The true thinking machine