Cerebellum and Basal Ganglia

Goals

To introduce the metaphor of neural machines
To discuss the anatomy, physiology, and function of the cerebellum
To discuss the anatomy, physiology, and function of the basal ganglia
To discuss how these structures may support different forms of learning

Topic slide

Suzanne Haber (b. 1946) is a neuroscientist at the University of Rochester who has done fundamental research on the neuroanatomy of basal ganglia circuitry.

David Marr (1945 – 1980) was a computational neuroscientist at MIT who created an influential model of cerebellar function in 1965.

Reading

Reading: PN6 Chapters 18

Concepts in motor control

Feedback control loop: In attempting to to produce a desired effect upon a controlled object, the difference between the desired outcome and actual outcome is used to change the system's effect upon the controlled object until the desired outcome is achieved.
- A good example of a feedback control loop is a thermostat which turns on a furnace that heats the room until the desired temperature is achieved and then shuts off.
- Feedback systems are problematic when the feedback signal is delayed or noisy. In such cases, the system might oscillate around the set point. If your thermostat is slow to respond to temperature changes, your heat will oscillate around the set point.
Feedforward control system: The system sends a command to the controlled object but does not take the response of the controlled object into account to modify the system.
- Feedforward systems must have a good internal model of the controlled object so it knows what commands to send. This might require learning, and learning might require the incorporation of a feedback controller.
- Feedforward control is related to the concept of a ballistic movement; i.e., one that occurs too quickly to be modified by feedback (e.g., a 'punch' is a ballistic movement. A person throwing a punch cannot modify its trajectory if the target of the punch moves).
- Some movements can be made in the dark without proprioceptive feedback (as in patients with severe sensory loss, i.e., a peripheral neuropathy). That is, the patient can make the movement accurately without feeling or seeing themselves make the movement.
  - However, targeted reaching cannot be made accurately without feedback.
Efference copy: A copy of a command send to a controlled object is also sent to an internal model that contains a representation of the current state of the controlled object and so that its future state can be predicted and modify the commands. For example, a copy of the motor commands sent to the muscles can also be sent to the cerebellum to be compared to an internal model of the state of the muscles as informed by proprioception. The future state of the muscles can be predicted by the internal model, and a change in the motor commands to the muscles can be made before feedback occurs.
Corollary discharge: A concept related to efference copy except that the role of the efference copy is to inhibit the sensory response to an internally generated movement. For example, when we move our eyes, a corollary discharge inhibits the sensory experience of the world moving. However, if we passively move our eyes with a finger (and thus no motor command signal is issued, and thus no corollary discharge) the world does seem to move.

One analogy that I have found useful for understanding feedback and feedforward systems is controlling the temperature of a shower in an unfamiliar hotel room. The first time you use the shower, you are dependent upon feedback. But because it takes a while for the water temperature to change after twisting the hot/cold knob, the feedback is delayed. So, you may be too cold, turn the knob for more hot water, wait a short while while continuing to freeze, turn the knob even further for more hot water, and then suddenly the water is too hot. You rapidly turn the knob the opposite way, but it stays too hot. You turn it more, and then suddenly the water is too cold. You oscillate between too hot and too cold, until you form an internal model of how the system responds, and the inherent lags in the water temperature change.

After a while, through feedback instructed learning, you are much better at anticipating the system's response, and setting the knob appropriately in the first instance. Indeed, if somebody starts running cold water at a nearby sink while you are showering, you can send a feedforward control to turn down the hot water as your internal model predicts an upcoming (but not yet felt) increase in the water temperature of the shower, since the sink is now drawing away cold water from the shower mixture. You make this compensatory command to the temperature control system without waiting for feedback.

I discussed the value of proprioceptive (or kinesthetic) feedback in the case of prosthetic arms. You can read about this research here. A YouTube video of a patient using a prosthetic limb with feedback can be found here.

Neural machines

I reviewed the four components of the movement control system that I first discussed in my last lecture; i.e. lower motor neurons, upper motor neurons, and the two 'consulting' systems, the basal ganglia and the cerebellum.

An important general point about the basal ganglia and cerebellum is that damage to these structures do not lead to paralysis. The patients can still move, although their movements may be clumsy, difficult to initiate, or difficult to suppress. This is consistent with the notion that the outputs of these structures modify motor plans that are generated in upper motor neurons.

I include the basal ganglia and cerebellum in my conception of a neural machine. I showed an image of Babbage’s Difference Engine to use to support my metaphor that the cerebellum and basal ganglia (and hippocampus) could be understood as neural machines.

There is a repetitive architecture in the Difference Engine – the same microstructure is repeated over and over. This suggests (at least to your instructor) a computation (performed in parallel) that is applied across an array of inputs. The repeating structure suggests a common computation. This computation is instantiated in the machinery (the hardware), suggesting that all that can change is the input.

The hippocampus, cerebellum, and basal ganglia are suggestive of neural machines (at least to your instructor). There is a repetitive stereotypical neural architecture that suggests that some computation is performed on an array of inputs. In each case, the output of the computation is directed back to the some of the same areas that provided the input. Thus, there is a closed loop organization in these systems as well. The metaphor is not perfect, but it is useful.

Perhaps given their stereotypical neural architecture, the basal ganglia, cerebellum, and hippocampus has attracted computational neuroscientists, who have attempted to reverse engineer their functions through an analysis of their circuits.

The problem is that the computation or computations that they instantiate are still not known, or at least not known precisely. Most of the knowledge we have of the functions of these exquisite structures is through observation of an organism's behavior following damage. Thus, I will discuss the effects of lesions upon these structures in human patients. As we will see, lesions to the basal ganglia and cerebellum result in obvious impairments in motor learning and in movement. Thus, for centuries, these structures have been considered to be exclusively motor. However, it is clear from their connectivity, that they may play an important role in cognitive functions as well.

Cerebellum

The cerebellum contains about 80% of all neurons in the brain, although it only accounts for about 10% of the brain volume. This is due to the enormous number of small granule cells that populate the cerebellum.

Rate of evolution of cerebellum

All vertebrates have a cerebellum, and some non-vertebrates have ganglia that may serve the functions of the cerebellum. Barton and Venditti(2014) studied relative evolution of the cerebellum. They concluded:

Our analyses indicate relative cerebellar expansion in apes and provide compelling evidence for a significant shift away from the otherwise tight evolutionary coupling between neocortex and cerebellum. The initial impetus may have been the demands of below-branch locomotion and arboreal route planning in large-bodied primates.

Their findings indicated:

The cerebellum expanded rapidly in parallel lineages of apes, including humans.
The cerebellum increased in absolute size and relative to neocortex size.
This expansion began at the origin of apes but accelerated in the great ape clade.
Cerebellar expansion may have been critical for technical intelligence.

Damage to cerebellum in humans

I started the discussion of the cerebellum by showing movie clips of two typical disorders related to cerebellar damage:

Ataxia and cerebellar gait disturbance
Intention tremor (note the oscillation as the patient gets closer to his reaching target).

Gross anatomy of the cerebellum

Recalling from lecture 2, the cerebellum and the pons form the metencephalon. The cerebellum is a three-layered structure with a very consistent neuronal architecture.

The major inputs to the cerebellum are from the mossy fiber system and the climbing fiber system which are discussed below. The major output is from the deep cerebellar nuclei. The major inputs and output fibers pass through the peduncles – which are massive white matter tracts.:

Superior Cerebellar Peduncle (also known as the Brachium conjunctivum)
- Major output from deep cerebellar nuclei
Middle Cerebellar Peduncle (Brachium pontis)
- Largest of the peduncles
- Input to cerebellum from the corticopontocerebellar tract (cerebral cortex > pons > cerebellum).
Inferior Cerebellar Peduncle (Restiform body)
- Proprioceptive input from the body via the dorsal spinocerebellar tract.

The sources of output from the cerebellum are the deep cerebellar nuclei (DCN) which receive an inhibitory input from the Purkinje cells (see below). There are three (per hemisphere) in humans:

Fastigial
Interposed
Dentate

In addition, the vestibular nuclei act as the DCN for the vestibulocerebellum (see below)

Three subsystems of the cerebellum

The cerebellum is composed of three subsystems, which are differentiated by their inputs, their outputs, and their functions.

Vestibulocerebellum
- Evolutionary old (fish)
- Mossy fiber input system: Receives input from brainstem vestibular nuclei and from visual regions.
- Output to vestibular nuclei in brainstem (acts like DCN)
- Plays an important role in balance and coordinating eye movements
Spinocerebellum
- Mossy fiber input system: Receives somatosensory and proprioceptive input from spinal cord (spinocerebellar tract) and sensory and vestibular information from head/body.
- Output through fastigial and interposed DCN to descending systems
  - reticulospinal tract and rubrospinal tract
- Needed for smooth movements – ataxia and intention tremor (can’t smoothly end movement)
- Particularly alcohol sensitive
  - a severely drunk individual displays similar motor deficits as an individual with spinocerebellar damage
Cerebrocerebellum (also know as the neocerebellum)
- Evolutionarily newest part of cerebellum
- Largest in apes and humans (as described earlier)
- Mossy fiber system: Receives efferent copy of output from motor cortex and other cortical regions such as dlPFC, somatosensory cortex, and parietal lobe from pontine nuclei.
- Output through dentate nucleus (the largest by far of the DCN)
- Projects to motor, premotor, and prefrontal cortex
- Planning, executing, and learning movements
- Lesions interfere with complex motions – throwing a ball.
- May have cognitive role (discussed below) beyond movement

Neural circuit

The cerebellum has a very stereotypic and relatively simple neural circuit. The principal cell is the Purkinje cell or Purkinje neuron. The Purkinje cell is unusual in that it has a huge dendritic tree that resembles a fan, in that it extends in a 2-dimensional space.

There are two major input fiber pathways providing input to the Purkinje cells:

Mossy fibers synapse onto granule cells, which then give rise to parallel fibers that run horizontally in the molecular layer and synapse on dendrites of Purkinje cells. There can be 150,000 or more synapses from parallel fibers to Purkinje cells.
- The source of the mossy fibers vary in different parts of the cerebellum.
  - The mossy fiber input to the spinocerebellum is primarily proprioceptive input from the spinocerebellar tract ascending from the spinal cord. This provides the cerebellum with an up-to-date status of the positions of every joint and the tension in every muscle.
  - The mossy fiber input to the cerebrocerebellum (or, neocerebellum) primarily comes from the pontine nuclei which receive collaterals from the descending corticospinal tract. Some of this input reflects efference copy of motor commands being sent to the spinal cord.
Climbing fibers originate in the inferior olive in the pons. Unlike the mossy fiber -> parallel fiber system, a single climbing fiber makes extensive contact with a Purkinje neuron as it wraps around the soma and dendritic tree. The climbing fiber input is very powerful due to the numerous synapses from a single climbing fiber.
- When a climbing fiber stimulates a Purkinje neuron, it fires a complex spike which includes opening calcium mediated voltage-gated channels in the dendritic tree. This causes long term depression (LTD) of the recently active parallel fiber synapses.

There are different sources of information that feed these two fiber systems – this allows for the possibility that two sources of information are compared, and an error, or learning signal, is generated.

Functions of cerebellum

The Inferior Olive receive input from many sources, including the DCN, and may itself act a comparator of intended and actual movements (a feedback controller). The Inferior Olive may thus generate an 'error signal' that weakens (through LTD) the synapses that resulted in the incorrect movement. The climbing fiber may thus provide a 'teaching signal' to the Purkinje cells (actually, more of an 'unlearning signal').

One 'textbook' summary (Adapted from Kandel, 5th edition) of the cerebellum consistent with my lecture reads as follows:

the ‘cerebellum compares internal feedback signals that report the intended movement with external feedback signals that report the actual motion’
the ‘cerebellum generates corrective signals that that … make movements accurate’
‘These corrective signals are feedforward or anticipatory actions that operate on the descending motor systems’
‘When these mechanisms fail because of lesions to the cerebellum, movement develops … oscillations and tremors.’
‘much about … movement must be well planned in advance, and planning necessarily incorporates adjustments to of motor programs based upon learning’.

This 'error correction' model of cerebellar function has been the dominant model in the past decade, and is supported by evidence derived mostly from research on the vestibulo-ocular reflex (VOR) (by Masao Ito) and long-term depression in Purkinje cells. However, there are two other competing hypotheses:

The 'timing hypothesis' suggests that cerebellum plays a dominant role in the precise timing of muscle groups needed for complex movements.
- Poor timing of muscle groups can lead to oscillations – such as intention tremor.
- Poor coordination of muscle groups in different joints can lead to movement problems – such as ataxia.
The 'learning hypothesis' is based upon the observation that the cerebellum is critical for delay conditioning in animal models of learning.

The 'error correction', 'timing', and 'learning' hypotheses are not mutually exclusive, and theorists have recently attempted to reconcile these models into a single model. It may be that the uniform architecture of the cerebellum and its neural circuitry supports different functions depending upon the nature of its inputs and the target of its output. At the conclusion of the lecture, I return to the notion that the cerebellum, the basal ganglia, and cerebral cortex instantiate different forms of learning.

Cognitive function and the cerebellum

One area of significant new interest is the potential role of the cerebellum in higher cognitive functions. Neural connectivity studies in monkeys and humans have demonstrated that cerebral cortical areas outside of typical motor areas show closed loop connections with the cerebrocerebellum. Functional MRI studies in humans have shown activation in the cerebellum in working memory and other cognitive tasks.

However, a 2011 review of cerebellar damage in cognitive tasks has emphasized the point that cognitive deficits are not obvious after cerebellar damage in humans. One area of cognitive ability that has been most consistently associated with cerebellar damage is sequencing (which resonates with the cerebellar timing hypothesis).

However, I also showed a MRI of an individual with a congenital absence of a cerebellum who, though impaired motorically and had slurred speech, was able to otherwise function relatively normally. Other such patients have been identified in the literature.

Basal ganglia

The Basal Ganglia have a complex three-dimensional (3D) shape and are located in the deep interior of the cerebral hemispheres. A nice overview of this anatomy using 3D animations can be found in this brief, narrated video.

The term 'ganglia' is a misnomer with regard to this structure. Ganglia are collections of neurons that are located outside of the central nervous system (e.g., the dorsal root ganglion are outside of the spinal cord, and are collections of sensory neurons whose axons enter the spinal cord through the dorsal roots).

Like the cerebellum, the basal ganglia has a uniform neural circuit with a closed loop connectivity pattern. It was initially thought to be an exclusively motor structure, but (like the cerebellum) a study of its neural connectivity pattern has shown that it includes closed loops with many different regions of cortex. Thus, its function may depend upon its particular inputs and outputs.

We will discuss the role of the basal ganglia in reward processing in an upcoming lecture.

The most important fact to know about the BG is that its main output is to the VL thalamus, and that this output is tonically inhibitory. The effect of this inhibition is to make it harder for movements to be initiated in the motor cortices. The BG circuits influence this tonic inhibition through transient disinhibition (inhibition of inhibition).

Major circuit of the basal ganglia (BG):

Input to the BG is topographically arranged though the Striatum
- Striatum = caudate + putamen
Caudate and Putamen are histologically and embryological similar, but appear as distinct structures because they are divided by internal capsule (recall, a white matter region through which courses the corticospinal and corticobulbar tracts).
Output from the BG is primarily through the Pallidum
- The Pallidum is also known as the globus pallidus which has two parts
  - Internal (Gpi)
  - External (Gpe)

There are two circuits through the Pallidum:

Direct route
- Cortex -> Striatum -> Gpi -> Thalamus
Indirect route (two varieties)
- Cortex -> Striatum -> Gpe -> Gpi -> Thalamus
- Cortex -> Striatum -> Gpe -> SubThalamic Nucleus (STN) -> Gpi -> Thalamus

The substantia nigra provides dopamine (DA) to the striatum, but the DA has different effects:

Excites the direct pathway
Inhibits the indirect pathway

The circuitry of the BG thus includes inhibition and excitation, and the interactions can be complex – such that inhibition is inhibited (i.e., disinhibition).

Generally, the thalamus excites cortex:

However, the major output from the BG to the thalamus is inhibitory. Therefore, strong output from the BG inhibits thalamus and thus inhibits cortex.
If the output of the BG is strengthened, then the thalamus is more inhibited and the cortex less excited.
If the inhibitory output of the BG is weakened, the thalamus is released from inhibition (disinhibited) and the cortex more excited.

This combination of excitation and inhibition against the background of tonic inhibition suggests response choice to computational neuroscientists

Diseases of the Basal Ganglia

There are two diseases that involve the basal ganglia – Parkinson’s disease and Huntington’s disease that can be instructive about the function of the BG. I presented example videos of each disease.

Parkinson’s disease:

Damage to direct pathway
Decreases inhibition of the Gpi
Therefore increases inhibition of thalamus by Gpi
Results in difficulties in initiating motion

Huntington’s disease:

Damage to indirect pathway
Increases inhibition of Gpi
Therefore decreases inhibition of thalamus by Gpi
Results in chorea – uncontrolled movements (dance-like, choreography)

I also discussed the role of BG in abulia and read a case report.

Forms of Learning

Kenji Doya, Peter Strick and others have advanced a theory that the Basal Ganglia, the Cerebellum, and cerebral cortex instantiate different forms of learning:

Basal Ganglia – reinforcement learning
Cerebellum – supervised learning
Cortex – unsupervised learning

Quoting from Doya (Current Opinion in Neurobiology, 2000):

Specialization of the cerebellum, the basal ganglia, and the cerebral cortex for different types of learning… The cerebellum is specialized for supervised learning, which is guided by the error signal encoded in the climbing fiber input from the inferior olive. The basal ganglia are specialized for reinforcement learning, which is guided by the reward signal encoded in the dopaminergic input from the substantia nigra. The cerebral cortex is specialized for unsupervised learning, which is guided by the statistical properties of the input signal itself, but may also be regulated by the ascending neuromodulatory inputs …

Videos

Prerecorded videos for Fall 2020

This first video below provides a brief introduction to the topic, followed by a lecture on the cerebellum.

This second video presents a lecture on the basal ganglia.

The Human Brain

Professor Gregory McCarthy

Lecture 12