Transcript OverviewCerebellum

Learning and Memory
This section of the course deals with learning and memory. I won’t define these terms
because this is a lot more difficult to do than you might expect.
This section is organized as follows.
1. An overview of the main current concepts of learning and memory.
2. Types of memory.
3. Specialized learning and memory systems. The definition of:
(a) Perceptual learning.
(b) Skill and habit learning.
(c) Declarative memory.
4. I will then concentrate on four systems: sensory cortices and perceptual learning
(brief); the cerebellum and motor learning; the amygdala and emotional learning;
the hippocampus and declarative memory.
In each case we will cover the behaviors measured to study learning in these
systems, the anatomy of the substrates and the physiological bases of the
underlying plasticity.
Overview 1
There have been many heuristic models of learning and memory. One popular model
is based on computer memory. A computer stores instructions and data on a hard disk
or in RAM; the processor then fetches and manipulates these files. In other words, the
memory location is different from the processing location.
The brain works entirely differently. Based on electrophysiological and behavioral
studies, there are many regions of the brain involved in learning and memory.
It is known that synapses are not fixed but can be changed by neuronal activity, a
process known as synaptic plasticity.
A major hypothesis of neuroscience is that synaptic plasticity is an important cellular
basis of memory storage.
Molecular neuroscience has revealed the molecular basis of many kinds of synaptic
plasticity (this material was covered in the NSC 5402: Cellular and Molecular
Neuroscience).
These molecular substrates of synaptic plasticity are present in most (all?) brain areas
although their expression levels vary.
This supports the idea that most (all?) parts of the brain can support synaptic plasticity
and might therefore be part of a distributed memory system.
Overview 2
There is something of a paradox here: memory is distributed over much
of the brain. Yet there are specialized memory subsystems.
We can understand this in the following way:
Different parts of the brain are specialized for different functions; in my
section you learned about visual processing in primary visual cortex. In
Dr. Fortier’s section you learned about motor systems.
Each of these specialized brain regions has its own synaptic plasticity
mechanisms and learning/memory involves synaptic changes in the
processing regions themselves.
We’ll see many examples throughout these lectures.
Types of Memory
The major subdivision of memory is into Declarative and Procedural
memory.
Declarative memory is what humans generally think of as memory. If an event
occurs we can describe what happened with some accuracy and related
it to other events.
Animals cannot “declare” what they have experienced. So a subtype of
declarative memory- episodic memory- is believed to play the same
role but without language. Here are two typical examples of episodic
memory in the rat:
(a) The rat learns to find food at a specific location in a maze.
(b) The monkey learns to select an item different from one presented a brief
time before (delayed non match to sample).
Procedural memory involves improving some perceptual (interpreting microphotographs of synapses) or motor skill (riding a bike, skiing etc); in
humans this is not associated with conscious awareness of what is being
learned.
We’ll look at examples of procedural learning next.
Perceptual Learning 1
Adams et al. 2004. Nature Neuroscience
Based on our everyday experience we
expect a light source (the sun) to be
coming from above. The patches that
are brighter on top are seen as
convex (bumps) while the patches
that are bright at the bottom are seen
as concave (dimples) cause of the
way light will be reflected from them.
This was thought to be “built in” to our
visual systems.
In this experiment the observation of
the images was paired with touch that
indicated that what appeared to be a
bump was really a dimple and vice
versa.
The observers then changed their
estimation of where the light was
coming from.
Visual perception was changed by
changed by somatosensory input.
The site of this learning is not known.
Perceptual Learning 2
In these experiments a subject has to decide whether
the central line is nearer the left or right line. As you
can see the threshold for this discrimination drops
over a period of many days. This perceptual
improvement is specific to this task- it does not
generalize to the vernier discrimination task: are two
lines collinear.
The authors of this study presented data that
suggested that this improvement was occurring in
primary visual cortex.
Other studies from this group also reinforce the idea
that synaptic plasticity in the visual cortices
themselves is the basis for this type of perceptual
learning.
This type of learning corresponds to an everyday
experience. After some time in a new environment,
you learn to better “see” what it is your are looking at,
e.g. learning to do microscopy.
Tsodyks and Gilbert (2004) Nature
Perceptual Learning 3
The perceptual changes seen in psychophysical
experiments can be correlated with
electrophysiological studies of single neurons.
Here we see a recording from a neuron in
auditory cortex. As you remember, the cell here
are tuned to specific frequencies: this cell gave
the strongest response to frequencies near 9.5
kHz. The animal then underwent classical
conditioning with a 9 kHz tone. As you can see
the neuron shifted its frequency tuning: its
response to 9.5 kHz was decreased and its
response to 9 kHz greatly increased.
Weinberger, 1993, Current Opinion in
Neurobiology.
Similar kinds of plasticity experiments have been
performed for orientation tuning in primary visual
cortex.
So the sensory cortex shows synaptic plasticity
that underlies perceptual learning.
This type of learning is unconscious and does
not generalize.
Motor Learning
Motor control finally derives from brainstem or spinal cord motorneurons and the
local interneurons that control them.
However the complex pattern of muscle activation that underlies posture and
movement is controlled by higher brain centers: the motor cortex, the basal ganglia
and the cerebellum. Due to lack of time, we won’t discuss the motor cortex or the
basal ganglia except to say that the basal ganglia are probably involved in the
learning of habitual movements that are “rewarded”.
Instead we’ll concentrate on cerebellum just because it has a fairly simple structure.
The cerebellum is hypothesized to be involved in two types of learning: error
corrected gain control and motor timing control.
First we’ll briefly review cerebellar anatomy and physiology and then discuss
examples of each type of learning: the vestibulo-ocular reflex (VOR) for error
correction and eyeblink conditioning for timing control.
Brodal, The Nervous System.
Cerebellum 1: Structure
The cerebellum is an ancient structure; its basic form is
already evident in sharks and remains basically the same
in all vertebrates. It sits dorsal to the medulla and pons
and derives embryologically from the rhombic lip.
The cerebellum has several subdivisions but they all
follow the same basic pattern of connectivity: sensory
input (directly from periphery on indirectly from cortex)
ends up in the cerebellar cortex. The cerebellar cortex
projects down to the cerebellar nuclei and these in turn
project to motor control regions (directly or via the motor
cortex).
The vestibulo-cerebellum receives vestibular input,
projects to the vestibular nuclei and regulates eye
movements.
This pathway is involved in plasticity of the vestibuloocular reflex (VOR).
The cerebellar hemispheres receive sensory input from
cortex (via the pontine nuclei) and project back to the red
nucleus and motor cortex (via thalamus).
This pathway is involved in learning of eyeblink
conditioning.
Bear et al.
Cerebellum 2: Circuitry
Climbing fibers responsive project to the
same input project to small sagittal strips
of cerebellar cortex.
The Purkinje cells within one strip project
to the same cluster of deep cerebellar
neurons.
These microzones and target DN cells are
therefore thought to be the functional
subunits of the cerebellum.
The principle players in the cerebellar cortex.
Granule cells: tiny neurons that receive most cerebellar input (mossy fibers). There axons
project up into the molecular layer and make a characteristic T branch to form parallel
fibers.
Purkinje cells: the output neurons of the cerebellum; the parallel fibers form numerous
synaptic contacts on the spines of Purkinje cells. These are excitatory and utilize the
glutamate as a transmitter and AMPA + metabatropic receptors.
Purkinje cells receive another excitatory input: climbing fibers that emanate from the
inferior olivary nucleus of the medulla.
Cerebellum 3: Input and Output
More important details: Purkinje cell
dendrites have a planar organization;
parallel fibers run perpendicular to the
Purkinje cell dendrites and contact many
P-cells.
Each P-cell receives only one climbing
fiber input.
Purkinje cells are the only output of the
cerebellar cortex. The inhibit the deep
nuclear neurons (DN). The DN neurons
are excitatory.
Cerebellum 4: Circuit details
A schematic of the cerebellar circuitry. Remember
that the parallel and climbing fiber input to
the P-cell are excitatory. The P-cell
projection to the DN neurons is inhibitory.
There are also inhibitory interneurons in the
cerebellum: stellate, basket and golgi
neurons.
There are multiple forms of synaptic plasticity in
the cerebellum:
1. Parallel fiber synapses onto P-cells have
presynaptic LTP and postsynaptic LTD.
2. Mossy fiber synapses onto granule cells
potentiate.
3. P-cell synapses onto DN neurons potentiate
and depress (complicated).
Cerebellum 5: LTD and LTP
The best studied form of cerebellar plasticity is LTD of parallel fibers. This is induced by pairing
stimulation of parallel and climbing fibers. The climbing fiber EPSP in unchanged by this protocol.
The parallel fiber EPSP is depressed for >30 minutes. So this is an associate kind of plasticitythis led to the idea that the climbing fiber was a “teacher” or carried an error code; the parallel
fiber “learned” from the climbing fiber. The model was that, for any Purkinje cell, a subset of
parallel fibers would be functional.
Stimulation of parallel fibers without climbing fiber co-activation produces LTP.
So PF synapses can go up or down in strength- this is believed to be a key element for cerebellar
based learning.
Cerebellum 6: LTD mechanism
The cellular mechanism for Purkinje cell postsynaptic associative LTD has been studied
extensively. The climbing fiber depolarizes the entire Purkinje cell dendritic tree causing
the opening of dendritic Ca2+ channels. Stimulation of parallel fibers results in glutamate
binding to metabatropic glutamate receptors causing the activation of PKC. The
simultaneous presence of dendritic Ca2+ and PKC causes the endocytosis of the
parallel fiber glutamate (AMPA) receptors producing a long lasting decrease in the
EPSPs at that synapse.
The Vestibulo-ocular Reflex (VOR)
G- gaze direction, H- head position, E- eye position, T- table velocity.
Animals often need to stabilize their gaze direction while moving. That is, they may need
to focus on one object even if their head and/or body is moving. So they use
compensatory eye movements to cancel the movement of their head and remain
focuses.
This is usually studied by rotating the animal horizontally while it focuses on some object
and observing the compensatory eye movements- the VOR.
VOR Circuit
The rotation of the head is signaled by the vestibular system (horizontal canals in this case)
and eye movements are controlled by the extraocular muscles. The vestibular nuclei
provide the link that controls the VOR.
This is a very simplified schematic of the VOR circuitry in a frog.
Cerebellar Control of the VOR
The VOR works quite well with just the brainstem
vestibular circuit. But it was discovered that the
cerebellum (an ancient lobe called the flocculus)
recieves vestibular input and project to the
vestibular nuclei responsible for the VOR.
Prism
For normal operation of the VOR, the flocculus is not very important. A major discovery was that
the VOR could change if the visual input was perturbed. The classic method for perturbing the
visual input was to put a pair of prisms over the eyes. When this is done a head rotation will result
in an increased (or decreased dependent on the optics) movement of images across the retina.
So the the VOR no longer stabilizes the image on the retina. After many hours to days the VOR
compensates and again stabilizes the retinal image. This VOR adaptation requires the flocculus.
The theory (based on extensive data) is that retinal slip information reaches P cells via climbing
fibers and vestibular input via parallel fibers. When the system is OK there is no retinal slip. When
visual input is perturbed a head movement results in a moving retinal image activating the
climbing fibers. As a result cerebellar plasticity leads to a modified VOR that removes retinal slip.
So perhaps the purpose of the cerebellum is to control the timing and/or gain of the VOR.
The Cerebellum as an Adaptive filter
Dean et al, 2010 NRN
In an adaptive filter the input goes into a bank of filters. For each filter its gain (amplification or reduction of the
input) is controlled as is its time constant (how long the output lasts). The outputs of the filters are then added up.
The “adaptive” part comes in when the gain can be regulated.
In the cerebellum it is assumed that the gain (weight=w) of the parallel fiber synapses are regulated by climbing
fiber input. This results in the correct matching of head velocity and stimulus.
This model explains a lot of the VOR data.
Eye blink conditioning
When air is puffed onto the eye
the animal blinks as a
protective mechanism. This is
an unconditioned reflex since it
happens in an untrained
animal. The air puff is the
unconditioned stimulus (US)
and the eyeblink the
unconditioned response.
A. When a tone precedes the air puff the animal
learns to associate the tone with the puff: they learn
that the tone predicts the air puff and so they will,
after many trials, do the eyeblink in response to the
tone. In other words the eyeblink will occur before
the air puff and the eye will be even better
protected. This is “delay conditioning” and the tone
is the conditioning stimulus (CS). The eyeblink is
now a conditioned response.
B. In some cases the CS does not overlap the US;
this is trace conditioning.
The cerebellum is essential for learning this
conditioned response.
The cerebellum and eye blink conditioning
Here is the circuitry for learning to blink in response to a tone. The climbing fibers give the
“teaching” US signal of the airpuff while the parallel fibers carry the CS- the tone. The two
are associated somewhere in cerebellum via synaptic plasticity.
The cerebellum: Summary
From these two examples we might conclude that the cerebellum is important for
controlling the timing of motor responses to sensory input: for the VOR it controls
the phase of the eye movement with respect to the vestibular stimulation; for the
eyeblink it controls the time of the eyeblink with respect to the tone.
The cerebellum can change this temporal relation. Many, but not all,
neuroscientists believe that this form of learning is based on P cell plasticity. It
has been very hard to prove this conclusively.
Firstly there are many forms of plasticity in cerebellum and it is hard to parse out
which one does what.
There are models that incorporate more than one form of plasticity and they do
make some good predictions, but there is still disagreements.
There have been knockouts of essential ingredients of P cell plasticity (PKC
isoforms); these knockouts result in animals that lack P cell LTD and also lack
adaptation of the VOR.
But these are in the mouse and it hardly has a VOR.
So, even in this “simple” system, it is very difficult to relate cellular plasticity to
memory and learnng.